Browse the corpus

Introduction Lung cancer is the most common cause of cancer death across the world.1 The clinical staging of non-small-cell lung cancer is an important process that identifies treatment options and guides disease prognosis. In patients with non-small-cell lung cancer who are fit for surgery and have no evidence of extrathoracic spread, the disease status of the mediastinal lymph nodes can be used to establish a patient's suitability for treatment with curative intent.2, 3 Several invasive and non-invasive techniques are available to support the diagnosis and staging of lung cancer. Patients with suspected lung cancer undergo a CT scan of the lower neck, thorax, and upper abdomen. About 50% of patients present with metastatic disease that is evident outside the thorax4 and, in these patients, a biopsy sample taken from the safest most accessible location is recommended. However, in patients with solely intrathoracic disease evident on the initial CT scan, the diagnostic and staging algorithm is more complex. A sample of the primary lesion is generally taken by bronchoscopy or CT-guided biopsy before attention turns to mediastinal nodal staging. PET-CT is reliable if mediastinal lymph nodes that are less than 1 cm in the short axis are negative. However, invasive sampling of mediastinal lymphadenopathy is recommended when lymph nodes are avid for 18F-fluorodeoxyglucose (18F-FDG), the tumour is central, there is a PET-positive hilar lymph node, or any mediastinal node is larger than 1 cm in the short axis (irrespective of 18F-FDG uptake).5

rt axis are negative. However, invasive sampling of mediastinal lymphadenopathy is recommended when lymph nodes are avid for 18F-fluorodeoxyglucose (18F-FDG), the tumour is central, there is a PET-positive hilar lymph node, or any mediastinal node is larger than 1 cm in the short axis (irrespective of 18F-FDG uptake).5 The diagnosis and staging of patients with intrathoracic disease can therefore need several investigative procedures, including bronchoscopy, radiology-guided biopsy sampling, PET-CT, and mediastinoscopy. This process often takes several weeks and is a time of great anxiety for patients. Additionally, 26% of patients with lung cancer report that their health deteriorates while waiting for an hospital appointment.6 Further time will elapse before a treatment decision has been made which could mean that they are unfit for oncological treatments by the time a treatment decision has been reached. The present approach to mediastinal staging of non-small-cell lung cancer (CT, PET-CT, and mediastinoscopy) can result in inaccurate nodal staging in 25% of operable patients,7 perhaps because the sensitivity for the detection of mediastinal metastases by CT scan is 51%, by PET-CT is 74%, and by mediastinoscopy is 78%.5, 8

present approach to mediastinal staging of non-small-cell lung cancer (CT, PET-CT, and mediastinoscopy) can result in inaccurate nodal staging in 25% of operable patients,7 perhaps because the sensitivity for the detection of mediastinal metastases by CT scan is 51%, by PET-CT is 74%, and by mediastinoscopy is 78%.5, 8 Endobronchial ultrasound-guided transbronchial needle aspiration (EBUS-TBNA) is a newer technique that allows minimally invasive sampling of all intrathoracic lymph nodes adjacent to the bronchial tree. A pooled analysis of 1299 patients9 with known or suspected non-small-cell lung cancer undergoing EBUS-TBNA showed that the procedure had a sensitivity of 90% for the detection of mediastinal nodal metastases. At the time of the inception of our trial in 2007, guidelines4 recommended EBUS-TBNA as an alternative to mediastinoscopy for patients who needed invasive mediastinal sampling after a PET-CT scan. Invasive mediastinal sampling is also recommended for staging patients with central tumours or patients with enlarged or 18F-FDG-avid hilar lymphadenopathy.

ur trial in 2007, guidelines4 recommended EBUS-TBNA as an alternative to mediastinoscopy for patients who needed invasive mediastinal sampling after a PET-CT scan. Invasive mediastinal sampling is also recommended for staging patients with central tumours or patients with enlarged or 18F-FDG-avid hilar lymphadenopathy. Therefore we aimed to investigate whether EBUS-TBNA could be used as an initial investigation for the diagnosis and staging of patients with suspected lung cancer because the procedure provides a tissue diagnosis and nodal staging in one investigation. Previous studies have shown that EBUS-TBNA might represent good value for money,10, 11 but there is a shortage of information about its efficacy or cost-effectiveness for patients with suspected lung cancer. We therefore did the Lung-BOOST (BronchOscopic or Oesophageal ultrasound for lung cancer diagnosis and STaging) trial—a pragmatic, multicentre, randomised controlled trial to test the hypothesis that EBUS-TBNA as an initial investigation after a staging CT scan would reduce the time to treatment decision, and reduce the number of investigations needed for the diagnosis and staging of patients with suspected lung cancer at no additional cost compared with conventional diagnosis and staging (CDS) techniques.

sis that EBUS-TBNA as an initial investigation after a staging CT scan would reduce the time to treatment decision, and reduce the number of investigations needed for the diagnosis and staging of patients with suspected lung cancer at no additional cost compared with conventional diagnosis and staging (CDS) techniques. Methods Study design and participants We did this randomised controlled trial in six centres in the UK (University College London Hospital, Whittington Hospital, North Middlesex University Hospital, Princess Alexandra Hospital, Barnet General Hospital, and Nottingham University Hospital). Patients at these centres who were suspected to have stage I to IIIA lung cancer on the basis of CT scans of the neck, thorax, and upper abdomen were eligible for trial entry. For inclusion into the trial, patients had to be aged at least 18 years and fit enough to undergo thoracotomy and lung resection. Exclusion criteria were significant concurrent malignant disease or any condition or concurrent medicine that contraindicated EBUS-TBNA or mediastinoscopy. Patients with known extrathoracic malignant disease, supraclavicular lymphadenopathy, or pleural effusion were also excluded. The 7th edition of the tumour, node, metastasis (TNM) staging system in lung cancer was used throughout.12 This investigator-initiated pragmatic trial was approved by the UK national research ethics service (reference 07/H0711/127) and the ethics committees of the six participating centres. Patients provided written informed consent.

Methods Study design and participants We did this randomised controlled trial in six centres in the UK (University College London Hospital, Whittington Hospital, North Middlesex University Hospital, Princess Alexandra Hospital, Barnet General Hospital, and Nottingham University Hospital). Patients at these centres who were suspected to have stage I to IIIA lung cancer on the basis of CT scans of the neck, thorax, and upper abdomen were eligible for trial entry. For inclusion into the trial, patients had to be aged at least 18 years and fit enough to undergo thoracotomy and lung resection. Exclusion criteria were significant concurrent malignant disease or any condition or concurrent medicine that contraindicated EBUS-TBNA or mediastinoscopy. Patients with known extrathoracic malignant disease, supraclavicular lymphadenopathy, or pleural effusion were also excluded. The 7th edition of the tumour, node, metastasis (TNM) staging system in lung cancer was used throughout.12 This investigator-initiated pragmatic trial was approved by the UK national research ethics service (reference 07/H0711/127) and the ethics committees of the six participating centres. Patients provided written informed consent. Randomisation and masking We randomly assigned participants (1:1) to either conventional diagnosis and staging (CDS group) or EBUS-TBNA as an initial investigation after a staging CT scan followed by further diagnosis and staging techniques if needed (EBUS group). We used a telephone randomisation method with permuted computer-generated blocks of four. Randomisation was stratified according to the presence of mediastinal lymph nodes that measured 1 cm or more in the short axis and by recruiting centre. An investigator undertook the informed consent process, followed by the telephone randomisation process done by research assistants. The random allocation sequence was kept in the randomisation centre and concealed from participants and investigators until the interventions were assigned. Because of the nature of the intervention, masking of participants and consenting investigators was not possible. However, pathologists and radiologists were unaware that patients were enrolled into a clinical trial. Data were obtained on paper-based case forms and entered by an independent clerk onto a secured trial database on a dedicated trial computer.

ion, masking of participants and consenting investigators was not possible. However, pathologists and radiologists were unaware that patients were enrolled into a clinical trial. Data were obtained on paper-based case forms and entered by an independent clerk onto a secured trial database on a dedicated trial computer. Procedures Participants allocated to CDS underwent investigations as determined by the local multidisciplinary team. We suggested an algorithm for CDS in the trial protocol based on the most recently available UK National Institute of Health and Clinical Excellence guidance (2005)13 at the time the trial started (appendix). The trial management group agreed that allowing the responsible multidisciplinary teams to determine the patients' investigations would provide the best comparator group. This allowed the control CDS group to emulate clinical practice, giving the trial strong external validity. Patients randomly assigned to the EBUS group underwent EBUS-TBNA as an initial procedure after a staging CT scan (appendix). The procedure was done in the outpatient setting with patients given moderate sedation with midazolam and fentanyl. We did EBUS-TBNA using a dedicated bronchoscope with a linear ultrasound probe integrated into the distal end (BF-UC160F-OL8, Olympus, Tokyo, Japan, or EB 1970UK, Pentax, Slough, UK). A systematic examination of all mediastinal and hilar lymph node stations was made. Nodes that we suspected were metastatic because of their size or location on the CT scan were aspirated with a 22-gauge or 21-gauge needle and labelled according to the Mountain–Dressler lymph node map. If no enlarged nodes were identified, samples were taken via EBUS-TBNA from the lymph node station that was most likely to drain the primary lesion.14 If a target node was inaccessible with EBUS-TBNA then endoscopic ultrasound-guided fine needle aspiration (EUS-FNA) as an alternative procedure was allowed. After EBUS-TBNA, any further investigations needed were established by the multidisciplinary team.

node station that was most likely to drain the primary lesion.14 If a target node was inaccessible with EBUS-TBNA then endoscopic ultrasound-guided fine needle aspiration (EUS-FNA) as an alternative procedure was allowed. After EBUS-TBNA, any further investigations needed were established by the multidisciplinary team. Patients were given treatment according to the recommendations of the multidisciplinary team. PET-CT was recommended for all patients before a decision to treat with curative intent. In the absence of symptoms, we did not undertake surveillance for brain or bone metastases unless radical treatment was planned.

node station that was most likely to drain the primary lesion.14 If a target node was inaccessible with EBUS-TBNA then endoscopic ultrasound-guided fine needle aspiration (EUS-FNA) as an alternative procedure was allowed. After EBUS-TBNA, any further investigations needed were established by the multidisciplinary team. Patients were given treatment according to the recommendations of the multidisciplinary team. PET-CT was recommended for all patients before a decision to treat with curative intent. In the absence of symptoms, we did not undertake surveillance for brain or bone metastases unless radical treatment was planned. Outcomes The primary endpoint was the time from first outpatient appointment with the respiratory specialist to treatment decision by the multidisciplinary team, after completion of the diagnosis and staging procedures. We prespecified that the primary outcome measure would be analysed in the subgroup of patients with non-small-cell lung cancer. We prespecified four secondary endpoints: (1) UK National Health Service (NHS) costs of diagnosing and staging lung cancer, (2) the number of investigations and outpatient attendances per patient, (3) the proportion of patients diagnosed and staged with one procedure, and (4) the number of avoidable thoracotomies. An avoidable thoracotomy was defined as an open and close procedure, unexpected mediastinal nodal metastases (pN2/pN3), pT4 or pM1a/b disease, resection of benign disease or disease recurrence, or death within 1 year of thoracotomy. We also documented the sensitivity, negative predictive value, and diagnostic accuracy of EBUS-TBNA a priori, and overall survival (post-hoc analysis) and complications from the different diagnostic and staging techniques.

isease, resection of benign disease or disease recurrence, or death within 1 year of thoracotomy. We also documented the sensitivity, negative predictive value, and diagnostic accuracy of EBUS-TBNA a priori, and overall survival (post-hoc analysis) and complications from the different diagnostic and staging techniques. The incremental cost of EBUS-TBNA as an initial investigation compared with CDS was calculated from the perspective of the NHS. We included the costs associated with all diagnostic and staging investigations. We also calculated the costs of treating patients diagnosed with lung cancer. Resource use data were obtained prospectively in the trial. Unit costs were obtained from NHS reference costs,15 NICE 2011 lung cancer guideline,2 and a published study;10 these were multiplied by the resource use and summed across all resource items. Statistical analysis A priori, we expected that 80% of patients would be diagnosed and staged with only one investigation in the EBUS group, compared with 33% in the CDS group. Before the trial began, clinical practice was assessed in a retrospective analysis of diagnostic and staging procedures in five of the participating centres (data not shown). On the basis of this analysis, we estimated that patients in the CDS group of the trial would need a median time to treatment decision of 30 days, and patients in the EBUS group would need a median of 14 days. A sample size of at least 130 patients was planned to give 99% power, assuming a type 1 error of 5%.

t shown). On the basis of this analysis, we estimated that patients in the CDS group of the trial would need a median time to treatment decision of 30 days, and patients in the EBUS group would need a median of 14 days. A sample size of at least 130 patients was planned to give 99% power, assuming a type 1 error of 5%. Analyses were done for the intention-to-diagnose population. The Kaplan-Meier method was used to analyse the primary endpoint (time-to-treatment decision). Hazard ratios (HRs) were calculated from a Cox model, and did not include adjustment for any baseline factors. We used standard definitions of sensitivity for the detection of nodal metastases. The final diagnosis of nodal staging was established in both groups by clinical follow-up of at least 1 year and pathological changes noted with EBUS-TBNA, conventional TBNA, EUS-FNA, mediastinoscopy, or dissection of mediastinal lymph nodes. The Fisher exact test was used to analyse categorical data, and unpaired t tests were used to compare groups of continuous normally distributed variables. All tests were two-sided and 5% was taken as the cutoff for statistical significance. The normal approximation method was used to calculate confidence intervals for the proportions. Final statistical analyses were done with STATA (version 10). This trial is registered on ClinicalTrials.gov, number NCT00652769.

The Fisher exact test was used to analyse categorical data, and unpaired t tests were used to compare groups of continuous normally distributed variables. All tests were two-sided and 5% was taken as the cutoff for statistical significance. The normal approximation method was used to calculate confidence intervals for the proportions. Final statistical analyses were done with STATA (version 10). This trial is registered on ClinicalTrials.gov, number NCT00652769. Role of the funding source The funders of the study had no role in study design, data collection, data analysis, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results Between June 10, 2008, and July 4, 2011, we randomly assigned 133 patients with suspected lung cancer to the CDS group (n=67) and EBUS group (n=66) (figure 1). One patient (randomly assigned to CDS) declined all further investigations and withdrew consent before any investigations were done. Both groups were well balanced for all major clinical characteristics (table 1). Lung cancer was diagnosed in 57 (86%) patients in the CDS group and 50 (76%) in the EBUS group (p=0·196), and clinical staging did not differ significantly between the groups in patients with non-small-cell lung cancer (table 2). The benign final diagnoses were pneumonia, organising pneumonia, lung abscess, and folded lung.

cancer was diagnosed in 57 (86%) patients in the CDS group and 50 (76%) in the EBUS group (p=0·196), and clinical staging did not differ significantly between the groups in patients with non-small-cell lung cancer (table 2). The benign final diagnoses were pneumonia, organising pneumonia, lung abscess, and folded lung. The median time-to-treatment decision was longer after CDS (29 days [95% CI 23–35]), than after EBUS (14 days [14–15]; HR 1·98, 95% CI 1·39–2·82, p<0·0001) in the intention-to-diagnose population. Therefore patients in the EBUS group of the trial were likely to receive a treatment decision twice as fast as patients in the CDS group (figure 2A). A greater proportion of patients had diagnosis and staging completed by 14 days in the EBUS group than in the CDS group (35 [53%] vs 8 [12%], p<0·0001). In the subset of patients with non-small-cell lung cancer (figure 2B), initial EBUS-TBNA resulted in a shorter time-to-treatment decision of 15 days (95% CI 14–16), compared with 30 days (95% CI 23–34) in the CDS group (HR 2·09, 95% CI 1·38–3·15, p=0·0002). The mean number of investigations per patient, PET scans (figure 1), and avoidable thoracotomies at 1 year (table 3). were all significantly lower in the EBUS group than in the CDS group, and the number of patients diagnosed and staged with one investigation was greater (table 3). There were fewer PET scans in the EBUS group (33 of 66 [50%]) than in the CDS group (48 of 66 [73%]; p=0·002); however, the number of patients having treatment with curative intent was similar in each group (appendix).

group, and the number of patients diagnosed and staged with one investigation was greater (table 3). There were fewer PET scans in the EBUS group (33 of 66 [50%]) than in the CDS group (48 of 66 [73%]; p=0·002); however, the number of patients having treatment with curative intent was similar in each group (appendix). In the CDS group, 44 (67%) of 66 patients initially underwent a bronchoscopy and 29 (44%) had a radiology-guided biopsy sample taken; in the EBUS group, 64 (97%) of 66 underwent EBUS and two (3%) had EUS-FNA as an initial procedure. Five (8%) of 66 patients had a subsequent radiology-guided biopsy sample taken (appendix). The number of mediastinoscopies did not differ between groups. In a post-hoc analysis, the median survival of patients with non-small-cell lung cancer in the EBUS group of 503 days (95% CI 312–715) was longer than the median survival in the CDS group of 312 days (95% CI 231–488; HR 0·60, 0·37–0·98, p=0·0382; figure 3). An exploratory analysis of lung cancer patients who underwent surgery suggested that postoperative survival was better in the EBUS group than in the CDS group (appendix).

503 days (95% CI 312–715) was longer than the median survival in the CDS group of 312 days (95% CI 231–488; HR 0·60, 0·37–0·98, p=0·0382; figure 3). An exploratory analysis of lung cancer patients who underwent surgery suggested that postoperative survival was better in the EBUS group than in the CDS group (appendix). 64 patients in the trial underwent EBUS-TBNA. The median size of lymph nodes sampled was 12 mm (IQR 7–20). The sensitivity of EBUS-TBNA in this trial was 92% (95% CI 78–98). The negative predictive value of EBUS-TBNA was 90% (72–97) and diagnostic accuracy was 95% (86–99). Two patients randomised to the EBUS group of the trial underwent EUS-FNA instead of EBUS-TBNA, both of whom had station 5 lymph nodes. The procedure yielded a diagnosis of malignant disease in both patients (one adenocarcinoma and one large cell lung cancer). In the CDS group of the study, five patients underwent conventional TBNA. Two of these patients had a benign final diagnosis, and in one patient conventional TBNA provided a diagnosis of squamous cell lung cancer. In the remaining two patients undergoing conventional TBNA, a negative procedure was followed by a mediastinoscopy that showed mediastinal metastases. One patient in each group had a pneumothorax from a CT-guided biopsy sample; the patient in the CDS group needed intercostal drainage and was admitted to hospital.

64 patients in the trial underwent EBUS-TBNA. The median size of lymph nodes sampled was 12 mm (IQR 7–20). The sensitivity of EBUS-TBNA in this trial was 92% (95% CI 78–98). The negative predictive value of EBUS-TBNA was 90% (72–97) and diagnostic accuracy was 95% (86–99). Two patients randomised to the EBUS group of the trial underwent EUS-FNA instead of EBUS-TBNA, both of whom had station 5 lymph nodes. The procedure yielded a diagnosis of malignant disease in both patients (one adenocarcinoma and one large cell lung cancer). In the CDS group of the study, five patients underwent conventional TBNA. Two of these patients had a benign final diagnosis, and in one patient conventional TBNA provided a diagnosis of squamous cell lung cancer. In the remaining two patients undergoing conventional TBNA, a negative procedure was followed by a mediastinoscopy that showed mediastinal metastases. One patient in each group had a pneumothorax from a CT-guided biopsy sample; the patient in the CDS group needed intercostal drainage and was admitted to hospital. The mean cost per patient for diagnostic and staging investigations was £2407 (SD £180·50) in the EBUS group and £2348 (192·20) in the CDS group (difference £59, 95% CI −£463 to £581; appendix). Mean initial treatment costs per patient in those diagnosed with lung cancer were £4452 (£180·00) and £4261 (£257·90), respectively (difference £191, 95% CI −447 to 829; appendix).

vestigations was £2407 (SD £180·50) in the EBUS group and £2348 (192·20) in the CDS group (difference £59, 95% CI −£463 to £581; appendix). Mean initial treatment costs per patient in those diagnosed with lung cancer were £4452 (£180·00) and £4261 (£257·90), respectively (difference £191, 95% CI −447 to 829; appendix). Discussion The results from our trial suggest that routine use of EBUS-TBNA as an initial investigation after a staging CT for suspected lung cancer scan results in a faster treatment decision, with fewer investigations at no significant difference in cost, and, in post-hoc analysis, seems to improve survival, compared with conventional diagnosis and staging methods (panel).

e use of EBUS-TBNA as an initial investigation after a staging CT for suspected lung cancer scan results in a faster treatment decision, with fewer investigations at no significant difference in cost, and, in post-hoc analysis, seems to improve survival, compared with conventional diagnosis and staging methods (panel). The primary endpoint of our Lung-BOOST trial was the time to treatment decision after the test, and the trial showed that routine and upfront use of EBUS-TBNA in the diagnostic pathway can reduce the median time-to-treatment decision from 29 days to 14 days. UK government initiatives in the NHS Cancer Plan have mandated since 2005 that patients receive treatment within 62 days of referral, with a maximum of 31 days between the decision to treat and the patient receiving treatment.16 The time that patients spend undergoing diagnostic and staging investigations is a time of great anxiety for patients, particularly because the median survival for all patients with lung cancer is poor (6·2 months). Importantly, 26% of patients self-report that their health deteriorates while awaiting a treatment decision.6 Therefore, the primary outcome measure in our trial of time-to-treatment decision is of great importance to patients and the multidisciplinary teams charged with their care. The results from the trial show that EBUS-TBNA can provide sufficient diagnostic and staging information in 45% of patients to define the treatment plan. PET-CT is recommended for all patients unless previous investigations have already shown that curative treatment is not an option. Many patients diagnosed with N2 disease by EBUS-TBNA will still need further investigations, including PET-CT scans if combination chemoradiotherapy or surgery are being considered. However, in this trial PET-CT was only needed for 19% of patients after a positive EBUS-TBNA. 25 patients with N2 disease in the EBUS group did not need a PET scan; of these, four could be candidates for chemoradiotherapy and might have a PET scan. Even if the wait for this scan took an extra week, this wait would not significantly affect the median time-to-treatment decision. Routine use of EBUS-TBNA was able to reduce time-to-treatment decision mainly by reducing the number of outpatient appointments and investigations (particularly PET-CT scans).

e a PET scan. Even if the wait for this scan took an extra week, this wait would not significantly affect the median time-to-treatment decision. Routine use of EBUS-TBNA was able to reduce time-to-treatment decision mainly by reducing the number of outpatient appointments and investigations (particularly PET-CT scans). The Lung-BOOST trial recruited patients over 3 years, and the rate of accrual was similar to a randomised trial of PET-CT in lung cancer staging.17 Since trial inception, EBUS-TBNA has become an important investigation for patients with lung cancer and is now preferred to mediastinoscopy as an initial investigation for nodal staging. However, much of the data that lend support to its usefulness are based on case series, many of which are retrospective and therefore limited by selection bias. The randomised design of our trial reduces bias because the EBUS-TBNA operators could not choose patients for the procedure. Despite not being able to select patients, the sensitivity of EBUS-TBNA in the study was high. The introduction of EBUS-TBNA in practice has shifted the methods of tissue acquisition in patients with lung cancer from flexible bronchoscopy and radiology-guided biopsy sampling to EBUS-TBNA. However, to our knowledge, this is the first randomised trial to show the effect of EBUS-TBNA alone on clinical outcomes for patients with lung cancer.

BNA in practice has shifted the methods of tissue acquisition in patients with lung cancer from flexible bronchoscopy and radiology-guided biopsy sampling to EBUS-TBNA. However, to our knowledge, this is the first randomised trial to show the effect of EBUS-TBNA alone on clinical outcomes for patients with lung cancer. The number of mediastinoscopies in the trial was low and did not differ between groups. However, the accuracy of preoperative mediastinal node staging was high, justifying the approach of the multidisciplinary teams. Only one patient in the trial had unexpected pathological N2 disease at thoracotomy and had previously undergone both EBUS and mediastinoscopy. Previous randomised trials of lung cancer staging have used mediastinoscopy routinely as part of clinical staging of lung cancer.17, 18 For example, ASTER18 showed that the combination of EUS with EBUS (followed by mediastinoscopy if EBUS or EUS was negative) was more effective than mediastinoscopy alone for diagnosis of mediastinal metastases. Our trial substantiates that mediastinoscopy is rarely needed for the preoperative staging of non-small-cell lung cancer in clinical practice. The results from this trial also suggest that EBUS-TBNA could be used as a primary diagnostic method in patients with suspected lung cancer rather than only as an alternative to mediastinoscopy.

tiates that mediastinoscopy is rarely needed for the preoperative staging of non-small-cell lung cancer in clinical practice. The results from this trial also suggest that EBUS-TBNA could be used as a primary diagnostic method in patients with suspected lung cancer rather than only as an alternative to mediastinoscopy. In this trial, EBUS-TBNA was done by systematic assessment of all visible lymph node stations. Biopsy samples were taken from enlarged lymph nodes and lymph nodes that anatomically drained the lung cancer primary lesion. This approach achieved a sensitivity of 92% for nodal staging. Application of PET-CT, with 74% sensitivity for nodal staging,5 is therefore not needed before EBUS-TBNA. PET-CT, however, remains essential for the accurate systemic staging of non-small-cell lung cancer before radical treatment. The management of advanced non-small-cell lung cancer now relies on phenotypic sub-classification and genotypic analysis of tumours. 41 patients (38%) were diagnosed with inoperable lung cancer by EBUS. Specimens obtained from EBUS-TBNA are accepted as suitable for the personalised approach to the management of non-small-cell lung cancer. In a multicentre study,19 specimens from EBUS-TBNA were suitable for EGFR mutation analysis in 90% of cases.

rs. 41 patients (38%) were diagnosed with inoperable lung cancer by EBUS. Specimens obtained from EBUS-TBNA are accepted as suitable for the personalised approach to the management of non-small-cell lung cancer. In a multicentre study,19 specimens from EBUS-TBNA were suitable for EGFR mutation analysis in 90% of cases. The results of the cost analysis suggested that use of EBUS-TBNA as an initial investigation after a CT scan was not more expensive than CDS. Because patients in the EBUS group of the trial had an earlier treatment decision (the primary outcome), we can conclude that EBUS-TBNA was more effective for the same cost, and was therefore cost-effective. There were more avoidable thoracotomies at 1 year in the CDS group (13 [76%]) than in the EBUS group (5 [29%]). The broader definition of avoidable thoracotomy in this trial accounts for the high proportion compared with unnecessary thoracotomies in previous studies.18

The results of the cost analysis suggested that use of EBUS-TBNA as an initial investigation after a CT scan was not more expensive than CDS. Because patients in the EBUS group of the trial had an earlier treatment decision (the primary outcome), we can conclude that EBUS-TBNA was more effective for the same cost, and was therefore cost-effective. There were more avoidable thoracotomies at 1 year in the CDS group (13 [76%]) than in the EBUS group (5 [29%]). The broader definition of avoidable thoracotomy in this trial accounts for the high proportion compared with unnecessary thoracotomies in previous studies.18 In a post-hoc analysis, survival was higher in patients who underwent EBUS-TBNA as part of their diagnostic and staging strategy. Further post-hoc analysis suggests that this difference in survival might be due to superior selection (ie, not attributable to health at baseline) of candidates for surgery in patients undergoing preoperative EBUS-TBNA (appendix). We postulate that routine preoperative use of EBUS-TBNA and sampling of mediastinal lymph nodes that anatomically drain the primary tumour might result in a refined population undergoing surgery, with improved survival in that patient group. Because the use of EBUS-TBNA halved the time-to-treatment decision, earlier treatment, when the patient is fitter, could also improve outcome. The result of a survival advantage for patients undergoing EBUS needs to be replicated.

in a refined population undergoing surgery, with improved survival in that patient group. Because the use of EBUS-TBNA halved the time-to-treatment decision, earlier treatment, when the patient is fitter, could also improve outcome. The result of a survival advantage for patients undergoing EBUS needs to be replicated. We recognise that the trial had several limitations. For example, the pragmatic nature of the trial meant that a consistent diagnostic and staging algorithm was not used across all the trial centres. However, the design of the study (which was undertaken at two teaching hospitals and four general hospitals) gives the results strong external validity. Advances in radiotherapy techniques during the trial period could mean that more patients with mediastinal metastases might now suitable for radical treatment and therefore would need PET-CT. In this pragmatic trial of patients with suspected lung cancer, 19% of the participants had a final diagnosis of disorders other than lung cancer, including metastatic melanoma and lung abscess. Despite this, the primary endpoint was statistically significant for both all patients and also those with non-small-cell lung cancer only. Finally, in this trial, EBUS-TBNA was undertaken by clinicians who were skilled in the procedure—the sensitivity of EBUS-TBNA might not be immediately reproducible in other centres.

espite this, the primary endpoint was statistically significant for both all patients and also those with non-small-cell lung cancer only. Finally, in this trial, EBUS-TBNA was undertaken by clinicians who were skilled in the procedure—the sensitivity of EBUS-TBNA might not be immediately reproducible in other centres. In conclusion, when EBUS-TBNA is used as an initial investigation method after a CT scan in patients with suspected lung cancer confined to the thorax, it can provide a diagnosis and accurate nodal stage in one investigation. This results in a reduction in the time-to-treatment decision and might improve survival in patients with lung cancer when compared with a conventional diagnostic and staging strategy, at no additional cost. Supplementary Material Supplementary appendix

In conclusion, when EBUS-TBNA is used as an initial investigation method after a CT scan in patients with suspected lung cancer confined to the thorax, it can provide a diagnosis and accurate nodal stage in one investigation. This results in a reduction in the time-to-treatment decision and might improve survival in patients with lung cancer when compared with a conventional diagnostic and staging strategy, at no additional cost. Supplementary Material Supplementary appendix Acknowledgments This trial was funded by a grant from the UK Medical Research Council to NN and SMJ (G0800465/1) and supported by researchers at the National Institute for Health Research (NIHR), University College London Hospitals Biomedical Research Centre. SMJ is a Wellcome Trust Senior Research Fellow in Clinical Science. We thank the patients and their families for participating in this trial. We are also grateful to the NIHR North London Cancer Research Network (Dr James Lyddiard, Aryana Chopra, Gerlinda Amor, Azmina Verjee, and colleagues) for trial support. We thank Cathy Read RN, for her advice on trial setup, Professor Allan Hackshaw and Dr Robert Rintoul for their advice on the manuscript and the following principal investigators: Dr Sajid Khan (Barnet General Hospital), Dr Supriya Sundaram (Princess Alexandra Hospital), Dr David Simcock (Barts and the London) and the multidisciplinary lung cancer teams in the recruiting centres, none of whom received compensation for their contributions.

the manuscript and the following principal investigators: Dr Sajid Khan (Barnet General Hospital), Dr Supriya Sundaram (Princess Alexandra Hospital), Dr David Simcock (Barts and the London) and the multidisciplinary lung cancer teams in the recruiting centres, none of whom received compensation for their contributions. Contributors NN coordinated the trial, and produced the manuscript. MN did the statistical analysis, and produced the figures and tables. SM did the health economics analysis and interpretation. DRL, SL, HM, and DRB contributed to data collection and data interpretation. RJS, MKP, SGS, and SMJ contributed to trial design and data interpretation. All authors reviewed the manuscript. Declaration of interests We declare no competing interests. Figure 1 Trial profile EBUS-TBNA=endobronchial ultrasound-guided transbronchial needle aspiration. EUS-FNA=endoscopic ultrasound-guided fine needle aspiration. NSCLC=non-small-cell lung cancer. Figure 2 Time to treatment decision in all patients (A) and in those with non-small-cell lung cancer (B) Kaplan-Meier plots for (A) all patients and (B) patients with non-small-cell lung cancer only undergoing CDS or EBUS-TBNA. CDS=conventional diagnosis and staging. NSCLC=non-small-cell lung cancer. EBUS-TBNA=endobronchial ultrasound-guided transbronchial needle aspiration. HR=hazard ratio. Figure 3 Overall survival of patients with non-small-cell lung cancer

Kaplan-Meier plots for (A) all patients and (B) patients with non-small-cell lung cancer only undergoing CDS or EBUS-TBNA. CDS=conventional diagnosis and staging. NSCLC=non-small-cell lung cancer. EBUS-TBNA=endobronchial ultrasound-guided transbronchial needle aspiration. HR=hazard ratio. Figure 3 Overall survival of patients with non-small-cell lung cancer Survival of patients with non-small-cell lung cancer undergoing CDS or EBUS-TBNA. NSCLC=non-small-cell lung cancer. CDS=conventional diagnosis and staging. EBUS-TBNA=endobronchial ultrasound-guided transbronchial needle aspiration. HR=hazard ratio. Table 1 Baseline characteristics Conventional diagnosis and staging (n=66) Endobronchial ultrasound-guided transbronchial needle aspiration (n=66) Age (years) 68 (IQR 61–73) 71 (IQR 62–78) Men 46 (70%) 43 (65%) Women 20 (30%) 23 (35%) Ethnic origin White 59 (89%) 51 (77%) Asian 2 (3%) 6 (9%) African 2 (3%) 4 (6%) Caribbean 2 (3%) 3 (5%) Other 1 (2%) 2 (3%) ECOG performance status 0 or 1 57 (96%) 60 (92%) Pack-years smoking history 42 (23·4) 42 (28·1) FEV1 (L) 1·9 (0·72) 1·9 (0·65) Clinical nodal staging on initial CT scan N0 20 (30%) 21 (32%) N1 9 (14%) 6 (9%) N2 33 (50%) 34 (51%) N3 4 (6%) 5 (8%) Data are median (range, IQR), n (%), or mean (SD), unless otherwise stated. FEV1=forced expiratory volume in 1 s. ECOG=Eastern Cooperative Oncology Group. Table 2 Final diagnoses and stages of non-small-cell lung cancer

Conventional diagnosis and staging (n=66) Endobronchial ultrasound-guided transbronchial needle aspiration (n=66) Age (years) 68 (IQR 61–73) 71 (IQR 62–78) Men 46 (70%) 43 (65%) Women 20 (30%) 23 (35%) Ethnic origin White 59 (89%) 51 (77%) Asian 2 (3%) 6 (9%) African 2 (3%) 4 (6%) Caribbean 2 (3%) 3 (5%) Other 1 (2%) 2 (3%) ECOG performance status 0 or 1 57 (96%) 60 (92%) Pack-years smoking history 42 (23·4) 42 (28·1) FEV1 (L) 1·9 (0·72) 1·9 (0·65) Clinical nodal staging on initial CT scan N0 20 (30%) 21 (32%) N1 9 (14%) 6 (9%) N2 33 (50%) 34 (51%) N3 4 (6%) 5 (8%) Data are median (range, IQR), n (%), or mean (SD), unless otherwise stated. FEV1=forced expiratory volume in 1 s. ECOG=Eastern Cooperative Oncology Group. Table 2 Final diagnoses and stages of non-small-cell lung cancer Conventional diagnosis and staging (n=66) Endobronchial ultrasound-guided transbronchial needle aspiration (n=66) Benign lesion 6 (9%) 14 (21%) Extrathoracic malignancy 3 (5%) 2 (3%) Small cell lung cancer 7 (11%) 4 (6%) Non-small-cell lung cancer 50 (76%) 46 (70%) Adenocarcinoma 21 (42%) 26 (57%) Squamous cell 21 (42%) 17 (37%) Large cell 3 (6%) 1 (2%) Adenosquamous 2 (4%) 1 (2%) Not otherwise specified 3 (6%) 1 (2%) Stage IA/B 11 (22%) 10 (22%) Stage IIA/B 10 (20%) 6 (13%) Stage IIIA 20 (40%) 22 (48%) Stage IIIB 6 (12%) 7 (15%) Stage IV 3 (6%) 1 (2%) Data are n (%). Staging is based on the 7th edition of TNM (tumour, node, metastasis) staging system for lung cancer. Table 3 Secondary outcomes

Conventional diagnosis and staging (n=66) Endobronchial ultrasound-guided transbronchial needle aspiration (n=66) Benign lesion 6 (9%) 14 (21%) Extrathoracic malignancy 3 (5%) 2 (3%) Small cell lung cancer 7 (11%) 4 (6%) Non-small-cell lung cancer 50 (76%) 46 (70%) Adenocarcinoma 21 (42%) 26 (57%) Squamous cell 21 (42%) 17 (37%) Large cell 3 (6%) 1 (2%) Adenosquamous 2 (4%) 1 (2%) Not otherwise specified 3 (6%) 1 (2%) Stage IA/B 11 (22%) 10 (22%) Stage IIA/B 10 (20%) 6 (13%) Stage IIIA 20 (40%) 22 (48%) Stage IIIB 6 (12%) 7 (15%) Stage IV 3 (6%) 1 (2%) Data are n (%). Staging is based on the 7th edition of TNM (tumour, node, metastasis) staging system for lung cancer. Table 3 Secondary outcomes Conventional diagnosis and staging (n=66) Endobronchial ultrasound-guided transbronchial needle aspiration (n=66) p value Investigations per patient 2·39 (0·78) 1·70 (0·72) <0·0001 Patients diagnosed and staged with one investigation 8 (12%) 30 (45%) <0·0001 Avoidable thoracotomies at 1 year 13 (76%) 5 (29%) 0·035 Data are mean (SD) or n (%). Panel Research in context Systematic review

Conventional diagnosis and staging (n=66) Endobronchial ultrasound-guided transbronchial needle aspiration (n=66) p value Investigations per patient 2·39 (0·78) 1·70 (0·72) <0·0001 Patients diagnosed and staged with one investigation 8 (12%) 30 (45%) <0·0001 Avoidable thoracotomies at 1 year 13 (76%) 5 (29%) 0·035 Data are mean (SD) or n (%). Panel Research in context Systematic review Before the trial started, we did a review of the scientific literature that was subsequently published.14 We compiled the information for this review by searching the PubMed and Medline databases for English-language articles published between Jan 1, 2000, and Jan 1, 2008. Electronic early-release publications were also included. The following search terms were used: “endobronchial ultrasound”, “endoscopic ultrasound”, “lung cancer staging”, “mediastinoscopy”, “positron emission tomography”, “PET-CT”, and “mediastinal staging”. Full articles were obtained and references were checked for additional material, as appropriate. References were chosen on the basis of the highest quality clinical evidence. No randomised trials of patients with suspected lung cancer undergoing EBUS-TBNA only were identified. Interpretation

Before the trial started, we did a review of the scientific literature that was subsequently published.14 We compiled the information for this review by searching the PubMed and Medline databases for English-language articles published between Jan 1, 2000, and Jan 1, 2008. Electronic early-release publications were also included. The following search terms were used: “endobronchial ultrasound”, “endoscopic ultrasound”, “lung cancer staging”, “mediastinoscopy”, “positron emission tomography”, “PET-CT”, and “mediastinal staging”. Full articles were obtained and references were checked for additional material, as appropriate. References were chosen on the basis of the highest quality clinical evidence. No randomised trials of patients with suspected lung cancer undergoing EBUS-TBNA only were identified. Interpretation EBUS-TBNA is recommended for the mediastinal staging of non-small-cell lung cancer and as an alternative to mediastinoscopy. The results from our trial show that routine use of EBUS-TBNA halved the amount of time between testing and treatment decision (compared with conventional diagnosis and staging) and should be considered as the initial investigation after a CT scan for patients with suspected lung cancer and intrathoracic disease at presentation. EBUS-TBNA has a high diagnostic sensitivity for detecting nodal metastases in patients with lung cancer and could reduce the time to treatment decision at no additional cost.

Research in context Evidence before this study One previous study assessed the efficacy of several doses of the saprophyte Mycobacterium vaccae against tuberculosis disease in adults infected with HIV-1, and showed a decreased risk of protocol-defined pulmonary tuberculosis. A previous study with the MVA85A, the candidate vaccine under assessment here, has showed that boosting with MVA85A did not enhance protective efficacy in BCG-vaccinated infants. Adults infected with HIV-1 are an important target population for a new tuberculosis vaccine, and in earlier studies, vaccine-induced immunogenicity in adults infected with HIV-1 was higher than in infants. Added value of this study This is the first time that a candidate tuberculosis vaccine has been assessed for efficacy against Mycobacterium tuberculosis infection in people infected with HIV-1. The results show that vaccinating adults infected with HIV-1 with MVA85A is safe, but does not confer protection against infection with M tuberculosis. Implications of all the available evidence

This is the first time that a candidate tuberculosis vaccine has been assessed for efficacy against Mycobacterium tuberculosis infection in people infected with HIV-1. The results show that vaccinating adults infected with HIV-1 with MVA85A is safe, but does not confer protection against infection with M tuberculosis. Implications of all the available evidence The safety of MVA85A in this large study population of adults with HIV infection is an important finding for tuberculosis vaccine development. The vector is safe to give to people without HIV testing; these safety data provide some generic reassurance that new candidate tuberculosis vaccines are safe in this higher risk population. Additionally, this study has shown that high-quality, multicentre tuberculosis vaccine trials in vulnerable populations are possible. The absence of efficacy despite immunogenicity in this and previous clinical trials of MVA85A suggests that the current parameters for selection of tuberculosis vaccine candidates are inadequate. Standardised preclinical animal models that better represent human infection and disease, and a greater understanding of immune mechanisms of protection in human tuberculosis are both urgently needed. Alternative approaches to vaccine development, including the delivery of candidate vaccines direct to the respiratory mucosa, merit assessment. Other lessons learnt from this trial include the characterisation of the epidemiology of M tuberculosis infection and disease associated with HIV-1 infection in a setting of antiretroviral therapy and isoniazid chemoprophylaxis.

ding the delivery of candidate vaccines direct to the respiratory mucosa, merit assessment. Other lessons learnt from this trial include the characterisation of the epidemiology of M tuberculosis infection and disease associated with HIV-1 infection in a setting of antiretroviral therapy and isoniazid chemoprophylaxis. Introduction Tuberculosis is a substantial global cause of mortality and morbidity, with 9 million new cases of active tuberculosis and 1·5 million deaths occurring in 2013.1 One third of the world's population is infected with Mycobacterium tuberculosis.1 HIV-1 co-infection is one of the most important risk factors for both infection with M tuberculosis and active tuberculosis disease,2 with an estimated 1·1 million of all new tuberculosis cases in 2013 occurring in people co-infected with HIV-1.1 The WHO African region accounts for 80% of HIV-1-associated tuberculosis.1 Additionally, the growing incidence of drug-resistant tuberculosis is associated with poor treatment outcome and increased mortality.3 The global Stop TB Partnership aims to eliminate tuberculosis as a public health problem by 2050. An agreed major component to advance this aim would be an effective vaccine.4 BCG is the only licensed tuberculosis vaccine—it provides protection against severe childhood tuberculosis,5, 6 but the protection conferred against pulmonary tuberculosis in adults and adolescents is highly variable.7, 8

lth problem by 2050. An agreed major component to advance this aim would be an effective vaccine.4 BCG is the only licensed tuberculosis vaccine—it provides protection against severe childhood tuberculosis,5, 6 but the protection conferred against pulmonary tuberculosis in adults and adolescents is highly variable.7, 8 At least 16 candidate tuberculosis vaccines have advanced to clinical testing.9 The modified vaccinia virus Ankara expressing the major M tuberculosis antigen 85A (MVA85A) is a clinically advanced candidate vaccine.10, 11, 12 MVA85A is well tolerated and immunogenic in adults infected and not infected with HIV-1, and in infants not exposed to HIV-1.10, 11, 12, 13, 14 MVA85A adds to BCG-induced protection against mycobacterial challenge in some preclinical animal models.15, 16, 17, 18, 19 However, boosting BCG with MVA85A in South African infants not infected with HIV-1 did not confer additional protection against tuberculosis disease or M tuberculosis infection.10 Administration of several doses of the saprophyte Mycobacterium vaccae to adults infected with HIV-1 was associated with a decreased risk of protocol-defined pulmonary tuberculosis,20 suggesting that vaccination might be effective in people infected with HIV-1. Here we report the results of a multisite, randomised, placebo-controlled, phase 2 trial to assess the safety, immunogenicity, and efficacy of MVA85A in healthy adults infected with HIV-1.

ed risk of protocol-defined pulmonary tuberculosis,20 suggesting that vaccination might be effective in people infected with HIV-1. Here we report the results of a multisite, randomised, placebo-controlled, phase 2 trial to assess the safety, immunogenicity, and efficacy of MVA85A in healthy adults infected with HIV-1. Methods Study design and participants We did a proof-of-concept, randomised, double-blind, placebo-controlled, phase 2 trial of MVA85A at two clinical sites, in Cape Town, South Africa and Dakar, Senegal. In Cape Town, participants were recruited in the community and from primary care clinics in Khayelitsha by use of radio and newspaper advertisements, flyers, pamphlets, and information campaigns at the clinics. Khayelitsha is a densely populated, low-income, peri-urban township. In 2010, antenatal HIV-1 prevalence was 33% and the tuberculosis case notification rate was at least 1500 per 100 000 population per year.21 In Dakar, participants were recruited from public service HIV clinics at the Centre de Traitement Ambulatoire and the Centre de Recherche Clinique et de Formation, Centre Hospitalier Universitaire de Fann. Senegal had an estimated HIV-1 prevalence in adults of less than 1% in 2012, and a reported tuberculosis incidence rate of 0·14% in 2013.1 The annual rate of M tuberculosis infection has not previously been estimated at either site. Eligibility criteria included participants aged 18–50 years with no evidence of active tuberculosis, and baseline CD4 counts greater than 350 cells per μL if they were not receiving antiretroviral therapy, or greater than 300 cells per μL (and with undetectable viral load before randomisation) if they were receiving antiretroviral therapy. Participants with latent tuberculosis infection were eligible for enrolment if they had completed at least 5 months of isoniazid preventive therapy, unless they had completed treatment for tuberculosis disease within 3 years before randomisation. The complete inclusion criteria are listed in the study protocol (appendix).

ticipants with latent tuberculosis infection were eligible for enrolment if they had completed at least 5 months of isoniazid preventive therapy, unless they had completed treatment for tuberculosis disease within 3 years before randomisation. The complete inclusion criteria are listed in the study protocol (appendix). The trial adhered to International Conference on Harmonisation Good Clinical Practice guidelines, and was approved by the University of Cape Town's Faculty of Health Sciences Human Research Ethics Committee and the Medicines Control Council of South Africa; the Senegalese National Ethics Committee for Research in Health; and the Oxford University Tropical Research Ethics Committee. All participants provided written informed consent before any study procedure.

wn's Faculty of Health Sciences Human Research Ethics Committee and the Medicines Control Council of South Africa; the Senegalese National Ethics Committee for Research in Health; and the Oxford University Tropical Research Ethics Committee. All participants provided written informed consent before any study procedure. Randomisation and masking Participants were randomly assigned (1:1) in blocks of four by a randomly generated sequence of participant identification numbers via an interactive voice response system to receive two intradermal injections of either 1 × 108 pfu MVA85A or placebo (Candida skin test antigen [Candin], Allermed Laboratories, San Diego, CA, USA). Randomisation was stratified by antiretroviral therapy status and study site. A statistician uninvolved with study analyses prepared the interactive voice response system randomisation schedule. Doses of vaccines were prepared and labelled in masked syringes by a pharmacist unmasked to group allocation. Participants, nurses (who were involved in assessment and follow-up), investigators, and laboratory staff were masked to group allocation. The second (booster) injection of MVA85A or placebo was given 6–12 months after the first vaccination and participants were actively followed up every 3 months until the last participant enrolled had completed 6 months of follow-up after the booster vaccination.

rs, and laboratory staff were masked to group allocation. The second (booster) injection of MVA85A or placebo was given 6–12 months after the first vaccination and participants were actively followed up every 3 months until the last participant enrolled had completed 6 months of follow-up after the booster vaccination. Procedures We collected data for the incidence of solicited and unsolicited adverse events, including both local injection-site reactions and systemic reactions. Participants reported solicited adverse events on diary cards for 7 days after each vaccination and in response to direct questioning by trained study staff on days 7 and 28 after each injection. Phlebotomy for routine haematological and biochemical analysis was done at screening, before booster vaccination, and on days 7 and 28 after each vaccination. Peripheral CD4 cell count and HIV-1 viral load were also measured at these timepoints and every 3 months until 6 months after booster vaccination. Serious adverse events were monitored by active surveillance throughout and until the end of the trial. The site investigators and local medical monitors determined the severity and seriousness of adverse events and the relation of these to the vaccine. An independent data monitoring committee assessed masked group safety data after 200 participants had been enrolled and unmasked after 600 participants had been enrolled.

The site investigators and local medical monitors determined the severity and seriousness of adverse events and the relation of these to the vaccine. An independent data monitoring committee assessed masked group safety data after 200 participants had been enrolled and unmasked after 600 participants had been enrolled. In a prespecified subset of 70 participants (35 from each group), immunology samples were obtained before each vaccination and on days 7 and 28 after each vaccination. All immunology tests were done masked to group allocation. We assessed vaccine immunogenicity with three assays. First, ex vivo interferon γ enzyme-linked immunospot (ELISPOT) analysis was done on fresh peripheral blood mononuclear cells.22 Cells were stimulated overnight with a single pool of 66 peptides of the antigen 85A (Ag85A), ESAT-6, and CFP-10. Second, Ag85A-specific intracellular cytokine staining assay was done on whole blood.23 Stimulated fixed whole blood samples were stained for CD3-positive, CD4-positive, CD8-positive, CD14-positive, and CD19-positive cells, interferon γ, tumour necrosis factor α, interleukin 17, and interleukin 2. Third, Ag85A-specific antibody response was measured on plasma. Ag85A-specific immunoglobulin G (IgG) antibodies were measured by ELISA on eight serial two-fold dilutions of plasma (1:25–1:3200), by use of affinity purified recombinant, histidine-tagged Ag85A24 (microwell plates coated with 50 ng per well of recombinant Ag85A in borate buffer, overnight at 4°C). Alkaline phosphatase-labelled goat anti-human IgG (Sigma, St Louis, MO, USA) was used as secondary antibody at a dilution of 1:1000 and optical density was read at 405 nm after development with phosphatase substrate (Sigma). Results were expressed in arbitrary units per mL (AU/mL), as compared with values of an internal tuberculosis serum standard of 2500 AU/mL.

n IgG (Sigma, St Louis, MO, USA) was used as secondary antibody at a dilution of 1:1000 and optical density was read at 405 nm after development with phosphatase substrate (Sigma). Results were expressed in arbitrary units per mL (AU/mL), as compared with values of an internal tuberculosis serum standard of 2500 AU/mL. Participants were screened to exclude active tuberculosis by symptom screen and chest radiography at both sites before enrolment. In Cape Town, participants also underwent sputum collection for tuberculosis smear microscopy, GeneXpert MTB/RIF (Cepheid, Sunnyvale, CA, USA), and mycobacterial liquid culture (MGIT; Becton Dickinson, Sparks, MD, USA) because of previously documented high frequencies of asymptomatic disease at this site.25 Latent M tuberculosis infection was defined as either a positive QuantiFERON-TB Gold In-Tube (QFT) test or a tuberculin purified protein derivative skin test (tuberculin skin test) reaction greater than 5 mm.

son, Sparks, MD, USA) because of previously documented high frequencies of asymptomatic disease at this site.25 Latent M tuberculosis infection was defined as either a positive QuantiFERON-TB Gold In-Tube (QFT) test or a tuberculin purified protein derivative skin test (tuberculin skin test) reaction greater than 5 mm. Participants were monitored throughout the trial for possible tuberculosis. Tuberculosis investigations were done in participants who had been in contact with a known case of active tuberculosis, in those who presented with at least one of cough for more than 1 week, fever for more than 1 week, drenching night sweats, unintentional weight loss of more than 3 kg, pleuritic chest pains, haemoptysis, or shortness of breath; and in those who converted to a positive QFT or tuberculin skin test (≤5 mm to >5 mm). Investigations included clinical examination, chest radiography, and collection of at least two sputum samples on which tuberculosis smear microscopy, GeneXpert MTB/RIF, and mycobacterial liquid culture were done. Chest radiographs were reviewed by two physicians, with a third reading to achieve consensus in the event of disagreement. QFT and tuberculin skin tests were repeated at the final study visit.

least two sputum samples on which tuberculosis smear microscopy, GeneXpert MTB/RIF, and mycobacterial liquid culture were done. Chest radiographs were reviewed by two physicians, with a third reading to achieve consensus in the event of disagreement. QFT and tuberculin skin tests were repeated at the final study visit. Outcomes Tuberculosis disease endpoint 1 was defined as culture or GeneXpert MTB/RIF positivity; endpoint 2 included endpoint 1 and a composite clinical endpoint (which included a single acid-fast bacilli smear from a sterile body site; two smears from pulmonary and gastric sampling, and compatible clinical symptoms and radiological signs); and endpoint 3 was participant commencement on anti-tubercular chemotherapy (see the study protocol for more information; appendix). The M tuberculosis infection endpoint was defined as conversion from negative QFT at baseline to positive QFT at the final visit. The primary study outcome was the safety of MVA85A in all participants who received at least one dose of study vaccine or placebo (the safety analysis population) as determined by the numbers and percentages of adverse events (including solicited, unsolicited, and serious adverse events).

Outcomes Tuberculosis disease endpoint 1 was defined as culture or GeneXpert MTB/RIF positivity; endpoint 2 included endpoint 1 and a composite clinical endpoint (which included a single acid-fast bacilli smear from a sterile body site; two smears from pulmonary and gastric sampling, and compatible clinical symptoms and radiological signs); and endpoint 3 was participant commencement on anti-tubercular chemotherapy (see the study protocol for more information; appendix). The M tuberculosis infection endpoint was defined as conversion from negative QFT at baseline to positive QFT at the final visit. The primary study outcome was the safety of MVA85A in all participants who received at least one dose of study vaccine or placebo (the safety analysis population) as determined by the numbers and percentages of adverse events (including solicited, unsolicited, and serious adverse events). The secondary outcome was the efficacy of MVA85A for the prevention of active tuberculosis in the per-protocol population (all randomly allocated participants who received at least one dose of study vaccine or placebo and had no major protocol deviations and no tuberculosis case definition endpoints within 28 days after study day 0 [first vaccination]), which was determined by the incidence of active tuberculosis meeting the definition of endpoint 1, calculated as the number of new cases of active tuberculosis with a date of diagnosis from 28 days after the first vaccination until the end of the study follow-up (May 19, 2014). An intention-to-treat analysis was also done for disease efficacy. In the per-protocol population, we also examined the efficacy of MVA85A by antiretroviral therapy status at the time of randomisation and by baseline isoniazid preventive therapy status.

t vaccination until the end of the study follow-up (May 19, 2014). An intention-to-treat analysis was also done for disease efficacy. In the per-protocol population, we also examined the efficacy of MVA85A by antiretroviral therapy status at the time of randomisation and by baseline isoniazid preventive therapy status. Other secondary outcomes were to assess CD4-positive lymphocyte counts and HIV-1 viral load before and after administration of MVA85A compared with placebo; to assess the immunogenicity of MVA85A compared with placebo as measured by the ex-vivo interferon γ ELISPOT assay; to assess the immunogenicity of MVA85A compared with placebo as measured by flow cytometric intracellular cytokine staining of CD4-positive and CD8-positive T cells after stimulation with a peptide pool of mycobacterial antigens; to identify potential immunological correlates of protection from tuberculosis in participants vaccinated with MVA85A and to assess the QFT conversion rate at final study assessment in MVA85A recipients compared with controls without a diagnosis of tuberculosis during the trial.

peptide pool of mycobacterial antigens; to identify potential immunological correlates of protection from tuberculosis in participants vaccinated with MVA85A and to assess the QFT conversion rate at final study assessment in MVA85A recipients compared with controls without a diagnosis of tuberculosis during the trial. Statistical analysis All sample size calculations assumed a loss to follow-up and death rate of 2%. The initial planned sample size for this trial was 1400 adult participants, to be followed up for 2 years after the last participant was enrolled. This sample size provided 80% power to detect a vaccine efficacy of 60% against tuberculosis disease. However, after review of the phase 2 infant efficacy data,10 the trial design was revised with safety as the primary objective and a smaller sample size and shorter follow-up of a minimum of 6 months. The revised sample size for this study was selected as adequate for a review of the safety profile. With 325 participants assigned to receive MVA85A, the revised sample would have a 90% probability of detecting at least one adverse event occurring at a rate of 0·71%. Because of the expected effect of antiretroviral therapy on tuberculosis disease, an estimated tuberculosis disease incidence ranging between 1·5% and 2% per year was used to estimate the power of the revised sample size for efficacy. Calculations were based on a one-sided log-rank test at a significance level of 0·10 and assumed completion of enrolment in 21 months, a follow-up period of about 15 months for the last patient enrolled, and a maximum of 36 months for the first patient enrolled. If the true efficacy was about 70%, 325 patients per treatment group (650 patients total) provided 81% power to show positive efficacy given an incidence rate of 2·0% in the control group per year, or 71% power given an incidence rate of 1·5% in the control group per year. At a true efficacy of about 60%, 325 patients per treatment group provided 67% power to show positive efficacy given an incidence rate of 2·0% per year, or 57% power given an incidence rate of 1·5% per year. Vaccine efficacy to prevent infection was a secondary endpoint: the recorded QFT conversion rate in the study provided 80% power to detect a vaccine efficacy of 50%.

atment group provided 67% power to show positive efficacy given an incidence rate of 2·0% per year, or 57% power given an incidence rate of 1·5% per year. Vaccine efficacy to prevent infection was a secondary endpoint: the recorded QFT conversion rate in the study provided 80% power to detect a vaccine efficacy of 50%. Statistical analyses were done using SAS version 9.2. All analyses were prespecified in the statistical analysis plan before locking of the database. For the safety analysis, we compared the proportion of participants with at least one adverse event in the MVA85A group versus those in the placebo using Fisher's exact test. We also calculated two-sided 95% CIs for proportions of adverse events within treatment groups and the differences between groups.

the database. For the safety analysis, we compared the proportion of participants with at least one adverse event in the MVA85A group versus those in the placebo using Fisher's exact test. We also calculated two-sided 95% CIs for proportions of adverse events within treatment groups and the differences between groups. The main statistical method used in the analysis of tuberculosis disease endpoints 1–3 was vaccine efficacy, estimated as 1 minus the estimated hazard ratio, based on a Cox regression analysis of time (days) to initial tuberculosis diagnosis, based on the per-protocol population. As supportive confirmatory analysis, we used the conditional binomial (Clopper-Pearson) method to estimate vaccine efficacy and its corresponding two-sided 95% CIs and p values. Time to initial diagnosis for each endpoint was compared by use of a two-sided log-rank test, stratified by study site and antiretroviral therapy status at randomisation. Analyses were summarised by antiretroviral therapy and treatment group for participants in the per-protocol population. Vaccine efficacy against M tuberculosis infection and the corresponding 95% CI, and p value were calculated with the conditional binomial method (Clopper-Pearson), identical to the tuberculosis case definition endpoint analysis.

antiretroviral therapy and treatment group for participants in the per-protocol population. Vaccine efficacy against M tuberculosis infection and the corresponding 95% CI, and p value were calculated with the conditional binomial method (Clopper-Pearson), identical to the tuberculosis case definition endpoint analysis. Other secondary endpoints were analysed in various ways. Median CD4 cell counts and associated two-sided 95% CIs were summarised by antiretroviral therapy status at randomisation, study site, treatment group, and timepoint. HIV-1 viral load (copies per mL) was summarised with medians (and associated 95% CIs) by antiretroviral therapy status at randomisation, study site, and treatment group, at each available timepoint. Both the CD4 cell counts and HIV-1 viral load values were log-transformed before any analysis was done. We used Wilcoxon paired analysis to compare within group before and after vaccination responses. Quintiles (Blomfontein, South Africa) did the statistical analysis, and Aeras paid for this service. The trial was registered with ClinicalTrials.gov, number NCT01151189. Role of the funding source Aeras was the trial sponsor and contributed to study design and data analysis. The other funders had no role in study design, data collection, data analysis, data interpretation, or writing of the report. BPN, FT, BSL, RJW, and HM had full access to all the data in the study. HM had final responsibility for the decision to submit for publication.

and contributed to study design and data analysis. The other funders had no role in study design, data collection, data analysis, data interpretation, or writing of the report. BPN, FT, BSL, RJW, and HM had full access to all the data in the study. HM had final responsibility for the decision to submit for publication. Results Between Aug 4, 2011, and April 24, 2013, 1233 adults infected with HIV-1 were screened and 650 were randomly assigned; 649 were included in the safety analysis and 645 in the per-protocol analysis (figure 1). 513 (71%) participants had CD4 counts greater than 300 cells per μL and were receiving antiretroviral therapy; 136 (21%) had CD4 counts above 350 cells per μL and had never received antiretroviral therapy. The results of the intention-to-treat analysis were not different and are not reported. 311 (96%) participants in the placebo group and 298 (92%) in the MVA85A group received the booster vaccination. One participant was randomly assigned to placebo but received MVA85A; this participant was included in the safety population for MVA85A but not in the per-protocol efficacy population. One participant was randomly assigned to MVA85A but withdrew consent before vaccination and was not vaccinated. This participant was excluded from both the safety and per-protocol populations. Baseline demographic characteristics were similar in the two study groups and across the two study sites (table 1; appendix). 625 participants completed the study; 14 participants were lost to follow-up (nine placebo, five MVA85A), five withdrew consent (two placebo, three MVA85A), and six died (four placebo, two MVA85A).

eline demographic characteristics were similar in the two study groups and across the two study sites (table 1; appendix). 625 participants completed the study; 14 participants were lost to follow-up (nine placebo, five MVA85A), five withdrew consent (two placebo, three MVA85A), and six died (four placebo, two MVA85A). In the per-protocol population, median follow-up was 655 days for the 320 recipients of MVA85A and 654 days for the 325 placebo participants. Other than the four participants shown in figure 1, all participants were included in the analysis.

eline demographic characteristics were similar in the two study groups and across the two study sites (table 1; appendix). 625 participants completed the study; 14 participants were lost to follow-up (nine placebo, five MVA85A), five withdrew consent (two placebo, three MVA85A), and six died (four placebo, two MVA85A). In the per-protocol population, median follow-up was 655 days for the 320 recipients of MVA85A and 654 days for the 325 placebo participants. Other than the four participants shown in figure 1, all participants were included in the analysis. At least one adverse event was reported in 312 (96%) of placebo recipients and 321 (99%) of MVA85A recipients (table 2). Solicited adverse events were more common in participants who received MVA85A than placebo (table 2). Most of these events were local injection-site reactions; other solicited adverse events included mild influenza-like symptoms and regional lymphadenopathy. We noted no significant difference between study groups in the frequency of serious adverse events. 34 serious adverse events occurred during the study, 17 in the placebo group and 17 in the MVA85A group (table 2; appendix). All but one of these events were judged to be unrelated to vaccination; a case of probable tuberculous meningitis that occurred 6 days after vaccination was judged to be possibly related to vaccination. The data monitoring committee reviewed this case, did not request unmasking, and recommended continuing with the study. The participant was treated for tuberculous meningitis and made a full recovery. At study completion, this participant was identified as having received MVA85A. 13 serious adverse events in the infections and infestations category occurred during the study (the only category with more than five serious adverse events in either group), eight in the placebo group and five in the MVA85A group; this difference was not significant (Fisher's exact test, p=0·58).

ing received MVA85A. 13 serious adverse events in the infections and infestations category occurred during the study (the only category with more than five serious adverse events in either group), eight in the placebo group and five in the MVA85A group; this difference was not significant (Fisher's exact test, p=0·58). The frequency of severe adverse events did not differ significantly between study groups (table 2). We noted no significant changes in CD4 cell count or HIV-1 viral load throughout the course of the trial in either study group (data not shown). Routine haematological and biochemical test results did not differ between study groups (data not shown). ELISPOT responses to Ag85A were significantly higher in participants from Dakar than in those from Cape Town at baseline (p=0·0016), but at no other timepoint. This difference was not seen with the less sensitive whole blood intracellular cytokine staining assay. MVA85A induced an Ag85A-specific T-cell response that peaked 7 days after the first and booster vaccinations (median spots per million: day 0 [first vaccination], 9·0 [IQR 2·3–51·0]; day 7 [first vaccination], 337·0 [139·3–993·8]; day 0 [booster vaccination], 103·5 [14·8–223·8]; day 7 [booster vaccination], 426·0 [150·0–745·0]; figure 2). Responses at each timepoint after vaccination did not differ by study site or by antiretroviral therapy status. Medians in the placebo group did not exceed 20 spots per million at any timepoint.

[139·3–993·8]; day 0 [booster vaccination], 103·5 [14·8–223·8]; day 7 [booster vaccination], 426·0 [150·0–745·0]; figure 2). Responses at each timepoint after vaccination did not differ by study site or by antiretroviral therapy status. Medians in the placebo group did not exceed 20 spots per million at any timepoint. Whole blood intracellular cytokine staining showed the most commonly measured cytokine from CD4 T cells was interferon γ, in agreement with the ELISPOT data. Tumour necrosis factor α and low concentrations of interleukins 2 and 17 were also detected (table 3, figure 2). Overall, numbers of antigen-specific CD8 T cells were very low and were only positive for interferon γ and tumour necrosis factor α. Multiparameter flow-cytometric analysis showed that mainly monofunctional Ag85A-specific CD4 T cells were present before and after vaccination (figure 3). Ag85A-specific antibody responses were less than twice the baseline value after vaccination in all but three participants.

interferon γ and tumour necrosis factor α. Multiparameter flow-cytometric analysis showed that mainly monofunctional Ag85A-specific CD4 T cells were present before and after vaccination (figure 3). Ag85A-specific antibody responses were less than twice the baseline value after vaccination in all but three participants. In the per-protocol population, the overall number of tuberculosis cases and incidence during study follow-up of tuberculosis cases (endpoint 1) was six (2%) in the MVA85A group and nine (3%) in the placebo group, for a vaccine efficacy of 32·8% (95% CI −111·5 to 80·3; table 4). Figure 4 shows the Kaplan-Meier time-to-disease analysis for endpoint 1. Stratification by antiretroviral therapy status showed no significant difference between treatment groups. Eight of the 15 endpoint 1 cases were QFT positive at enrolment. No additional participants met endpoint 2 who did not already meet endpoint 1. Vaccine efficacy for endpoint 3 was 10·5% (−161·3 to 70.0). Disease incidence did not differ by site. Median time to diagnosis of endpoint 1 was 249 days in the MVA85A group and 236 days in the placebo group. 159 (50%) of 320 MVA85A recipients and 148 (46%) of 325 placebo recipients were investigated for tuberculosis during the study. The study was insufficiently powered to assess the efficacy of MVA85A for the prevention of tuberculosis disease in the subset of participants receiving antiretroviral therapy or isoniazid prophylaxis. The absence of efficacy also made it impossible to identify potential immunological correlates of protection from tuberculosis in participants vaccinated with MVA85A.

he efficacy of MVA85A for the prevention of tuberculosis disease in the subset of participants receiving antiretroviral therapy or isoniazid prophylaxis. The absence of efficacy also made it impossible to identify potential immunological correlates of protection from tuberculosis in participants vaccinated with MVA85A. The number of QFT-negative participants who converted to QFT positive by the end of the study was 38 (20%) in the MVA85A group and 40 (23%) in the placebo group, for a vaccine efficacy of 11·7% (95% CI −41·3 to 44·9). QFT conversion did not differ by antiretroviral therapy status (data not shown), but it did differ by site. In Cape Town, 41 (31%) of 132 participants converted, whereas in Dakar, 37 (16%) of 227 converted (χ2 10·89, p=0·001). Frequency of QFT reversion (participants who were positive at baseline and negative at end of study) was similar in the two treatment groups (17 [14%] of 124 for MVA85A and 27 [19%] of 139 for placebo; p=0·22), and did not differ by antiretroviral therapy status (data not shown). Tuberculin skin test conversion was not a prespecified endpoint and is not reported here, but will be the subject of further analysis. Discussion This phase 2 trial in 650 adult participants infected with HIV-1 showed that MVA85A was well tolerated and immunogenic, with safety and immunogenicity profiles similar to those reported elsewhere for other populations in which this candidate vaccine has been assessed.10, 11, 12, 13, 14 However, we did not identify any significant efficacy against tuberculosis disease or M tuberculosis infection.

VA85A was well tolerated and immunogenic, with safety and immunogenicity profiles similar to those reported elsewhere for other populations in which this candidate vaccine has been assessed.10, 11, 12, 13, 14 However, we did not identify any significant efficacy against tuberculosis disease or M tuberculosis infection. Both first and booster vaccination with MVA85A induced a significant increase in Ag85A-specific T cells. Responses did not differ by antiretroviral therapy status. A probable explanation for this finding is the high baseline median CD4 count (571 cells per mm3; table 1, appendix) in participants who had not received antiretroviral therapy. Unlike the previously reported infant efficacy trial of MVA85A,10 baseline ELISPOT responses were detected in this trial and were significantly higher in participants from Dakar than in those from Cape Town. This result might be due to greater exposure to environmental mycobacteria; and the finding is unlikely to be due to a technical issue because it was only recorded at this timepoint, and there was a robust quality control programme in place for these assays. Furthermore, the median response 7 days after vaccination in this trial exceeded that seen in the infant trial (337 vs 136 spots per million).10 Additionally, the functional phenotype of the dominant T-cell population in this trial was monofunctional by contrast with the infant trial, in which the dominant phenotype was polyfunctional.10 In both trials, the recorded response was insufficient to be associated with protection. It is not clear whether a quantitatively greater or a qualitatively different immune response is needed for protection. Alternative approaches, including the delivery of candidate vaccines direct to the respiratory mucosa, might be more potent routes of immunisation. For example, we have previously reported that delivery of MVA85A by aerosol to HIV-negative, BCG-vaccinated adults in the UK is well tolerated and induces potent mucosal and systemic immunity.26 Further assessment is needed before this route can be examined in countries with a high burden of tuberculosis. This approach, together with other strategies to improve the immunogenicity of MVA85A, are currently under investigation.

adults in the UK is well tolerated and induces potent mucosal and systemic immunity.26 Further assessment is needed before this route can be examined in countries with a high burden of tuberculosis. This approach, together with other strategies to improve the immunogenicity of MVA85A, are currently under investigation. The recorded annual incidence of tuberculosis (endpoint 1) was substantial (1·43% across treatment groups) and did not differ between sites. However, this incidence was lower than previously reported in Cape Town.27 The numbers of participants receiving antiretroviral therapy was greater than originally envisaged, because of the increased availability of this therapy during the study period and the change in national and international guidelines on the provision of antiretroviral therapy. These factors, combined with the redesign of this study upon availability of the infant trial results,10 led to a reduction in statistical power to detect a difference in tuberculosis disease incidence between treatment groups, leading to wide CIs for our estimates of vaccine efficacy.

vision of antiretroviral therapy. These factors, combined with the redesign of this study upon availability of the infant trial results,10 led to a reduction in statistical power to detect a difference in tuberculosis disease incidence between treatment groups, leading to wide CIs for our estimates of vaccine efficacy. In this trial, the incidence of infection determined by QFT conversion was much higher than the incidence of tuberculosis disease, so CIs around the estimates of efficacy against infection are narrower. The overall recorded annual QFT conversion rate of about 12% meant that we had about 80% power to detect a vaccine efficacy of 50% against M tuberculosis infection. In view of the cost and complexity of human efficacy studies, there is now increased focus on infection as an endpoint rather than disease in proof-of-concept studies before progression to prevention-of-disease efficacy trials.9 However, this approach presupposes that the immune mechanisms needed to prevent infection and disease are similar. Our poor understanding of the biology underlying dynamic QFT conversion and reversion further complicates this shift in emphasis. The rate of QFT reversion was almost as high as the rate of conversion: whether this finding represents a true biological effect or technical variability in the assay cannot be determined from these data.

standing of the biology underlying dynamic QFT conversion and reversion further complicates this shift in emphasis. The rate of QFT reversion was almost as high as the rate of conversion: whether this finding represents a true biological effect or technical variability in the assay cannot be determined from these data. In this study, we have shown that high-quality, multicentre tuberculosis vaccine trials are possible in Africa, and have succeeded in the characterisation of the epidemiology of tuberculosis associated with HIV-1 in two African cities. Nevertheless, the disappointing finding with respect to vaccine efficacy for MVA85A suggests the need for standardised preclinical animal models that better represent human disease and an improved understanding of immune mechanisms of protection in human tuberculosis. Such advances would greatly enhance the ability to efficiently translate clinical research capacity into the development and deployment of an effective vaccine. Supplementary Material Supplementary appendix Acknowledgments The study was funded by the European & Developing Countries Clinical Trials Partnership (IP.07.32080.002), Aeras, Bill & Melinda Gates Foundation, the Wellcome Trust (095780, 084323, and 088316), and the Oxford-Emergent Tuberculosis Consortium. Quintiles (Bloemfontein, South Africa) were used for the statistical analysis, and Aeras paid for this service. The appendix includes a complete list of acknowledgments. We dedicate this study to the memory of Robyn Louw.

the Wellcome Trust (095780, 084323, and 088316), and the Oxford-Emergent Tuberculosis Consortium. Quintiles (Bloemfontein, South Africa) were used for the statistical analysis, and Aeras paid for this service. The appendix includes a complete list of acknowledgments. We dedicate this study to the memory of Robyn Louw. Contributors BPN, FT, SD, HE, RG, VJ, IN, TO, AT, MRa, BSL, SM, and RJW were responsible for implementation of the study and supervision at the study sites. MC, TND, KH, MRo, IS, and KAW did the immunological analysis. MO, RJW, SM, and HM raised the funding and wrote the protocol. All authors contributed to data analysis and contributed to the writing of the report. Declaration of interests HM was previously a shareholder in the Oxford-Emergent Tuberculosis Consortium (OETC), a joint venture established for the development of MVA85A (OETC no longer exists). KH has a patent (US 5736524 A) related to the development of a DNA vaccine against Mycobacterium tuberculosis. RJW received grants from the European & Developing Countries Clinical Trials Partnership, the Wellcome Trust, the UK Medical Research Council, and the European Union during the conduct of the study, and personal fees from GlaxoSmithKline unrelated to this work. All other authors declare no competing interests. Figure 1 Trial profile Figure 2 Vaccine immunogenicity (both study sites combined)

Declaration of interests HM was previously a shareholder in the Oxford-Emergent Tuberculosis Consortium (OETC), a joint venture established for the development of MVA85A (OETC no longer exists). KH has a patent (US 5736524 A) related to the development of a DNA vaccine against Mycobacterium tuberculosis. RJW received grants from the European & Developing Countries Clinical Trials Partnership, the Wellcome Trust, the UK Medical Research Council, and the European Union during the conduct of the study, and personal fees from GlaxoSmithKline unrelated to this work. All other authors declare no competing interests. Figure 1 Trial profile Figure 2 Vaccine immunogenicity (both study sites combined) (A) Antigen 85A (Ag85A) interferon γ enzyme-linked immunospot analysis responses. Data are presented as spot-forming cells (SFC) per million peripheral blood mononuclear cells (PBMCs). p values were calculated with Wilcoxon matched-pair signed-rank tests. Box and whisker plots show median, IQR, and minimum and maximum values. (B) Whole blood intracellular cytokine staining assay of total cytokines. Data are presented as frequency of CD4 and CD8 T cells producing cytokines. Box and whisker plots show median, IQR, and minimum and maximum values. IFNγ=interferon γ. TNFα=tumour necrosis factor α. IL=interleukin. V1=vaccination 1. V2=vaccination 2. Figure 3 Polyfunctional CD4 T cells

(A) Antigen 85A (Ag85A) interferon γ enzyme-linked immunospot analysis responses. Data are presented as spot-forming cells (SFC) per million peripheral blood mononuclear cells (PBMCs). p values were calculated with Wilcoxon matched-pair signed-rank tests. Box and whisker plots show median, IQR, and minimum and maximum values. (B) Whole blood intracellular cytokine staining assay of total cytokines. Data are presented as frequency of CD4 and CD8 T cells producing cytokines. Box and whisker plots show median, IQR, and minimum and maximum values. IFNγ=interferon γ. TNFα=tumour necrosis factor α. IL=interleukin. V1=vaccination 1. V2=vaccination 2. Figure 3 Polyfunctional CD4 T cells Plots show frequency of CD4 T cells producing combinations of the studied cytokines. Bars are median values and dots represent individual volunteers. IFNγ=interferon γ. TNFα=tumour necrosis factor α. IL=interleukin. V1=vaccination 1. V2=vaccination 2. Figure 4 Cumulative incidence of diagnosis of tuberculosis endpoint 1 by treatment group Endpoint 1 was defined as a positive finding from culture or GeneXpert MTB/RIF assay. Table 1 Demographic and baseline characteristics (safety analysis population)

Plots show frequency of CD4 T cells producing combinations of the studied cytokines. Bars are median values and dots represent individual volunteers. IFNγ=interferon γ. TNFα=tumour necrosis factor α. IL=interleukin. V1=vaccination 1. V2=vaccination 2. Figure 4 Cumulative incidence of diagnosis of tuberculosis endpoint 1 by treatment group Endpoint 1 was defined as a positive finding from culture or GeneXpert MTB/RIF assay. Table 1 Demographic and baseline characteristics (safety analysis population) Placebo (n=325) MVA85A (n=324) Median age, years (range) 39·0 (22–41) 38·0 (21–49) Women 255 (78%) 265 (82%) Ethnic origin Black 304 (94%) 302 (93%) Mixed 21 (6%) 22 (7%) QFT test result Positive 150 (46%) 135 (42%) Negative 173 (53%) 188 (58%) Indeterminate 2 (1%) 1 (<1%) TST result >5 mm 128 (39%) 124 (38%) ≤5 mm 191 (59%) 190 (59%) Missing data 6 (2%) 10 (3%) Latent tuberculosis infection 178 (55%) 164 (51%) 5–6 months IPT before enrolment 144 (44%) 133 (41%) Receiving antiretroviral therapy 256 (79%) 257 (79%) CD4 count (cells per mm3) Participants not receiving antiretroviral therapy 564 (169·8) 571 (187·5) Participants receiving antiretroviral therapy 599 (199·6) 598 (220·7) HIV viral load (copies per mL) Participants not receiving antiretroviral therapy 41 371 (92 456·9) 62 168 (166 912·1) Participants receiving antiretroviral therapy 29 (27·1) 34 (63·7) Data are n (%) or mean (SD), unless otherwise stated. QFT=QuantiFERON-TB Gold In-Tube. TST=tuberculin skin test. IPT=isoniazid preventive therapy. Table 2 Overview of adverse events (safety analysis population)

Placebo (n=325) MVA85A (n=324) Median age, years (range) 39·0 (22–41) 38·0 (21–49) Women 255 (78%) 265 (82%) Ethnic origin Black 304 (94%) 302 (93%) Mixed 21 (6%) 22 (7%) QFT test result Positive 150 (46%) 135 (42%) Negative 173 (53%) 188 (58%) Indeterminate 2 (1%) 1 (<1%) TST result >5 mm 128 (39%) 124 (38%) ≤5 mm 191 (59%) 190 (59%) Missing data 6 (2%) 10 (3%) Latent tuberculosis infection 178 (55%) 164 (51%) 5–6 months IPT before enrolment 144 (44%) 133 (41%) Receiving antiretroviral therapy 256 (79%) 257 (79%) CD4 count (cells per mm3) Participants not receiving antiretroviral therapy 564 (169·8) 571 (187·5) Participants receiving antiretroviral therapy 599 (199·6) 598 (220·7) HIV viral load (copies per mL) Participants not receiving antiretroviral therapy 41 371 (92 456·9) 62 168 (166 912·1) Participants receiving antiretroviral therapy 29 (27·1) 34 (63·7) Data are n (%) or mean (SD), unless otherwise stated. QFT=QuantiFERON-TB Gold In-Tube. TST=tuberculin skin test. IPT=isoniazid preventive therapy. Table 2 Overview of adverse events (safety analysis population) Overall Participants not receiving antiretroviral therapy Participants receiving antiretroviral therapy Placebo (n=325) MVA85A (n=324) Difference (MVA85A minus placebo) (95% CI) Placebo (n=69) MVA85A (n=67) Difference (MVA85A minus placebo) (95% CI) Placebo (n=256) MVA85A (n=257) Difference (MVA85A minus placebo) (95% CI) Any adverse event 312 (96·0%; 93·3–97·7) 321 (99·1%; 97·3–99·7) 3·1 (0·7 to 5·4) 67 (97·1%; 90·0–99·2) 66 (98·5%; 92·0–99·7) 1·4 (−3·5 to 6·3) 245 (95·7%; 92·5–97·6) 255 (99·2%; 97·2–99·8) 3·5 (0·8 to 6·2) Solicited adverse event 235 (72·3%; 67·2–76·9) 288 (88·9%; 85·0–91·9) 16·6 (10·6 to 22·5) 50 (72·5%; 61·0–81·6) 63 (94·0%; 85·6–97·7) 21·6 (9·6 to 33·5) 185 (72·3%; 66·5–77·4) 225 (87·5%; 83·0–91·0) 15·3 (8·5 to 22·1) Serious adverse event 17 (5·2%; 3·9–8·2) 17 (5·2%; 3·3–8·2) 0·02 (−3·4 to 3·4) 2 (2·9%; 0·8–10·0) 9 (13·4%; 7·2–23·6) 10·5 (1·5 to 19·6) 15 (5·9%; 3·6–9·4) 8 (3·1%; 1·6–6·0) −2·7 (−6·3 to 0·8) Related adverse event 307 (94·5%; 91·4–96·5) 318 (98·1%; 96·0–99·2) 3·7 (0·8 to 6·6) 66 (95·7%; 88·0–98·5) 66 (98·5%; 92·0–99·7) 2·9 (−2·8 to 8·5) 241 (94·1%; 90·6–96·4) 252 (98·1%; 95·5–99·1) 3·9 (0·6 to 7·2) Severe adverse event 84 (25·8%; 21·4–30·9) 100 (30·9%; 26·1-36·1) 5·0 (−1·9 to 11·9) 15 (21·7%; 13·6–32·8) 22 (32·8%; 22·8–44·8) 11·1 (−3·8 to 26) 69 (27·0%; 21·7–32·9) 78 (30·4%; 25·1–36·2) 3·4 (−4·4 to 11·2) Data are n (%; 95% CI), unless otherwise stated. Serious adverse events were coded with Medical Dictionary for Regulatory Activities version 14.0. Patients with multiple events in each category are counted only once in each category.

44·8) 11·1 (−3·8 to 26) 69 (27·0%; 21·7–32·9) 78 (30·4%; 25·1–36·2) 3·4 (−4·4 to 11·2) Data are n (%; 95% CI), unless otherwise stated. Serious adverse events were coded with Medical Dictionary for Regulatory Activities version 14.0. Patients with multiple events in each category are counted only once in each category. Table 3 Total intracellular cytokine response, presented as frequency of CD4 T cells and CD8 T cells producing specific cytokines

44·8) 11·1 (−3·8 to 26) 69 (27·0%; 21·7–32·9) 78 (30·4%; 25·1–36·2) 3·4 (−4·4 to 11·2) Data are n (%; 95% CI), unless otherwise stated. Serious adverse events were coded with Medical Dictionary for Regulatory Activities version 14.0. Patients with multiple events in each category are counted only once in each category. Table 3 Total intracellular cytokine response, presented as frequency of CD4 T cells and CD8 T cells producing specific cytokines MVA85A (n=28) MVA85A timepoint comparisons (p values) Placebo (n=29) Day 0 (vaccination 1) Day 7 (vaccination 1) Day 0 (vaccination 2) Day 7 (vaccination 2) Day 0 (vaccination 1) vs day 7 (vaccination 1) Day 0 (vaccination 1) vs day 0 (vaccination 2) Day 0 (vaccination 1) vs day 7 (vaccination 2) Day 0 (vaccination 2) vs day 7 (vaccination 2) Day 0 (vaccination 1) Day 7 (vaccination 1) Day 0 (vaccination 2) Day 7 (vaccination 2) CD4 IFNγ 0·01 (0–0·07) 0·1 (0–1·12) 0·03 (0–0·28) 0·11 (0·02–0·82) <0·0001 0·0015 <0·0001 <0·0001 0·02 (0–0·12) 0·01 (0–0·08) 0 (0–0·08) 0·01 (0–0·18) CD4 TNFα 0·02 (0–0·12) 0·11 (0–0·53) 0·05 (0–0·57) 0·11 (0–0·46) <0·0001 0·0403 <0·0001 <0·0001 0·02 (0–0·15) 0·02 (0-0·14) 0·02 (0–0·11) 0·02 (0–0·23) CD4 IL-2 0·021 (0–0·11) 0·07 (0–0·68) 0·04 (0–0·27) 0·1 (0·03–0·44) <0·0001 0·0421 <0·0001 <0·0001 0·02 (0–0·08) 0·017 (0–0·08) 0·02 (0–0·09) 0·018 (0–0·06) CD4 IL-17 0·09 (0·01–0·28) 0·12 (0·03–0·27) 0·09 (0–0·37) 0·1 (0·03–0·23) 0·0946 0·5425 0·4047 0·2843 0·07 (0–0·27) 0·06 (0·02–0·27) 0·08 (0·01–0·26) 0·078 (0–0·25) CD8 IFNγ 0 (0–0·21) 0·02 (0–0·94) 0 (0–0·58) 0·01 (0–0·3) 0·0101 0·5499 0·2264 0·2897 0 (0–0·35) 0 (0–0·19) 0 (0–0·33) 0 (0–0·24) CD8 TNFα 0 (0–0·28) 0 (0–0·24) 0 (0–0·48) 0 (0–0·05) 0·4513 0·7615 0·7337 0·3953 0 (0–0·09) 0 (0–0·38) 0 (0–0·2) 0 (0–0·13) Data are median (minimum to maximum) of total cytokines at each of the study timepoints, unless otherwise stated. Population is the immunology substudy (the first 70 participants), of which complete data were available for 57 participants. Statistical comparison of total cytokine responses in MVA85A study group used Wilcoxon matched-pairs signed-rank test. IL=interleukin. IFNγ=interferon γ. TNFα=tumour necrosis factor α.

e stated. Population is the immunology substudy (the first 70 participants), of which complete data were available for 57 participants. Statistical comparison of total cytokine responses in MVA85A study group used Wilcoxon matched-pairs signed-rank test. IL=interleukin. IFNγ=interferon γ. TNFα=tumour necrosis factor α. Table 4 Primary and secondary efficacy results (per-protocol population) Overall Participants not receiving antiretroviral therapy Participants receiving antiretroviral therapy Placebo MVA85A Vaccine efficacy (95% CI) Placebo MVA85A Vaccine efficacy (95% CI) Placebo MVA85A Vaccine efficacy (95% CI) Disease endpoint 1 (primary efficacy endpoint) 9/325 (2·8%) 6/320 (1·9%) 32·8%(−111·5 to 80·3) 1/69 (1·4%) 2/65 (3·1%) −114·1%(−12 528·3 to 88·9) 8/256 (3·1%) 4/255 (1·6%) 50·3%(−85·4 to 89·1) Disease endpoint 3 9/325 (2·8%) 8/320 (2·5%) 10·5%(−161·3 to 70·0) 1/69 (1·4%) 3/65 (4·6%) −224·7%(−16 947·7 to 73·9) 8/256 (3·1%) 5/255 (2·0%) 38·2%(−114·1 to 84·1) QFT positive conversion 40/173 (23·1%) 38/186 (20·4%) 11·7%(−41·3 to 44·9) 11/36 (30·6%) 6/38 (15·8%) 44·2%(−64·8 to 83·0) 29/137 (21·2%) 32/148 (21·6%) −0·1%(−71·5 to 41·4) Data are n/N (%), unless otherwise stated. Disease endpoint 1 was defined as culture or GeneXpert MTB/RIF positivity; disease endpoint 2 included endpoint 1 and a composite clinical endpoint; and disease endpoint 3 was commencement on anti-tubercular chemotherapy. No additional participants met endpoint 2 who did not already meet endpoint 1. QFT=QuantiFERON-TB Gold In-Tube.

Introduction Cystic fibrosis has been a target for gene therapy since the CFTR gene was cloned in 1989.1 Lung disease is the main cause of morbidity and mortality in individuals with cystic fibrosis, with a median age at death of 29 years (95% CI 27–31).2 Early expectations of a rapid breakthrough were based on supposed ease of access to the target respiratory epithelium via inhaled aerosols. These hopes were tempered by the subsequent realisation that the airways are well defended, in keeping with their predominant function as conducting passages, rather than absorptive surfaces. Research in context Evidence before this study We searched PubMed between June 1, 1992, and March 1, 2015, for studies published that included the terms “non-viral, gene therapy, cystic fibrosis” or “liposome, gene therapy, cystic fibrosis”. Added value of this study We report the first trial of non-viral CFTR gene therapy for patients with cystic fibrosis that is powered to detect clinically relevant pulmonary changes. Our study has progressed this field of research from phase 1 and 2a studies showing changes in molecular surrogates of CFTR function, to a phase 2b setting assessing changes in lung function in patients with a broad range of CFTR mutations. Additionally, our study shows that monthly repeated application of non-viral gene therapy can be safely administered to the lungs over a 1 year period. Implications of all the evidence

We report the first trial of non-viral CFTR gene therapy for patients with cystic fibrosis that is powered to detect clinically relevant pulmonary changes. Our study has progressed this field of research from phase 1 and 2a studies showing changes in molecular surrogates of CFTR function, to a phase 2b setting assessing changes in lung function in patients with a broad range of CFTR mutations. Additionally, our study shows that monthly repeated application of non-viral gene therapy can be safely administered to the lungs over a 1 year period. Implications of all the evidence By providing the first proof of concept that non-viral gene therapy can beneficially affect lung function, follow-up studies can assess optimum dose, dosing interval, and patient stratification at trial entry. Our findings are likely to catalyse earlier translation of more efficient vectors into first-in-man trials.

Implications of all the evidence By providing the first proof of concept that non-viral gene therapy can beneficially affect lung function, follow-up studies can assess optimum dose, dosing interval, and patient stratification at trial entry. Our findings are likely to catalyse earlier translation of more efficient vectors into first-in-man trials. Various vectors for delivery of the CFTR gene into respiratory epithelial cells have been assessed. Viral approaches, including adenoviruses, adeno-associated viruses, and retroviruses, have faltered because of inefficient transduction from the luminal surface and immune responses restricting the efficacy of repeated application.3 As such, research from the UK Cystic Fibrosis Gene Therapy Consortium has initially focused on non-viral vectors. Formulation and delivery of plasmid DNA–liposome complexes have been refined in a large series of preclinical studies,4, 5 and safety,6, 7 molecular efficacy, and practical doses have been assessed in several phase 1 and 2a studies in patients with cystic fibrosis.1, 3 We did this study to assess the clinical efficacy of the non-viral CFTR gene–liposome complex pGM169/GL67A8 after repeated delivery to the airways. Methods Study design and participants We did this randomised, double-blind, placebo-controlled, phase 2b trial in two cystic fibrosis centres with patients recruited from 18 sites in the UK. Eligible participants had diagnosed cystic fibrosis, were aged 12 years or older, had a forced expiratory volume in 1 s (FEV1) of 50–90% predicted, and had any combination of CFTR mutations.

double-blind, placebo-controlled, phase 2b trial in two cystic fibrosis centres with patients recruited from 18 sites in the UK. Eligible participants had diagnosed cystic fibrosis, were aged 12 years or older, had a forced expiratory volume in 1 s (FEV1) of 50–90% predicted, and had any combination of CFTR mutations. The protocol was approved by the National Research Ethics Committee and the local Research Committees at the two dosing sites and the 16 other referral centres. Each patient, or a parent, provided written informed consent, and children provided assent. Randomisation and masking We randomly assigned patients (1:1), via a computer-based randomisation system, to receive nebulised pGM169/GL67A or 0·9% saline (placebo). Randomisation was stratified by % predicted FEV1 (<70 vs ≥70%), age (<18 vs ≥18 years), inclusion in the mechanistic substudy, and dosing site (London or Edinburgh). Participants in the mechanistic substudy were randomly assigned (2:1) to receive nebulised pGM169/GL67A or placebo, and could participate as part of either a nasal or bronchoscopy group, or both. Participants and investigators were masked to treatment allocation, with the randomisation code known only by pharmacy staff at the two dosing sites.

nistic substudy were randomly assigned (2:1) to receive nebulised pGM169/GL67A or placebo, and could participate as part of either a nasal or bronchoscopy group, or both. Participants and investigators were masked to treatment allocation, with the randomisation code known only by pharmacy staff at the two dosing sites. Procedures Patients received 5 mL of either 0·9% saline or pGM169/GL67A complex nebulised through a Trudell AeroEclipse II device (Trudell Medical International, London, ON, Canada) at 28 day intervals (plus or minus 5 days) for 12 months. Each 5 mL dose of pGM169/GL67A contained 13·3 mg of plasmid DNA and 75 mg of the GL67A lipid mixture. Routine treatments were continued throughout the study, except for DNase, which was withheld for 24 h before and after dosing. In addition to the nebulised dose, patients in the nasal group of the mechanistic substudy received 2 mL of placebo or pGM169/GL67A divided between nasal cavities via a nasal spray device at the time of each lung dose. Patients in the bronchoscopy group followed the standard protocol, but also underwent a bronchoscopy under general anaesthesia before the first dose and 28 days (plus or minus 5 days) after the final dose.

of placebo or pGM169/GL67A divided between nasal cavities via a nasal spray device at the time of each lung dose. Patients in the bronchoscopy group followed the standard protocol, but also underwent a bronchoscopy under general anaesthesia before the first dose and 28 days (plus or minus 5 days) after the final dose. Outcomes The primary efficacy endpoint was the relative change in % predicted FEV1, calculated from the mean of two baseline values (at screening and before dosing on day of the first dose) to the mean of two values (2 and 4 weeks after last dose) at study completion. Secondary outcomes included additional measurements of lung function, CT scans, and Cystic Fibrosis Questionnaire-Revised (CFQ-R) scores.9 Exploratory endpoints included exercise testing, activity monitoring, and sputum inflammatory markers. Mechanistic endpoints were nasal or bronchial vector-specific DNA, mRNA, and electrophysiological assessment of CFTR function. We did extensive safety assessments.

ystic Fibrosis Questionnaire-Revised (CFQ-R) scores.9 Exploratory endpoints included exercise testing, activity monitoring, and sputum inflammatory markers. Mechanistic endpoints were nasal or bronchial vector-specific DNA, mRNA, and electrophysiological assessment of CFTR function. We did extensive safety assessments. Statistical analysis The statistical analyses were prespecified in a statistical analysis plan. With use of pilot data, we estimated the standard deviation of the relative change in % predicted FEV1 in the target cystic fibrosis population to be 10% over 12 months. A total sample size of 120 assessable patients would provide 90% power to detect a 6% difference between groups in the mean change from baseline at a two-sided 5% significance level. This power calculation was conservative because covariate adjustment can be expected to increase statistical power. We did analyses in the per-protocol population (primary analysis), predefined as participants who received at least nine doses of pGM169/GL67A or placebo, and in the intention-to-treat population, who received at least one dose of pGM169/GL67A or placebo.

ariate adjustment can be expected to increase statistical power. We did analyses in the per-protocol population (primary analysis), predefined as participants who received at least nine doses of pGM169/GL67A or placebo, and in the intention-to-treat population, who received at least one dose of pGM169/GL67A or placebo. We compared outcomes between groups with an ANCOVA model, with inclusion of the relevant baseline value, treatment allocation, and stratification factors (baseline predicted FEV1, age, dosing site, inclusion in substudy). Results are reported as adjusted mean differences with corresponding 95% CIs. We assessed subgroup effects by including the relevant interaction term in the ANCOVA model. To allow results from different endpoints to be plotted on a common scale, the estimated treatment effects were standardised and presented as multiples of the underlying SD. No adjustment was made to the p values to allow for multiplicity because the secondary endpoints were supportive and the corresponding p values were interpreted conservatively. We assessed bronchial and nasal biomarkers with a Mann–Whitney U test. A two-sided p value less than 0·05 was considered statistically significant. The trial was overseen by an independent Data Monitoring and Ethics Committee and a Trial Steering Committee. This trial is registered with ClinicalTrials.gov, number NCT01621867.

We compared outcomes between groups with an ANCOVA model, with inclusion of the relevant baseline value, treatment allocation, and stratification factors (baseline predicted FEV1, age, dosing site, inclusion in substudy). Results are reported as adjusted mean differences with corresponding 95% CIs. We assessed subgroup effects by including the relevant interaction term in the ANCOVA model. To allow results from different endpoints to be plotted on a common scale, the estimated treatment effects were standardised and presented as multiples of the underlying SD. No adjustment was made to the p values to allow for multiplicity because the secondary endpoints were supportive and the corresponding p values were interpreted conservatively. We assessed bronchial and nasal biomarkers with a Mann–Whitney U test. A two-sided p value less than 0·05 was considered statistically significant. The trial was overseen by an independent Data Monitoring and Ethics Committee and a Trial Steering Committee. This trial is registered with ClinicalTrials.gov, number NCT01621867. Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

The trial was overseen by an independent Data Monitoring and Ethics Committee and a Trial Steering Committee. This trial is registered with ClinicalTrials.gov, number NCT01621867. Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results Figure 1 shows the trial profile. Between June 12, 2012, and June 24, 2013, we randomly assigned 140 patients to receive placebo (n=62) or pGM169/GL67A (n=78), of whom 136 (97%) patients comprised the intention-to-treat population and 116 (83%) patients comprised the per-protocol population (figure 1). Reasons for discontinuation in the intention-to-treat population were similar between groups (appendix). Baseline characteristics were similar between the two groups (table 1). Unless indicated otherwise, all subsequent details relate to the per-protocol population.

nts comprised the per-protocol population (figure 1). Reasons for discontinuation in the intention-to-treat population were similar between groups (appendix). Baseline characteristics were similar between the two groups (table 1). Unless indicated otherwise, all subsequent details relate to the per-protocol population. 114 (98%) patients had paired pre-treatment and post-treatment measurements of % predicted FEV1. Of the two patients (both in the placebo group) who did not have paired measurements, one patient could not do the test because of a surgery-related pneumothorax and one withdrew because of time commitments and was unavailable for follow-up measurements. We recorded a significant ANCOVA-adjusted treatment effect in the pGM169/GL67A group versus placebo at 12 months' follow-up (3·7%, 95% CI 0·1–7·3; p=0·046; figure 2) The relative changes within each of the individual groups were −4·0% (95% CI −6·6 to −1·4) in the placebo group and −0·4% (−2·8 to 2·1) in the pGM169/GL67A group (figure 2). Post-hoc analysis showed that 21 (18%) patients (n=6 in the placebo group and n=15 in the pGM169/GL67A group) had an improvement in % predicted FEV1 of 5% or more of their individual baseline values. For comparison, the treatment effect in patients in the intention-to-treat population who had spirometry measurements both before dosing and within the protocol-defined window after their final dose (n=56 in the placebo group and n=65 in the pGM169/GL67A group) was 3·6% (95% CI 0·2–7·0; p=0·039), with the 20 patients included in the intention-to-treat, but not per-protocol, analysis, receiving a mean of 3·7 doses (SD 1·9).

easurements both before dosing and within the protocol-defined window after their final dose (n=56 in the placebo group and n=65 in the pGM169/GL67A group) was 3·6% (95% CI 0·2–7·0; p=0·039), with the 20 patients included in the intention-to-treat, but not per-protocol, analysis, receiving a mean of 3·7 doses (SD 1·9). Figure 3 summarises changes in a range of secondary outcomes. The treatment effect was significant for FVC (p=0·031; appendix) and CT gas trapping (p=0·048), but not for other measures of lung function, imaging, and quality of life (figure 3). We assessed whether a responder subgroup could be identified; the appendix summarises the prespecified subgroups. We noted no significant differences in the primary outcome treatment effect with respect to sex, age, CFTR mutation (phe508del homozygous vs other), Pseudomonas colonisation, predominant smaller or larger airway disease on CT at presentation, concurrent drugs, or treatment-associated adverse events (appendix). Although some subgroups had larger treatment effects than others, these results were typically due to a greater decline in FEV1 in the placebo group, rather than to any difference of effect in the pGM169/GL67A group (appendix). Stratification by baseline % predicted FEV1 suggested a difference, albeit non-significant, in treatment effect between patients with more severe disease (FEV1 49·6–69·2% predicted), who had a treatment effect of 6·4% (95% CI 0·8–12·1), and those with less severe disease (69·6–89·9% predicted), who had a treatment effect of 0·2% (−4·6 to 4·9; pinteraction=0·065; appendix). In patients with more severe disease, post-trial and pre-trial changes in both the placebo group (−4·9%) and the pGM169/GL67A group (1·5%) contributed to the treatment effect. Secondary outcomes showed a similar trend favouring the more severe category (appendix).

ct of 0·2% (−4·6 to 4·9; pinteraction=0·065; appendix). In patients with more severe disease, post-trial and pre-trial changes in both the placebo group (−4·9%) and the pGM169/GL67A group (1·5%) contributed to the treatment effect. Secondary outcomes showed a similar trend favouring the more severe category (appendix). Patients in both treatment groups received a median of three (IQR one to five) courses of oral or intravenous antibiotics during the trial. Specifically, we assessed co-administered antibiotics during the critical analysis period from dose 11 to the end of the trial. Numbers of patients receiving any additional antibiotics were 26 (48%) in the placebo group and 30 (51%) in the pGM169/GL67A group (χ2 p=0·774). Thus, the observed FEV1 treatment effect was considered to be independent of concurrent antibiotic courses.

the critical analysis period from dose 11 to the end of the trial. Numbers of patients receiving any additional antibiotics were 26 (48%) in the placebo group and 30 (51%) in the pGM169/GL67A group (χ2 p=0·774). Thus, the observed FEV1 treatment effect was considered to be independent of concurrent antibiotic courses. No clinically relevant pattern of changes could be distinguished in the exploratory outcomes of activity and exercise monitoring and serum and sputum inflammatory markers (appendix). In the bronchoscopy group of the substudy, vector-specific DNA increased in 12 (86%) of 14 patients in the pGM169/GL67A group and was below the limit of quantification in all (n=7) placebo samples (p=0·001; figure 4A); vector-specific mRNA was below the level of sensitivity in both groups (appendix). Changes in basal post-trial and pre-trial potential difference values did not differ significantly in either group (appendix). Figure 4B shows bronchial chloride responses using the mean of all interpretable tracings for each patient; a negative value indicates a change in the non-cystic fibrosis direction. Patients in the placebo group (n=7) had a median change (post-trial minus pre-trial) of 3·1 mV (range 9·3 to −1·2) and those in the pGM169/GL67A group (n=10) had a change of −1·3 mV (4·0 to −5·8; p=0·032; figure 4B). Five (50%) of ten patients in the pGM169/GL67A group had values that were more negative than the largest response in the placebo group (figure 4). In the same analysis with only the most negative value recorded for each patient at any timepoint, patients in the placebo group had a median post-trial minus pre-trial change of 2·6 mV (range 9·3 to −1·2) and those in the pGM169/GL67A group had a change of −2·8 mV (4·0 to −16·8 mV; p=0·088; figure 4C). Six (60%) patients in the pGM169/GL67A group had values that were more negative than the largest response in the placebo group (figure 4). The appendix shows absolute bronchial potential difference values.

ge 9·3 to −1·2) and those in the pGM169/GL67A group had a change of −2·8 mV (4·0 to −16·8 mV; p=0·088; figure 4C). Six (60%) patients in the pGM169/GL67A group had values that were more negative than the largest response in the placebo group (figure 4). The appendix shows absolute bronchial potential difference values. In patients in the nasal group of the substudy, vector-specific DNA increased in all the 17 patients given pGM169/GL67A. Despite apparent pGM169 contamination in some samples, the change in pGM169 concentrations differed significantly between the groups (appendix); no vector-specific mRNA was quantifiable in either group. We noted no significant changes in the baseline, zero chloride, or isoprenaline responses (appendix). Four (29%) of 14 pGM169/GL67A patients had mean post-trial minus pre-trial treatment responses (ranging from −3·4 mV to −7·0 mV) that were more negative than the largest response in the placebo group (n=6; appendix). The appendix shows absolute nasal potential difference values.

e, or isoprenaline responses (appendix). Four (29%) of 14 pGM169/GL67A patients had mean post-trial minus pre-trial treatment responses (ranging from −3·4 mV to −7·0 mV) that were more negative than the largest response in the placebo group (n=6; appendix). The appendix shows absolute nasal potential difference values. All patients had adverse events, with no significant difference between groups for either total events or within the nine predefined adverse event categories (table 2). One patient in the placebo group and one patient in the pGM169/GL67A group discontinued study treatment because of adverse events (fatigue and increased respiratory symptoms and flu-like symptoms, respectively). We recorded six serious adverse events, all in the pGM169/GL67A group (appendix). Neither the Data Monitoring and Ethics Committee nor the Trial Steering Committee regarded any serious adverse event as related to study drug; however, one event was considered to be possibly related to a trial procedure (bronchoscopy). We noted no clinically relevant changes in haematology, biochemistry, conversion of anti-CFTR T cells, anti-DNA antibodies, histology, or lipid staining (appendix) and no patients died during the study.

related to study drug; however, one event was considered to be possibly related to a trial procedure (bronchoscopy). We noted no clinically relevant changes in haematology, biochemistry, conversion of anti-CFTR T cells, anti-DNA antibodies, histology, or lipid staining (appendix) and no patients died during the study. Discussion We report the first trial of non-viral based gene therapy for cystic fibrosis, powered to detect clinically relevant pulmonary changes. After monthly dosing for 1 year, we recorded evidence of a beneficial effect of gene therapy versus placebo on FEV1. No effect of sex, age, or whether patients were homozygous for the most common F508del CFTR mutation could be detected. No clinically important adverse events attributable to treatment with pGM169/GL67A were reported.

dosing for 1 year, we recorded evidence of a beneficial effect of gene therapy versus placebo on FEV1. No effect of sex, age, or whether patients were homozygous for the most common F508del CFTR mutation could be detected. No clinically important adverse events attributable to treatment with pGM169/GL67A were reported. Although these findings are encouraging, they should be put into perspective. We noted a stabilisation of FEV1 in the pGM169/GL67A group rather than an improvement. This stabilisation took place over a 1 year period and further work will be needed to see if this effect is maintained. The reduction in FEV1 in the placebo group was within the range reported in some other prospective trials10, 11, 12 and is consistent with a median survival of 29 years, but is greater than would be expected from registry data.2 Three factors are likely to have influenced this difference. First, the requirement for clinical stability at trial entry meant that patients might have been at their optimum respiratory health at this stage. Second, the enthusiasm of patients to enter the trial, accompanied by a focus on self-care, might have resulted in short-term improvements in lung function during the recruitment period. Both factors are likely to lead to a subsequent decline in lung function as patients regress to their mean values. Third, we included all available data, whether from stable patients or those with exacerbations, by contrast with registry data, which focuses on measurements obtained at annual review. Stabilisation of lung disease in itself is a worthwhile aim and we would caution against the bar being set too high for novel therapeutics in cystic fibrosis populations with an unselected range of mutations. The large response to ivacaftor in patients with class III mutations takes place in the context of correctly localised CFTR protein. By contrast, much smaller improvements in lung function were shown in the ivacaftor–lumacaftor trial for the most common mutation (phe508del) in which the CFTR protein is misfolded and mislocalised.13

ponse to ivacaftor in patients with class III mutations takes place in the context of correctly localised CFTR protein. By contrast, much smaller improvements in lung function were shown in the ivacaftor–lumacaftor trial for the most common mutation (phe508del) in which the CFTR protein is misfolded and mislocalised.13 The response in our study was heterogeneous, with apparent responders and non-responders. The data suggest that an approximate doubling of treatment effect was achieved in patients with more severe disease stratified by baseline FEV1, supported by trends in other clinically relevant secondary measures. A larger trial with a stratified trial entry design, powered to assess subgroups, and that addresses the mechanisms of response heterogeneity, will be important to verify or refute these data. This differential response could relate to the dose deposited in the airways; in patients with lower baseline FEV1 the relatively more obstructed smaller airways result in a larger proportion of the 5 mL dose being deposited in the larger airways. In pre-trial studies we assessed airway deposition in patients with cystic fibrosis with varying FEV1 severity with technetium-99m labelled human serum albumin of similar droplet size (3–4 μm, using a different nebuliser system) to the pGM169/GL67A formulation. Bronchial airway (generations 2–8) fractional deposition was 2·9% of delivered dose (standard error of the mean [SEM] 0·2; n=33) in patients with 70–90% predicted FEV1 and roughly twice as great (6·0%, SEM 1·0; n=23) in those with 50–70% predicted FEV1. An additional contributory factor to this enhanced efficacy might be the increased mitotic rate of more severely affected tissues,14 which decreases the proportion of time that the nuclear membrane is intact, the membrane acting as a barrier to plasmid DNA entry to the nucleus.

=23) in those with 50–70% predicted FEV1. An additional contributory factor to this enhanced efficacy might be the increased mitotic rate of more severely affected tissues,14 which decreases the proportion of time that the nuclear membrane is intact, the membrane acting as a barrier to plasmid DNA entry to the nucleus. We cannot rule out that the changes recorded in the present study are the result of a non-specific response to the pGM169/GL67A formulation. The placebo was 0·9% saline rather than a scrambled or CFTR-deleted plasmid–liposome complex. We selected 0·9% saline partly on the basis of pragmatic financial considerations, but mainly for ethical considerations, not wishing to expose patients with cystic fibrosis to first-in-man repeated pulmonary dosing of an untested product that might direct the expression of an immunologically active peptide or novel non-coding RNA molecule with deleterious biological functions. Furthermore, we wanted to compare progression on therapy with the natural history of the disease. In terms of alternative explanations for the effects we noted, we know of no evidence that monthly nebulisation of 0·9% saline is deleterious to lung function, nor that liposome alone produces physiological improvements in either patients without,15 or those with16 cystic fibrosis. Delivery of non-CFTR encoding plasmid DNAs to the human airways has not been associated with a gain in CFTR chloride-channel function, nor improvement in any cystic fibrosis-related assay,17, 18 and plasmid DNA is generally associated with pro-inflammatory, rather than non-specific, beneficial effects.19 We did not identify any pathophysiological changes in the airways, such as inflammation or remodelling, nor any changes in bacterial species that might otherwise explain the outcomes. Nevertheless, we cannot exclude that DNA–liposome complexes augment host defences, stimulate mucus clearance, or enhance bacterial killing to an extent undetectable on semi-quantitative routine culture.

such as inflammation or remodelling, nor any changes in bacterial species that might otherwise explain the outcomes. Nevertheless, we cannot exclude that DNA–liposome complexes augment host defences, stimulate mucus clearance, or enhance bacterial killing to an extent undetectable on semi-quantitative routine culture. Results showing more robust changes in molecular CFTR surrogates would have been reassuring. Despite extensive optimisation of quantitative realtime-PCR assays, the pGM169-derived mRNA assay has poor sensitivity and is adversely affected by the inclusion of high levels of total RNA or modest concentrations of pGM169 plasmid DNA. In ovine studies we have shown that a 20 mL nebulised dose of pGM169/GL67A, four times that used in the present trial, is the lower threshold for reproducible detection of mRNA with this assay in airway tissue samples (unpublished).6 Thus, our inability to detect pGM169-derived mRNA after delivery of 5 mL of pGM169/GL67A to the human airways, although disappointing, was not surprising. In human tissues, we have noted the low sensitivity of assays assessing vector-specific mRNA from human samples in vivo,16, 20, 21 and have noted the greater sensitivity of detection of electrophysiological changes, consistent with findings in this study.17, 18, 22

n airways, although disappointing, was not surprising. In human tissues, we have noted the low sensitivity of assays assessing vector-specific mRNA from human samples in vivo,16, 20, 21 and have noted the greater sensitivity of detection of electrophysiological changes, consistent with findings in this study.17, 18, 22 The ratio of area sampled to area dosed is small. Although we recorded significant chloride secretory changes in the bronchial, but not the nasal, epithelium, we caution against placing undue weight on either observation. The size of the groups in the mechanistic substudy was limited by both the practicality of the procedures and the acceptability to patients of the additional invasive tests, leading to low statistical power for these measures. We would instead conclude that modest variable changes can be shown with currently available assays that remain insufficiently sensitive to detect changes in low levels of CFTR function when assessed in vivo in humans; further optimisation in these or other assays is needed.

to low statistical power for these measures. We would instead conclude that modest variable changes can be shown with currently available assays that remain insufficiently sensitive to detect changes in low levels of CFTR function when assessed in vivo in humans; further optimisation in these or other assays is needed. Although we are encouraged by the first demonstration of a significant beneficial effect in lung function compared with placebo associated with gene therapy in patients with cystic fibrosis, the mean difference was modest, only recorded in some individuals, and at the lower end of the range of results seen in clinical trials which result in changes in patient-related care.23, 24 We did not formally assess infective exacerbations in view of the fairly small patient numbers in our study, but use of antibiotic courses as a surrogate identified no obvious treatment advantage. The treatment effect is consistent with a clinically meaningful benefit from the perspective of the European Medicine Agency;25 however, further improvements in efficacy and consistency of response to the current formulation, or its combination with CFTR potentiators, are needed before gene therapy is suitable for clinical practice. Furthermore, our findings should encourage the rapid introduction of more potent gene transfer vectors into early phase trials, now that much of the groundwork has been established.

e to the current formulation, or its combination with CFTR potentiators, are needed before gene therapy is suitable for clinical practice. Furthermore, our findings should encourage the rapid introduction of more potent gene transfer vectors into early phase trials, now that much of the groundwork has been established. The data reported here provide the first proof of concept that repeated administration of non-viral CFTR gene therapy can safely change clinically relevant parameters, providing another step along the path of translational cystic fibrosis gene therapy. For the UK Cystic Fibrosis Gene Therapy Consortium see http://www.cfgenetherapy.org.uk This online publication has been corrected. The corrected version first appeared at thelancet.com/respiratory on September 7, 2015 Supplementary Material Supplementary appendix

The data reported here provide the first proof of concept that repeated administration of non-viral CFTR gene therapy can safely change clinically relevant parameters, providing another step along the path of translational cystic fibrosis gene therapy. For the UK Cystic Fibrosis Gene Therapy Consortium see http://www.cfgenetherapy.org.uk This online publication has been corrected. The corrected version first appeared at thelancet.com/respiratory on September 7, 2015 Supplementary Material Supplementary appendix Acknowledgments We thank the patients, carers, and families who gave so much to this trial; those who so generously supported the programme of work that led to this trial by giving to the Cystic Fibrosis Trust; the Medicor Foundation and the Cystic Fibrosis Trust for part-funding the cost of the plasmid DNA and lipid; the National Institute for Health Research (NIHR) Clinical Research Network and Just Gene Therapy for additional funding; Sandra Griffiths, Nayan Das, Amanda Bravery, Juan Gonzales-Maffe, and Ginny Picot (Imperial Clinical Trials Unit); Jermaine Wright, Fabricio Ghiraldi, Stephen Man, Judith Foy, and Viba Teli (Royal Brompton Hospital Pharmacy) and Peter Brown (Royal Marsden Pharmacy); Leticia Brown, Michelle Watson-Dotchin, and Michelle Pugh for administrative support; the Efficacy and Mechanism Evaluation Programme team, including Lucy Knight and Dani Preedy, who guided us throughout the trial; Tony Hickson and Gursharan Randhawa for help with the intellectual property and commercialisation aspects; the Data Monitoring and Ethics Committee (Colin Wallace, Caroline Elston, George Dickson, Julie Morris) and the Trial Steering Committee (Ashley Woodcock, Nikki Samsa, Brandon Wainwright, Pierre Lehn, Janet Allen) for their wise and helpful advice; and the teams at the referring sites who gave so generously of their time: Jeremy Hull and Catherine McKenny (Oxford Churchull); Nick Bell and Kathryn Bateman (Bristol); Judy Ryan (Papworth); Maya Desai, Michelle Taberner, and Edward Nash (Birmingham); Paul Aurora and Ammani Prasad (Great Ormond Street); Judith Duguid, Alexandra Highton, and Timothy Ho (Frimley); Siobhan Carr and Catherine Lambert (Royal London); Thomas Daniels and Victoria Brown (Southampton); Daniel Peckham and Amy Driffill (Leeds); Felicity Perrin and Mike Waller (King's College London); Gordon MacGregor, Steve Bicknell, Ewan Ross, and Anne Devenny (Glasgow); Carol Dryden (Wishaw); Richard Brooker, Graham Devereux, Owen Dempsey (Aberdeen); Helen Rodgers and Jonathon McCormick (Dundee); and Steven Bourke and David Spencer (Dundee).

iffill (Leeds); Felicity Perrin and Mike Waller (King's College London); Gordon MacGregor, Steve Bicknell, Ewan Ross, and Anne Devenny (Glasgow); Carol Dryden (Wishaw); Richard Brooker, Graham Devereux, Owen Dempsey (Aberdeen); Helen Rodgers and Jonathon McCormick (Dundee); and Steven Bourke and David Spencer (Dundee). This study was funded by the Efficacy and Mechanism Evaluation (EME) Programme, a Medical Research Council and NIHR partnership. We acknowledge the financial support of NHS Research Scotland, through the Edinburgh Clinical Research Facility. The research was supported by the NIHR Respiratory Biomedical Research Unit at the Royal Brompton and Harefield NHS Foundation Trust, and the NIHR Imperial Biomedical Research Centre at Imperial College Healthcare NHS Trust and Imperial College London. The UK Cystic Fibrosis Gene Therapy Consortium would not have come together, and this trial would not have happened, without the vision and leadership of Rosie Barnes during her tenure as Chief Executive of the Cystic Fibrosis Trust.

Research Centre at Imperial College Healthcare NHS Trust and Imperial College London. The UK Cystic Fibrosis Gene Therapy Consortium would not have come together, and this trial would not have happened, without the vision and leadership of Rosie Barnes during her tenure as Chief Executive of the Cystic Fibrosis Trust. Contributors EWFWA, ACB, SC, JCD, DMG, DRG, APG, UG, SCH, TEH, JAI, GDM, and DJP conceived, designed, and analysed the overall study. DA and GDM designed and coordinated data collection and statistical analysis. DKA, KJB, DB, PC, GD, NSD, HIE, RFF, JG, JSRG, DMH, KH, SLH, JJ, BFK, MM, EKP, ALQ, CJS, SSh, NJS, NS, EJS, SNS, RPU, and MDW assessed patient outcomes and undertook and analysed individual in-vivo assays. EVB, MHD, and SSo coordinated and undertook the administration of the trial. RB, NJ, PL-E, GR, and KS oversaw receipt, preparation, and dispensing of study drug. JB, RC, MC, HED, AD, JD, SG-S, LH, MPL, AWM, MCM, DM, CM, MAM, HM, LJM, AGN, TO, JP-L, IAP, KMP, BJS, SGS-J, MT, MYW, and JMW designed, undertook, and analysed in-vitro assays. SHC, RKS, and PW-H coordinated the production of lipid 67A. DDSC, LAD, and GM designed, undertook, and analysed studies of study drug delivery.

JB, RC, MC, HED, AD, JD, SG-S, LH, MPL, AWM, MCM, DM, CM, MAM, HM, LJM, AGN, TO, JP-L, IAP, KMP, BJS, SGS-J, MT, MYW, and JMW designed, undertook, and analysed in-vitro assays. SHC, RKS, and PW-H coordinated the production of lipid 67A. DDSC, LAD, and GM designed, undertook, and analysed studies of study drug delivery. Declaration of interests ACB, AD, APG, AGN, AWM, BFK, BJS, CM, CJS, DKA, DA, DB, DM, DMH, DMG, DDSC, DJP, DRG, EKP, EJS, EVB, EWFWA, GD, GM, GDM, GR, HED, HM, HIE, IAP, JAI, JB, JCD, JD, JG, JSRG, JJ, JP-L, JMW, KJB, KH, KMP, KS, LAD, LH, LJM, MC, MCM, MHD, MM, MT, MYW, MAM, MDW, MPL, NSD, NJ, NJS, NS, PC, PL-E, PW-H, RB, RC, RFF, RPU, SC, SCH, SG-S, SLH, SSh, SSo, SGS-J, SNS, TEH, TO, and UG report grants from the National Institute for Health Research, the Cystic Fibrosis Trust, Just Gene Therapy, Medicor Foundation, and other support from Genzyme, a Sanofi company, related to the submitted work during the conduct of the study. ALQ, JCD, JMW, MCM, NJS, RKS, SC, and SHC report fees, grants, honoraria, and patents outside of the submitted work. ACB, DJP, DRG, EWFWA, JAI, JCD, LAD, SCH, and UG report patents related to the submitted work. Figure 1 Trial profile Numbers of patients in the intention-to-treat population are unequal because of the 2:1 allocation in the mechanistic substudy. FEV1=forced expiratory volume in 1 s. Figure 2 Timecourse of the primary outcome response to either placebo or pGM169/GL67A (A) and the individual patient responses in the pGM169/GL67A (B) and placebo (C) groups

Numbers of patients in the intention-to-treat population are unequal because of the 2:1 allocation in the mechanistic substudy. FEV1=forced expiratory volume in 1 s. Figure 2 Timecourse of the primary outcome response to either placebo or pGM169/GL67A (A) and the individual patient responses in the pGM169/GL67A (B) and placebo (C) groups Error bars in panel A show the standard error of the mean. Primary outcome measurements were taken at each treatment visit before administration of study drugs. Pre and post values indicate the mean of two measurements at the respective timepoints. Positive values in panels B and C show an improvement. FEV1=forced expiratory volume in 1 s. Figure 3 Forest plot showing secondary outcome responses to placebo or pGM169/GL67A Data are mean (SD) or mean (95% CI), unless otherwise indicated. The size of the circles is proportional to the number of patients represented and the error bars show 95% CIs. Values shown for FEV1 are the relative change in the % predicted FEV1. To allow results from different endpoints to be plotted on a common scale, the estimated treatment effects were standardised to be presented as multiples of the underlying SD (standardised treatment effect). FEV1=forced expiratory volume in 1 s. MEF25–75=mid-expiratory flow between 25% and 75% of FVC. KCOc=diffusion capacity of the alveolar capillary membrane, corrected for haemoglobin concentrations. TLCOc=transfer factor of the lung for carbon monoxide, corrected for haemoglobin concentrations. *Refers to scores from the Cystic Fibrosis Questionnaire-Revised.

EF25–75=mid-expiratory flow between 25% and 75% of FVC. KCOc=diffusion capacity of the alveolar capillary membrane, corrected for haemoglobin concentrations. TLCOc=transfer factor of the lung for carbon monoxide, corrected for haemoglobin concentrations. *Refers to scores from the Cystic Fibrosis Questionnaire-Revised. Figure 4 Assessment of DNA from bronchial brushings in the placebo (n=7) and pGM169/GL67A (n=14) subgroups (A) and the response of the bronchial epithelium to perfusion with a zero chloride solution containing isoprenaline 10 μM (B, C) Horizontal bars show median values. Each circle in panel A represents an individual patient. Each symbol in panels B and C shows the change in response from trial start to finish for the relevant treatment in an individual patient. Of the 16 participants in the bronchoscopy subgroup, 15 individuals had post-dose bronchoscopies, of whom 14 individuals generated samples for DNA and mRNA molecular analysis. The plotted value in panel B is the mean of all interpretable recordings (range 1–3), and in panel C is the most negative value obtained from all interpretable recordings, at each timepoint for that patient. A more negative value is in the non-cystic fibrosis direction. LOQ=limit of quantification, PBNQ=positive but not quantifiable. Table 1 Baseline and demographic characteristics

Horizontal bars show median values. Each circle in panel A represents an individual patient. Each symbol in panels B and C shows the change in response from trial start to finish for the relevant treatment in an individual patient. Of the 16 participants in the bronchoscopy subgroup, 15 individuals had post-dose bronchoscopies, of whom 14 individuals generated samples for DNA and mRNA molecular analysis. The plotted value in panel B is the mean of all interpretable recordings (range 1–3), and in panel C is the most negative value obtained from all interpretable recordings, at each timepoint for that patient. A more negative value is in the non-cystic fibrosis direction. LOQ=limit of quantification, PBNQ=positive but not quantifiable. Table 1 Baseline and demographic characteristics Placebo group (n=54) pGM169/GL67A group (n=62) Age (years) 26·0 (13·0) 23·6 (10·8) <18 years old 17 (31%) 23 (37%) ≥18 years old 37 (69%) 39 (63%) Sex Female 25 (46%) 31 (50%) Male 29 (54%) 31 (50%) Centre distribution number Edinburgh 24 (44%) 22 (35%) London 30 (56%) 40 (65%) Height (cm) 165·0 (10·6) 163·6 (10·9) Weight (kg) 61·6 (15·6) 61·0 (15·7) FEV1 (% predicted) 69·0 (9·9) 69·9 (11·1) Body-mass index (kg/m2) 22·4 (4·4) 22·4 (4·5) Mutation class Phe508del/Phe508del 26 (48%) 31 (50%) Phe508del/class 1–6 22 (41%) 23 (37%) Not Phe508del/class 1 1 (2%) 3 (5%) Heterozygous/homozygous class 3–6 2 (4%) 2 (3%) Phe508del/unknown class 3 (6%) 3 (5%) Data are mean (SD) or n (%), unless otherwise indicated. Table 2 Adverse events

Placebo group (n=54) pGM169/GL67A group (n=62) Age (years) 26·0 (13·0) 23·6 (10·8) <18 years old 17 (31%) 23 (37%) ≥18 years old 37 (69%) 39 (63%) Sex Female 25 (46%) 31 (50%) Male 29 (54%) 31 (50%) Centre distribution number Edinburgh 24 (44%) 22 (35%) London 30 (56%) 40 (65%) Height (cm) 165·0 (10·6) 163·6 (10·9) Weight (kg) 61·6 (15·6) 61·0 (15·7) FEV1 (% predicted) 69·0 (9·9) 69·9 (11·1) Body-mass index (kg/m2) 22·4 (4·4) 22·4 (4·5) Mutation class Phe508del/Phe508del 26 (48%) 31 (50%) Phe508del/class 1–6 22 (41%) 23 (37%) Not Phe508del/class 1 1 (2%) 3 (5%) Heterozygous/homozygous class 3–6 2 (4%) 2 (3%) Phe508del/unknown class 3 (6%) 3 (5%) Data are mean (SD) or n (%), unless otherwise indicated. Table 2 Adverse events Placebo group (n=54) pGM169/GL67A group (n=62) Lower airway respiratory symptoms 7·9 9·0 Gastrointestinal symptoms 2·1 1·8 Fever or flu-like symptoms 1·1 1·4 Headache 1·2 1·1 Upper airway symptoms 2·3 3·4 Elevated liver function tests 0·3 0·4 Haematuria 0·2 0·2 Isolated raised inflammatory markers 0·8 0·7 Other 3·2 3·3 Total 19·1 21·2 Data are mean number of times the respective symptom was experienced by each patient during the trial. Values were calculated by dividing the total number of the relevant adverse event by the total number of relevant patients in that group.

Introduction Pulmonary arterial hypertension (PAH) is a rare disorder characterised by progressive remodelling of the small pulmonary arteries resulting in increased pulmonary vascular resistance and ultimately right ventricular failure and death.1, 2 The diagnosis of PAH requires a mean pulmonary artery pressure of 25 mm Hg or more with a pulmonary artery wedge pressure of 15 mm Hg or less at right heart catheterisation in the absence of chronic thromboembolic, left heart, or respiratory disease.3 The classification of PAH includes idiopathic and heritable forms. Additionally, PAH might occur after drug or toxin exposure (eg, anorexigens) or in association with other conditions such as congenital heart disease, connective tissue disease, liver disease, or HIV infection.4

lic, left heart, or respiratory disease.3 The classification of PAH includes idiopathic and heritable forms. Additionally, PAH might occur after drug or toxin exposure (eg, anorexigens) or in association with other conditions such as congenital heart disease, connective tissue disease, liver disease, or HIV infection.4 In 2000, heterozygous germline mutations in the gene encoding the bone morphogenetic protein receptor type II (BMPR2) were identified as the main genetic cause of familial PAH.5, 6 Over 300 different BMPR2 mutations have been identified with a prevalence of greater than 75% in families with PAH.7, 8 BMPR-II is a receptor for the bone morphogenetic proteins (members of the transforming growth factor-β superfamily). Mutations in the BMPR2 gene cause loss-of-function and reduced signalling downstream of the receptor. Subsequently, BMPR2 mutations have been identified in apparently sporadic cases of idiopathic PAH with a frequency ranging from 11%9 to 40%.10 The occurrence of BMPR2 mutations in sporadic PAH cases in the absence of a family history can be accounted for by the relatively low penetrance of BMPR2 mutations (20–30%) and the occurrence of de novo mutations.11 Research in context Evidence before this study

In 2000, heterozygous germline mutations in the gene encoding the bone morphogenetic protein receptor type II (BMPR2) were identified as the main genetic cause of familial PAH.5, 6 Over 300 different BMPR2 mutations have been identified with a prevalence of greater than 75% in families with PAH.7, 8 BMPR-II is a receptor for the bone morphogenetic proteins (members of the transforming growth factor-β superfamily). Mutations in the BMPR2 gene cause loss-of-function and reduced signalling downstream of the receptor. Subsequently, BMPR2 mutations have been identified in apparently sporadic cases of idiopathic PAH with a frequency ranging from 11%9 to 40%.10 The occurrence of BMPR2 mutations in sporadic PAH cases in the absence of a family history can be accounted for by the relatively low penetrance of BMPR2 mutations (20–30%) and the occurrence of de novo mutations.11 Research in context Evidence before this study In the year 2000, mutations in the BMPR2 gene were identified as the major genetic cause of pulmonary arterial hypertension (PAH). Some small studies have examined the effect of BMPR2 mutations on the presentation, haemodynamic profile, and outcomes in patients with PAH. These studies suggested that those with BMPR2 mutations present at a younger age with more severe derangements of cardiopulmonary haemodynamics. Due to a lack of statistical power, lack of adjustment for important factors such as age and sex, and confounding from inclusion of prevalent cases without necessary adjustments in many of these studies, the effect of BMPR2 mutations on long-term outcomes has not been reliably established.

ardiopulmonary haemodynamics. Due to a lack of statistical power, lack of adjustment for important factors such as age and sex, and confounding from inclusion of prevalent cases without necessary adjustments in many of these studies, the effect of BMPR2 mutations on long-term outcomes has not been reliably established. Added value of this study By harmonising individual participant data from 1550 patients in eight published and unpublished studies, with updated follow-up, this study provides the most definitive assessment of the effect of BMPR2 mutations on the haemodynamic profile at diagnosis and long-term outcomes in patients with PAH. This study has shown that possession of a BMPR2 mutation is associated with an increased risk of death or transplantation and all-cause mortality. This association appears to be mediated by a more severe haemodynamic profile measured at diagnosis with the greatest proportion of the risk accounted for by the lower cardiac index in BMPR2 mutation carriers. There was a strong interaction between the effect of a BMPR2 mutation and age at diagnosis, such that the increased risk of death or transplantation and all-cause mortality associated with possession of a BMPR2 mutation was greater in younger patients. Implications of all the available evidence Patients with PAH with underlying BMPR2 mutations are younger at diagnosis, have more severe disease, and have a worse prognosis than patients without BMPR2 mutations. The role of routine genetic testing for BMPR2 mutations on the management of patients with PAH deserves further study.

Implications of all the available evidence Patients with PAH with underlying BMPR2 mutations are younger at diagnosis, have more severe disease, and have a worse prognosis than patients without BMPR2 mutations. The role of routine genetic testing for BMPR2 mutations on the management of patients with PAH deserves further study. Recent European guidelines for the management of PAH recommend offering genetic counselling and screening for BMPR2 mutations to patients diagnosed with idiopathic, heritable, and anorexigen-associated PAH, mainly to enable predictive genetic testing of relatives.12 Studies have suggested that patients with PAH carrying causal BMPR2 mutations present at an earlier age with more severe haemodynamic compromise.13, 14, 15, 16, 17 Although this might be expected to confer a worse survival, robust evidence describing the effect of BMPR2 mutations on long-term outcomes in these patients is lacking, primarily due to the limited power of individual studies and survival bias.18, 19 We established the BMPR2 Studies Collaboration to investigate the effect of BMPR2 mutations on clinical phenotypes and long-term outcomes in patients with PAH. This international consortium has allowed central collation and harmonisation of participant data on 1550 patients with PAH from eight cohorts based in six different countries.

he BMPR2 Studies Collaboration to investigate the effect of BMPR2 mutations on clinical phenotypes and long-term outcomes in patients with PAH. This international consortium has allowed central collation and harmonisation of participant data on 1550 patients with PAH from eight cohorts based in six different countries. Methods Data sources We sought individual participant data from studies identified through systematic searches of the published literature using MEDLINE and Embase (search terms “BMPR2” and “pulmonary hypertension”, up to March 18, 2014), searches of conference proceedings (restricted to the English language), and discussions with investigators. Cohort studies were eligible for inclusion if they met the following criteria: included patients with idiopathic PAH, heritable PAH, or anorexigen-associated PAH; sequenced patients for BMPR2 mutations; recorded baseline information on demographic and haemodynamic data at diagnosis, and for analysis of survival, recorded information on outcomes (death or transplantation). Inclusion of patients recruited to cohorts and registries since publication of original manuscripts was provided where available, including updated survival data. Details of contributing cohorts are presented in the appendix. Data from each study were obtained using a standardised spreadsheet; raw data were examined, and inconsistencies or irregularities were clarified with the relevant investigators.

nal manuscripts was provided where available, including updated survival data. Details of contributing cohorts are presented in the appendix. Data from each study were obtained using a standardised spreadsheet; raw data were examined, and inconsistencies or irregularities were clarified with the relevant investigators. Ethical approval was obtained for each of the individual studies included in this analysis and all participants provided informed written consent. This study followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses of Individual Participants Data (PRISMA-IPD) guidelines (checklist available in the appendix).20 Study participants All contributing studies used international criteria for the diagnosis of PAH.21 For the purpose of this study, the expert physician diagnosis of idiopathic PAH, heritable PAH, or anorexigen-associated PAH in a specialist centre was sufficient and data pertaining to the full range of investigations at diagnosis were not collected.

ting studies used international criteria for the diagnosis of PAH.21 For the purpose of this study, the expert physician diagnosis of idiopathic PAH, heritable PAH, or anorexigen-associated PAH in a specialist centre was sufficient and data pertaining to the full range of investigations at diagnosis were not collected. Vasodilator responsiveness was defined as a reduction in mean pulmonary arterial pressure of at least 10 mm Hg to a level below 40 mm Hg with no reduction in cardiac output after administration of inhaled nitric oxide, although some centres have historically used inhaled prostacyclin or intravenous prostacyclin, according to local practice. Treatment for PAH was prescribed consistent with prevailing international guidelines at the time of recruitment, at the discretion of the clinical team in each institution. Data regarding initial and subsequent PAH targeted therapy were not available for this analysis. Cohorts comprised a combination of incident patients, defined for the purposes of this study as those who were enrolled in their respective study and thus committed to genotyping within 6 months of PAH diagnosis, and prevalent patients who were enrolled more than 6 months after PAH diagnosis.

ailable for this analysis. Cohorts comprised a combination of incident patients, defined for the purposes of this study as those who were enrolled in their respective study and thus committed to genotyping within 6 months of PAH diagnosis, and prevalent patients who were enrolled more than 6 months after PAH diagnosis. Patients were excluded from the analysis if they had PAH associated with conditions such as connective tissue disease, HIV, congenital heart disease, or portal hypertension. Furthermore, to avoid potential confounding from mutations in other genes or undetected BMPR2 mutations, patients with a family history of PAH but with no identifiable BMPR2 mutation were also excluded. Patients with a history of anorexigen exposure were included since BMPR2 mutations have been recorded in these patients, and the disease is indistinguishable from idiopathic PAH.4, 22 Outcomes The primary outcome was the composite of death or lung transplantation. All-cause mortality was the secondary outcome. Patients contributed only the first outcome recorded during follow-up (ie, deaths preceded by transplantation were not included) because data regarding post-transplant survival were not available. Outcomes were censored if a patient was lost to follow-up or reached the end of the follow-up period. In analysis of all-cause mortality, patients were censored at the time of transplantation. Date of PAH diagnosis was defined as the date of diagnostic right heart catheterisation.

ransplant survival were not available. Outcomes were censored if a patient was lost to follow-up or reached the end of the follow-up period. In analysis of all-cause mortality, patients were censored at the time of transplantation. Date of PAH diagnosis was defined as the date of diagnostic right heart catheterisation. Statistical analysis Baseline characteristics of patients according to BMPR2 mutation status were compared using t test for continuous variables and χ2 test for categorical variables. Associations of BMPR2 mutation status with risk of death or transplantation and all-cause mortality recorded during follow-up were assessed using Cox proportional hazards regression models stratified by cohort and timing of study entry (ie, incident or prevalent). We used a one-stage stratified model rather than two-stage random effects model for our primary analysis because of the small number of participants in some studies. As a sensitivity analysis, we pooled data using a two-stage random effects model and assessed for heterogeneity between studies by calculating the I2 statistic and assessed statistical significance based on Cochran's Q test p value. Unless stated otherwise, hazard ratios (HRs) were adjusted for age at diagnosis and sex. The proportional hazards assumption was satisfied for the composite of death or transplantation and all-cause mortality. Survival curves comparing patients with and without BMPR2 mutations were calculated using unadjusted Kaplan-Meier estimates and compared using the log-rank test.

djusted for age at diagnosis and sex. The proportional hazards assumption was satisfied for the composite of death or transplantation and all-cause mortality. Survival curves comparing patients with and without BMPR2 mutations were calculated using unadjusted Kaplan-Meier estimates and compared using the log-rank test. To restrict the scope for potential bias due to inclusion of prevalent patients (ie, those diagnosed with PAH more than 6 months before study enrolment), Cox proportional hazards regression models and survival curves were fitted allowing for left truncation arising from the interval between diagnosis and enrolment. These patients were only included in the risk set from the time of study entry and were excluded if they entered the study more than 10 years after diagnosis. In patients for whom the date of enrolment in the study was not available, patients entered the risk set on the date they were genotyped for BMPR2 mutations. Given that worse survival has been reported in incident patients compared with prevalent patients in observational studies,18 and a higher risk of PAH-related death or admission to hospital was reported in incident patients in a clinical trial population,23 Cox models were also stratified by timing of study entry. Data pertaining to familial clustering of individuals and mutations were not available; however, to account for this, survival models were adjusted for clustering by sets of individuals with the same mutation from the same cohort. Two studies were not included in the survival analysis (due to insufficient survival data) but were included in the description of demographic and haemodynamic characteristics.

however, to account for this, survival models were adjusted for clustering by sets of individuals with the same mutation from the same cohort. Two studies were not included in the survival analysis (due to insufficient survival data) but were included in the description of demographic and haemodynamic characteristics. We explored interactions between the effect of BMPR2 mutation with age at diagnosis and sex within the one-stage stratified Cox models. The interaction with age at diagnosis was assessed with age as a continuous variable, with cases separated into three post-hoc groups according to age at diagnosis for illustrative purposes (<30 years, 30–50 years, and >50 years). To assess the degree to which the association of BMPR2 mutations with outcome was mediated by haemodynamic characteristics at diagnosis, we calculated the percentage of excess risk mediated (PERM) by three mediators thought likely to be in the causal pathway: pulmonary vascular resistance, cardiac index, and vasodilator responsiveness. Each of these mediators was added to the age-adjusted and sex-adjusted Cox proportional hazards models individually, and then all three simultaneously. The PERM, that is the degree to which the HR is attenuated by addition of the mediator in question, was calculated using the equation: (HRwithout mediator-HRwith mediatorHRwithout mediator-1)×100

To assess the degree to which the association of BMPR2 mutations with outcome was mediated by haemodynamic characteristics at diagnosis, we calculated the percentage of excess risk mediated (PERM) by three mediators thought likely to be in the causal pathway: pulmonary vascular resistance, cardiac index, and vasodilator responsiveness. Each of these mediators was added to the age-adjusted and sex-adjusted Cox proportional hazards models individually, and then all three simultaneously. The PERM, that is the degree to which the HR is attenuated by addition of the mediator in question, was calculated using the equation: (HRwithout mediator-HRwith mediatorHRwithout mediator-1)×100 For this analysis, missing covariate data were imputed using multiple imputation by chained equations in those individuals included in the survival analysis, to generate ten datasets with complete covariates. Cox proportional hazards models were fitted within each imputed dataset and combined using Rubin's rules. This analysis was repeated using only cases that had complete data for all three mediators as a sensitivity analysis. In an exploratory analysis, we compared haemodynamic parameters at presentation (compared using one-way ANOVA) and survival in patients by BMPR2 mutation type. Mutations were assigned to one of five categories (frameshift, stop-gained, splice site or intronic, large deletions or exonal duplications, missense).

For this analysis, missing covariate data were imputed using multiple imputation by chained equations in those individuals included in the survival analysis, to generate ten datasets with complete covariates. Cox proportional hazards models were fitted within each imputed dataset and combined using Rubin's rules. This analysis was repeated using only cases that had complete data for all three mediators as a sensitivity analysis. In an exploratory analysis, we compared haemodynamic parameters at presentation (compared using one-way ANOVA) and survival in patients by BMPR2 mutation type. Mutations were assigned to one of five categories (frameshift, stop-gained, splice site or intronic, large deletions or exonal duplications, missense). A two-sided p value less than 0·05 was considered statistically significant throughout. Statistical analyses were done using Stata (version 14; StataCorp, College Station, TX, USA). Role of the funding source No funding bodies had any role in the design, conduct, analysis of this study, or writing of the manuscript. The corresponding author had full access to the data and had the final responsibility for the decision to submit this manuscript with the permission of all coauthors.

A two-sided p value less than 0·05 was considered statistically significant throughout. Statistical analyses were done using Stata (version 14; StataCorp, College Station, TX, USA). Role of the funding source No funding bodies had any role in the design, conduct, analysis of this study, or writing of the manuscript. The corresponding author had full access to the data and had the final responsibility for the decision to submit this manuscript with the permission of all coauthors. Results Figure 1 shows the inclusion and exclusion of studies and patients. Of ten studies identified, one eligible cohort (that involved 61 patients and was available only in abstract form)24 did not contribute data to the current analysis, and one cohort was excluded because it exclusively included 47 patients younger than 16 years.25 We analysed data from a total of 1550 patients with idiopathic, heritable, and anoroxigen-associated PAH from eight cohorts (appendix).13, 14, 15, 16, 17, 26, 27, 28 The mean age at diagnosis was 40·1 (SD 17·2) years, 72% (1105/1545 [data for five patients were unavailable]) were women, 60% (931/1550) were from western Europe, 18% (276/1550) from North America, and 22% (343/1550) from east Asia. Overall, 448 (29%) of 1550 patients had an identified BMPR2 mutation and 86 (6%) of 1550 patients had a recorded history of anorexigen exposure. Histograms of the dates during which patients included in the survival analyses were diagnosed and recruited into these studies are shown in the appendix.

550) from east Asia. Overall, 448 (29%) of 1550 patients had an identified BMPR2 mutation and 86 (6%) of 1550 patients had a recorded history of anorexigen exposure. Histograms of the dates during which patients included in the survival analyses were diagnosed and recruited into these studies are shown in the appendix. The proportion of patients with BMPR2 mutations varied between studies (appendix). In patients with no recorded family history of PAH, a BMPR2 mutation was identified in 17% (200/1174), whereas in patients with a family history of PAH a mutation was identified in 82% (202/247). Patients with a BMPR2 mutation were younger at diagnosis (mean age 35·4 years vs 42·0 years, p<0·0001) and the proportion of patients with a BMPR2 mutation was greater in those diagnosed at a younger age (37% [162/434] in those aged <30 years, 33% [187/562] in those aged 30–50 years, and 17% [75/451] in those aged >50 years at diagnosis; p<0·0001). A comparison of haemodynamic and functional parameters measured at the time of diagnosis between carriers and non-carriers of a BMPR2 mutation is shown in table 1. Those carrying a BMPR2 mutation had a higher mean pulmonary artery pressure and pulmonary vascular resistance, and lower cardiac index. No difference was recorded in the severity of symptoms assessed by New York Heart Association functional class or exercise capacity assessed by 6 min walk distance. Patients with a BMPR2 mutation were less likely to respond to acute vasodilator testing (table 1).

pulmonary vascular resistance, and lower cardiac index. No difference was recorded in the severity of symptoms assessed by New York Heart Association functional class or exercise capacity assessed by 6 min walk distance. Patients with a BMPR2 mutation were less likely to respond to acute vasodilator testing (table 1). Characteristics of the 1164 patients from the six studies that contributed to the survival analysis are shown in the appendix. Survival curves by BMPR2 mutation status in the combined dataset are shown in figure 2. Of the 1164 patients, 723 (62%) were incident cases and 441 (38%) were prevalent cases. The median interval between diagnosis and study entry in the prevalent patients was 1·8 years (IQR 1·1–4·5). During 5870 person-years at risk (median duration of follow-up from diagnosis 5·4 years [IQR 3·0–8·2]), there were 354 deaths and 74 patients underwent lung transplantation. Age-adjusted and sex-adjusted HRs comparing BMPR2 mutation carriers with non-carriers were 1·42 (95% CI 1·15–1·75; p=0·0011) for the composite of death or lung transplantation and 1·27 (1·00–1·60; p=0·046) for all-cause mortality (table 2). HRs were unchanged after adjusting for previous exposure to anorexigens. Addition of each of the three mediators assessed to the age-adjusted and sex-adjusted models attenuated the HRs associated with BMPR2 mutation for both death or transplantation and all-cause mortality (table 2). Cardiac index mediated the greatest proportion of excess risk, accounting for 65% and 79% of the increased HR for death or transplantation and all-cause mortality, respectively. The PERM by the combination of pulmonary vascular resistance, cardiac index, and vasodilator responsiveness was 71% for death or transplantation and 100% for all-cause mortality. In the complete case sensitivity analysis (913 patients; appendix) the PERM by each mediator was similar.

all-cause mortality, respectively. The PERM by the combination of pulmonary vascular resistance, cardiac index, and vasodilator responsiveness was 71% for death or transplantation and 100% for all-cause mortality. In the complete case sensitivity analysis (913 patients; appendix) the PERM by each mediator was similar. HRs associated with possession of a BMPR2 mutation were similar in men and women (p=0·576 for death or transplantation, p=0·636 for all-cause mortality), but higher in patients with a younger age at diagnosis (p=0·0030 for death or transplantation, p=0·011 for all-cause mortality; figure 3, appendix). The interaction of BMPR2 and age at diagnosis persisted after adjustment for potential mediators (appendix). Similar results were recorded with meta-analysis using a two-stage approach with random effects (appendix). Between-study heterogeneity was modest, both for death or transplantation (I2=36·9% [95% CI 0–70]; p=0·16) and all-cause mortality (I2=20·1% [0–65]; p=0·28). There were no significant differences in haemodynamic parameters at diagnosis between those with different mutation subtypes (appendix). Patients with missense mutations were slightly younger at diagnosis. No significant difference was recorded in the risk of death or transplantation or all-cause mortality among carriers of different types of BMPR2 mutations (appendix).

c parameters at diagnosis between those with different mutation subtypes (appendix). Patients with missense mutations were slightly younger at diagnosis. No significant difference was recorded in the risk of death or transplantation or all-cause mortality among carriers of different types of BMPR2 mutations (appendix). Discussion To our knowledge, this meta-analysis of individual participant data from published and unpublished studies provides the most comprehensive standardised assessment of associations of BMPR2 mutations with long-term outcomes in patients with idiopathic, heritable, and anorexigen-associated PAH. We have shown that patients diagnosed with PAH have an increased risk of death or transplantation and all-cause mortality if they possess a mutation in the BMPR2 gene. HRs associated with possession of a BMPR2 mutation were similar in males and females, but greater with younger age at diagnosis. Furthermore, we have confirmed in this analysis the previously reported observations that patients with BMPR2 mutations present at a younger age, have more severe haemodynamic compromise at diagnosis with higher mean pulmonary artery pressure and pulmonary vascular resistance and lower cardiac index, and are less likely to respond to acute vasodilator testing and more likely to undergo lung transplantation.13, 14, 15, 16, 17, 26

tations present at a younger age, have more severe haemodynamic compromise at diagnosis with higher mean pulmonary artery pressure and pulmonary vascular resistance and lower cardiac index, and are less likely to respond to acute vasodilator testing and more likely to undergo lung transplantation.13, 14, 15, 16, 17, 26 The precise mechanisms underlying the difference in survival in those with a BMPR2 mutation remain unclear. We have found that after adjusting for pulmonary vascular resistance, cardiac index, and vasodilator responsiveness, HRs for death or lung transplantation and all-cause mortality were attenuated. The low number of BMPR2 mutation carriers having vasodilator responsiveness, a phenotype associated with a good prognosis when treated with calcium-channel blocker therapy,29, 30 is consistent with a predominance of extensive vascular remodelling rather than vasoconstriction. Given that the greatest percentage of the excess risk associated with a BMPR2 mutation was accounted for by reduced cardiac index at diagnosis, impaired right ventricular adaptation to increased afterload in those with BMPR2 mutations could be an important factor, as has been suggested in preclinical studies.31

iction. Given that the greatest percentage of the excess risk associated with a BMPR2 mutation was accounted for by reduced cardiac index at diagnosis, impaired right ventricular adaptation to increased afterload in those with BMPR2 mutations could be an important factor, as has been suggested in preclinical studies.31 Our finding of a greater proportion of BMPR2 mutations in younger age groups is consistent with the common observation that diseases with a major genetic contribution tend to present with an earlier onset. The strong interaction between BMPR2 mutation status and age at diagnosis is of great interest and has not been reported before. Patients carrying a BMPR2 mutation in which PAH manifests at a younger age might have a more severe mutation that not only results in more extensive pulmonary vascular remodelling or impaired right ventricular adaptation at diagnosis, but also results in more rapid progression of the disease process. This hypothesis is supported by the observation that the worse prognosis associated with a BMPR2 mutation in patients diagnosed before the age of 30 years in this study was not completely attenuated after adjustment for pulmonary vascular resistance, cardiac index, and vasoreactivity.

rogression of the disease process. This hypothesis is supported by the observation that the worse prognosis associated with a BMPR2 mutation in patients diagnosed before the age of 30 years in this study was not completely attenuated after adjustment for pulmonary vascular resistance, cardiac index, and vasoreactivity. Data from the UK PAH registry32 suggest that patients with PAH diagnosed after the age of 50 years are phenotypically distinct compared with younger patients. Older patients have a higher prevalence of cardiovascular comorbidities including systemic hypertension, atrial fibrillation, and diabetes. In the present study, although we did not collect data on comorbidities, we found that the proportion of patients with a BMPR2 mutation is lower in those diagnosed after the age of 50 years than in younger patients. Additionally, we show that after adjusting for age and sex, mutations do not affect survival in these older patients, and prognosis might even be better in those with mutations. BMPR2 mutations present in those who do not develop PAH until later in life might be less deleterious. Alternatively, in the older age group, comorbidities might outweigh the effect of BMPR2 mutations on survival.

o not affect survival in these older patients, and prognosis might even be better in those with mutations. BMPR2 mutations present in those who do not develop PAH until later in life might be less deleterious. Alternatively, in the older age group, comorbidities might outweigh the effect of BMPR2 mutations on survival. Our study confirms the relatively high frequency of BMPR2 mutations in idiopathic and heritable PAH, and supports a central role for the BMPR2 pathway in the initiation of this disease. Moreover, the effect of BMPR2 mutation on survival suggests a role for BMPR-II dysfunction in the clinical progression of the disease. Both of these observations support further investigations into the potential targeting of the BMPR-II pathway for therapeutic intervention in PAH.33, 34 The main reason to test for the presence or absence of a BMPR2 mutation in a patient with PAH is to guide predictive genetic testing in unaffected relatives. Although our findings show that BMPR2 mutations are associated with a worse survival, the usefulness of this result for prognostic purposes might be restricted in the clinic, since the majority of this risk appears to be accounted for by the known haemodynamic predictors of mortality measured during the diagnostic assessment during right heart catheterisation. Despite this, in younger patients, in which the increased risk appears to persist after adjustment for these factors, albeit only in subgroup analyses, screening for mutations might add value, and this warrants further investigation.

ortality measured during the diagnostic assessment during right heart catheterisation. Despite this, in younger patients, in which the increased risk appears to persist after adjustment for these factors, albeit only in subgroup analyses, screening for mutations might add value, and this warrants further investigation. Our analysis has major strengths. We had access to data for more than 95% of participants from eligible cohorts. We analysed individual participant data to avoid limitations of literature-based reviews. We had information on both all-cause mortality and death or transplantation. We studied clinically relevant subpopulations (such as by age and sex) reliably, exploiting the study's considerable statistical power. We avoided potential over-adjustment in the primary analysis by not adjusting for variables (eg, pulmonary vascular resistance, cardiac index, and vasoreactivity) that could mediate associations between BMPR2 and death or transplantation and all-cause mortality. We ensured generalisability by studying cohorts located across east Asia, Europe, and North America.

primary analysis by not adjusting for variables (eg, pulmonary vascular resistance, cardiac index, and vasoreactivity) that could mediate associations between BMPR2 and death or transplantation and all-cause mortality. We ensured generalisability by studying cohorts located across east Asia, Europe, and North America. Our analysis has some limitations. Studies included differed in their methods of recruitment and data collection, and in the proportion of familial cases and individuals with BMPR2 mutations, which might explain the heterogeneity recorded between studies. Nevertheless, we obtained similar results to those in our primary analysis based on a stratified Cox proportional hazards model when we used a two-stage random effects meta-analysis model in sensitivity analysis. Additionally, given the evidence for interaction recorded between mutation status and age, the differences in age at diagnosis in different studies could partly explain the heterogeneity recorded in two-stage meta-analyses. The inclusion of prevalent patients in survival analyses can introduce bias; however, we addressed this in the Cox proportional hazards model by allowing for left truncation arising from the interval between diagnosis and study entry and also stratifying by timing of enrolment. Additionally, we observed no interaction between BMPR2 mutation status and timing of enrolment. Finally, the lack of data regarding the timing and use of PAH-directed therapies might introduce some bias, although we believe any effect is likely to be very small. Indeed, if patients with BMPR2 mutations were treated more aggressively due to their more severe haemodynamic derangements at diagnosis, this could have resulted in an attenuation of the association we have recorded.

irected therapies might introduce some bias, although we believe any effect is likely to be very small. Indeed, if patients with BMPR2 mutations were treated more aggressively due to their more severe haemodynamic derangements at diagnosis, this could have resulted in an attenuation of the association we have recorded. By harnessing data from observational studies done worldwide, we have shown that in patients with idiopathic, familial, and anorexigen-associated PAH, the presence of a mutation in the BMPR2 gene is associated with an increased risk of death or lung transplantation and all-cause mortality, particularly in those diagnosed at a younger age. Our analysis suggests that this association is largely mediated by the more severe haemodynamic derangements and low frequency of vasodilator responsiveness at diagnosis seen in those with BMPR2 mutations. Supplementary Material Supplementary appendix

By harnessing data from observational studies done worldwide, we have shown that in patients with idiopathic, familial, and anorexigen-associated PAH, the presence of a mutation in the BMPR2 gene is associated with an increased risk of death or lung transplantation and all-cause mortality, particularly in those diagnosed at a younger age. Our analysis suggests that this association is largely mediated by the more severe haemodynamic derangements and low frequency of vasodilator responsiveness at diagnosis seen in those with BMPR2 mutations. Supplementary Material Supplementary appendix Acknowledgments The authors would like to acknowledge Laurence Rottat (APHP, Centre de Référence de l'Hypertension Pulmonaire Sévère, Service de Pneumologie, Hôpital de Bicêtre, Le Kremlin Bicêtre, France), Chiara Barozzi and Luciana Tomasi (University of Bologna, Bologna, Italy), Hiroki Kabata (Keio University Hospital, Tokyo, Japan), Melanie Eyries (Université Pierre et Marie Curie-Paris, Paris, France), Robyn Barst and Jane Morse (Columbia University, New York, NY, USA). We acknowledge funding from Cambridge NIHR Biomedical Research Centre, Medical Research Council (MR/K020919/1), British Heart Foundation (CH/09/001/25945), Assistance Publique-Hôpitaux de Paris, INSERM, Université Paris-Sud, National Natural Science Foundation of China (81320108005), Beijing Natural Science Foundation (7141009), Intermountain Research and Medical Foundation (1007044), National Center for Advancing Translational Sciences (UL1TR000445), and National Institutes of Health (R01 HL060056, P01 HL108800, and K23 HL098743). The contents of this paper are solely the responsibility of the authors and do not necessarily represent official views of the National Center for Advancing Translational Sciences or the National Institutes of Health.

ences (UL1TR000445), and National Institutes of Health (R01 HL060056, P01 HL108800, and K23 HL098743). The contents of this paper are solely the responsibility of the authors and do not necessarily represent official views of the National Center for Advancing Translational Sciences or the National Institutes of Health. Contributors JDWE and NWM conceived the study. JDWE, EDAn, MH, and NWM designed the study. BG, DM, X-JW, NG, EDAu, GE, KA, EG, YY, Z-CJ, AM, MP, LAW, IN, TS, CE, KH, MW, EBR, WKC, FS, GS, OS, and MH collected the data. JDWE and NWM coordinated the study. JDWE, SG, and SK analysed the data. All authors interpreted the results. JDWE, EDAn, MH, and NWM drafted the manuscript with critical revisions for important intellectual content from all authors. All authors approved the final version.

BR, WKC, FS, GS, OS, and MH collected the data. JDWE and NWM coordinated the study. JDWE, SG, and SK analysed the data. All authors interpreted the results. JDWE, EDAn, MH, and NWM drafted the manuscript with critical revisions for important intellectual content from all authors. All authors approved the final version. Declaration of interests BG reports personal fees from Actelion, GlaxoSmithKline, Bayer, and Pfizer, outside of the submitted work. DM reports grants and personal fees from Actelion and Bayer, and personal fees from GlaxoSmithKline, Pfizer, Novartis, and BMS, outside of the submitted work. GE reports grants from NIH, during the conduct of the study; personal fees from Bellerophon, Actelion, and Bayer, and grants from Actelion, Gilead, and United Therapeutics, outside of the submitted work. GS reports grants and personal fees from Actelion, GlaxoSmithKline, and Bayer, and personal fees from Pfizer, outside of the submitted work. OS reports grants, personal fees, and non-financial support from Actelion, GlaxoSmithKline, and Bayer, grants and personal fees from Pfizer, and personal fees and non-financial support from United Therapeutics, outside of the submitted work. EDAn reports grants from The British Heart Foundation, Medical Research Council, NHS Blood and Transplant, National Institute for Health Research, and European Union, and personal fees from Elsevier (France), outside of the submitted work. MH reports personal fees and non-financial support from Actelion and Pfizer, and grants, personal fees, and non-financial support from Bayer and GlaxoSmithKline, outside of the submitted work. JDWE, X-JW, NG, EDAu, KA, EG, YY, Z-CJ, AM, MP, LAW, IN, TS, CE, KH, MW, EBR, WKC, FS, SG, SK, and NWM declare no competing interests.

rts personal fees and non-financial support from Actelion and Pfizer, and grants, personal fees, and non-financial support from Bayer and GlaxoSmithKline, outside of the submitted work. JDWE, X-JW, NG, EDAu, KA, EG, YY, Z-CJ, AM, MP, LAW, IN, TS, CE, KH, MW, EBR, WKC, FS, SG, SK, and NWM declare no competing interests. Figure 1 Study and patient selection Figure 2 Kaplan-Meier survival curves according to BMPR2 mutation status (A) Transplant-free survival, all patients (p=0·0016). (B) Overall survival, all patients (p=0·32). (C) Transplant-free survival, younger than 50 years at diagnosis (p<0·0001). (D) Overall survival, younger than 50 years at diagnosis (p=0·0032). (E) Transplant-free survival, older than 50 years at diagnosis (p=0·27). (F) Overall survival, 50 years or older at diagnosis (p=0·16). Survival curves are not adjusted for age at diagnosis or sex and are not stratified by study cohort. Figure 3 Hazard ratios (HRs) for the effect of a BMPR2 mutation on death or transplantation and all-cause mortality by age at diagnosis and sex p value for interaction of BMPR2 and age at diagnosis calculated with age at diagnosis as a continuous variable. Table 1 Demographics and clinical measurements at diagnosis

(A) Transplant-free survival, all patients (p=0·0016). (B) Overall survival, all patients (p=0·32). (C) Transplant-free survival, younger than 50 years at diagnosis (p<0·0001). (D) Overall survival, younger than 50 years at diagnosis (p=0·0032). (E) Transplant-free survival, older than 50 years at diagnosis (p=0·27). (F) Overall survival, 50 years or older at diagnosis (p=0·16). Survival curves are not adjusted for age at diagnosis or sex and are not stratified by study cohort. Figure 3 Hazard ratios (HRs) for the effect of a BMPR2 mutation on death or transplantation and all-cause mortality by age at diagnosis and sex p value for interaction of BMPR2 and age at diagnosis calculated with age at diagnosis as a continuous variable. Table 1 Demographics and clinical measurements at diagnosis All patients BMPR2 mutation status Non-carriers (N=1102) Carriers (N=448) p value Age at diagnosis (N=1447), years 40·1 (17·2) 42·0 (17·8) 35·4 (14·8) <0·0001 Male sex 440/1545 (28%) 302/1097 (28%) 138/448 (31%) 0·20 Family history of PAH 202/1376 (15%) .. 202/402 (50%) .. Body-mass index (N=1206), kg/m2 24·9 (9·1) 24·9 (10·6) 24·9 (5·9) 0·99 6-min walk distance (N=1072), m 378 (124) 374 (128) 388 (113) 0·088 NYHA functional class 0·38 I–II 423/1426 (30%) 313/1031 (30%) 110/394 (28%) III 896/1426 (63%) 647/1031 (63%) 249/394 (63%) IV 107/1426 (8%) 72/1031 (7%) 35/394 (9%) Mean pulmonary artery pressure (N=1503), mm Hg 57·6 (15·0) 56·4 (15·3) 60·5 (13·8) <0·0001 Pulmonary vascular resistance (N=1300), Wood units 14·0 (8·4) 12·9 (8·3) 16·6 (8·3) <0·0001 Right atrial pressure (N=1253), mm Hg 8·2 (5·5) 8·0 (5·7) 8·6 (5·2) 0·065 Cardiac output (N=1202), L/min 3·98 (1·44) 4·20 (1·50) 3·50 (1·17) <0·0001 Cardiac index (N=1358), L/min per m2 2·40 (0·88) 2·51 (0·92) 2·11 (0·69) <0·0001 Vasodilator responder 157/1287 (12%) 147/907 (16%) 10/380 (3%) <0·0001 Data are n/N (%) or mean (SD), unless otherwise stated. PAH=pulmonary arterial hypertension. NYHA=New York Heart Association.

·98 (1·44) 4·20 (1·50) 3·50 (1·17) <0·0001 Cardiac index (N=1358), L/min per m2 2·40 (0·88) 2·51 (0·92) 2·11 (0·69) <0·0001 Vasodilator responder 157/1287 (12%) 147/907 (16%) 10/380 (3%) <0·0001 Data are n/N (%) or mean (SD), unless otherwise stated. PAH=pulmonary arterial hypertension. NYHA=New York Heart Association. Table 2 Proportion of excess risk mediated by haemodynamic variables at diagnosis

·98 (1·44) 4·20 (1·50) 3·50 (1·17) <0·0001 Cardiac index (N=1358), L/min per m2 2·40 (0·88) 2·51 (0·92) 2·11 (0·69) <0·0001 Vasodilator responder 157/1287 (12%) 147/907 (16%) 10/380 (3%) <0·0001 Data are n/N (%) or mean (SD), unless otherwise stated. PAH=pulmonary arterial hypertension. NYHA=New York Heart Association. Table 2 Proportion of excess risk mediated by haemodynamic variables at diagnosis HR (95% CI) for BMPR2 mutation p value PERM Death or transplantation Pulmonary vascular resistance Adjusted for age and sex 1·42 (1·15–1·75) 0·0011 Adjusted for age, sex, and pulmonary vascular resistance 1·28 (1·03–1·58) 0·024 34% Cardiac index Adjusted for age and sex 1·42 (1·15–1·75) 0·0011 Adjusted for age, sex, and cardiac index 1·18 (0·95–1·47) 0·14 65% Vasoreactivity Adjusted for age and sex 1·42 (1·15–1·75) 0·0011 Adjusted for age, sex, and vasoreactivity 1·26 (1·02–1·57) 0·036 37% Combined model Adjusted for age and sex 1·42 (1·15–1·75) 0·0011 Adjusted for age, sex, pulmonary vascular resistance, cardiac index, and vasoreactivity 1·12 (0·89–1·41) 0·33 71% All-cause mortality Pulmonary vascular resistance Adjusted for age and sex 1·27 (1·00–1·60) 0·046 Adjusted for age, sex, and pulmonary vascular resistance 1·13 (0·89–1·43) 0·33 53% Cardiac index Adjusted for age and sex 1·27 (1·00–1·60) 0·046 Adjusted for age, sex, and cardiac index 1·06 (0·83–1·35) 0·67 79% Vasoreactivity Adjusted for age and sex 1·27 (1·00–1·60) 0·046 Adjusted for age, sex, and vasoreactivity 1·14 (0·89–1·45) 0·29 49% Combined model Adjusted for age and sex 1·27 (1·00–1·60) 0·046 Adjusted for age, sex, pulmonary vascular resistance, cardiac index, and vasoreactivity 1·00 (0·77–1·29) 0·98 100% Hazard ratios (HRs) associated with possession of a BMPR2 mutation after addition of each mediator individually to age-adjusted and sex-adjusted Cox proportional hazards models with the percentage of excess risk mediated (PERM) by each mediator. Missing data for mediators generated by multiple imputation.

Introduction Multidrug-resistant tuberculosis—present in 3–4% of new tuberculosis cases and 20% of previously treated cases worldwide (with much higher prevalence in some countries)—causes 190 000 deaths each year and is a major challenge to clinicians and policy makers.1 Fewer than half of all notified cases with underlying multidrug resistance are identified as such, and with the scale-up of Xpert MTB/RIF, many patients diagnosed with rifampin resistance have no access to appropriate treatment. In individuals appropriately treated for multidrug-resistant tuberculosis, conventional, 20–24 month regimens (subsequently referred to as longer therapy) have a success rate of only 50% worldwide2 because of factors such as low drug effectiveness,2, 3 lengthy and toxic regimens that are difficult to complete,4 and high rates of prevalent5 and acquired resistance6 to second-line drugs. Treatment of multidrug-resistant tuberculosis is also resource intensive, costing thousands of US dollars per patient7 and consuming up to half of tuberculosis control budgets in high-burden countries.1

regimens that are difficult to complete,4 and high rates of prevalent5 and acquired resistance6 to second-line drugs. Treatment of multidrug-resistant tuberculosis is also resource intensive, costing thousands of US dollars per patient7 and consuming up to half of tuberculosis control budgets in high-burden countries.1 A potential solution to these challenges is the use of a shorter, cheaper, more effective, and more tolerable new regimen to expand treatment capacity and improve treatment success. In May, 2016, WHO made a conditional recommendation for a new short-course regimen that can treat most patients with multidrug-resistant tuberculosis in 9–12 months.8 This regimen consists of an initial 4–6 month phase of seven drugs including a second-line injectable, followed by a 5 month continuation of four of the oral drugs including pyrazinamide and a fluoroquinolone. It costs less than US$1000 per patient and has shown promising effectiveness, with more than 80% of patients cured in initial observational cohorts.9, 10, 11, 12 WHO now recommends this short-course regimen for patients with multidrug-resistant pulmonary tuberculosis without confirmed or probable resistance to key drugs in the regimen, while acknowledging the low capacity to test for such resistance in many settings.13 Research in context Evidence before this study

A potential solution to these challenges is the use of a shorter, cheaper, more effective, and more tolerable new regimen to expand treatment capacity and improve treatment success. In May, 2016, WHO made a conditional recommendation for a new short-course regimen that can treat most patients with multidrug-resistant tuberculosis in 9–12 months.8 This regimen consists of an initial 4–6 month phase of seven drugs including a second-line injectable, followed by a 5 month continuation of four of the oral drugs including pyrazinamide and a fluoroquinolone. It costs less than US$1000 per patient and has shown promising effectiveness, with more than 80% of patients cured in initial observational cohorts.9, 10, 11, 12 WHO now recommends this short-course regimen for patients with multidrug-resistant pulmonary tuberculosis without confirmed or probable resistance to key drugs in the regimen, while acknowledging the low capacity to test for such resistance in many settings.13 Research in context Evidence before this study Multidrug-resistant tuberculosis has a tremendous toll on patients who have to endure nearly 2 years of treatment, while exerting pressure on the budgets of tuberculosis control programmes and posing a major barrier to tuberculosis elimination worldwide. In May, 2016, WHO recommended a short-course regimen on the basis of promising individual-level effectiveness in several observational studies; however, to the best of our knowledge, the population-level implications of this recommendation have not been assessed. Added value of this study

Multidrug-resistant tuberculosis has a tremendous toll on patients who have to endure nearly 2 years of treatment, while exerting pressure on the budgets of tuberculosis control programmes and posing a major barrier to tuberculosis elimination worldwide. In May, 2016, WHO recommended a short-course regimen on the basis of promising individual-level effectiveness in several observational studies; however, to the best of our knowledge, the population-level implications of this recommendation have not been assessed. Added value of this study In this study, we estimated the epidemiological benefit of adopting the newly endorsed short-course regimen for multidrug-resistant tuberculosis. We also explored the extent to which the anticipated effect depends on characteristics of the regimen that remain to be determined, such as treatment success under programmatic conditions, durability of effectiveness, exclusions on the basis of additional drug resistance, treatment outcomes after such exclusions, and the extent to which cost savings from the new regimen can be used to expand treatment access. We provided a numerical estimate of the potential population-level effect of the short-course regimen in a representative setting—a 23% reduction in incidence after 8 years—and explored factors that modify this projection under different conditions. Under some reasonable sets of assumptions (eg, lower effectiveness of the short-course regimen than that suggested in initial observational studies or a higher prevalence of resistance to second-line drugs), the new regimen was projected to result in minimal, or even negative, effects on the incidence of multidrug-resistant tuberculosis.

le sets of assumptions (eg, lower effectiveness of the short-course regimen than that suggested in initial observational studies or a higher prevalence of resistance to second-line drugs), the new regimen was projected to result in minimal, or even negative, effects on the incidence of multidrug-resistant tuberculosis. Implications of all the available evidence The new short-course regimen can potentially have an important role in the control of multidrug-resistant tuberculosis. However, this effect needs to be balanced against uncertainties related to long-term effectiveness and the importance of additional drug resistance. To optimise the effect of this new regimen, early-adopter countries should simultaneously expand diagnosis and treatment of multidrug-resistant tuberculosis and closely monitor treatment outcomes in both patients receiving the regimen and those ineligible because of additional drug resistance. An important, positive population-level effect of introducing this regimen is realistic but cannot be assumed without further evidence on the role of resistance to second-line drugs and long-term efficacy data from ongoing clinical trials.

eceiving the regimen and those ineligible because of additional drug resistance. An important, positive population-level effect of introducing this regimen is realistic but cannot be assumed without further evidence on the role of resistance to second-line drugs and long-term efficacy data from ongoing clinical trials. However, unknowns about this new short-course regimen exist. First, although this regimen seemed highly effective in observational cohorts,8 the first rigorous comparison of its efficacy with that of longer therapy will not be completed until 2018.14 Second, use of this regimen necessitates testing for susceptibility to additional drugs (fluoroquinolones and second-line injectables, resistance to which is common in some populations with multidrug-resistant tuberculosis15). Moreover, great uncertainty exists regarding treatment outcomes in patients with resistance to other components of the regimen8 (particularly pyrazinamide, to which half or more multidrug-resistant strains are resistant5), raising concerns about the regimen's effectiveness and usefulness in geographical settings with more extensive resistance than the settings where the regimen was developed and first tested.16 Furthermore, whether tuberculosis programmes and health systems can truly use resources freed by a shorter regimen to expand treatment access remains uncertain. Finally, the effect of this regimen might differ substantially from one epidemiological setting to another.

where the regimen was developed and first tested.16 Furthermore, whether tuberculosis programmes and health systems can truly use resources freed by a shorter regimen to expand treatment access remains uncertain. Finally, the effect of this regimen might differ substantially from one epidemiological setting to another. In light of these uncertainties, we aimed to use a dynamic transmission model to investigate the potential effect of this new short-course regimen and to project outcomes under different assumptions regarding regimen effectiveness, treatment access, treatment outcomes in patients with additional drug resistance, and underlying epidemiology of multidrug-resistant tuberculosis.

transmission model to investigate the potential effect of this new short-course regimen and to project outcomes under different assumptions regarding regimen effectiveness, treatment access, treatment outcomes in patients with additional drug resistance, and underlying epidemiology of multidrug-resistant tuberculosis. Methods Model overview In this population modelling analysis, we developed a compartmental transmission model of a multidrug-resistant tuberculosis epidemic, similar to previous tuberculosis models,17, 18 with explicit representation of diagnosis and treatment of multidrug resistance (figure 1; see appendix for description of the full model). In brief, both drug-susceptible and multidrug-resistant strains circulate in a population, with multidrug resistance emerging during treatment of drug-susceptible disease19 and subsequently also spreading through person-to-person transmission. Active tuberculosis, once symptomatic, is identified and treated at a given rate, but only a proportion of patients are tested for multidrug resistance and treated accordingly. Treatment is either apparently effective (ie, symptoms and infectiousness resolve, followed by lasting cure or by temporary resolution with subsequent relapse to active disease) or ineffective (ie, associated with ongoing tuberculosis mortality risk and infectiousness). Longer therapy was modelled as lasting a median of 20 months and representing a full attempt at treatment, including any changes made to the initial regimen based on clinical response or drug susceptibility testing results; outcomes were based on results in observational cohorts.3 We assumed that those who do not respond to a full treatment attempt for multidrug-resistant tuberculosis remain infectious until either death or spontaneous resolution.

de to the initial regimen based on clinical response or drug susceptibility testing results; outcomes were based on results in observational cohorts.3 We assumed that those who do not respond to a full treatment attempt for multidrug-resistant tuberculosis remain infectious until either death or spontaneous resolution. Calibration To explore a large and representative number of scenarios consistent with these data, we considered 2 million sets of model parameters drawn from distributions based on the literature (table 1; appendix pp 7–8). We used log-normal distributions for continuous measures bounded from 0 to infinity, logit-normal distributions for continuous measures bounded from 0 to 1, and uniform distributions when data to suggest a most likely value were missing or sparse. In the primary analysis, we calibrated the model to a setting characterised by WHO estimates of incidence, prevalence, and mortality of tuberculosis, as well as prevalence of multidrug resistance in new and retreatment tuberculosis notifications, in people aged 15 years and older for the WHO southeast Asian region—ie, Bangladesh, Bhutan, North Korea, India, Indonesia, Maldives, Myanmar, Nepal, Sri Lanka, Thailand, and Timor-Leste—in 2014 (table 2; appendix pp 10–11).1

well as prevalence of multidrug resistance in new and retreatment tuberculosis notifications, in people aged 15 years and older for the WHO southeast Asian region—ie, Bangladesh, Bhutan, North Korea, India, Indonesia, Maldives, Myanmar, Nepal, Sri Lanka, Thailand, and Timor-Leste—in 2014 (table 2; appendix pp 10–11).1 To model the expansion of diagnosis and treatment of multidrug-resistant tuberculosis in the past decade, we linearly increased the proportions of patients who are identified as having drug resistance (eg, by Xpert MTB/RIF) and considered for treatment over time, from zero in 2004 to reported levels (3·8% of new tuberculosis cases and 67% of retreatment cases) in 2014. For the primary analysis, we assumed that, in absence of a short-course regimen, the probability of receiving multidrug-resistant tuberculosis treatment would subsequently remain constant (reflecting a relatively fixed treatment budget), whereas the short-course regimen allows expansion of case detection and treatment, reflecting the lower cost and resource requirement of the new regimen. Modelling of short-course regimen We modelled the introduction of a short-course regimen as an instantaneous switch from the conventional, longer therapy in 2016 for patients who are diagnosed with multidrug-resistant tuberculosis and not found to have additional drug resistance that makes them ineligible. This scenario reflects a simulated policy change with rapid restructuring of the treatment programme for multidrug-resistant tuberculosis.

onventional, longer therapy in 2016 for patients who are diagnosed with multidrug-resistant tuberculosis and not found to have additional drug resistance that makes them ineligible. This scenario reflects a simulated policy change with rapid restructuring of the treatment programme for multidrug-resistant tuberculosis. To estimate the number of additional patients who could be treated in a budget-neutral introduction of the new regimen, we compared the costs of drugs and clinical care for each regimen. Drug costs for the short-course regimen are less than half of those of longer therapy, and the shortened durations of the intensive phase and the overall treatment course also reduce other associated health-care costs,8 whereas added costs of second-line drug susceptibility testing are small relative to the total cost of treatment.26 For simplicity, we assumed in the primary analysis that introduction of the short-course regimen would allow twice as many patients to be treated on the same multidrug-resistant tuberculosis treatment budget. We implemented this doubling by expanding the number of patients with multidrug-resistant tuberculosis offered treatment, first to previously treated patients and then to new patients.

he short-course regimen would allow twice as many patients to be treated on the same multidrug-resistant tuberculosis treatment budget. We implemented this doubling by expanding the number of patients with multidrug-resistant tuberculosis offered treatment, first to previously treated patients and then to new patients. In the primary analysis, we modelled a scenario in which roughly 10% of patients with multidrug-resistant tuberculosis have additional drug resistance (or suspected resistance) that disqualifies them from the short-course regimen, leading to very poor outcomes. We used a median duration of 10 months for the short-course regimen and 20 months for longer therapy,9 and assumed that loss to follow-up is reduced by half with the short-course regimen. Treatment success for people remaining in treatment was set at 92·5% for the short-course regimen9 and 66–85% for longer therapy;3 these percentages include only those who are not lost to follow-up and are therefore higher than reported figures that do not distinguish between loss to follow-up and other adverse outcomes. Relapse risk after successful treatment was set at 1% for the short-course regimen and 1–10% for longer therapy. The estimated outcomes of the short-course regimen were based on results from an observational cohort study in Bangladesh;9 similar results were obtained elsewhere.8, 12 The estimated 10% ineligibility for the short-course regimen is based on the assumptions that patients would be screened for second-line drug resistance with a line probe assay (of imperfect sensitivity),27 moxifloxacin resistance would be similar to levels observed in Pakistan and Bangladesh,5 and monoresistance to second-line injectables would be rare.28, 29 We also assumed, conservatively, that patients found to have such disqualifying additional drug resistance would have very poor outcomes, comparable to those reported for extensively drug-resistant tuberculosis1 and to tuberculosis outcomes in the pre-antibiotic era30 (ie, that half of these patients will ultimately die of tuberculosis, although such deaths might occur well after treatment is completed).

nal drug resistance would have very poor outcomes, comparable to those reported for extensively drug-resistant tuberculosis1 and to tuberculosis outcomes in the pre-antibiotic era30 (ie, that half of these patients will ultimately die of tuberculosis, although such deaths might occur well after treatment is completed). We explored several alternative scenarios to the above assumptions (table 3). Alternatives involving inter-related aspects of prevalence, diagnosis, and associated treatment outcomes of second-line drug resistance were explored combinatorially (table 4). The primary outcome for each scenario was the percentage reduction in multidrug-resistant tuberculosis incidence in 2024, compared with projections under continued use of longer therapy. Results are reported as the median simulated value and corresponding 95% uncertainty range (UR), reflecting the 2·5th to 97·5th percentile of data-consistent simulations. Sensitivity analyses We assessed the sensitivity of the primary results to the value of all underlying model parameters. We also assessed the sensitivity of our results to assumptions about ongoing scale-up of drug susceptibility testing even in the absence of the short-course regimen, to underlying dynamics of acquisition, transmission, and reactivation of multidrug-resistant tuberculosis, and to alternative epidemiological scenarios reflecting a range of tuberculosis incidence and multidrug-resistant tuberculosis prevalence (appendix).

bility testing even in the absence of the short-course regimen, to underlying dynamics of acquisition, transmission, and reactivation of multidrug-resistant tuberculosis, and to alternative epidemiological scenarios reflecting a range of tuberculosis incidence and multidrug-resistant tuberculosis prevalence (appendix). Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results Our model generated 11 289 data-consistent simulations, which fitted well with our epidemiological calibration targets (table 2). Posterior distributions of model parameters favoured lower rates of acquisition and transmission of multidrug-resistant tuberculosis (reflecting that multidrug resistance is present in only 2% of new tuberculosis notifications, despite decades of treatment with isoniazid and rifampin) but otherwise suggested no strong support for specific parameter values within the ranges of the specified prior distributions (appendix p 13).

-resistant tuberculosis (reflecting that multidrug resistance is present in only 2% of new tuberculosis notifications, despite decades of treatment with isoniazid and rifampin) but otherwise suggested no strong support for specific parameter values within the ranges of the specified prior distributions (appendix p 13). Assuming that current practices continue, we projected that the incidence of multidrug-resistant tuberculosis would decrease by a median of 14% (95% UR −36 to 39) from 4·9 [95% UR 4·2–5·9] per 100 000 population in 2014 to 4·3 [2·9–7·6] per 100 000 population in 2024 (figure 2A), reflecting higher levels of treatment than in the past. However, the large 95% UR reflects the paucity of longitudinal data. Despite this uncertainty in the overall trajectory of multidrug-resistant tuberculosis incidence, the short-course regimen was consistently projected to have benefit under the assumptions of the primary scenario. We projected that, 8 years after introduction of the short-course regimen, the incidence of multidrug-resistant tuberculosis would be 3·3 (2·2–5·6) per 100 000 population—ie, the incidence in 2024 would be 23% (10–38) lower with the short-course regimen (figure 2B). A slightly larger reduction in multidrug-resistant tuberculosis mortality (31%, 14–46) was projected than for incidence.

egimen, the incidence of multidrug-resistant tuberculosis would be 3·3 (2·2–5·6) per 100 000 population—ie, the incidence in 2024 would be 23% (10–38) lower with the short-course regimen (figure 2B). A slightly larger reduction in multidrug-resistant tuberculosis mortality (31%, 14–46) was projected than for incidence. The magnitude of the short-course regimen's effect on the incidence of multidrug-resistant tuberculosis was dependent on several key assumptions (Figure 3, Figure 4). If the short-course regimen only improved outcomes in patients treated but did not facilitate treatment access (alternative scenario 1), it was projected to reduce incidence by only 14% (95% UR 4–28). Similarly, if the short-course regimen's benefit was restricted to expansion of treatment access alone and did not change the average treatment outcome (alternative scenario 2), then the incidence in 2024 was projected to fall by 11% (3–24). Furthermore, if we assumed that a finding of equivalent efficacy between the two regimens was dependent on the short-course regimen only being used in those without additional resistance—while those excluded from the short-course regimen had very poor outcomes (alternative scenario 3)—then the short-course regimen could have a minimal effect on the multidrug-resistant tuberculosis epidemic as a whole (relative change in incidence −3%, −16 to 9), despite doubling the number of people treated. Similarly pessimistic projections were seen when the prevalence of disqualifying drug resistance was increased to 30% (alternative scenario 5). Figure 4 shows the projected effects of the short-course regimen in a given setting as a function of three measurable parameters: treatment outcomes in those who take the short-course regimen; treatment outcomes in those excluded from the regimen (and instead given longer therapy); and the proportion of the population excluded from the short-course regimen.

the short-course regimen in a given setting as a function of three measurable parameters: treatment outcomes in those who take the short-course regimen; treatment outcomes in those excluded from the regimen (and instead given longer therapy); and the proportion of the population excluded from the short-course regimen. In sensitivity analyses, the relative effect of the short-course regimen did not depend substantially on the degree of future scale-up of drug susceptibility testing (appendix p 14). Other variables that strongly influenced the effect of the short-course regimen included the long-term efficacy of longer therapy and assumptions about the duration and trajectory of the multidrug-resistant tuberculosis epidemic (appendix pp 15–16). The regimen's effect showed little sensitivity to the balance of acquired versus transmitted multidrug resistance and was only moderately sensitive to the balance of recent versus remote transmission (appendix p 17). Similarly, the short-course regimen had a greater potential effect in high-prevalence settings; results were otherwise similar across a range of simulated epidemiological settings (appendix p 18).

multidrug resistance and was only moderately sensitive to the balance of recent versus remote transmission (appendix p 17). Similarly, the short-course regimen had a greater potential effect in high-prevalence settings; results were otherwise similar across a range of simulated epidemiological settings (appendix p 18). Discussion This epidemic model suggests that implementation of the short-course regimen could have an important effect on the multidrug-resistant tuberculosis epidemic, with an estimated 23% reduction in incidence over 8 years relative to continued use of longer therapy. This effect depends on key assumptions, including improved long-term effectiveness, the ability to use resource savings to expand access, and minimised poor outcomes resulting from additional drug resistance. If these assumptions prove incorrect, then the short-course regimen could have minimal or even detrimental effect—eg, possibly having no effect on the incidence of multidrug-resistant tuberculosis even if the number of people treated could be doubled. These findings emphasise the need for additional data collection as the short-course regimen is rolled out and highlight that implementation of this regimen could have important population-level effects, but also that this result is far from certain.

t tuberculosis even if the number of people treated could be doubled. These findings emphasise the need for additional data collection as the short-course regimen is rolled out and highlight that implementation of this regimen could have important population-level effects, but also that this result is far from certain. More effective regimens for multidrug-resistant tuberculosis are sorely needed, and a substantial proportion of the projected impact of a shorter regimen derives from the assumption of superior efficacy in those treated. The high treatment success rates (>80%) and low relapse risks (<1%)8 of the short-course regimen observed in initial cohorts are promising compared with longer therapy (50% success rate worldwide1 and 62% in those who would have met inclusion criteria for the short-course regimen).8 Whether efficacy of this new regimen is truly superior (and durable) awaits the results of an ongoing clinical trial.14 Our results suggest that if the short-course regimen is not more efficacious than longer therapy in eligible patients, then its impact will largely depend on whether it can facilitate expansion of treatment access and whether patients with disqualifying resistance can be appropriately triaged and successfully treated. In hotspots of more extensive drug resistance, the conditions under which the short-course regimen offers benefit will be more limited and will depend even more on the achievable gains in efficacy and resource use.

and whether patients with disqualifying resistance can be appropriately triaged and successfully treated. In hotspots of more extensive drug resistance, the conditions under which the short-course regimen offers benefit will be more limited and will depend even more on the achievable gains in efficacy and resource use. Because of the high cost of traditional care, the potential to diagnose and treat more patients within constrained budgets contributes strongly to the short-course regimen's potential effects. Our projections are similar to an estimate of the effect of universal Xpert use in India, accompanied by gradual improvement in treatment outcomes (ie, 25% reduction in incidence of multidrug-resistant tuberculosis over a decade).31 However, unlike that analysis, we explored a mechanism (short-course regimen) by which such increased treatment access and improved treatment outcomes could potentially be achieved in a budget-neutral manner, if per-patient savings were used to identify and treat more patients. If resources were reallocated elsewhere, the effect of the short-course regimen on incidence would shrink, but the overall impact on burdened tuberculosis control programmes and health systems, as well as on patients for whom multidrug-resistant tuberculosis can be economically devastating, could remain substantial. Future analyses to explicitly assess the economic effects of the short-course regimen are warranted. We also assumed that availability of clofazimine will meet demands, that second-line drug susceptibility can be tested before patients are lost to follow-up, and that the short-course regimen will be scaled up rapidly. To the extent that scale-up is slow, incomplete, or associated with increased pretreatment losses to follow-up, the effect will be diminished. Moreover, although children and extrapulmonary tuberculosis contribute little to tuberculosis transmission, the still-uncertain usefulness of the short-course regimen in such populations will affect its ability to reduce morbidity and mortality of multidrug-resistant tuberculosis.

low-up, the effect will be diminished. Moreover, although children and extrapulmonary tuberculosis contribute little to tuberculosis transmission, the still-uncertain usefulness of the short-course regimen in such populations will affect its ability to reduce morbidity and mortality of multidrug-resistant tuberculosis. Our model highlights an important drawback of the short-course regimen: its reliance on component drugs to which resistance is prevalent in some populations.5, 32 At baseline, we assumed that 10% of people without previous treatment for multidrug-resistant tuberculosis would be identified as having resistance to fluoroquinolones or second-generation aminoglycosides (ie, contraindications to the short-course regimen) and excluded on that basis. In settings where this proportion is 30%33 or higher,34 we projected a substantially diminished effect of the new regimen. Similarly, settings that implement the short-course regimen without sufficient capacity for rapid second-line drug susceptibility testing might experience reduced effectiveness and diminished short-term benefit, as well as long-term risk of amplified second-line drug resistance. Pyrazinamide resistance also could limit the effect of the short-course regimen. Pyrazinamide is included for the whole duration of the regimen and might be important for ensuring good treatment outcomes or preventing additional drug resistance, but 37–81% of multidrug-resistant strains might be pyrazinamide resistant.5 Therefore, assessment of pyrazinamide's role is urgently needed; if further study determines that individuals with resistance to pyrazinamide should also be excluded from this regimen (resulting in exclusion of nearly 50% of patients with multidrug-resistant tuberculosis in southeast Asia5 and a greater proportion in some other settings35), then the regimen's population-level effect is likely to be very small.

that individuals with resistance to pyrazinamide should also be excluded from this regimen (resulting in exclusion of nearly 50% of patients with multidrug-resistant tuberculosis in southeast Asia5 and a greater proportion in some other settings35), then the regimen's population-level effect is likely to be very small. As with all modelling studies, our analysis has some limitations. Our model projections reflect uncertainty related to trends in resistance to first-line and second-line drugs, rapidly changing diagnostic and treatment practices, and the scarcity of data on the population dynamics of multidrug-resistant tuberculosis. Importantly, our homogeneously mixed model could have overestimated the effect of this regimen in specific settings. We also simplified the dynamic representation of drug resistance to only two strains. Resistance to other drugs was implicitly factored into treatment outcomes, but transmission of multiple drug-resistant strains was not explicitly modelled. For this reason, we limited projections to a relatively short (<10 year) timeframe over which the selection of resistance to second-line drugs is expected to have relatively little epidemiological effect. Mounting second-line resistance, if it occurs, could lead to worse outcomes over time than those projected here, especially in the long term. Future modelling analyses could assess the effect of the short-course regimen on the acquisition and emergence of fluoroquinolone resistance. We were also unable to model all the complexities of tuberculosis epidemics—for example, we did not explicitly model individual heterogeneity in HIV or diabetes status or variation in tuberculosis-associated or background mortality over time, but these factors might be important considerations in certain settings.

nce. We were also unable to model all the complexities of tuberculosis epidemics—for example, we did not explicitly model individual heterogeneity in HIV or diabetes status or variation in tuberculosis-associated or background mortality over time, but these factors might be important considerations in certain settings. In summary, this modelling analysis illustrates the potential important effects of a newly recommended short-course regimen on the multidrug-resistant tuberculosis epidemic. However, it also highlights that this effect is dependent on certain key factors, including the regimen's long-term efficacy, the ability to facilitate scale-up of treatment access through resource savings, and the number and outcomes of patients who are excluded on the basis of additional drug resistance. Crucial data in estimating the ultimate effect of this regimen include evidence of durable efficacy from randomised controlled trials and data for the effect of pyrazinamide resistance, which is highly prevalent in patients with multidrug resistance. Additional research to develop improved regimens in the future will be essential, in view of the key limitations of the present short-course regimen. Ultimately, in making urgent decisions about whether to implement this new regimen at the country and global levels, the potential to reduce incidence by 20% or more needs to be weighed against the substantial uncertainty still surrounding the long-term effects of this regimen on the population dynamics of multidrug-resistant tuberculosis.

urgent decisions about whether to implement this new regimen at the country and global levels, the potential to reduce incidence by 20% or more needs to be weighed against the substantial uncertainty still surrounding the long-term effects of this regimen on the population dynamics of multidrug-resistant tuberculosis. This online publication has been corrected. The corrected version first appeared at thelancet.com/respiratory on Jan 5, 2017 Supplementary Material Supplementary appendix Acknowledgments This work was supported by the US National Institutes of Health (5T32AI007291-25 to EAK) and the Bill & Melinda Gates Foundation (Work Order 10 to DWD). Contributors DWD and EAK conceived the study. EAK developed the model, analysed the data, and wrote the first draft of the report. DWD and ATF contributed to study design, data interpretation, and critical review of the report. Declaration of interests We declare no competing interests. Figure 1 Model structure

Acknowledgments This work was supported by the US National Institutes of Health (5T32AI007291-25 to EAK) and the Bill & Melinda Gates Foundation (Work Order 10 to DWD). Contributors DWD and EAK conceived the study. EAK developed the model, analysed the data, and wrote the first draft of the report. DWD and ATF contributed to study design, data interpretation, and critical review of the report. Declaration of interests We declare no competing interests. Figure 1 Model structure Possible movements between the main states of the model are shown. Once first-line drug susceptibility testing is performed and indicates first-line resistance, assignment to an MDR-TB treatment regimen depends on the availability of the short-course regimen and the prevalence of additional resistance as detected by the accompanying second-line drug susceptibility tests. Mortality from all compartments (higher during active tuberculosis) and stratification by tuberculosis treatment history were also modelled but not shown. DS-TB=drug-susceptible tuberculosis. MDR-TB=multidrug-resistant tuberculosis. *Individuals on treatment with apparent response have improvement in symptoms followed either by durable cure (ie, no further active disease unless reinfected) or by temporary recovery only (ie, with relapse at some point after treatment). New acquisition of MDR-TB might occur during both apparently effective and ineffective first-line treatment of DS-TB. †Patients who are suspected or documented as having additional drug resistance are ineligible for the short-course regimen—which patients fall into this category depends on the prevalence of additional resistance and the second-line drug susceptibility tests that accompany the short-course regimen. ‡Includes patients who failed MDR-TB treatment with either regimen, who relapsed after treatment with either regimen, and those with known MDR-TB who never initiated treatment because of pretreatment loss to follow-up or low capacity of tuberculosis control programmes. These individuals—like those in other active, untreated tuberculosis compartments—were modelled as having an ongoing risk of tuberculosis-related mortality and an ongoing possibility of spontaneous resolution.

tiated treatment because of pretreatment loss to follow-up or low capacity of tuberculosis control programmes. These individuals—like those in other active, untreated tuberculosis compartments—were modelled as having an ongoing risk of tuberculosis-related mortality and an ongoing possibility of spontaneous resolution. Figure 2 Projected incidence of multidrug-resistant tuberculosis in the primary scenario (A) Continued use of longer therapy. (B) Implementation of the short-course regimen in 2016. Figure 3 Projected change in incidence of multidrug-resistant tuberculosis in 2024 under the short-course regimen relative to longer therapy, by scenario Median (95% uncertainty range) is shown beside each plot; the height of each plot corresponds to the probability density of the model projections. See table 3 for further descriptions of each scenario. *Improved long-term efficacy, doubling of treatment access, very poor outcomes for the 10% of patients with multidrug-tuberculosis who are ineligible for the short-course regimen, halving of losses to follow-up, and reduced relapse risk. Figure 4 Relative change in incidence of multidrug-resistant tuberculosis in 2024, under different combinations of apparent response to short-course regimen and regimen exclusions, by treatment outcome in those excluded

Median (95% uncertainty range) is shown beside each plot; the height of each plot corresponds to the probability density of the model projections. See table 3 for further descriptions of each scenario. *Improved long-term efficacy, doubling of treatment access, very poor outcomes for the 10% of patients with multidrug-tuberculosis who are ineligible for the short-course regimen, halving of losses to follow-up, and reduced relapse risk. Figure 4 Relative change in incidence of multidrug-resistant tuberculosis in 2024, under different combinations of apparent response to short-course regimen and regimen exclusions, by treatment outcome in those excluded (A) Fair—77% apparent response, with 4% relapse. (B) Poor—50% durably cured at end of therapy. (C) Very poor—20% cured at 2 years. Data are median (95% uncertainty range). Differences in modelled treatment outcomes reflect not only the efficacy of the regimen but also the prevalence of additional drug resistance in the population and the drug susceptibility test used. Since these underlying values are difficult to measure, this figure provides decision makers with projections of impact according to three measurable parameters. The rationale for these characteristics is explained in table 4. Specific scenarios of interest, which assume published point estimates of efficacy of the short-course regimen in different subgroups,8 are indicated as follows. *Second-line line probe assay screening in an area of very low prevalence of additional resistance. †Full phenotypic drug susceptibility tests (or full rapid drug susceptibility tests if available in the future) used in an area of moderate prevalence of second-line resistance. ‡Regimen implemented without having second-line drug susceptibility tests available in an area of moderate prevalence of second-line resistance. §Second-line line probe assay screening in an area of moderate prevalence of additional resistance. ¶Second-line line probe assay screening in an area of high prevalence of additional resistance.

without having second-line drug susceptibility tests available in an area of moderate prevalence of second-line resistance. §Second-line line probe assay screening in an area of moderate prevalence of additional resistance. ¶Second-line line probe assay screening in an area of high prevalence of additional resistance. Table 1 Select model parameters

without having second-line drug susceptibility tests available in an area of moderate prevalence of second-line resistance. §Second-line line probe assay screening in an area of moderate prevalence of additional resistance. ¶Second-line line probe assay screening in an area of high prevalence of additional resistance. Table 1 Select model parameters Median estimate Distribution Sampling range* Probability of rapid progression after initial tuberculosis infection20 0·14 Logit-normal 0·08–0·25 Protection against rapid progression after reinfection, if latently infected21 0·5 Logit-normal 0·1–0·9 Reactivation rate from latent to early (asymptomatic) active tuberculosis,22 per year 0·001 Log-normal 0·0005–0·002 Rate of tuberculosis diagnosis and treatment initiation,1 per year 1 Log-normal 0·7–1·5 Proportion failing to initiate treatment for multidrug-resistant tuberculosis after diagnosis (in excess of loss to follow-up of patients with drug-susceptible tuberculosis)1 0·05 Logit-normal 0·03–0·10 Proportion of treated patients who have an apparent treatment response† Newly diagnosed patients with drug-susceptible tuberculosis, first-line therapy1 0·98 Logit-normal 0·96–0·99 Patients with multidrug-resistant tuberculosis, longer therapy3 0·77 Logit-normal 0·66–0·85 Proportion who relapse, among those with apparent treatment response Newly diagnosed patients with drug-susceptible tuberculosis, first-line therapy19 0·040 Logit-normal 0·026–0·060 Patients with multidrug-resistant tuberculosis, longer therapy23 0·040 Logit-normal 0·015–0·100 Probability of loss to follow-up during therapy First-line therapy1 0·06 Logit-normal 0·03–0·1 Longer therapy for multidrug-resistant tuberculosis1 0·19 Logit-normal 0·14–0·25 Relative transmissibility of multidrug-resistant strain24 0·60 Log-normal 0·38–0·94 Risk of acquiring multidrug resistance during first-line therapy19 0·005 Logit-normal 0·0025–0·01 Proportion of patients with multidrug resistance disqualified from the short-course regimen5, 25 0·1 Logit-normal 0·07–0·15 See the appendix pp 7–8 for a complete list of parameters and additional references, and p 9 for an illustration of how the values of treatment-related parameters translate to observed treatment outcomes.

on of patients with multidrug resistance disqualified from the short-course regimen5, 25 0·1 Logit-normal 0·07–0·15 See the appendix pp 7–8 for a complete list of parameters and additional references, and p 9 for an illustration of how the values of treatment-related parameters translate to observed treatment outcomes. * 2·5th to 97·5th percentiles of unbounded distributions. † Including those who might later be lost to follow-up or relapse, or both. Table 2 Calibration targets and model fit Reported values for southeast Asia* Median model values (95% uncertainty range) Tuberculosis incidence per 100 000 adult population per year 203 (192–232) 203 (191–207) Annual change in incidence –2% –2·2% (1·8–2·8) Tuberculosis prevalence per 100 000 adult population 275 (224–330) 271 (228–323) Tuberculosis mortality per 100 000 adult population per year 26·2 (20·9–32·6) 26·7 (21·2–32·3) Proportion of new notifications with multidrug resistance 2·2% (1·9–2·6) 2·1% (1·9–2·5) Proportion of retreatment notifications with multidrug resistance 16% (14–18) 16·7% (14·4–17·9) * We derived these estimates from WHO-reported point estimates (uncertainty intervals),1 adjusted to reflect the burden of pulmonary tuberculosis in the adult population (ie, those aged 15 years and older), on the basis of the proportion of cases that are pulmonary, the proportion estimated to occur in those aged 15 years and older, and the proportion of the southeast Asian population aged 15 years and older. Table 3 Assumptions in primary scenario and alternative scenarios, by variable

Reported values for southeast Asia* Median model values (95% uncertainty range) Tuberculosis incidence per 100 000 adult population per year 203 (192–232) 203 (191–207) Annual change in incidence –2% –2·2% (1·8–2·8) Tuberculosis prevalence per 100 000 adult population 275 (224–330) 271 (228–323) Tuberculosis mortality per 100 000 adult population per year 26·2 (20·9–32·6) 26·7 (21·2–32·3) Proportion of new notifications with multidrug resistance 2·2% (1·9–2·6) 2·1% (1·9–2·5) Proportion of retreatment notifications with multidrug resistance 16% (14–18) 16·7% (14·4–17·9) * We derived these estimates from WHO-reported point estimates (uncertainty intervals),1 adjusted to reflect the burden of pulmonary tuberculosis in the adult population (ie, those aged 15 years and older), on the basis of the proportion of cases that are pulmonary, the proportion estimated to occur in those aged 15 years and older, and the proportion of the southeast Asian population aged 15 years and older. Table 3 Assumptions in primary scenario and alternative scenarios, by variable Primary scenario Alternative scenarios Baseline Short-course regimen Level of treatment initiation for multidrug-resistant tuberculosis Maintain existing levels Double existing levels Scenario 1—maintain existing levels in baseline and with short-course regimen; Gradual increase in baseline, doubled with short-course regimen (appendix); Gradual increase in baseline and with short-course regimen (appendix); or Immediate optimisation of drug susceptibility testing in baseline and with short-course regimen (appendix) Proportion of patients with apparent treatment response, among those who receive the short-course regimen* Not applicable 92·5% Scenario 2—same as longer therapy (roughly 77%†) Scenario 3—same as longer therapy (roughly 77%†), combined with improved “fair” outcomes for those ineligible (see below) Outcome of longer therapy for patients ineligible for the short-course regimen* Not applicable 20% cured at end of therapy (“very poor”)‡ Scenario 4—same apparent response (roughly 77%†) as the average patient with multidrug-resistant tuberculosis in the baseline scenario (“fair”) Scenario 3—”fair” response as in Scenario 4, combined with reduced response in those who receive the short-course regimen (see above) Proportion of patients ineligible for short-course regimen on the basis of second-line resistance and drug susceptibility testing practices* Not applicable 10% Scenario 5—30% Scenario 6—0% Relapse risk in those with apparent treatment response who finish treatment course Roughly 4% for longer therapy† 1% for short-course regimen Scenario 7—roughly 8% (twice that of longer therapy) for short-course regimen Loss to follow-up Roughly 19% for longer therapy† 10% for short-course regimen Scenario 8—roughly 19% for both regimens† * See table 2 for further details.

who finish treatment course Roughly 4% for longer therapy† 1% for short-course regimen Scenario 7—roughly 8% (twice that of longer therapy) for short-course regimen Loss to follow-up Roughly 19% for longer therapy† 10% for short-course regimen Scenario 8—roughly 19% for both regimens† * See table 2 for further details. † Sampled from distributions shown in table 1. ‡ Apparent treatment response not explicitly modelled; see appendix p 9 for further details of calculation. Table 4 Values used in combinatorial analyses, by variable

who finish treatment course Roughly 4% for longer therapy† 1% for short-course regimen Scenario 7—roughly 8% (twice that of longer therapy) for short-course regimen Loss to follow-up Roughly 19% for longer therapy† 10% for short-course regimen Scenario 8—roughly 19% for both regimens† * See table 2 for further details. † Sampled from distributions shown in table 1. ‡ Apparent treatment response not explicitly modelled; see appendix p 9 for further details of calculation. Table 4 Values used in combinatorial analyses, by variable Rationale Proportion of patients receiving short-course regimen with apparent treatment response 98%, with 1% relapse As reported with short-course regimen for patients susceptible to both fluoroquinolone and pyrazinamide8 92·5%, with 1% relapse Average across all patients with multidrug-resistant tuberculosis receiving the short-course regimen in Bangladesh9 85%, with 2·5% relapse As reported with short-course regimen (with large uncertainty) for patients resistant to pyrazinamide only8 77%, with 4% relapse Average outcomes of longer therapy Outcome of longer therapy for patients ineligible for short-course regimen Fair: 77% apparent response, with 4% relapse Outcomes equivalent to the average outcome of longer therapy in the baseline scenario; might reflect effective individualisation of treatment or limited drug resistance in ineligible patients (eg, resistance to pyrazinamide only) Poor: 50% durably cured at end of therapy* Corresponds to outcomes of longer therapy for patients with multidrug resistance and fluoroquinolone resistance8 Very poor: 20% cured at 2 years† Most conservative assumption; those excluded have typical outcomes of extensively drug-resistant tuberculosis on longer therapy Proportion of patients ineligible for short-course regimen on the basis of second-line resistance and drug susceptibility testing practices 0% No second-line drug susceptibility testing or no second-line resistance 10% Line probe assay (for fluoroquinolones and second-line injectables); levels of resistance similar to Pakistan and Bangladesh 30% Line probe assay; levels of resistance similar to eastern Europe 50% Exclusion of all patients with pyrazinamide or second-line drug resistance in a typical setting * Corresponds to two-thirds having “fair” outcomes and a third having “very poor” outcomes.

s); levels of resistance similar to Pakistan and Bangladesh 30% Line probe assay; levels of resistance similar to eastern Europe 50% Exclusion of all patients with pyrazinamide or second-line drug resistance in a typical setting * Corresponds to two-thirds having “fair” outcomes and a third having “very poor” outcomes. † Modelled as an ongoing probability of roughly 13% per year of cure and an ongoing tuberculosis mortality risk, resulting in roughly 23% death, 20% cure, and 57% persisting with active disease at 2 years.

Introduction Idiopathic and heritable pulmonary arterial hypertension constitute a rare disease group characterised by an imbalance in endothelial-derived vasoactive factors, inflammation, and structural remodelling of pulmonary vessels.1 The resultant pressure load on the right ventricle causes premature death from heart failure.1, 2 The incidence of pulmonary arterial hypertension is estimated at 1–7·6 per million per year and cases of idiopathic or heritable pulmonary arterial hypertension account for 0·9–2·6 per million per year.3 Estimated 3-year survival is 58–74%,2, 4 but the disease is heterogeneous; several biological mechanisms1 and a range of variants in several genes5 have been linked to pathogenesis, and life expectancy is variable. Regular assessment of disease severity and prognosis is necessary to guide clinical management. The existing guidelines recommend a combination of established prognostic parameters on the basis of clinical assessment, imaging, and biochemistry.6 These clinical parameters are not always available for each patient visit and existing risk assessments have poor accuracy (with C statistics ranging between 0·57 for the US National Institutes of Health,7 0·59 for the French Registries,2 and 0·77 for the REVEAL equation8), leaving considerable scope for improvement.3 Research in context Evidence before this study

This study is compliant with the Guidelines for Accurate and Transparent Health Estimates Reporting (GATHER) with details provided in the appendix (pp 58–60).38 Role of the funding source This research was supported by funding from the Bill & Melinda Gates Foundation. The funders had no role in the study design, data collection and analysis, interpretation of data, decision to publish, or preparation of the manuscript. Results In 2015, 3·2 million people (95% UI 3·1 million to 3·3 million) died from COPD worldwide, an increase of 11·6% (5·3–19·8) compared with 1990, despite a decrease in the age-standardised death rate of 41·9% (37·7–45·1). Population growth and ageing of the global population outweighed the downward trend in age-standardised death rates. The greatest reduction in age-standardised death rates occurred in countries in the high-middle-SDI quintile and middle-SDI quintile. From 1990 to 2015, the prevalence of COPD increased by 44·2% (95% UI 41·7–46·6) to 174·5 million individuals (160·2 million to 189·0 million). The decrease in age-standardised prevalence of 14·7% (13·5–15·9) was much smaller than the decrease in age-standardised death rates. The greatest decrease in age-standardised prevalence was seen in countries in the high-middle-SDI quintile and the middle-SDI quintile (table 1).Table 1 Deaths due to asthma and COPD and number of prevalent cases of disease in 2015 and percentage change in all-age and age-standardised rates in locations grouped by SDI quintile

Introduction Idiopathic and heritable pulmonary arterial hypertension constitute a rare disease group characterised by an imbalance in endothelial-derived vasoactive factors, inflammation, and structural remodelling of pulmonary vessels.1 The resultant pressure load on the right ventricle causes premature death from heart failure.1, 2 The incidence of pulmonary arterial hypertension is estimated at 1–7·6 per million per year and cases of idiopathic or heritable pulmonary arterial hypertension account for 0·9–2·6 per million per year.3 Estimated 3-year survival is 58–74%,2, 4 but the disease is heterogeneous; several biological mechanisms1 and a range of variants in several genes5 have been linked to pathogenesis, and life expectancy is variable. Regular assessment of disease severity and prognosis is necessary to guide clinical management. The existing guidelines recommend a combination of established prognostic parameters on the basis of clinical assessment, imaging, and biochemistry.6 These clinical parameters are not always available for each patient visit and existing risk assessments have poor accuracy (with C statistics ranging between 0·57 for the US National Institutes of Health,7 0·59 for the French Registries,2 and 0·77 for the REVEAL equation8), leaving considerable scope for improvement.3 Research in context Evidence before this study We searched PubMed for relevant articles (before March 1, 2017) with search terms including “pulmonary arterial hypertension”, “prognostic”, “proteomics”, and “biomarker”. Several studies report single biomarkers for pulmonary arterial hypertension, usually derived from other diseases, but none have undertaken unbiased screening of large numbers of plasma markers and related these to outcomes in pulmonary arterial hypertension. The prognostication of pulmonary arterial hypertension remains poor. Pulmonary arterial hypertension is diagnosed at cardiac catheterisation. Thereafter a combination of exercise capacity (eg, 6-min walk test), patient-reported symptoms (eg, functional class assessment), echocardiography, and circulating NT-proBNP concentrations—captured in prognostic equations such as the REVEAL score—are used to follow disease progression, response to treatment, and make clinical management decisions. Not all these measurements are made at each visit and some (eg, 6-min walk test) are subject to confounding factors. Better, non-invasive, and objective methods of assessment are needed that can be deployed in the clinical setting.

llow disease progression, response to treatment, and make clinical management decisions. Not all these measurements are made at each visit and some (eg, 6-min walk test) are subject to confounding factors. Better, non-invasive, and objective methods of assessment are needed that can be deployed in the clinical setting. Added value of this study We did an unbiased screen of 1129 proteins measured in plasma samples collected on routine clinic visits. Measurement of circulating concentrations of nine proteins in combination predicted survival, which outperformed traditional clinical assessments. A prognostic score on the basis of plasma concentrations of the nine proteins was validated in independent cohorts from three countries (UK, France, and Germany) and is relevant to both incident and prevalent cases of pulmonary arterial hypertension. An increase in the panel score over time is associated with increased mortality. Implications of all the available evidence

We did an unbiased screen of 1129 proteins measured in plasma samples collected on routine clinic visits. Measurement of circulating concentrations of nine proteins in combination predicted survival, which outperformed traditional clinical assessments. A prognostic score on the basis of plasma concentrations of the nine proteins was validated in independent cohorts from three countries (UK, France, and Germany) and is relevant to both incident and prevalent cases of pulmonary arterial hypertension. An increase in the panel score over time is associated with increased mortality. Implications of all the available evidence The guidelines for prognostication in pulmonary arterial hypertension recommend the use of only one blood biomarker, BNP or NT-proBNP, and consideration of a multitude of clinical measures, which when formalised into risk equations perform only moderately well in predicting outcomes. These data suggest that a panel of nine proteins, which report on different pathogenic mechanisms linked to pulmonary arterial hypertension, can be used to stratify patients according to risk and assess response to treatment better than existing clinical tools. Further investigation of the pathways represented in the protein panel might also offer new insights for the development of novel therapies.

enic mechanisms linked to pulmonary arterial hypertension, can be used to stratify patients according to risk and assess response to treatment better than existing clinical tools. Further investigation of the pathways represented in the protein panel might also offer new insights for the development of novel therapies. The existing management guidelines include the measurement of plasma brain natriuretic peptide (BNP) or N-terminal proBNP (NT-proBNP) concentrations, an indicator of right ventricular function, in the assessment of patients with pulmonary arterial hypertension.6 Biomarkers reporting other components of the pathophysiology of idiopathic pulmonary arterial hypertension, such as inflammation (interleukin 6 and growth differentiation factor 15), renal function (creatinine), and iron status (red cell distribution width), also predict clinical outcome,9, 10, 11, 12 but none are used routinely. The use of multiple biomarkers could improve risk assessment. Proteomics offers an unbiased approach to identifying and quantifying multiple biomarkers representative of disease processes. Mass spectrometry-based proteomic analysis of lung tissue, plasma, and cultured cells from patients with pulmonary arterial hypertension has identified a small number of dysregulated proteins.13, 14, 15 Alternative high-throughput strategies exploiting targeted peptide-binding reagents in a multiplex manner permit the screening of large numbers of identifiable proteins. One such technology uses DNA-based aptamer reagents, known as SOMAmers, that are modified to improve binding kinetics.16, 17

lated proteins.13, 14, 15 Alternative high-throughput strategies exploiting targeted peptide-binding reagents in a multiplex manner permit the screening of large numbers of identifiable proteins. One such technology uses DNA-based aptamer reagents, known as SOMAmers, that are modified to improve binding kinetics.16, 17 We used a SomaScan array to measure concentrations of 1129 proteins in plasma to identify and validate circulating proteomic signatures that predict survival of patients with idiopathic or heritable pulmonary arterial hypertension.

lated proteins.13, 14, 15 Alternative high-throughput strategies exploiting targeted peptide-binding reagents in a multiplex manner permit the screening of large numbers of identifiable proteins. One such technology uses DNA-based aptamer reagents, known as SOMAmers, that are modified to improve binding kinetics.16, 17 We used a SomaScan array to measure concentrations of 1129 proteins in plasma to identify and validate circulating proteomic signatures that predict survival of patients with idiopathic or heritable pulmonary arterial hypertension. Methods Study design and participants In this multicentre, observational cohort study, we identified and analysed four cohorts of patients with idiopathic or heritable pulmonary arterial hypertension from three expert centres recognised internationally as centres of excellence for pulmonary arterial hypertension diagnosis and management in London (UK; cohorts 1 and 2), Giessen (Germany; cohort 3), and Paris (France; cohort 4). The diagnostic criteria for idiopathic or heritable pulmonary arterial hypertension over the course of this study were stable: raised mean pulmonary artery pressure of more than 25 mm Hg, with pulmonary capillary wedge pressure less than 15 mm Hg (and pulmonary vascular resistance [PVR] >3 Wood units) at rest with exclusion of known associated diseases. The guidelines quoted are internationally agreed. Samples from 25 healthy controls were also collected at Hammersmith Hospital for comparison of proteomic and alternative assay measurements.6 All samples and data were obtained with informed consent and local research ethics committee approval.

nown associated diseases. The guidelines quoted are internationally agreed. Samples from 25 healthy controls were also collected at Hammersmith Hospital for comparison of proteomic and alternative assay measurements.6 All samples and data were obtained with informed consent and local research ethics committee approval. Procedures Patients were not fasting and were sampled at their routine clinical appointment visits (397 in total for all patients). Peripheral venous blood samples were collected using EDTA (edetic acid) for cohorts 1, 2, and 3 or sodium citrate Vacutainer tubes (BD Biosciences, Oxford, UK) for cohort 4, immediately put on ice, centrifuged (1300 × g, 15 min) within 30 min of collection, and plasma aliquots were stored at −80°C until required. The plasma samples underwent one freeze–thaw cycle to aliquot 120 μL for the SomaScan assay and provide other aliquots for NT-proBNP and targeted assays. Clinical data were collected within 30 days of blood sampling and biochemical data were collected within 7 days of blood sampling.11 We calculated the REVEAL prognostic equation,8 and fitted it to the study cohorts. The equation includes categories on the basis of sub-diagnosis, age, sex, renal insufficiency, WHO functional class, systolic blood pressure, heart rate, 6-min walk distance, NT-proBNP, presence of pericardial effusion, percentage predicted diffusing capacity for carbon monoxide, mean right atrial pressure, and PVR. Proteomic analysis was done with SOMAscanV3 (Somalogic Inc, Boulder, CO, USA)16 and patient status was concealed to the technicians. The list of 1129 targeted proteins has been reported previously.17 To minimise between-experiment variation, bridging samples were included in all experiments; specifically, to check consistency of the measurements we included samples from 24 patients from cohort 1 in all experiments to verify that the measurements from experiment to experiment were comparable. Median variation in relative fluorescence units between experiments was less than 10%, and more than 90% of analytes showed less than 20% variation in average levels between experiments.

ples from 24 patients from cohort 1 in all experiments to verify that the measurements from experiment to experiment were comparable. Median variation in relative fluorescence units between experiments was less than 10%, and more than 90% of analytes showed less than 20% variation in average levels between experiments. Following selection of the proteins of interest from analyses of cohort 1 samples, we measured the same proteins again in the same samples used in the proteomic analysis by alternative commercially available assays, each specific for the protein of interest, to check that the two methods agreed; the ELISA and Luminex assays used to validate the SOMAscan measurements are detailed in the appendix (p 3). Outcomes All-cause mortality was the primary endpoint. In a secondary analysis, lung transplantation or death was used as a composite endpoint. Statistical analysis We present differences in protein concentrations by subtraction of log relative fluorescence units. We assessed the association between patient characteristics and biomarkers by Spearman's rank test or Mann–Whitney U test and Kruskal–Wallis test for categorical variables.

Outcomes All-cause mortality was the primary endpoint. In a secondary analysis, lung transplantation or death was used as a composite endpoint. Statistical analysis We present differences in protein concentrations by subtraction of log relative fluorescence units. We assessed the association between patient characteristics and biomarkers by Spearman's rank test or Mann–Whitney U test and Kruskal–Wallis test for categorical variables. We did survival analyses using time from sampling to death or censoring. We compared the protein concentrations of survivors and non-survivors (overall survival in cohort 1 and at 2·5 years' follow-up in cohort 2) with Mann–Whitney tests to maximise the power of the protein validation analysis. We used random sample analysis to assess robustness of differences between survivors and non-survivors: we repeated Mann–Whitney analyses 18 times in both discovery and validation (cohorts 1 and 2), each time removing one patient out of six patients in three randomised blocks, with each sample left out of three analyses. We tested two panel scoring systems. One system was based on a simple count of proteins indicating risk on the basis of receiver operating characteristic (ROC) cutoffs, and the second was a Cox regression model, in which predicted hazard based on continuous biomarker concentrations was calculated on the basis of fitting to the discovery cohort.

g systems. One system was based on a simple count of proteins indicating risk on the basis of receiver operating characteristic (ROC) cutoffs, and the second was a Cox regression model, in which predicted hazard based on continuous biomarker concentrations was calculated on the basis of fitting to the discovery cohort. For assessment of discrimination, we used ROC curves to compare prognostic discriminatory power of biomarkers. Kaplan–Meier plots illustrated events (deaths) in relation to biomarker levels and predicted risk in Cox models, assessed by the log-rank test. We fitted the simple panel score and REVEAL equation (an accepted clinical score derived from a variety of clinical parameters)8 to Cox models to test the additional value and potential clinical use of the panel. For reclassification, we calculated indices—net reclassification index (NRI) and the relative integrated discrimination improvement (IDI) statistic18—with R package PredictABEL19 for the addition of the prognostic panel score to the REVEAL equation.8 We developed the models in cohorts 1 and 2 combined and validated them in cohort 4. Cohort 3 was used to test the performance of the proteins and panel score longitudinally, before and after initiation of targeted therapy. We assessed calibration of the Cox models by comparing predicted mortality of patients against observed mortality using Harrell's rms package in R. We converted variables to Z-scores (SD around mean) before testing by Cox regression. We did calculations with SPSS version 21.0 and R version 3.0.2.

n of targeted therapy. We assessed calibration of the Cox models by comparing predicted mortality of patients against observed mortality using Harrell's rms package in R. We converted variables to Z-scores (SD around mean) before testing by Cox regression. We did calculations with SPSS version 21.0 and R version 3.0.2. Role of the funding source The funders of the study had no role in the study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had the final responsibility for the decision to submit for publication. Results Cohorts 1 and 2 were censored on May 15, 2014, cohort 3 on May 21, 2015, and cohort 4 on June 1, 2014. We assessed patients for eligibility between Oct 25, 2011, and Aug 13, 2013, for cohort 1, between Oct 24, 2002, and June 22, 2011, for cohort 2, between Aug 27, 2003, and Nov 19, 2012, for cohort 3, and between June 2, 2003, and Dec 23, 2011, for cohort 4 (table 1, figure 1). At the end of the follow-up periods, 18 patients died in cohort 1, 37 patients died in cohort 2, 17 patients died in cohort 3, and 39 patients died in cohort 4; no patients were lost to follow-up. Nine patients (n=3 cohort 1 and n=6 cohort 2) underwent lung or heart and lung transplantation; seven of nine patients who had undergone transplantation died during follow-up.Figure 1 Study design Table 1 Baseline characteristics

Results Cohorts 1 and 2 were censored on May 15, 2014, cohort 3 on May 21, 2015, and cohort 4 on June 1, 2014. We assessed patients for eligibility between Oct 25, 2011, and Aug 13, 2013, for cohort 1, between Oct 24, 2002, and June 22, 2011, for cohort 2, between Aug 27, 2003, and Nov 19, 2012, for cohort 3, and between June 2, 2003, and Dec 23, 2011, for cohort 4 (table 1, figure 1). At the end of the follow-up periods, 18 patients died in cohort 1, 37 patients died in cohort 2, 17 patients died in cohort 3, and 39 patients died in cohort 4; no patients were lost to follow-up. Nine patients (n=3 cohort 1 and n=6 cohort 2) underwent lung or heart and lung transplantation; seven of nine patients who had undergone transplantation died during follow-up.Figure 1 Study design Table 1 Baseline characteristics Cohort 1 (n=143) Cohort 2 (n=75) Cohort 3 (n=43) Cohort 4 (n=93) Recruitment period 2011–13 2002–11 2004–11 2003–11 Age, years 53 (41–69) 55 (39–70) 50 (29–61) 54 (39–66) Sex Females 100 (70%) 45 (60%) 28 (65%) 58 (62%) Males 43 (30%) 30 (40%) 15 (35%) 35 (38%) Ethnic origin White 117 (82%) 59 (79%) 43 (100%) 78 (84%) Asian 14 (10%) 11 (15%) 0 5 (5%) Black 4 (3%) 3 (4%) 0 8 (9%) Other ethnicity or not stated 8 (6%) 2 (3%) 0 2 (2%) Idiopathic pulmonary arterial hypertension 140 (98%) 71 (95%) 43 (100%) 77 (83%) Heritable pulmonary arterial hypertension 3 (2%) 4 (5%) 0 16 (17%) WHO FC Class I 6 (4%) 1 (1%) 0 4 (4%) Class II 32 (22%) 14 (19%) 6 (14%) 28 (30%) Class III 91 (64%) 41 (55%) 28 (65%) 55 (59%) Class IV 14 (10%) 19 (25%) 9 (21%) 6 (6%) 6-min walk, m 339 (144–432) 258 (120–369) 359 (251–425) 390 (300–433) mPAP, mm Hg 52 (43–62) 51 (46–62) 50 (45–58) 51 (44–61) mRAP, mm Hg 10 (6–13) 12 (8–17·5) 7 (3–10) 6 (3·5–10) PAWP, mm Hg 10 (8–14) 10 (7–13) 8 (5–9) 8 (6–10) CI, L/min/kg/m2 2·13 (1·71–2·65) 2·2 (1·71–2·59) 2·23 (1·89–2·60) 2·54 (2·06–3·40) CO, L/min 4·16 (3·18–5·39) 4·13 (3·00–5·20) 3·80 (3·23–4·41) 4·30 (3·47–5·50) PVR, Wood units 10·0 (6·0–14·5) 9·3 (7·5–13·1) 11·4 (8·4–15·0) 9·9 (6·4–14·3) Treatment naive 13 (9%) 19 (25%) 43 (100%) 24 (26%) Monotherapy CCB 5 (3%) 0 0 1 (1%) PDE5 32 (22%) 14 (19%) 0 7 (8%) ERA 14 (10%) 16 (21%) 0 26 (28%) Prost 1 (1%) 4 (5%) 0 2 (2%) Dual therapy ERA and PDE5 53 (37%) 10 (13%) 0 21 (23%) Prost and ERA 2 (1%) 5 (7%) 0 2 (2%) Prost and PDE5 7 (5%) 5 (7%) 0 3 (3%) Triple therapy 16 (11%) 2 (3%) 0 7 (8%) Estimated survival 1-year follow-up 96% 89% 98% 91% 2-year follow-up 88% 63% 88% 88% 3-year follow-up 0 45% 86% 77% Time after diagnosis sampled, years 3·16 (0·54–7·3) 1·11 (0·37–2·45) 0·41 (0·32–0·89)* 0·88 (0·15–1·82) Follow-up, years 2·0 (1·6–2·2) 2·5 (1·5–4·9) 6·5 (4·3–9·9) 4·4 (3·0–5·7) Data are median (IQR), n (%), or n. WHO FC=WHO/New York Heart Association Functional Classification. Shuttle walk=incremental shuttle walk test. mPAP=mean pulmonary artery pressure. mRAP=mean right atrial pressure. PAWP=pulmonary artery wedge pressure. CI=cardiac index. CO=cardiac output. PVR=pulmonary vascular resistance.

are median (IQR), n (%), or n. WHO FC=WHO/New York Heart Association Functional Classification. Shuttle walk=incremental shuttle walk test. mPAP=mean pulmonary artery pressure. mRAP=mean right atrial pressure. PAWP=pulmonary artery wedge pressure. CI=cardiac index. CO=cardiac output. PVR=pulmonary vascular resistance. CCB=calcium channel blocker. ERA=endothelin receptor antagonist. PDE5=phosphodiesterase 5 inhibitors. Prost=prostanoid analogues. IQR=interquartile range. * Years after diagnosis sampled for second sample shown; baseline samples were taken at diagnosis. Concentrations of 134 proteins were associated with overall survival in cohort 1 (figure 2A). 40 of these proteins were validated as able to differentiate between survivors and non-survivors in cohort 2. 20 prognostic proteins, including BNP, were prioritised by random sampling analysis as the most robust (appendix p 4). These proteins had good specificity and sensitivity in ROC analysis (figure 2B and appendix p 5), and protein concentrations that distinguished between survivors and non-survivors in cohort 1 and performed well in cohort 2 were identified (figure 2C). To ensure the small number of patients with heritable pulmonary arterial hypertension were not confounding, we did an analysis excluding these seven patients and found that the 20 proteins were again significant in both discovery (cohort 1) and validation (cohort 2) analyses (appendix p 6).Figure 2 Prognostic protein panel analysis

small number of patients with heritable pulmonary arterial hypertension were not confounding, we did an analysis excluding these seven patients and found that the 20 proteins were again significant in both discovery (cohort 1) and validation (cohort 2) analyses (appendix p 6).Figure 2 Prognostic protein panel analysis (A) Volcano plot illustrating differences in protein expression between survivors and non-survivors. (B) ROC analysis of 20 selected proteins showing sensitivity and 1 – specificity at cutoffs. (C) Kaplan–Meier survival analysis of patients with idiopathic pulmonary arterial hypertension in cohort 2 divided by TIMP-2 cutoff derived from ROC analysis of cohort 1. (D) Hazard ratios and 95% CI from Cox regression analysis comparing 20 prognostic proteins with established prognostic marker, NT-proBNP. (E) Commercially available ELISA or Luminex assays targeting the 14 independently prognostic proteins used to validate SomaScan measurements in a subset of 80 plasma samples selected from cohort 1 (n=55) and healthy controls (n=25), with samples with high and low concentrations of the analytes chosen. Nine proteins were validated and further studied in cohort 3 (serial samples) and cohort 4 (validation cohort). This scatter-plot illustrates TIMP-1 measurements by SomaScan and Luminex assays in idiopathic pulmonary arterial hypertension cohort 4. Cutoffs for SomaScan and Luminex values derived by percentile equalling ROC-derived cutoff in cohort 1 are indicated by dashed lines. Statistics indicate Spearman's rank correlation. ROC=receiver operating characteristic. RFU=relative fluorescence unit.

nex assays in idiopathic pulmonary arterial hypertension cohort 4. Cutoffs for SomaScan and Luminex values derived by percentile equalling ROC-derived cutoff in cohort 1 are indicated by dashed lines. Statistics indicate Spearman's rank correlation. ROC=receiver operating characteristic. RFU=relative fluorescence unit. We investigated whether the 20 prognostic proteins offered an improvement in risk estimation in addition to the only prognostic protein biomarker currently in use in pulmonary arterial hypertension, namely NT-proBNP. 14 of 20 proteins were each prognostic independent of NT-proBNP in Cox models with death as the primary endpoint (all p<0·05; figure 2D). With transplantation or death as a composite endpoint, all 14 proteins remained significant and independent of NT-proBNP (data not shown). A significant correlation between values in the SomaScan and the independent ELISA or Luminex assay was shown for nine protein measurements (all p<0·05, Spearman's rank [data not shown])—interleukin-1 receptor-like 1 (IL1R1/ST2), tissue inhibitors of metalloproteinases (TIMP-1 and TIMP-2), plasminogen, apolipoprotein-E (ApoE), erythropoietin (EPO), complement factor H and factor D, and insulin-like growth factor binding protein-1 (IGFBP-1). The measurements for these nine proteins allowed us to derive threshold protein concentrations associated with survival (figure 2E, appendix p 7), and we derived threshold concentrations for each protein in cohort 1 and validated these in cohort 2.

d factor D, and insulin-like growth factor binding protein-1 (IGFBP-1). The measurements for these nine proteins allowed us to derive threshold protein concentrations associated with survival (figure 2E, appendix p 7), and we derived threshold concentrations for each protein in cohort 1 and validated these in cohort 2. We used the prognostic thresholds for each of these nine proteins to produce a protein panel score for each patient, whereby each protein indicating risk (ie, when the plasma concentration was above or below the threshold cutoff for survival) added 1 to a patient's score. This calculation produced scores ranging from 0 to 9 for each patient and discriminated non-survivors in discovery (for cohort 1, area under the curve [AUC] 0·93, 95% CI 0·88–0·99) and validation (for cohort 2, 0·86, 0·77–0·94). The simplified scoring of each protein based on a cutoff performed as well as an equation using continuous protein concentrations, which was also derived in cohort 1 and tested in cohort 2 (AUC 0·83, 95% CI 0·75–0·92; appendix p 9). Increasing panel scores clearly distinguished risk groups (figure 3A, appendix p 10).Figure 3 Survival analysis of panel score and established prognostic factors

as an equation using continuous protein concentrations, which was also derived in cohort 1 and tested in cohort 2 (AUC 0·83, 95% CI 0·75–0·92; appendix p 9). Increasing panel scores clearly distinguished risk groups (figure 3A, appendix p 10).Figure 3 Survival analysis of panel score and established prognostic factors Kaplan–Meier survival estimates in patients with different panel scores, in all patients with idiopathic pulmonary arterial hypertension from (A) cohorts 1 and 2 and (B) cohort 4. ROC analysis of Cox models before and after addition of the prognostic protein panel to the established equation, in all patients with idiopathic pulmonary arterial hypertension from (C) cohorts 1 and 2 and (D) cohort 4. ROC=receiver operating characteristic. AUC=area under the curve. Removal of any two proteins did not impair the performance of the remaining panel, suggesting no protein was dominant, and emphasising the discriminating power of the combination (appendix p 11). The panel score was also prognostic in a sub-analysis of samples obtained before initiation of therapy, comprising 77 (35%) of 218 patients from cohorts 1 and 2 (appendix p 12) and in groups of patients from cohorts 1 and 2 stratified by age (above and below 50 years; appendix p 13) and bilirubin concentration (21 μmol/L, the upper limit of the normal range in the clinical assay; appendix p 13).

re initiation of therapy, comprising 77 (35%) of 218 patients from cohorts 1 and 2 (appendix p 12) and in groups of patients from cohorts 1 and 2 stratified by age (above and below 50 years; appendix p 13) and bilirubin concentration (21 μmol/L, the upper limit of the normal range in the clinical assay; appendix p 13). 43 patients were sampled at diagnosis and after initiating therapy in cohort 3 (median time between samples 4 months, IQR 3–10). Although changes in the concentrations of any individual protein, including NT-proBNP, were not associated with outcome, an increasing panel score was prognostic (p=0·0186; figure 4A). Patients whose protein panel score was higher at follow-up than at baseline showed poorer survival than those whose scores remained stable or improved (p=0·0133; figure 4B), which identified patients who had not responded to therapy. These patients had similar clinical characteristics at baseline (appendix p 8). Changes in the panel score appear more sensitive than other measures—eg, the small changes in pulmonary vascular resistance recorded at repeat catheterisation were not associated with survival (appendix p 14).Figure 4 Prognostic value of changes in the protein panel score from diagnosis to after initiation of therapy

. Changes in the panel score appear more sensitive than other measures—eg, the small changes in pulmonary vascular resistance recorded at repeat catheterisation were not associated with survival (appendix p 14).Figure 4 Prognostic value of changes in the protein panel score from diagnosis to after initiation of therapy (A) Cox proportional hazard estimates associated with changes in individual proteins and the overall panel score, showing only the combination of proteins into the score is significantly associated with outcomes. (B) Kaplan–Meier survival estimates in patients with serial panel score measurements (cohort 3), showing an increase in the panel score from diagnosis to after initiation of therapy is associated with poor outcomes. The protein panel score was further validated in an independent group of 93 patients with idiopathic or heritable pulmonary arterial hypertension from cohort 4 (4·4 years' [IQR 3·0–5·7] follow-up; table 1); an increasing panel score distinguished risk groups (figure 3B, appendix p 10). In panel development (cohorts 1 and 2) and panel validation analyses (in cohort 4), the panel score predicted survival independent of NT-proBNP measurements (appendix p 8).

ension from cohort 4 (4·4 years' [IQR 3·0–5·7] follow-up; table 1); an increasing panel score distinguished risk groups (figure 3B, appendix p 10). In panel development (cohorts 1 and 2) and panel validation analyses (in cohort 4), the panel score predicted survival independent of NT-proBNP measurements (appendix p 8). Results from Cox models confirmed that the REVEAL equation was prognostic and that the protein panel score provided independent prognostic information in both panel development and panel validation (table 2). The categorical NRI indicated that the protein panel score reclassified more patients who died during follow-up at above-average risk and vice versa, while the relative IDI showed a 50–223% relative improvement after addition of the panel score to the model. This outcome means that the protein panel is changing the risk estimates for patients in a significant proportion of individuals, and is providing prognostic information additional to the established equation. In both the panel development and panel validation analyses, the protein panel score improved the C statistic by a similar margin (0·08–0·09; for REVEAL risk score AUC 0·83, 95% CI 0·77–0·89; p<0·0001; for panel and REVEAL 0·91, 0·87–0·96; p<0·0001; figure 3C, D, table 2). Calibration in both model development and validation was similar before and after addition of the panel score (appendix p 15). A combination of the panel score and NT-proBNP as a continuous variable performed very similarly to the combination of the panel score and REVEAL equation in both analyses (appendix p 16).Table 2 Model performance

model development and validation was similar before and after addition of the panel score (appendix p 15). A combination of the panel score and NT-proBNP as a continuous variable performed very similarly to the combination of the panel score and REVEAL equation in both analyses (appendix p 16).Table 2 Model performance REVEAL equation Panel of nine proteins Performance of equation and panel in combined model C statistic Development 0·83 (0·77–0·89) 0·89 (0·84–0·94) Validation 0·72 (0·59–0·84) 0·83 (0·72–0·94) Hazard ratio in model Development 1·73 (1·36–2·21) 2·44 (1·79–3·33) Validation 1·42 (1·01–1·99) 1·9 (1·33–2·72) Effect on performance of adding panel to equation Categorical NRI (above/below overall event rate) Development Reference 0·20 (0·09–0·31) Validation Reference 0·39 (0·07–0·70) IDI Development Reference 0·17 (0·09–0·24) Validation Reference 0·13 (0·06–0·20) Relative IDI Development Reference 0·50 (0·28–0·72) Validation Reference 2·23 (1·05–3·41) Δ C statistic Development Reference 0·083 (0·052–0·114) Validation Reference 0·095 (0·026–0·164) Hazard ratios from Cox regression analyses of panel score and REVEAL equation, categorical NRI based on overall death—25% (55 deaths in 218 patients) at 2·5 years in development analyses (cohorts 1 and 2) and 23% (21 deaths in 93 patients) at 3 years in validation analyses (cohort 3)—relative IDI (IDI/discrimination slope), and improvement in C statistic (Δ C statistic) after addition of the panel to the REVEAL equation. Model development was performed in cohorts 1 and 2 combined, and validation in cohort 4. NRI=net reclassification index. IDI=integrated discrimination improvement.

lyses (cohort 3)—relative IDI (IDI/discrimination slope), and improvement in C statistic (Δ C statistic) after addition of the panel to the REVEAL equation. Model development was performed in cohorts 1 and 2 combined, and validation in cohort 4. NRI=net reclassification index. IDI=integrated discrimination improvement. Discussion To our knowledge, this study is the first to apply high-throughput analysis of the plasma proteome to patients with idiopathic or heritable pulmonary arterial hypertension. The importance of robust statistical interrogation of novel biomarkers versus established criteria in risk prediction models has been emphasised before.20 We applied these methods extensively and identified nine proteins that predict survival independent of the established circulating prognostic factor, NT-proBNP. A panel score on the basis of plasma concentrations of these nine proteins, whereby a score increasing from 0 to 9 was associated with increased risk in an individual, improved clinical risk prediction based on NT-proBNP and the REVEAL prognostic equation.8 The protein panel was informative when used in incident or prevalent patients, and changes in the panel score after initiating therapy had clinical use by the identification of patients who had not responded to treatment. The protein panel improved model discrimination and reclassification without skewing calibration, and was validated in an independent cohort of patients from a separate expert centre, again independent of established clinical measurements.

inical use by the identification of patients who had not responded to treatment. The protein panel improved model discrimination and reclassification without skewing calibration, and was validated in an independent cohort of patients from a separate expert centre, again independent of established clinical measurements. BNP was one of the initial 20 prognostic proteins identified following discovery (cohort 1), validation (cohort 2), and random sampling analysis (cohorts 1 and 2), which gave us confidence in our approach. In the final analyses, a panel of nine proteins provided information independent of the most up-to-date risk equation incorporating many clinical variables. The protein components of this panel report on different pathways recognised in the pathophysiology of pulmonary arterial hypertension and collectively are more informative than when used individually. ST2, secreted in response to stretching myocardiocytes,21 is a potential biomarker in chronic heart failure,22 and circulating concentrations can also reflect inflammation.23 Increased TIMP expression and imbalance in matrix metalloproteinase activity is implicated in pulmonary vascular remodelling and disease progression in patients with pulmonary arterial hypertension.24 Pulmonary ApoE expression is reduced in patients with idiopathic pulmonary arterial hypertension,25 and in experimental models ApoE inhibits the proliferation of pulmonary artery smooth muscle cells and protects against the development of pulmonary arterial hypertension.26 The IGF-1 system is known to have a role in vascular pathologies, such as pulmonary arterial hypertension, and IGFBP-1 is one of a family of proteins modulating the effects of IGF-1 on vascular smooth muscle and endothelial cells.27 Increased complement factor D expression and loss of the inhibitory factor H predicts dysregulation of the complement system and overactivation of inflammation and innate immunity.28 Finally, increased EPO and reduced plasminogen concentrations might reflect the abnormal iron status29 and prothrombotic state30 of patients with idiopathic pulmonary arterial hypertension. The proteins identified in this study appear to have biological relevance to pulmonary arterial hypertension. None of the proteins identified in this study were the same as those identified, using the same platform, to predict cardiovascular risk in patients with stable coronary heart disease.17

arterial hypertension. The proteins identified in this study appear to have biological relevance to pulmonary arterial hypertension. None of the proteins identified in this study were the same as those identified, using the same platform, to predict cardiovascular risk in patients with stable coronary heart disease.17 Effective clinical management of patients with idiopathic or heritable pulmonary arterial hypertension requires the early recognition of patients who are failing to respond to treatment and need alternative or additional targeted therapies. The existing assessment of patients is dependent on subjective reporting of wellbeing to assign functional class, functional tests such as the 6-min walk test, which lacks specificity, and the availability of data from echocardiography.6 The nine-protein panel provides an objective measure that improves the prognostic accuracy of clinical evaluation.8 Biomarkers should always be used in clinical context, but it is relevant that the single addition of plasma NT-BNP concentration (a component of the REVEAL equation) as a continuous variable to the panel score performs as well as the panel score plus the REVEAL equation. This observation presents the possibility that the use of the protein panel with NT-BNP might provide a useful point-of-care test for the assessment and early referral of patients with idiopathic or heritable pulmonary arterial hypertension for specialist intervention.

well as the panel score plus the REVEAL equation. This observation presents the possibility that the use of the protein panel with NT-BNP might provide a useful point-of-care test for the assessment and early referral of patients with idiopathic or heritable pulmonary arterial hypertension for specialist intervention. This study represents a comprehensive analysis of the circulating proteome in patients with idiopathic or heritable pulmonary arterial hypertension, leading to the discovery of a panel of proteins capable of predicting mortality more accurately than established measurements. Although the number of proteins assayed represents a broad range of proteins with disparate functions, the proteins studied are limited by the aptamers developed for the assay. Pulmonary arterial hypertension is a rare disease, and the numbers of patients in this study preclude the analysis of interactions between proteins that might improve prognostication. The importance of changes in protein concentrations over time and in response to various therapeutic strategies requires further prospective study, with assessments of the effect of interval between samples. The heterogeneity of pulmonary arterial hypertension and of the patients studied in these cohorts (eg, in terms of age, gender, disease severity, and genetic background) means that the proteins and biological pathways identified in this study might not always be the most important in each individual with pulmonary arterial hypertension. We did the analyses on independent cohorts of patients recruited at three distinct international centres of expertise. The differences in occurrence of death were primarily because of different durations of follow-up and preclusion of long-term survivors in cohort 1 (2011–13) from cohort 2 (2002–11). The effect of different therapies on the concentration of the proteins was not studied and would require sampling of the same subjects before and after initiation of specific therapies at set timepoints. Fasting and dietary status and the method of blood sampling are known to affect proteomic measurements, but we did not study this here. We validated proteomic measurements with alternative experimental methods and similar results were observed in both EDTA-preserved and sodium citrate-preserved plasma. Blood samples were collected alongside routine clinical plasma samples, showing the practical deployment of this protein panel in a clinical setting.

e. We validated proteomic measurements with alternative experimental methods and similar results were observed in both EDTA-preserved and sodium citrate-preserved plasma. Blood samples were collected alongside routine clinical plasma samples, showing the practical deployment of this protein panel in a clinical setting. The cost of these blood protein measurements would be relatively low compared with more complex clinical procedures. The development of a dedicated assay to measure all nine proteins together would simplify its clinical use. The prognostic nine-protein panel score powerfully selects subgroups of patients that are likely to have events (death or transplantation), which could be beneficial for targeting aggressive therapeutic strategies or maximising the power of clinical trials. The proteins identified warrant mechanistic evaluation in addition to examination in other distinct disease groups, including other forms of pulmonary hypertension and cardiovascular disease. Supplementary Material Supplementary appendix

The prognostic nine-protein panel score powerfully selects subgroups of patients that are likely to have events (death or transplantation), which could be beneficial for targeting aggressive therapeutic strategies or maximising the power of clinical trials. The proteins identified warrant mechanistic evaluation in addition to examination in other distinct disease groups, including other forms of pulmonary hypertension and cardiovascular disease. Supplementary Material Supplementary appendix Acknowledgments This paper presents independent research that was supported by the National Institute for Health Research (NIHR)/Wellcome Trust Imperial Clinical Research Facility, at Imperial College Healthcare National Health Service (NHS) Trust, London, UK; the Sheffield NIHR Clinical Research Facility; and funding through the NIHR Rare Diseases Translational Research Collaboration. The views expressed are those of the authors and not necessarily those of the NHS, the NIHR, or the Department of Health. CJR was supported by an Imperial College Junior Research Fellowship and British Heart Foundation Intermediate Basic Science Research Fellowship (FS/15/59/31839). PG was supported by a Fellowship from the Wellcome Trust (103378/Z/13/Z). AL is supported by British Heart Foundation Project Grant (PG/11/11629288) and a British Heart Foundation Senior Basic Science Fellowship (FS/13/48/30453). MRW is supported by a British Heart Foundation programme grant (RG/10/16/28575). This work was supported in part by the Assistance Publique-Hôpitaux de Paris, Inserm, Université Paris-Sud, and Agence Nationale de la Recherche (Département Hospitalo-Universitaire Thorax Innovation; LabEx LERMIT, ANR-10-LABX-0033; and RHU BIO-ART LUNG 2020, ANR-15-RHUS-0002). We are indebted to Souad Ali for blood sample collection and to George Villas, Lavanya Ranganathan, and the TRIPHIC (Translational Research in Pulmonary Hypertension at Imperial College) system for the processing and pseudonymisation of patient information.

NR-10-LABX-0033; and RHU BIO-ART LUNG 2020, ANR-15-RHUS-0002). We are indebted to Souad Ali for blood sample collection and to George Villas, Lavanya Ranganathan, and the TRIPHIC (Translational Research in Pulmonary Hypertension at Imperial College) system for the processing and pseudonymisation of patient information. Contributors All authors collected data and gave constructive criticism of the study manuscript. CJR, JW, PG, HG, RTS, HAG, AL, MH, and MRW did the study design and interpretation, and wrote the manuscript. CJR analysed data.

NR-10-LABX-0033; and RHU BIO-ART LUNG 2020, ANR-15-RHUS-0002). We are indebted to Souad Ali for blood sample collection and to George Villas, Lavanya Ranganathan, and the TRIPHIC (Translational Research in Pulmonary Hypertension at Imperial College) system for the processing and pseudonymisation of patient information. Contributors All authors collected data and gave constructive criticism of the study manuscript. CJR, JW, PG, HG, RTS, HAG, AL, MH, and MRW did the study design and interpretation, and wrote the manuscript. CJR analysed data. Declaration of interests All authors declare no competing interests relevant to the submitted work. JSRG reports grants, personal fees, and other from Actelion Pharmaceuticals, personal fees and other from Bayer and GlaxoSmithKline (GSK), and personal fees from Bellerophon, Merck, Sharp, and Dohme (MSD), Pfizer, and AOP Orphan, outside of the submitted work. LSH reports personal fees and other from Actelion Pharmaceuticals, grants and personal fees from Bayer, and personal fees from MSD and GSK, outside of the submitted work. HAG reports personal fees from Actelion Pharmaceuticals, Bayer, GSK, Novartis, Pfizer, Bellerophon Pulse Technologies, MSD, and grants from Deutsche Forschungsgemeinschaft, outside of the submitted work. GS reports grants and personal fees from Actelion Pharmaceuticals, Bayer, and GSK, and personal fees from Pfizer and MSD, outside of the submitted work. MH reports personal fees from Actelion Pharmaceuticals and Pfizer, and grants and personal fees from Bayer and GSK, during the conduct of the study; and personal fees from Novartis, outside of the submitted work. OS reports grants, personal fees, and non-financial support from Actelion Pharmaceuticals, Bayer, and Merck, and grants and personal fees from GSK, outside of the submitted work. DM reports grants and personal fees from Actelion Pharmaceuticals and Bayer, and personal fees from GSK, Pfizer, Novartis, BMS, and MSD, outside of the submitted work. BG reports non-financial support from Actelion Pharmaceuticals, Bayer, and GSK, and grants and non-financial support from Pfizer, outside of the submitted work. DGK reports grants and personal fees from Actelion Pharmaceuticals, Bayer, and GSK, and personal fees from MSD, outside of the submitted work. RC reports personal fees from Actelion Pharmaceuticals, Bayer, and GSK, outside of the submitted work. CAE reports personal fees and grants from Actelion Pharmaceuticals and Bayer, personal fees from GSK, and grants from Pfizer, outside of the submitted work. AL reports grants from British Heart Foundation, Medical Research Council UK, and Novartis, and other from Actelion Pharmceuticals and Bayer Healthcare, outside of the submitted work.

fees and grants from Actelion Pharmaceuticals and Bayer, personal fees from GSK, and grants from Pfizer, outside of the submitted work. AL reports grants from British Heart Foundation, Medical Research Council UK, and Novartis, and other from Actelion Pharmceuticals and Bayer Healthcare, outside of the submitted work. MRW reports personal fees from Bayer and GSK received for scientific advice outside the scope of the submitted work.

Research in context Evidence before this study Chronic obstructive pulmonary disease (COPD) and asthma have been identified as important contributors to fatal and non-fatal disease burden in all iterations of the Global Burden of Disease study (GBD). Since the 1990s, two landmark epidemiological studies in asthma, the International Study of Asthma and Allergies in Childhood and the European Community Respiratory Health Survey, provided comparable evidence of the asthma prevalence in children and adults, respectively, but in a limited number of countries. Similarly, for COPD, international initiatives such as PREPOCOL, PLATINO, BOLD, IBERPOC, and EPI-SCAN used standardised population spirometry to quantify COPD and its severity. An absence of consensus on case definitions and other sources of measurement bias between data sources complicates their estimations. Added value of this study

Chronic obstructive pulmonary disease (COPD) and asthma have been identified as important contributors to fatal and non-fatal disease burden in all iterations of the Global Burden of Disease study (GBD). Since the 1990s, two landmark epidemiological studies in asthma, the International Study of Asthma and Allergies in Childhood and the European Community Respiratory Health Survey, provided comparable evidence of the asthma prevalence in children and adults, respectively, but in a limited number of countries. Similarly, for COPD, international initiatives such as PREPOCOL, PLATINO, BOLD, IBERPOC, and EPI-SCAN used standardised population spirometry to quantify COPD and its severity. An absence of consensus on case definitions and other sources of measurement bias between data sources complicates their estimations. Added value of this study In this study, we provide details on the methods used in GBD to minimise measurement error introduced by heterogeneous cause of death and prevalence data on COPD and asthma. We also provide an analysis of how sociodemographic development has a different effect on the burden of COPD and asthma. We show that mortality but not prevalence of asthma is strongly related to sociodemographic development. For COPD, the burden increases from low sociodemographic development to the mid-range of our Socio-demographic Index and decreases with increasing development, most likely through the pathways of exposure to smoking and environmental risks. We also present risk factor estimates and discuss potential new risks that can be added in future GBD iterations.

m low sociodemographic development to the mid-range of our Socio-demographic Index and decreases with increasing development, most likely through the pathways of exposure to smoking and environmental risks. We also present risk factor estimates and discuss potential new risks that can be added in future GBD iterations. Implications of all the available evidence COPD and asthma are important contributors to the burden of non-communicable diseases. Although much of the burden is either preventable or treatable with affordable interventions, these diseases have received less attention than other non-communicable diseases. Up-to-date population information on these diseases is key to policy making to improve access to and quality of existing intervention strategies.

lthough much of the burden is either preventable or treatable with affordable interventions, these diseases have received less attention than other non-communicable diseases. Up-to-date population information on these diseases is key to policy making to improve access to and quality of existing intervention strategies. Introduction Chronic respiratory diseases are among the leading causes of mortality and morbidity worldwide. Of all chronic respiratory diseases, chronic obstructive pulmonary disease (COPD) and asthma are the most common. These diseases ranked among the top 20 conditions causing disability globally and were ranked eighth (COPD) and 23rd (asthma) as causes of disease burden as measured by disability-adjusted life years (DALYs) in 2015.1, 2 Yet the measurement of mortality, prevalence, and other population indicators of these two diseases is complicated by misclassification and an absence of consensus about case definitions. Both death rates and prevalence of COPD steeply increase with age. The age pattern of asthma mortality resembles that of COPD rather than the relatively steady prevalence in adults seen in asthma surveys and health service encounter data. This difference in age patterns between cause of death and prevalence data sources has been attributed to a range of factors including the commonly reported misclassification of asthma in the elderly as COPD, variable and temporal effects of smoking, and an actual overlap of asthma and COPD (asthma COPD overlap; ACO).3, 4 However, no consensus exists on the definition of ACO to date.5 Also, evidence from a longitudinal study6 did not show a larger reduction in lung function in those patients with COPD and asthma than those without asthma, whereas others have challenged the concept of ACO altogether.7, 8

D (asthma COPD overlap; ACO).3, 4 However, no consensus exists on the definition of ACO to date.5 Also, evidence from a longitudinal study6 did not show a larger reduction in lung function in those patients with COPD and asthma than those without asthma, whereas others have challenged the concept of ACO altogether.7, 8 Spirometry is the fundamental tool used to define and stage COPD and, accordingly, establish population prevalence in surveys. Under the umbrella of the burden of obstructive lung disease (BOLD) initiative, surveys have been done in 29 countries, with surveys in a further nine countries still in progress.9 Two previous initiatives in five Colombian cities (PREPOCOL)10 and in five Latin American capital cities (PLATINO)11 provided more population estimates. Although all these studies used comparable methods, there is still no universal consensus about the thresholds of spirometry findings to define COPD.12, 13 The two dominant case definitions for airflow limitation compatible with COPD are a value of less than 0·70 for the ratio of FEV1 and forced vital capacity (FVC), or the lower limit of normal (LLN) method of deriving a threshold as the fifth percentile of FEV1:FVC in a healthy reference population.14 No universal LLN threshold exists because it is thought to vary between populations.15 Because most people identified with COPD based on spirometry findings report not having been diagnosed prior to survey, population screening and case-finding in symptomatic smokers have been suggested to provide an opportunity for smoking cessation interventions before the disease has progressed.16, 17

n populations.15 Because most people identified with COPD based on spirometry findings report not having been diagnosed prior to survey, population screening and case-finding in symptomatic smokers have been suggested to provide an opportunity for smoking cessation interventions before the disease has progressed.16, 17 Most surveys of asthma use a case definition based on self-report of a diagnosis of asthma by a physician and wheeze (with other respiratory symptoms) in the past 12 months.18 Others have suggested that wheezing symptoms in the past year and bronchial hyperresponsiveness to inhalation of methacholine or histamine that is reversible with a bronchodilator is a better case definition for clinically relevant asthma.19 This case definition has been used to measure asthma prevalence in a few surveys, but has not been universally adopted, partly for logistical reasons, but also because of concern about poor specificity and poor prediction of future risk of asthma in individuals without symptoms.20 However, the use of biological measurements to improve the validity of the asthma definition depends on the aim of the study. For instance, bronchial hyper-responsiveness has similar or better specificity, but much worse sensitivity, than symptom questionnaires, making it a less suitable method for the measurement of prevalence.21, 22 Misclassification and varying case definitions are commonly encountered in population health measurement.2 A key component of the Global Burden of Disease (GBD) analyses is to identify and correct for such sources of measurement bias.

Most surveys of asthma use a case definition based on self-report of a diagnosis of asthma by a physician and wheeze (with other respiratory symptoms) in the past 12 months.18 Others have suggested that wheezing symptoms in the past year and bronchial hyperresponsiveness to inhalation of methacholine or histamine that is reversible with a bronchodilator is a better case definition for clinically relevant asthma.19 This case definition has been used to measure asthma prevalence in a few surveys, but has not been universally adopted, partly for logistical reasons, but also because of concern about poor specificity and poor prediction of future risk of asthma in individuals without symptoms.20 However, the use of biological measurements to improve the validity of the asthma definition depends on the aim of the study. For instance, bronchial hyper-responsiveness has similar or better specificity, but much worse sensitivity, than symptom questionnaires, making it a less suitable method for the measurement of prevalence.21, 22 Misclassification and varying case definitions are commonly encountered in population health measurement.2 A key component of the Global Burden of Disease (GBD) analyses is to identify and correct for such sources of measurement bias. In this Article, we present the results of estimating mortality, prevalence, and disease burden in DALYs and years lived with disability (YLDs) for COPD and asthma from the GBD 2015 study. We also report on the attribution of risk factors for these diseases and the relation between disease burden and the Socio-demographic Index (SDI), a compound measure of income, years of education, and total fertility rate.

den in DALYs and years lived with disability (YLDs) for COPD and asthma from the GBD 2015 study. We also report on the attribution of risk factors for these diseases and the relation between disease burden and the Socio-demographic Index (SDI), a compound measure of income, years of education, and total fertility rate. Methods Mortality The methods of the GBD 2015 study have been extensively reported elsewhere.1, 2, 23 Briefly, deaths, incidence, prevalence, and DALY rates were estimated for 310 diseases and injuries for 195 countries and territories by age group and sex from 1990 to 2015. All-cause mortality was derived from vital registration systems, censuses, and surveys, and analysed with demographic methods to correct for incompleteness. Causes of death, derived from an extensive database of vital registration and verbal autopsy data, were analysed using GBD's Cause Of Death Ensemble modeling (CODEm) tool to calculate mixed effects or spatiotemporal Gaussian process regression models of rates or cause fractions with varying combinations of predictive covariates. Predictive validity testing determined the optimal ensemble of models. Covariates included smoking prevalence, cigarettes per capita, the proportion of the population exposed to household air pollution, mean exposure to ambient particulate matter (defined as the population-weighted annual average mass concentration of particles with a diameter less than 2·5 μm [PM2·5] in a m3 of air) from outdoor air pollution, a scalar of the combined exposure to risks for COPD (and asthma), and SDI. Because the sensitivity of verbal autopsy algorithms to detect specific chronic respiratory diseases is poor, we only modelled data on deaths from all chronic respiratory diseases in CODEm and constrained the estimates for specific chronic respiratory diseases to the estimates for all chronic respiratory deaths. We constrained estimates for all individual causes to the all-cause mortality rates derived from demographic estimation.

nly modelled data on deaths from all chronic respiratory diseases in CODEm and constrained the estimates for specific chronic respiratory diseases to the estimates for all chronic respiratory deaths. We constrained estimates for all individual causes to the all-cause mortality rates derived from demographic estimation. Non-fatal estimation for COPD Non-fatal estimates for COPD were based on systematic reviews of published papers, unpublished reports, surveys available in GBD's Global Health Data Exchange repository, and health service encounter data from the USA (coded in International Classification of Diseases [ICD]-9 to 490–492, 494, and 496). We used 7301 prevalence datapoints and 22 incidence datapoints covering 15 of 21 GBD world regions. No data were available for Andean Latin America, the Caribbean, central Asia, central and east sub-Saharan Africa, and Oceania. We used the Global Initiative for Chronic Obstructive Pulmonary Disease (GOLD) spirometry-based definition for COPD (a ratio of FEV1:FVC <0·70 after bronchodilation)14 and modelled overall prevalence and the proportions in COPD spirometry stages mild (FEV1 ≥80% of normal), moderate (FEV1 50–79% of normal), and severe or very severe combined (FEV1 <50% of normal) in DisMod-MR 2.1, a Bayesian meta-regression tool. DisMod-MR 2.1 takes all available data on prevalence, incidence, remission (defined in GBD as the cure rate), and cause of death rates jointly into account and forces a consistent set of estimates for each parameter. Before entering data into DisMod-MR 2.1, we adjusted survey data using different spirometry case definitions. We adjusted datapoints from 14 studies reporting on the GOLD case definition without a bronchodilator after fitting an exponential curve to age-specific ratios of both measurements from three studies.24, 25, 26 Using a similar approach, we adjusted datapoints from six studies reporting LLN pre-bronchodilator data based on one study,24 three studies with LLN post-bronchodilator data based on five studies,12, 24, 27, 28, 29 and two studies using an older version of LLN by the European Respiratory Society based on two studies.30, 31 We used the meta-regression component of DisMod-MR 2.1 to determine an adjustment factor for data based on physician diagnosis and the US health service encounter data. We included a scalar for the combined exposure to all risks estimated for COPD as a predictive covariate.

spiratory Society based on two studies.30, 31 We used the meta-regression component of DisMod-MR 2.1 to determine an adjustment factor for data based on physician diagnosis and the US health service encounter data. We included a scalar for the combined exposure to all risks estimated for COPD as a predictive covariate. We included corresponding datapoints for excess mortality rate estimated as the ratio of cause-specific mortality rate and prevalence corresponding to the same year and age range of the datapoint. We used lag-distributed income per capita as a predictive covariate for excess mortality, forcing a negative coefficient on the assumption that case fatality decreases with increasing wealth in a country. Prevalence by GOLD class was available from only 24 countries in 14 GBD world regions.

age range of the datapoint. We used lag-distributed income per capita as a predictive covariate for excess mortality, forcing a negative coefficient on the assumption that case fatality decreases with increasing wealth in a country. Prevalence by GOLD class was available from only 24 countries in 14 GBD world regions. The proportions of people in GOLD classes I, II, and III or IV were modelled separately in DisMod-MR 2.1 and then scaled to a sum of 1 and multiplied by the overall prevalence of COPD. In GBD, severity of COPD is classified into health states (appendix p 4). To map the prevalence by GOLD class into health states representing symptoms, we used the Medical Expenditure Panel Survey (MEPS) data32 for 2001–11 from the USA. MEPS is an ongoing data collection project with new panels recruited every 2 years. Respondents report on all health service contacts and the reasons for those contacts. We identified individuals with an ICD-9 diagnosis of COPD. We translated scores from a generic quality-of-life instrument, the 12-Item Short Form Health Survey (SF-12),33 into GBD disability weight values based on convenience samples of research fellows at the Institute for Health Metrics and Evaluation and annual GBD workshop participants filling in SF-12 for a selection of 60 of the 235 health states used in GBD 2015. Health states were presented as lay descriptions that had been the basis of the pairwise comparisons presented to respondents to the GBD disability weight surveys. After controlling for comorbidity, we assigned a specific disability weight to each individual with a diagnosis of COPD. We categorised cases into asymptomatic (disability weight value of 0), mild COPD (disability weight value between 0 and the midpoint of GBD disability weights for mild and moderate COPD), moderate COPD (disability weight value greater than the midpoint between mild and moderate and midpoint between moderate and severe COPD disability weights), and severe COPD (the remainder). We took the prevalence estimates for the USA in 2005 (at the midpoint of MEPS data range) and mapped the distribution of cases by GOLD classes into the distribution of severity from MEPS (appendix p 5). This gave us a mapping from GOLD class into GBD health states, which could then be applied to the prevalence data by GOLD class from all other countries and time periods.

(at the midpoint of MEPS data range) and mapped the distribution of cases by GOLD classes into the distribution of severity from MEPS (appendix p 5). This gave us a mapping from GOLD class into GBD health states, which could then be applied to the prevalence data by GOLD class from all other countries and time periods. Non-fatal estimation for asthma The main data sources for asthma were population surveys and US health service encounter data on the diagnoses for any health service contact for 42 million people. We used 9219 prevalence, 29 incidence, and 32 remission datapoints and population death rates from asthma estimated in CODEm and scaled to total death rates with all other cause-specific estimates. Data on prevalence were available for 121 countries covering all 21 GBD world regions. Our case definition for asthma was a reported diagnosis by a physician, with wheezing in the past 12 months. In DisMod-MR 2.1, we adjusted data based on reported wheezing only and US health service encounter data, and used a scalar of the combined exposure to risk factors for asthma. Similar to the COPD model, we added excess mortality rates corresponding to all prevalence datapoints with lag-distributed income per capita as a predictive covariate. The health states and disability weights for three asthma health states are listed in the appendix (p 4). The distribution between the three asthma health states and an asymptomatic health state was analysed in MEPS. In the absence of comparable epidemiological severity distribution data, a simplifying assumption had to be made that the US distribution of severity for asthma can be generalised to all countries. Additional details on the estimation process for COPD and asthma can be found in the appendix (pp 15–28).

nalysed in MEPS. In the absence of comparable epidemiological severity distribution data, a simplifying assumption had to be made that the US distribution of severity for asthma can be generalised to all countries. Additional details on the estimation process for COPD and asthma can be found in the appendix (pp 15–28). Risk estimation Estimates were made of six risk factors for COPD (smoking, second-hand smoke, household air pollution, ambient particulate matter, ozone, and occupational particulates) and two risk factors for asthma (smoking and occupational asthmagens). Sufficient evidence of causality, availability of exposure data, potential for modification, and policy interest are criteria for choosing risks and associated outcomes in GBD. Population-attributable fractions of disease outcomes were estimated from exposure data, relative risks of outcomes, and a theoretical minimum level of exposure. Population surveys were the main source of exposure data on smoking, second-hand smoke, and household air pollution. Exposure to PM2·5 was measured from satellite data on aerosols in the atmosphere and calibrated to observations from ground monitors. We based exposure to ozone on a chemical transport model of satellite data.34 Occupational exposures were based on the proportion of the working population exposed to asthmagens and particulates based on distribution of the population in nine occupational groups as reported by the International Labor Organization.35 We derived relative risks from meta-analyses of cohort studies. The theoretical minimum exposure level was set as zero for smoking, second-hand smoke, and the occupational exposures. For household air pollution, the minimum was defined as no household reporting use of solid fuel for cooking. For ambient particulate matter, the minimum was set as a uniform distribution between the lowest and fifth percentile exposure level from all data sources. For ozone, the minimum was set as a uniform distribution between the lowest and fifth percentile exposure measured in the American Cancer Society's Cancer Prevention Study II.36 Unlike disease estimates that are mutually exclusive and collectively exhaustive in GBD, risk estimates are based on a counterfactual analysis (what if past exposure to a risk had been at the theoretical minimum level?) and are, therefore, not additive.

red in the American Cancer Society's Cancer Prevention Study II.36 Unlike disease estimates that are mutually exclusive and collectively exhaustive in GBD, risk estimates are based on a counterfactual analysis (what if past exposure to a risk had been at the theoretical minimum level?) and are, therefore, not additive. Estimates of combinations of risks take mediation into account based on the difference in relative risks from cohort and trial data that did and did not control for another risk as a confounder. After adjustment for mediation, risks were combined using a multiplicative function to avoid the sum of risks exceeding the total amount of disease.23 Additional details on the estimation process for COPD and asthma risks can be found in the appendix (pp 29–57).

did and did not control for another risk as a confounder. After adjustment for mediation, risks were combined using a multiplicative function to avoid the sum of risks exceeding the total amount of disease.23 Additional details on the estimation process for COPD and asthma risks can be found in the appendix (pp 29–57). DALY estimation We calculated years of life lost (YLLs) by multiplying the number of deaths for a cause by the remaining life expectancy in GBD's standard life table based on the lowest observed mortality rates at each age in any population over 5 million.1 We calculated YLDs by multiplying the prevalence of each sequela by the disability weight that quantifies the relative severity of the sequela on a scale between 0 and 1. We derived disability weights from nine population surveys and an open-access internet survey using pairwise comparison methods.37 DALYs are the sum of YLLs and YLDs. We estimated uncertainty by recalculating every outcome of interest 1000 times, drawing from distributions of the sampling error around input data, corrections for measurement error, and estimates of residual non-sampling error and, in the case of cause of death estimates, model selection. Uncertainty intervals (UIs) were defined as the 25th and 975th values of the posterior distributions. We computed differences between estimates at the 1000-draw level and reported them as significant if more than 95% of values for the difference were either positive or negative. We computed age-standardised rates using the GBD standard population.1

re defined as the 25th and 975th values of the posterior distributions. We computed differences between estimates at the 1000-draw level and reported them as significant if more than 95% of values for the difference were either positive or negative. We computed age-standardised rates using the GBD standard population.1 SDI is an index of sociodemographic development consisting of lagged distributed income per capita, mean years of education over the age of 15 years, and total fertility rate.1 Each component was given equal weight and rescaled from 0 (for the lowest value observed during 1980–2015) to 1 (for the highest value observed) for income per capita and average years of schooling, and the reverse for the total fertility rate. The final SDI score was computed as the geometric mean of each of the components. We classified countries into five quintiles based on the entire distribution of location-year combinations between 1980 and 2015. We present results on each country's position based on its 2015 SDI value. A LOESS regression on all data from 1980 to 2015 was done to define the expected relationship between SDI and each health outcome. We contrast observed disease rates against this expected level to identify world regions performing better or worse than expected based on their development status. This study is compliant with the Guidelines for Accurate and Transparent Health Estimates Reporting (GATHER) with details provided in the appendix (pp 58–60).38

SDI is an index of sociodemographic development consisting of lagged distributed income per capita, mean years of education over the age of 15 years, and total fertility rate.1 Each component was given equal weight and rescaled from 0 (for the lowest value observed during 1980–2015) to 1 (for the highest value observed) for income per capita and average years of schooling, and the reverse for the total fertility rate. The final SDI score was computed as the geometric mean of each of the components. We classified countries into five quintiles based on the entire distribution of location-year combinations between 1980 and 2015. We present results on each country's position based on its 2015 SDI value. A LOESS regression on all data from 1980 to 2015 was done to define the expected relationship between SDI and each health outcome. We contrast observed disease rates against this expected level to identify world regions performing better or worse than expected based on their development status. This study is compliant with the Guidelines for Accurate and Transparent Health Estimates Reporting (GATHER) with details provided in the appendix (pp 58–60).38 Role of the funding source This research was supported by funding from the Bill & Melinda Gates Foundation. The funders had no role in the study design, data collection and analysis, interpretation of data, decision to publish, or preparation of the manuscript.

test decrease in age-standardised prevalence was seen in countries in the high-middle-SDI quintile and the middle-SDI quintile (table 1).Table 1 Deaths due to asthma and COPD and number of prevalent cases of disease in 2015 and percentage change in all-age and age-standardised rates in locations grouped by SDI quintile Number of deaths (thousands) Percentage change in all-age deaths, 1990–2015 Percentage change in age-standardised death rates, 1990–2015 Number of prevalent cases (thousands) Percentage change in all-age prevalence, 1990–2015 Percentage change in age-standardised prevalence, 1990–2015 COPD Global 3188 (3084 to 3293) 11·6 (5·3 to 19·8) −41·9 (−45·1 to −37·7) 174 483 (160 205 to 188 952) 44·2 (41·7 to 46·6) −14·7 (−15·9 to −13·5) High SDI quintile 482 (468 to 505) 31·6 (27·8 to 38·2) −26·2 (−28·2 to −22·5) 43 105 (39 912 to 46 414) 35·3 (31·8 to 39·1) −7·3 (−9·3 to −4·9) High-middle SDI quintile 626 (602 to 651) −11·1 (−17·4 to −4·3) −57·8 (−60·8 to −54·7) 44 923 (41 215 to 48 803) 42·3 (39·3 to 45·1) −20·2 (−21·5 to −19·0) Middle SDI quintile 1110 (1055 to 1169) −3·4 (−11·0 to 5·8) −53·5 (−57·2 to −49·2) 52 209 (47 430 to 57 154) 104·8 (101·6 to 108·0) −22·6 (−23·9 to −21·4) Low-middle SDI quintile 907 (850 to 965) 51·5 (24·1 to 89·5) −25·7 (−38·1 to −7·9) 30 058 (27 495 to 32 719) 74·3 (71·6 to 77·2) −3·7 (−4·8 to −2·7) Low SDI quintile 61 (52 to 71) 68·8 (40·3 to 111·3) −16·3 (−30·1 to 3·5) 4223 (3795 to 4656) 36·1 (33·2 to 38·9) −1·6 (−3·1 to −0·1) Asthma Global 397 (363 to 439) −26·7 (−43·7 to 7·2) −58·8 (−69·0 to −39·0) 358 198 (323 134 to 393 466) 12·6 (9·0 to 16·4) −17·7 (−19·9 to −15·1) High SDI quintile 22 (20 to 24) −53·2 (−56·8 to −48·9) −71·8 (−73·7 to −69·5) 63 883 (59 724 to 68 309) −13·8 (−17·0 to −10·2) −26·0 (−28·4 to −23·0) High-middle SDI quintile 54 (50 to 61) −3·2 (−12·9 to 9·0) −49·4 (−54·9 to −43·0) 76 935 (69 650 to 84 654) 8·4 (4·1 to 13·0) −15·2 (−18·0 to −12·1) Middle SDI quintile 120 (110 to 132) −12·2 (−28·7 to 16·0) −38·4 (−49·5 to −19·4) 91 375 (82 505 to 100 370) 8·3 (4·2 to 12·6) −14·5 (−16·3 to −12·3) Low-middle SDI quintile 159 (136 to 186) −40·5 (−61·4 to 19·0) −69·6 (−81·3 to −32·9) 90 605 (79 887 to 101 371) 28·9 (24·6 to 33·4) −18·4 (−20·7 to −15·9) Low SDI quintile 41 (34 to 51) 22·1 (2·1 to 55·4) −54·3 (−63·9 to −38·3) 35 011 (30 065 to 40 255) 94·8 (86·2 to 106·1) −7·3 (−11·2 to −3·0) Data in parentheses are 95% uncertainty intervals.

(136 to 186) −40·5 (−61·4 to 19·0) −69·6 (−81·3 to −32·9) 90 605 (79 887 to 101 371) 28·9 (24·6 to 33·4) −18·4 (−20·7 to −15·9) Low SDI quintile 41 (34 to 51) 22·1 (2·1 to 55·4) −54·3 (−63·9 to −38·3) 35 011 (30 065 to 40 255) 94·8 (86·2 to 106·1) −7·3 (−11·2 to −3·0) Data in parentheses are 95% uncertainty intervals. SDI is calculated for each location (all 188 countries, seven territories, and 519 subnational locations estimated in GBD 2015) as a function of lag-distributed income per capita, average educational attainment in the population aged over 15 years, and the total fertility rate. SDI of 0 represents the lowest level of income per capita, educational attainment, and highest total fertility rate observed from 1980 to 2015, and SDI of 1 represents the highest income per capita, educational attainment, and lowest total fertility rate with an effect on health over the same period. Cutoffs on the SDI scale for the quintiles have been selected based on their 2015 values by location. COPD=chronic obstructive pulmonary disease. SDI=Socio-demographic Index.

represents the highest income per capita, educational attainment, and lowest total fertility rate with an effect on health over the same period. Cutoffs on the SDI scale for the quintiles have been selected based on their 2015 values by location. COPD=chronic obstructive pulmonary disease. SDI=Socio-demographic Index. In 2015, 0·40 million people (95% UI 0·36 million to 0·44 million) died from asthma, a decrease of 26·7% (−7·2 to 43·7) compared with 1990. The decrease in age-standardised death rates was 58·8% (39·0–69·0) between 1990 and 2015. The greatest reduction in age-standardised death rates occurred in countries in the high-SDI and low-middle-SDI quintiles. From 1990 to 2015, the prevalence of asthma increased by 12·6% (9·0–16·4) to 358·2 million individuals (323·1 million to 393·5 million). The decrease in age-standardised prevalence by 17·7% (15·1–19·9) was smaller than the overall decrease in age-standardised death rates. The age-standardised death rate for asthma in 2015 was higher in males (6·7 [5·9–7·5] per 100 000 people) than in females (5·6 [4·8–6·4] per 100 000 people). A greater reduction in age-standardised prevalence was seen in countries in the high-SDI and low-middle-SDI quintiles (table 1).

in age-standardised death rates. The age-standardised death rate for asthma in 2015 was higher in males (6·7 [5·9–7·5] per 100 000 people) than in females (5·6 [4·8–6·4] per 100 000 people). A greater reduction in age-standardised prevalence was seen in countries in the high-SDI and low-middle-SDI quintiles (table 1). Globally, COPD affected 104·7 million males (95% UI 96·0 million to 113·8 million) and 69·7 million females (64·2 million to 75·4 million) in 2015. Age-standardised prevalence was 3·2% (2·9–3·5) in males and 2·0% (1·8–2·1) in females. Age-standardised DALY rates in males (1273·0 [95% UI 1215·5–1328·3] per 100 000 people) were almost twice as high as those in females (717·4 [677·7–759·3] per 100 000 people) reflecting a higher male-to-female ratio for deaths than for prevalence. Conversely, age-standardised DALY rates due to asthma were similar between male individuals (365 [290–451] per 100 000 people) and female individuals (368 [286–461] per 100 000 people). In 2015, more females (190·2 million [172·2 million to 208·9 million]) than males (168·0 million [150·8 million to 185·1 million]) had asthma; a reversal of the higher male-to-female ratio during adolescence.

male individuals (365 [290–451] per 100 000 people) and female individuals (368 [286–461] per 100 000 people). In 2015, more females (190·2 million [172·2 million to 208·9 million]) than males (168·0 million [150·8 million to 185·1 million]) had asthma; a reversal of the higher male-to-female ratio during adolescence. YLLs contributed more than 80% of DALYs due to COPD. Conversely, asthma is highly prevalent at all ages and leads to fewer deaths than COPD and thus YLDs formed the larger component of DALYs, at just over 60%. The 63·9 million DALYs (95% UI 61·2 million to 66·3 million) due to COPD represented 2·6% (95% UI 2·4–2·8) of the entire global burden of disease in 2015. 26·2 million DALYs (20·5 million to 32·6 million) due to asthma contributed 1·1% (0·9–1·3) of the total burden in 2015 (table 2). The greatest decrease in age-standardised DALY rates due to COPD occurred in countries in the high-middle-SDI and middle-SDI quintiles. The biggest reduction in age-standardised asthma DALY rates occurred in the low-middle-SDI quintile (table 2).Table 2 YLLs, YLDs, and DALYs due to asthma and COPD in 2015 and percentage change in all-age counts and age-standardised DALY rates from 1990 to 2015 in locations grouped by SDI quintiles

middle-SDI quintiles. The biggest reduction in age-standardised asthma DALY rates occurred in the low-middle-SDI quintile (table 2).Table 2 YLLs, YLDs, and DALYs due to asthma and COPD in 2015 and percentage change in all-age counts and age-standardised DALY rates from 1990 to 2015 in locations grouped by SDI quintiles Number of YLLs, all ages (thousands) Number of YLDs, all ages (thousands) Number of DALYs, all ages (thousands) Percentage change in DALYs, 1990–2015, all ages Percentage change in age-standardised DALY rates, 1990–2015 COPD Global 51 803 (49 898 to 53 611) 12 047 (10 207 to 13 725) 63 850 (61 215 to 66 289) −1·0 (−7·1 to 6·2) −43·7 (−47·0 to −39·8) High SDI quintile 5914 (5762 to 6180) 2214 (1890 to 2545) 8128 (7755 to 8530) 12·7 (9·9 to 16·6) −28·2 (−30·1 to −25·9) High-middle SDI quintile 9058 (8693 to 9446) 2500 (2103 to 2882) 11 661 (11 093 to 12 226) −19·8 (−25·0 to −14·2) −58·5 (−61·2 to −55·6) Middle SDI quintile 17 918 (16 979 to 18 887) 4050 (3422 to 4623) 21 812 (20 738 to 22 908) −16·4 (−22·3 to −9·2) −55·8 (−58·9 to −52·1) Low-middle SDI quintile 17 444 (16 260 to 18 652) 2954 (2493 to 3345) 20 399 (19 079 to 21 673) 32·0 (8·0 to 61·0) −27·0 (−40·0 to −10·6) Low SDI quintile 1433 (1203 to 1685) 374 (316 to 429) 1806 (1559 to 2064) 55·7 (31·3 to 86·9) −18·0 (−30·5 to −0·5) Asthma Global 10 270 (9369 to 11 448) 15 899 (10 371 to 22 344) 26 169 (20 501 to 32 583) −14·6 (−26·0 to 2·1) −42·8 (−52·0 to −29·5) High SDI quintile 384 (366 to 408) 2818 (1838 to 3905) 3203 (2221 to 4299) −25·4 (−29·9 to −21·9) −35·9 (−40·2 to −32·6) High-middle SDI quintile 1227 (1130 to 1400) 3419 (2228 to 4795) 4766 (3508 to 6154) −3·8 (−9·1 to 0·9) −30·3 (−35·6 to −25·9) Middle SDI quintile 2912 (2655 to 3223) 4061 (2657 to 5704) 6855 (5464 to 8453) −12·5 (−22·9 to −0·7) −40·7 (−49·3 to −30·2) Low-middle SDI quintile 4327 (3728 to 5073) 4020 (2621 to 5682) 8350 (6705 to 10 088) −27·4 (−44·6 to 9·1) −60·8 (−71·8 to −31·9) Low SDI quintile 1402 (1162 to 1692) 1563 (1009 to 2238) 2961 (2343 to 3687) 45·9 (27·2 to 67·9) −31·4 (−41·2 to −18·4) Data in parentheses are 95% uncertainty intervals. YLLs=years of life lost. YLDs=years lived with disability. DALYs=disability-adjusted life years. COPD=chronic obstructive pulmonary disease. SDI=Socio-demographic Index.

2 (1162 to 1692) 1563 (1009 to 2238) 2961 (2343 to 3687) 45·9 (27·2 to 67·9) −31·4 (−41·2 to −18·4) Data in parentheses are 95% uncertainty intervals. YLLs=years of life lost. YLDs=years lived with disability. DALYs=disability-adjusted life years. COPD=chronic obstructive pulmonary disease. SDI=Socio-demographic Index. Age-standardised DALY rates due to COPD in 2015 were estimated to exceed 2000 per 100 000 people in Papua New Guinea, India, Lesotho, and Nepal. Rates below 300 per 100 000 people were seen in some countries in high-income Asia Pacific, central Europe, north Africa and Middle East, the Caribbean, western Europe, and Andean Latin America (figure 1). Age-standardised asthma DALY rates in excess of 1200 per 100 000 people were estimated for Afghanistan, Central African Republic, Fiji, Kiribati, Lesotho, Papua New Guinea, and Swaziland. Countries in eastern and central Europe, China, Italy, and Japan had asthma DALY rates between 100 and 200 per 100 000 people (figure 2). DALY estimates for COPD and asthma by country and the percentage change in DALYs and age-standardised DALY rates between 1990 and 2015 are presented in the appendix (p 61–67).Figure 1 Age-standardised DALY rate per 100 000 people due to COPD by country, both sexes, 2015 DALYs=disability-adjusted life years. COPD=chronic obstructive pulmonary disease. ATG=Antigua and Barbuda. FSM=Federated States of Micronesia. Isl=islands. LCA=Saint Lucia. TLS=Timor-Leste. TTO=Trinidad and Tobago. VCT=Saint Vincent and the Grenadines.

Age-standardised DALY rates due to COPD in 2015 were estimated to exceed 2000 per 100 000 people in Papua New Guinea, India, Lesotho, and Nepal. Rates below 300 per 100 000 people were seen in some countries in high-income Asia Pacific, central Europe, north Africa and Middle East, the Caribbean, western Europe, and Andean Latin America (figure 1). Age-standardised asthma DALY rates in excess of 1200 per 100 000 people were estimated for Afghanistan, Central African Republic, Fiji, Kiribati, Lesotho, Papua New Guinea, and Swaziland. Countries in eastern and central Europe, China, Italy, and Japan had asthma DALY rates between 100 and 200 per 100 000 people (figure 2). DALY estimates for COPD and asthma by country and the percentage change in DALYs and age-standardised DALY rates between 1990 and 2015 are presented in the appendix (p 61–67).Figure 1 Age-standardised DALY rate per 100 000 people due to COPD by country, both sexes, 2015 DALYs=disability-adjusted life years. COPD=chronic obstructive pulmonary disease. ATG=Antigua and Barbuda. FSM=Federated States of Micronesia. Isl=islands. LCA=Saint Lucia. TLS=Timor-Leste. TTO=Trinidad and Tobago. VCT=Saint Vincent and the Grenadines. Figure 2 Age-standardised DALY rate per 100 000 people due to asthma, by country, both sexes, 2015 DALYs=disability-adjusted life years. ATG=Antigua and Barbuda. FSM=Federated States of Micronesia. Isl=islands. LCA=Saint Lucia. TLS=Timor-Leste. TTO=Trinidad and Tobago. VCT=Saint Vincent and the Grenadines.

DALYs=disability-adjusted life years. COPD=chronic obstructive pulmonary disease. ATG=Antigua and Barbuda. FSM=Federated States of Micronesia. Isl=islands. LCA=Saint Lucia. TLS=Timor-Leste. TTO=Trinidad and Tobago. VCT=Saint Vincent and the Grenadines. Figure 2 Age-standardised DALY rate per 100 000 people due to asthma, by country, both sexes, 2015 DALYs=disability-adjusted life years. ATG=Antigua and Barbuda. FSM=Federated States of Micronesia. Isl=islands. LCA=Saint Lucia. TLS=Timor-Leste. TTO=Trinidad and Tobago. VCT=Saint Vincent and the Grenadines. Examining the expected relationship between SDI and all-age DALY rates showed a reduction in asthma rates with increasing SDI in both sexes, whereas DALY rates due to COPD increased up until around 0·5 SDI, decreased to the lowest values at an SDI value of 0·75, after which they slightly increased (figure 3). These patterns reflect a combined effect of population growth, ageing, and variation in prevalence. The change in age-standardised DALY rates with SDI shows an increase in DALY rates due to COPD until the middle range of SDI values and then a sharp decline. DALY rates due to asthma in both sexes decreased monotonically with rising SDI (figure 4). The relationship between DALY rates due to asthma and SDI largely reflected variation in YLLs, whereas DALY rates due to COPD varied similarly for YLLs and YLDs across the SDI continuum (figure 5).Figure 3 Expected relationship between all-age DALY rates due to COPD and asthma and SDI by sex, 2015

ising SDI (figure 4). The relationship between DALY rates due to asthma and SDI largely reflected variation in YLLs, whereas DALY rates due to COPD varied similarly for YLLs and YLDs across the SDI continuum (figure 5).Figure 3 Expected relationship between all-age DALY rates due to COPD and asthma and SDI by sex, 2015 DALYs=disability-adjusted life years. COPD=chronic obstructive pulmonary disease. SDI=Socio-demographic Index. Figure 4 Expected relationship between age-standardised DALY rates due to COPD and asthma and SDI by sex, 2015 DALYs=disability-adjusted life years. COPD=chronic obstructive pulmonary disease. SDI=Socio-demographic Index. Figure 5 Expected relationship between age-standardised DALY rates due to COPD and asthma and SDI by YLLs and YLDs, 2015 DALYs=disability-adjusted life years. COPD=chronic obstructive pulmonary disease. SDI=Socio-demographic Index. YLLs=years of life lost. YLDs=years lived with disability.

DALYs=disability-adjusted life years. COPD=chronic obstructive pulmonary disease. SDI=Socio-demographic Index. Figure 5 Expected relationship between age-standardised DALY rates due to COPD and asthma and SDI by YLLs and YLDs, 2015 DALYs=disability-adjusted life years. COPD=chronic obstructive pulmonary disease. SDI=Socio-demographic Index. YLLs=years of life lost. YLDs=years lived with disability. The GBD regions of Oceania, east Asia, south Asia, and high-income North America had higher age-standardised DALY rates due to COPD in both sexes than expected based on their SDI. Male individuals in eastern Europe also had higher than expected DALY rates. Regions with better-than-expected COPD DALY rates included eastern, central, and western sub-Saharan Africa; central and Andean Latin America and the Caribbean; and north Africa and the Middle East (figure 6). Age-standardised DALY rates due to asthma in Oceania were much higher than expected based on SDI. Australasia, southeast Asia, the Caribbean, and southern sub-Saharan Africa also had higher DALY rates than expected. The asthma DALY rates in south Asia were higher than expected in 1990 (when SDI was lowest), but converged with expected values in 2015. Central Europe, east Asia, and western and eastern sub-Saharan Africa had lower than expected asthma DALY rates (figure 7).Figure 6 Age-standardised DALY rates due to COPD by 21 GBD world regions and the expected value based on the SDI by sex, 1990–2015

(when SDI was lowest), but converged with expected values in 2015. Central Europe, east Asia, and western and eastern sub-Saharan Africa had lower than expected asthma DALY rates (figure 7).Figure 6 Age-standardised DALY rates due to COPD by 21 GBD world regions and the expected value based on the SDI by sex, 1990–2015 The black line represents the expected value of a disease rate based on a LOESS regression of all years of estimates by GBD locations and their SDI value. DALYs=disability-adjusted life years. COPD=chronic obstructive pulmonary disease. GBD=Global Burden of Disease. SDI=Socio-demographic Index. LOESS=locally weighted regression and smoothing scatterplots. Figure 7 Age-standardised DALY rates due to asthma by 21 GBD world regions and the expected value based on the SDI by sex, 1990–2015 The black line represents the expected value of a disease rate based on a LOESS regression of all years of estimates by GBD locations and their SDI value. DALYs=disability-adjusted life years. GBD=Global Burden of Disease. SDI=Socio-demographic Index. LOESS=locally weighted regression and smoothing scatterplots.

Figure 7 Age-standardised DALY rates due to asthma by 21 GBD world regions and the expected value based on the SDI by sex, 1990–2015 The black line represents the expected value of a disease rate based on a LOESS regression of all years of estimates by GBD locations and their SDI value. DALYs=disability-adjusted life years. GBD=Global Burden of Disease. SDI=Socio-demographic Index. LOESS=locally weighted regression and smoothing scatterplots. Smoking and ambient particulate matter were the main risks for COPD followed by household air pollution, occupational particulates, ozone, and second-hand smoke (figure 8). Together, these risks explained 73·3% (95% UI 65·8–80·1) of DALYs due to COPD. Smoking and occupational asthmagens were the only risks quantified for asthma in GBD, explaining just 16·5% (14·6–18·7) of the asthma DALYs. The contribution of risks to the burden of COPD varied by SDI quintiles. In high-SDI countries, the behavioural risks (smoking and second-hand smoke) were the most important, whereas environmental risks and, to a lesser extent, occupational risks explained most of the burden in lower-SDI quintiles. The proportions of COPD burden not explained by any of the GBD risks showed little variation between SDI quintiles (figure 9, table 3).Figure 8 Age-standardised DALY rates due to COPD and asthma attributable to seven risk factors, both sexes, 2015 COPD=chronic obstructive pulmonary disease. DALYs=disability-adjusted life years.

Smoking and ambient particulate matter were the main risks for COPD followed by household air pollution, occupational particulates, ozone, and second-hand smoke (figure 8). Together, these risks explained 73·3% (95% UI 65·8–80·1) of DALYs due to COPD. Smoking and occupational asthmagens were the only risks quantified for asthma in GBD, explaining just 16·5% (14·6–18·7) of the asthma DALYs. The contribution of risks to the burden of COPD varied by SDI quintiles. In high-SDI countries, the behavioural risks (smoking and second-hand smoke) were the most important, whereas environmental risks and, to a lesser extent, occupational risks explained most of the burden in lower-SDI quintiles. The proportions of COPD burden not explained by any of the GBD risks showed little variation between SDI quintiles (figure 9, table 3).Figure 8 Age-standardised DALY rates due to COPD and asthma attributable to seven risk factors, both sexes, 2015 COPD=chronic obstructive pulmonary disease. DALYs=disability-adjusted life years. Figure 9 Contribution of behavioural and environmental and occupational risks to DALYs due to COPD per 100 000 people in locations grouped by SDI quintiles, 2015 Environmental and occupational: ambient particulate matter, household air pollution, occupational particulates, and ozone. Behavioural: smoking and second-hand smoke. Behavioural environmental: ambient particulate matter, household air pollution, occupational particulates, ozone, smoking, and second-hand smoke. DALYs=disability-adjusted life years. COPD=chronic obstructive pulmonary disease. SDI=Socio-demographic Index.

culates, and ozone. Behavioural: smoking and second-hand smoke. Behavioural environmental: ambient particulate matter, household air pollution, occupational particulates, ozone, smoking, and second-hand smoke. DALYs=disability-adjusted life years. COPD=chronic obstructive pulmonary disease. SDI=Socio-demographic Index. Table 3 Proportional contribution of behavioural and environmental and occupational risks for COPD in DALYs per 100 000 people in locations grouped by SDI quintiles, 2015 Percentage contribution from environmental and occupational risks Percentage contribution from behavioural and environmental and occupational risks Percentage contribution from behavioural risks Percentage unattributed High SDI quintile 7·0 15·1 54·8 23·2 High-middle SDI quintile 23·5 20·3 27·1 29·2 Middle SDI quintile 32·5 21·3 17·5 28·7 Low-middle SDI quintile 37·0 26·3 12·7 24·0 Low SDI quintile 40·4 17·8 7·9 34·0 COPD=chronic obstructive pulmonary disease. DALYs=disability-adjusted life years. SDI=Socio-demographic Index. More detailed GBD 2015 results are available for download online and in online visualisation tools, and the GBD 2015 code can be accessed online.

Percentage contribution from environmental and occupational risks Percentage contribution from behavioural and environmental and occupational risks Percentage contribution from behavioural risks Percentage unattributed High SDI quintile 7·0 15·1 54·8 23·2 High-middle SDI quintile 23·5 20·3 27·1 29·2 Middle SDI quintile 32·5 21·3 17·5 28·7 Low-middle SDI quintile 37·0 26·3 12·7 24·0 Low SDI quintile 40·4 17·8 7·9 34·0 COPD=chronic obstructive pulmonary disease. DALYs=disability-adjusted life years. SDI=Socio-demographic Index. More detailed GBD 2015 results are available for download online and in online visualisation tools, and the GBD 2015 code can be accessed online. Discussion Asthma was the most prevalent chronic respiratory disease, affecting an estimated 358 million people in 2015. COPD was half as common, with 174 million people affected in 2015. Deaths from COPD were eight times more common than deaths from asthma. YLLs contributed 81·2% of the 63·8 million global DALYs due to COPD, whereas YLDs represented the largest proportion of the 26·2 million global DALYs due to asthma. COPD ranked eighth (2·6% of global DALYs) and asthma 23rd (1·1% of global DALYs) among the 315 GBD causes in 2015. Age-standardised DALY rates from COPD and asthma declined significantly by 43·7% (39·8–47·0) for COPD and by 42·8% (29·5–52·0) for asthma annually between 1990 and 2015. Most of the reductions have come from a reduction in mortality, by 41·9% for COPD and 58·8% for asthma. The reductions in YLDs have been much smaller. This finding reflects greater improvements in reducing case fatality rather than a change in incidence and prevalence. The absence of a relationship between asthma prevalence and asthma death rates, and our knowledge about asthma pathophysiology and clinical trial findings, add evidence that most asthma deaths at all ages are preventable by treatment with low-dose inhaled corticosteroids and other management strategies or, to a lesser extent, avoidance of risk factors. Indeed, the observed low asthma mortality in high-income countries reflects better access to health services and better treatment options following international asthma guidance.39 This fact is reflected by a strong relationship between SDI and mortality, but not prevalence of asthma. The relationship between SDI and COPD is less monotonic. Higher COPD death rates and prevalence at the middle range of SDI values reflect the increase in smoking and outdoor air pollution observed in countries going through the demographic and epidemiological transition.

SDI and mortality, but not prevalence of asthma. The relationship between SDI and COPD is less monotonic. Higher COPD death rates and prevalence at the middle range of SDI values reflect the increase in smoking and outdoor air pollution observed in countries going through the demographic and epidemiological transition. Between the GBD 2013 and GBD 2015 iterations, a methodological change led to a significant difference in prevalence and YLDs due to COPD. In the GBD 2013 study, we estimated 328·5 million prevalent cases and 26·1 million YLDs for the year 2013, whereas the GBD 2015 estimates for 2015 were 174·5 cases and 12·0 million YLDs. This difference is due to a shift from taking LLN estimates as the reference case definition to using the fixed-ratio definition of GOLD. This change in the methods led to much lower prevalence estimates in 2015 than in 2013 because the GOLD criteria identified lower prevalence in younger adults than older adults. Younger adults represent a much larger proportion of the world's population, and therefore a higher estimate of prevalence at these ages affects the total prevalence more than an equivalent change in the prevalence at older ages. The main motivation to revert to the fixed-ratio GOLD definition for GBD 2015 was that we aimed to estimate symptomatic disease. Our severity distributions are derived from epidemiological data on GOLD classes, which fit better with the estimation of prevalence based on GOLD's fixed ratio criteria than LLN. Most of the arguments for using an LLN case definition are based on future risk of disease to identify people with early signs of disease, who could be prevented from developing symptomatic disease by measures such as smoking cessation. Both in primary and secondary care, clinicians rely on respiratory symptoms, exposure to major known risks, and airflow limitation to diagnose COPD clinically. Accurate LLN estimation requires prediction reference equations, and to date there are no universal prediction equations for LLN because spirometry is not only variable by age and sex, but is also race-related and affected by different local environments.15, 40 To date, the fixed ratio has gained more popular use in clinical practice because it is easy to calculate, thus helping to remove barriers to the widespread use of spirometry and diagnosis of fixed airflow obstruction.

variable by age and sex, but is also race-related and affected by different local environments.15, 40 To date, the fixed ratio has gained more popular use in clinical practice because it is easy to calculate, thus helping to remove barriers to the widespread use of spirometry and diagnosis of fixed airflow obstruction. Furthermore, nearly all evidence on efficacy and safety of respiratory drugs and other treatments comes from randomised trials with patients identified using a fixed ratio definition.41 No comparable estimates of the global prevalence of COPD exist other than those made for previous iterations of GBD. A decade ago, findings from a meta-analysis42 of COPD prevalence studies showed large heterogeneity depending on case definitions based on spirometry, physician diagnosis, symptoms, and radiology. No attempt was made to pool estimates between different study methods and diagnostic thresholds. The various initiatives (BOLD, PLATINO, EPISCAN, and PREPOCOL) for population-representative COPD spirometry surveys have been combined for pooled analyses to compare prevalence estimates between sites, but estimation of global prevalence has not been attempted.16

ween different study methods and diagnostic thresholds. The various initiatives (BOLD, PLATINO, EPISCAN, and PREPOCOL) for population-representative COPD spirometry surveys have been combined for pooled analyses to compare prevalence estimates between sites, but estimation of global prevalence has not been attempted.16 An estimate of 300 million prevalence cases of asthma was made in 2004 as part of the Global Initiative for Asthma (GINA) based on the International Study of Asthma and Allergies in Childhood (ISAAC) and European Community Respiratory Health Survey (ECRHS) estimates of wheezing prevalence and an arbitrary 50% reduction of these estimates for clinical asthma.43 This estimate is a lower than our estimate of 329 million prevalent cases in 2000, and 327 million cases in 2005, and our current 2015 estimate of 358 million cases. A comparison between the ISAAC study estimates in children and the ECRHS study in adults showed a high correlation between childhood and adult prevalence estimates within countries, but large variation between countries.18 In an analysis of the World Health Surveys done in the early 2000s in 70 countries,44 estimates of the prevalence of wheeze, reported physician diagnosis, and clinical asthma (based on questions of a physician diagnosis and ever having been treated for asthma or currently using asthma medication) were pooled. This showed a difference of around two times between higher estimates for wheeze compared with a reported physician diagnosis, whereas prevalence of clinical asthma was only marginally higher than for physician diagnosis. However, the pooling method in this study was not explained.44 ISAAC Phase Three was completed in 233 centres in 97 countries (80% low-income and middle-income countries), and ISAAC repeated surveys of prevalence of asthma have been done in 106 centres in 56 countries to date. We have made use of all publicly available survey results from ISAAC and ECRHS.

tudy was not explained.44 ISAAC Phase Three was completed in 233 centres in 97 countries (80% low-income and middle-income countries), and ISAAC repeated surveys of prevalence of asthma have been done in 106 centres in 56 countries to date. We have made use of all publicly available survey results from ISAAC and ECRHS. We estimated the highest age-standardised DALY rates due to COPD in 2015 in Papua New Guinea, India, Lesotho, and Nepal. Findings from three verbal autopsy studies in the 1980s in Papua New Guinea showed high chronic respiratory disease mortality. We decided to retain these three studies as they provide the only data on causes of death for this country; there were no reasons to exclude them based on an assessment of their quality. The high rates of mortality and morbidity in Lesotho and Nepal were based on predictive covariates as we do not have primary data for COPD from these two countries. The high rates in India were driven by mortality data sources and two small spirometry studies in Pune and Mumbai.45

sed on an assessment of their quality. The high rates of mortality and morbidity in Lesotho and Nepal were based on predictive covariates as we do not have primary data for COPD from these two countries. The high rates in India were driven by mortality data sources and two small spirometry studies in Pune and Mumbai.45 Our knowledge of the natural history of COPD and asthma is extensive yet incomplete. For asthma, over 100 cohorts focusing on asthma and allergy have been initiated worldwide over the past 30 years.46 These long-term birth cohort studies are essential to understanding the life course and childhood predictors of asthma and allergy and the complex interplay between genes and the environment (including lifestyle and socioeconomic determinants). However, information to quantify population-level exposure to allergens in a comparable manner is incomplete, and it has therefore not been possible to add it as a risk in GBD. Such natural history evidence is mostly missing for COPD.47

een genes and the environment (including lifestyle and socioeconomic determinants). However, information to quantify population-level exposure to allergens in a comparable manner is incomplete, and it has therefore not been possible to add it as a risk in GBD. Such natural history evidence is mostly missing for COPD.47 The contribution of modifiable risk factors to COPD is large, yet much less for asthma. There are preventive interventions to reduce exposure to smoking, second-hand smoke, air pollution, biomass for cooking or heating, occupational exposures, or any combination of these factors. Additionally, other risk factors have been identified such as parental or sibling history of asthma and atopy, low birthweight, lower respiratory infections in childhood, education, day care, pet ownership, and other exposures, among others suggested. However, we are still far from eliminating these as major contributors to the burden of COPD. Smoking is the largest contributor to the COPD burden in countries at the higher end of the SDI (69·4% of COPD burden in high-SDI quintile countries), whereas the proportion of COPD explained by environmental exposures is highest in countries with low SDI (58·1% of COPD burden in low-SDI quintile countries). Given the importance of smoking as a risk factor of disease, monitoring national and international trends and projections in smoking remains paramount for worldwide health surveillance.48, 49 Smoking prevalence has decreased in men and women since 1990 worldwide, but progress in tobacco control has not been universal.50 Globally, exposure to household air pollution from solid fuels has decreased since 1990, but exposure to ambient air pollution has increased since 1990.23

orldwide health surveillance.48, 49 Smoking prevalence has decreased in men and women since 1990 worldwide, but progress in tobacco control has not been universal.50 Globally, exposure to household air pollution from solid fuels has decreased since 1990, but exposure to ambient air pollution has increased since 1990.23 However, a considerable proportion of COPD remains unexplained and cannot be attributed to the risks quantified in GBD. In the next iteration of GBD, we plan to quantify a past history of pulmonary tuberculosis as an additional risk factor for COPD as there is growing evidence for a causal relationship.51, 52 We did not estimate air pollution as a risk for asthma, because we had insufficient evidence for an increased risk of disease. Repeated lower respiratory infections in childhood and the long-term effects of asthma have been reported as other explanatory factors, but it is not quite apparent how estimation of these effects can be operationalised in GBD.53, 54 In future GBD iterations, ambient and household air pollution will be re-evaluated as risk factors for asthma.55, 56 Other newly established individual risk factors of COPD, such as low level of physical activity, could also have contributed to the unexplained COPD burden, as it has been related both to an increased risk of COPD among smokers57 and to a higher risk of COPD mortality.58

However, a considerable proportion of COPD remains unexplained and cannot be attributed to the risks quantified in GBD. In the next iteration of GBD, we plan to quantify a past history of pulmonary tuberculosis as an additional risk factor for COPD as there is growing evidence for a causal relationship.51, 52 We did not estimate air pollution as a risk for asthma, because we had insufficient evidence for an increased risk of disease. Repeated lower respiratory infections in childhood and the long-term effects of asthma have been reported as other explanatory factors, but it is not quite apparent how estimation of these effects can be operationalised in GBD.53, 54 In future GBD iterations, ambient and household air pollution will be re-evaluated as risk factors for asthma.55, 56 Other newly established individual risk factors of COPD, such as low level of physical activity, could also have contributed to the unexplained COPD burden, as it has been related both to an increased risk of COPD among smokers57 and to a higher risk of COPD mortality.58 We estimate only a small proportion of asthma burden due to risks quantified in GBD: 10·1% from occupational asthmagens and 7·8% due to smoking.59, 60 Evidence from long-term observational studies and birth cohorts have rendered three hypotheses on other causes and triggers of asthma, namely the hygiene, westernisation, and obesity or sedentarism hypotheses. Comparative studies61, 62 of rural and urban populations gave rise to the hygiene theory that exposure to infections in early childhood explains the lower prevalence of asthma in rural areas. The second theory is that socioeconomic development or westernisation predisposes to the development of asthma, but it is not clear which pathways other than those described in the hygiene theory have a role.63 Obesity has been linked to a higher prevalence of asthma in children64 and an increased risk of developing new asthma in adults.65

that socioeconomic development or westernisation predisposes to the development of asthma, but it is not clear which pathways other than those described in the hygiene theory have a role.63 Obesity has been linked to a higher prevalence of asthma in children64 and an increased risk of developing new asthma in adults.65 As concluded by Fuchs and colleagues in their 2017 Review,66 we need to better understand underlying mechanisms of associations of asthma onset or remission with risk and protective factors, and future asthma research should integrate both paediatric and adult populations and longitudinal studies. The general limitations of GBD studies have been reported elsewhere and apply to estimates of obstructive airways disease as well,1, 2, 23 and there are a number of limitations specific to COPD and asthma. The first concerns the poor consensus on a case definition of COPD. There was a difference of more than two times in YLDs between the GBD 2013 study, which used LLN as its case definition, and the current study's YLD results based on the fixed ratio of the GOLD definition. As the survey data on GOLD class distributions is largely based on a fixed ratio estimate of overall prevalence, we advocate that for GBD estimation purposes, use of the fixed ratio is preferable. Additionally, defining the cutoff value for LLN as the fifth percentile in a healthy reference population makes the arbitrary assumption that prevalence cannot be lower than 5%.

ely based on a fixed ratio estimate of overall prevalence, we advocate that for GBD estimation purposes, use of the fixed ratio is preferable. Additionally, defining the cutoff value for LLN as the fifth percentile in a healthy reference population makes the arbitrary assumption that prevalence cannot be lower than 5%. Second, to make use of all spirometry surveys that reported COPD prevalence using different thresholds, and with or without bronchodilation, we had to adjust data sources to the expected values of our reference case definition. We used a limited set of surveys that presented data with the reference and alternative case definitions. Each of those adjustments showed a strong age pattern, which we tried to capture with regression methods. Such adjustments add uncertainty, which would be avoided if estimates were all reported in a standard manner.

e used a limited set of surveys that presented data with the reference and alternative case definitions. Each of those adjustments showed a strong age pattern, which we tried to capture with regression methods. Such adjustments add uncertainty, which would be avoided if estimates were all reported in a standard manner. Third, because no physiological measurement is considered a gold standard, diagnosis of asthma relies on clinical assessment and self-report, a physician diagnosis, or both. Thus, measurement of asthma prevalence can be affected by the limitations of recall bias, access to health services, and different interpretations of survey questions inherent in self-reported health measurements.67 Access to clinical care is a challenge in low-income and middle-income countries, as well as in rural settings, therefore, defining asthma by reported symptoms and a doctor diagnosis could lead to an underestimate of asthma prevalence. We refined our assessment of asthma studies in GBD 2015 to better deal with nuances in self-report measures and adjusted for three instead of just one non-reference case definition. However, we cannot exclude that some residual measurement bias has affected the comparability of estimates between countries.

prevalence. We refined our assessment of asthma studies in GBD 2015 to better deal with nuances in self-report measures and adjusted for three instead of just one non-reference case definition. However, we cannot exclude that some residual measurement bias has affected the comparability of estimates between countries. Fourth, mapping severity in MEPS to COPD GOLD class prevalence assumes this relationship can be generalised from the USA to the rest of the world. However, we only use the relationship between epidemiological data on GOLD class distributions in the USA to the severity pattern of cases with COPD in MEPS as reflected in respondents' answers to the SF-12. Our epidemiological models of the GOLD class distribution allow us to differentiate severity by age, sex, year, and location in as far as the sparse information on GOLD class prevalence allows. Our measurements of COPD severity would benefit from increased use standardised measures in surveys that reflect the lay descriptions on which the GBD disability weights are based or that use a generic quality-of-life measure like SF-12.

and location in as far as the sparse information on GOLD class prevalence allows. Our measurements of COPD severity would benefit from increased use standardised measures in surveys that reflect the lay descriptions on which the GBD disability weights are based or that use a generic quality-of-life measure like SF-12. Fifth, our measurement of asthma severity completely relies on MEPS data and therefore, unlike COPD, assumes the same distribution for every location, year, age, and sex. This assumption is highly unlikely as treatments have a large effect on severity of asthma. For this reason, we found no relationship between SDI and YLDs from asthma, counter to the expectation that increased access to treatment, particularly steroid inhalers, would impact asthma severity and hence disability. Researchers are encouraged, in future surveys, to collect information on the proportion of cases that would fall into the lay description categories for controlled, partially controlled, and uncontrolled asthma. Sixth, for many countries in the world that do not have functional vital registration systems, we had to rely on death estimates of all chronic respiratory diseases from verbal autopsy studies because these studies cannot distinguish between asthma, COPD, or other chronic respiratory diseases. Initiatives to strengthen vital registration systems are key to improving population health measurement because verbal autopsy can only identify a restricted set of diseases.68

tory diseases from verbal autopsy studies because these studies cannot distinguish between asthma, COPD, or other chronic respiratory diseases. Initiatives to strengthen vital registration systems are key to improving population health measurement because verbal autopsy can only identify a restricted set of diseases.68 Seventh, the estimate of the global prevalence of asthma changed from 242 million in 2013 based on GBD 2013 to 358 million in 2015 for GBD 2015. In GBD 2015, cause-specific mortality rates were added to the DisMod-MR 2.1 model with income per capita as a covariate to differentiate excess mortality rates based on a country's income. This addition had little effect on the estimates in high-income countries, but increased estimates in low-income and middle-income countries considerably. We believe that this approach is an improvement in the estimation strategy and that future estimates of global prevalence of asthma will be more consistent with the GBD 2015 finding.

little effect on the estimates in high-income countries, but increased estimates in low-income and middle-income countries considerably. We believe that this approach is an improvement in the estimation strategy and that future estimates of global prevalence of asthma will be more consistent with the GBD 2015 finding. COPD and asthma are important contributors to the burden of non-communicable disease. Although much of the burden is either preventable or treatable with affordable interventions, these diseases have received less attention than other prominent non-communicable diseases like cardiovascular disease, cancer, or diabetes. Up-to-date population information on these common diseases is key to policy decision making to improve access to and quality of existing intervention strategies. We call for greater standardisation in data collection with regard to case definition and severity distributions of all non-communicable diseases in general, and of asthma and COPD in particular. More, and updated, population measurements of COPD and asthma are needed to better quantify the size of the problem, to benchmark with other chronic conditions associated with smoking and ageing, and with any other environmental and air pollution exposures. Correspondence to: Prof Theo Vos, Institute for Health Metrics and Evaluation, University of Washington, Seattle, WA 98121, USA tvos@uw.edu For the GBD 2015 results see http://ghdx.healthdata.org/gbd-results-tool For the online visualisation tools see https://vizhub.healthdata.org/gbd-compare

COPD and asthma are important contributors to the burden of non-communicable disease. Although much of the burden is either preventable or treatable with affordable interventions, these diseases have received less attention than other prominent non-communicable diseases like cardiovascular disease, cancer, or diabetes. Up-to-date population information on these common diseases is key to policy decision making to improve access to and quality of existing intervention strategies. We call for greater standardisation in data collection with regard to case definition and severity distributions of all non-communicable diseases in general, and of asthma and COPD in particular. More, and updated, population measurements of COPD and asthma are needed to better quantify the size of the problem, to benchmark with other chronic conditions associated with smoking and ageing, and with any other environmental and air pollution exposures. Correspondence to: Prof Theo Vos, Institute for Health Metrics and Evaluation, University of Washington, Seattle, WA 98121, USA tvos@uw.edu For the GBD 2015 results see http://ghdx.healthdata.org/gbd-results-tool For the online visualisation tools see https://vizhub.healthdata.org/gbd-compare For the GBD 2015 code see http://ghdx.healthdata.org/gbd-2015-code Supplementary Material Supplementary appendix

Correspondence to: Prof Theo Vos, Institute for Health Metrics and Evaluation, University of Washington, Seattle, WA 98121, USA tvos@uw.edu For the GBD 2015 results see http://ghdx.healthdata.org/gbd-results-tool For the online visualisation tools see https://vizhub.healthdata.org/gbd-compare For the GBD 2015 code see http://ghdx.healthdata.org/gbd-2015-code Supplementary Material Supplementary appendix GBD 2015 Chronic Respiratory Disease Collaborators Joan B Soriano, Amanuel Alemu Abajobir, Kalkidan Hassen Abate, Semaw Ferede Abera, Anurag Agrawal, Muktar Beshir Ahmed, Amani Nidhal Aichour, Ibtihel Aichour, Miloud Taki Eddine Aichour, Khurshid Alam, Noore Alam, Juma M Alkaabi, Fatma Al-Maskari, Nelson Alvis-Guzman, Alemayehu Amberbir, Yaw Ampem Amoako, Mustafa Geleto Ansha, Josep M Antó, Hamid Asayesh, Tesfay Mehari Atey, Euripide Frinel G Arthur Avokpaho, Aleksandra Barac, Sanjay Basu, Neeraj Bedi, Isabela M Bensenor, Adugnaw Berhane, Addisu Shunu Beyene, Zulfiqar A Bhutta, Stan Biryukov, Dube Jara Boneya, Michael Brauer, David O Carpenter, Daniel Casey, Devasahayam Jesudas Christopher, Lalit Dandona, Rakhi Dandona, Samath D Dharmaratne, Huyen Phuc Do, Florian Fischer, Ayele Geleto, Aloke Gopal Ghoshal, Richard F Gillum, Ibrahim Abdelmageem Mohamed Ginawi, Vipin Gupta, Simon I Hay, Mohammad T Hedayati, Nobuyuki Horita, H Dean Hosgood, Mihajlo (Michael) B Jakovljevic, Spencer Lewis James, Jost B Jonas, Amir Kasaeian, Yousef Saleh Khader, Ibrahim A Khalil, Ejaz Ahmad Khan, Young-Ho Khang, Jagdish Khubchandani, Luke D Knibbs, Soewarta Kosen, Parvaiz A Koul, G Anil Kumar, Cheru Tesema Leshargie, Xiaofeng Liang, Hassan Magdy Abd El Razek, Azeem Majeed, Deborah Carvalho Malta, Treh Manhertz, Neal Marquez, Alem Mehari, George A Mensah, Ted R Miller, Karzan Abdulmuhsin Mohammad, Kedir Endris Mohammed, Shafiu Mohammed, Ali H Mokdad, Mohsen Naghavi, Cuong Tat Nguyen, Grant Nguyen, Quyen Le Nguyen, Trang Huyen Nguyen, Dina Nur Anggraini Ningrum, Vuong Minh Nong, Jennifer Ifeoma Obi, Yewande E Odeyemi, Felix Akpojene Ogbo, Eyal Oren, Mahesh PA, Eun-Kee Park, George C Patton, Katherine Paulson, Mostafa Qorbani, Reginald Quansah, Anwar Rafay, Mohammad Hifz Ur Rahman, Rajesh Kumar Rai, Salman Rawaf, Nik Reinig, Saeid Safiri, Rodrigo Sarmiento-Suarez, Benn Sartorius, Miloje Savic, Monika Sawhney, Mika Shigematsu, Mari Smith, Fentaw Tadese, George D Thurston, Roman Topor-Madry, Bach Xuan Tran, Kingsley Nnanna Ukwaja, Job F M van Boven, Vasiliy Victorovich Vlassov, Stein Emil Vollset, Xia Wan, Andrea Werdecker, Sarah Wulf Hanson, Yuichiro Yano, Hassen Hamid Yimam, Naohiro Yon

torius, Miloje Savic, Monika Sawhney, Mika Shigematsu, Mari Smith, Fentaw Tadese, George D Thurston, Roman Topor-Madry, Bach Xuan Tran, Kingsley Nnanna Ukwaja, Job F M van Boven, Vasiliy Victorovich Vlassov, Stein Emil Vollset, Xia Wan, Andrea Werdecker, Sarah Wulf Hanson, Yuichiro Yano, Hassen Hamid Yimam, Naohiro Yon emoto, Chuanhua Yu, Zoubida Zaidi, Maysaa El Sayed Zaki, Christopher J L Murray, and Theo Vos.

torius, Miloje Savic, Monika Sawhney, Mika Shigematsu, Mari Smith, Fentaw Tadese, George D Thurston, Roman Topor-Madry, Bach Xuan Tran, Kingsley Nnanna Ukwaja, Job F M van Boven, Vasiliy Victorovich Vlassov, Stein Emil Vollset, Xia Wan, Andrea Werdecker, Sarah Wulf Hanson, Yuichiro Yano, Hassen Hamid Yimam, Naohiro Yon emoto, Chuanhua Yu, Zoubida Zaidi, Maysaa El Sayed Zaki, Christopher J L Murray, and Theo Vos. Affiliations Instituto de Investigación Hospital Universitario de la Princesa (IISP), Madrid, Spain (Prof J B Soriano MD); Universidad Autónoma de Madrid, Madrid, Spain (Prof J B Soriano MD); School of Public Health, University of Queensland, Brisbane, QLD, Australia (A A Abajobir MPH, L D Knibbs PhD); Department of Epidemiology, College of Health Sciences (M B Ahmed MPH), Jimma University, Jimma, Ethiopia (K H Abate MS); School of Public Health, College of Health Sciences (S F Abera MSc), College of Health Sciences (K E Mohammed MPH), Mekelle University, Mekelle, Ethiopia (T M Atey MS); Food Security and Institute for Biological Chemistry and Nutrition, University of Hohenheim, Stuttgart, Germany (S F Abera MSc); CSIR—Institute of Genomics and Integrative Biology, Delhi, India (A Agrawal PhD); Department of Internal Medicine, Baylor College of Medicine, Houston, TX, USA (A Agrawal PhD); University Ferhat Abbas of Setif, Setif, Algeria (A N Aichour BS); National Institute of Nursing Education, Setif, Algeria (I Aichour MS); High National School of Veterinary Medicine, Algiers, Algeria (M T Aichour MD); Murdoch Childrens Research Institute, The University of Melbourne, Parkville, VIC, Australia (K Alam PhD); The University of Sydney, Sydney, NSW, Australia (K Alam PhD); Department of Health, Queensland, Brisbane, QLD, Australia (N Alam MAppEpid); College of Medicine and Health Sciences, United Arab Emirates University, Al Ain, United Arab Emirates (J M Alkaabi MD, Prof F Al-Maskari PhD); Universidad de Cartagena, Cartagena de Indias, Colombia (Prof N Alvis-Guzman PhD); Dignitas International, Zomba, Malawi (A Amberbir PhD); Department of Medicine, Komfo Anokye Teaching Hospital, Kumasi, Ghana (Y A Amoako MD); West Hararghe Zonal Health Department, Chiro, Ethiopia (M G Ansha MPH); Barcelona Institute for Global Health (IS Global), Barcelona, Spain (Prof J M Antó MD); Department of Medical Emergency, School of Paramedic, Qom University of Medical Sciences, Qom, Iran (H Asayesh MS); Institut de Recherche Clinique du Bénin (IRCB), Cotonou, Benin (E F G A Avokpaho MPH); Laboratoire d'Etudes et de Recherche-Action en Santé (LERAS Afrique), Parakou, Benin (E F G A Avokp

rof J M Antó MD); Department of Medical Emergency, School of Paramedic, Qom University of Medical Sciences, Qom, Iran (H Asayesh MS); Institut de Recherche Clinique du Bénin (IRCB), Cotonou, Benin (E F G A Avokpaho MPH); Laboratoire d'Etudes et de Recherche-Action en Santé (LERAS Afrique), Parakou, Benin (E F G A Avokp aho MPH); Faculty of Medicine, University of Belgrade, Belgrade, Serbia (A Barac PhD); Stanford University, Stanford, CA, USA (S Basu PhD); College of Public Health and Tropical Medicine, Jazan, Saudi Arabia (N Bedi MD); University of São Paulo, São Paulo, Brazil (I M Bensenor PhD); College of Health Sciences, Debre Berhan University, Debre Berhan, Ethiopia (A Berhane PhD); College of Health and Medical Sciences (A S Beyene MPH), Haramaya University, Harar, Ethiopia (A Geleto MPH); Centre of Excellence in Women and Child Health, Aga Khan University, Karachi, Pakistan (Z A Bhutta PhD); Centre for Global Child Health, The Hospital for Sick Children, Toronto, ON, Canada (Z A Bhutta PhD); Institute for Health Metrics and Evaluation (S Biryukov BS, Prof M Brauer ScD, D Casey MPH, Prof L Dandona MD, Prof R Dandona PhD, Prof S I Hay DSc, I A Khalil MD, T Manhertz BA, N Marquez BA, Prof A H Mokdad PhD, Prof M Naghavi PhD, G Nguyen MPH, K Paulson BS, N Reinig BS, M Smith MPA, Prof S E Vollset DrPH, X Wan PhD, S Wulf Hanson MPH, Prof C J L Murray DPhil, Prof T Vos PhD), Center for Health Trends and Forecasts, Institute for Health Metrics and Evaluation (Prof M B Jakovljevic PhD), University of Washington, Seattle, WA, USA; Department of Public Health (D J Boneya MPH), Debre Markos University, Debre Markos, Ethiopia (C T Leshargie MPH); University of British Columbia, Vancouver, BC, Canada (Prof M Brauer ScD); University at Albany, Rensselaer, NY, USA (Prof D O Carpenter MD); Christian Medical College, Vellore, India (Prof D J Christopher MD); Public Health Foundation of India, Gurugram, India (Prof L Dandona MD, Prof R Dandona PhD, G A Kumar PhD); Department of Community Medicine, Faculty of Medicine, University of Peradeniya, Peradeniya, Sri Lanka (S D Dharmaratne MD); Institute for Global Health Innovations, Duy Tan University, Da Nang, Vietnam (H P Do MSc, C T Nguyen MSc, Q L Nguyen MD, T H Nguyen MSc, V M Nong MSc); School of Public Health, Bielefeld University, Bielefeld, Germany (F Fischer PhD); University of Newcastle, Newcastle, NSW, Australia (A Geleto MPH); National Allergy Asthma Bronchitis Institute, Kolkata, India (A G Ghoshal DNB); College of Medicine (A Meh

, C T Nguyen MSc, Q L Nguyen MD, T H Nguyen MSc, V M Nong MSc); School of Public Health, Bielefeld University, Bielefeld, Germany (F Fischer PhD); University of Newcastle, Newcastle, NSW, Australia (A Geleto MPH); National Allergy Asthma Bronchitis Institute, Kolkata, India (A G Ghoshal DNB); College of Medicine (A Meh ari MD), Howard University, Washington, DC, USA (R F Gillum MD); College of Medicine, University of Hail, Hail, Saudi Arabia (I A Ginawi MD); Department of Anthropology, University of Delhi, Delhi, India (V Gupta PhD); Oxford Big Data Institute, Li Ka Shing Centre for Health Information and Discovery, University of Oxford, Oxford, UK (Prof S I Hay DSc); Department of Medical Mycology and Parasitology, School of Medicine, Mazandaran University of Medical Sciences, Sari, Iran (Prof M T Hedayati PhD); Department of Pulmonology, Yokohama City University Graduate School of Medicine, Yokohama, Japan (N Horita MD); Albert Einstein College of Medicine, Bronx, NY, USA (Prof H D Hosgood PhD); Faculty of Medical Sciences, University of Kragujevac, Kragujevac, Serbia (Prof M B Jakovljevic PhD); Denver Health/University of Colorado, Denver, CO, USA (S L James MD); Department of Ophthalmology, Medical Faculty Mannheim, Ruprecht-Karls-University Heidelberg, Mannheim, Germany (Prof J B Jonas MD); Endocrinology and Metabolism Population Sciences Institute (A Kasaeian PhD), Hematology-Oncology and Stem Cell Transplantation Research Center, Tehran University of Medical Sciences, Tehran, Iran (A Kasaeian PhD); Department of Community Medicine, Public Health and Family Medicine, Jordan University of Science and Technology, Irbid, Jordan (Prof Y S Khader ScD); Health Services Academy, Islamabad, Pakistan (E A Khan MD); Department of Health Policy and Management, Seoul National University College of Medicine, Seoul, South Korea (Prof Y Khang MD); Institute of Health Policy and Management, Seoul National University Medical Center, Seoul, South Korea (Prof Y Khang MD); Department of Nutrition and Health Science, Ball State University, Muncie, IN, USA (J Khubchandani PhD); Center for Community Empowerment, Health Policy and Humanities, National Institute of Health Research & Development, Jakarta, Indonesia (S Kosen MD); Sher-i-Kashmir Institute of Medical Sciences, Srinagar, India (Prof P A Koul MD); Chinese Center for Disease Control and Prevention, Beijing, China (Prof X Liang MD); Mansoura Faculty of Medicine, Mansoura, Egypt (H Magdy Abd El Razek MBBCH); Department of Primary Care &

& Development, Jakarta, Indonesia (S Kosen MD); Sher-i-Kashmir Institute of Medical Sciences, Srinagar, India (Prof P A Koul MD); Chinese Center for Disease Control and Prevention, Beijing, China (Prof X Liang MD); Mansoura Faculty of Medicine, Mansoura, Egypt (H Magdy Abd El Razek MBBCH); Department of Primary Care & Public Health (Prof A Majeed MD), Imperial College London, London, UK (Prof S Rawaf MD); Universidade Federal de Minas Gerais, Belo Horizonte, Brazil (Prof D C Malta PhD); Center for Translation Research and Implementation Science, National Heart, Lung, and Blood Institute, National Institutes of Health, Bethesda, MD, USA (G A Mensah MD); Pacific Institute for Research & Evaluation, Calverton, MD, USA (T R Miller PhD); Centre for Population Health, Curtin University, Perth, WA, Australia (T R Miller PhD); University of Salahaddin, Erbil, Iraq (K A Mohammad PhD); ISHIK University, Erbil, Iraq (K A Mohammad PhD); Health Systems and Policy Research Unit, Ahmadu Bello University, Zaria, Nigeria (S Mohammed PhD); Institute of Public Health, Heidelberg University, Heidelberg, Germany (S Mohammed PhD); Department of Public Health, Semarang State University, Semarang City, Indonesia (D N A Ningrum MPH); Graduate Institute of Biomedical Informatics, College of Medical Science and Technology, Taipei Medical University, Taipei City, Taiwan (D N A Ningrum MPH); Pulmonary Medicine, Howard University Hospital, Washington, DC, USA (J I Obi MBBS, Y E Odeyemi MBBS); Centre for Health Research, Western Sydney University, Sydney, NSW, Australia (F A Ogbo MPH); University of Arizona, Tucson, AZ, USA (Prof E Oren PhD); JSS Medical College, JSS University, Mysore, India (Prof M PA DNB); Department of Medical Humanities and Social Medicine, College of Medicine, Kosin University, Busan, South Korea (E Park PhD); Murdoch Childrens Research Institute, Department of Paediatrics, University of Melbourne, Melbourne, VIC, Australia (Prof G C Patton MD); Non-Communicable Diseases Research Center, Alborz University of Medical Sciences, Karaj, Iran (M Qorbani PhD); University of Ghana, Accra, Ghana (R Quansah PhD); Noguchi Memorial Institute of Medical Research, Accra, Ghana (R Quansah PhD); Contech International Health Consultants, Lahore, Pakistan (A Rafay MS); Contech School of Public Health, Lahore, Pakistan (A Rafay MS); International Institute for Population Sciences, Mumbai, India (M H U Rahman MPhil); Society for Health and Demographic Surveillance, Suri, India (R K Rai MPH); Managerial

ansah PhD); Contech International Health Consultants, Lahore, Pakistan (A Rafay MS); Contech School of Public Health, Lahore, Pakistan (A Rafay MS); International Institute for Population Sciences, Mumbai, India (M H U Rahman MPhil); Society for Health and Demographic Surveillance, Suri, India (R K Rai MPH); Managerial Epidemiology Research Center, Department of Public Health, School of Nursing and Midwifery, Maragheh University of Medical Sciences, Maragheh, Iran (S Safiri PhD); Universidad Ciencias Aplicadas y Ambientales, Bogotá DC, Colombia (R Sarmiento-Suarez MPH); Public Health Medicine, School of Nursing and Public Health, University of KwaZulu-Natal, Durban, South Africa (Prof B Sartorius PhD); UKZN Gastrointestinal Cancer Research Centre, South African Medical Research Council, Durban, South Africa (Prof B Sartorius PhD); Center for Disease Burden (Prof S E Vollset DrPH), Norwegian Institute of Public Health, Oslo, Norway (M Savic PhD); Department of Public Health, Marshall University, Huntington, WV, USA (M Sawhney PhD); National Institute of Infectious Diseases, Tokyo, Japan (M Shigematsu PhD); Sandia National Laboratories, Albuquerque, NM, USA (M Shigematsu PhD); Department of Public Health, Wollo University, Dessie, Ethiopia (F Tadese MPH); Nelson Institute of Environmental Medicine, School of Medicine, New York University, Tuxedo, NY, USA (Prof G D Thurston ScD); Institute of Public Health, Faculty of Health Sciences, Jagiellonian University Medical College, Kraków, Poland (R Topor-Madry PhD); Faculty of Health Sciences, Wroclaw Medical University, Wroclaw, Poland (R Topor-Madry PhD); Johns Hopkins University, Baltimore, MD, USA (B X Tran PhD); Hanoi Medical University, Hanoi, Vietnam (B X Tran PhD); Department of Internal Medicine, Federal Teaching Hospital, Abakaliki, Nigeria (K N Ukwaja MD); University of Groningen, Groningen, Netherlands (J F M van Boven PhD); National Research University Higher School of Economics, Moscow, Russia (Prof V V Vlassov MD); Department of Global Public Health and Primary Care, University of Bergen, Bergen, Norway (Prof S E Vollset DrPH); Institute of Basic Medical Sciences, Chinese Academy of Medical Sciences, Beijing, China (X Wan PhD); Competence Center Mortality-Follow-Up of the German National Cohort, Federal Institute for Population Research, Wiesbaden, Germany (A Werdecker PhD); Department of Preventive Medicine, Northwestern University, Chicago, IL, USA (Y Yano MD); Mizan Tepi University, Mizan Teferi, Ethiopia (H H Yimam

ng, China (X Wan PhD); Competence Center Mortality-Follow-Up of the German National Cohort, Federal Institute for Population Research, Wiesbaden, Germany (A Werdecker PhD); Department of Preventive Medicine, Northwestern University, Chicago, IL, USA (Y Yano MD); Mizan Tepi University, Mizan Teferi, Ethiopia (H H Yimam MPH); Department of Biostatistics, School of Public Health, Kyoto University, Kyoto, Japan (N Yonemoto MPH); Department of Epidemiology and Biostatistics, School of Public Health (Prof C Yu PhD), Global Health Institute (Prof C Yu PhD), Wuhan University, Wuhan, China; University Hospital of Setif, Setif, Algeria (Prof Z Zaidi DSc); and Faculty of Medicine, Mansoura University, Mansoura, Egypt (Prof M E Zaki PhD). Contributors TV and JBS prepared the first draft. CJLM conceived the study and provided overall guidance. All other authors provided data, developed models, reviewed results, initiated modelling infrastructure, or reviewed and contributed to the report. Declaration of interests We declare no competing interests.

Alton EWFW, Armstrong DK, Ashby D, et al, on behalf of the UK Cystic Fibrosis Gene Therapy Consortium. Repeated nebulisation of non-viral CFTR gene therapy in patients with cystic fibrosis: a randomised, double-blind, placebo-controlled, phase 2b trial. Lancet Respir Med 2015; published online July 3. http://dx.doi.org/10.1016/S2213-2600(15)00245-3—In this Article, Catherine McKenny (Oxford Churchill) was missing from the Acknowledgments, as was reference to financial support from NHS Research Scotland, through the Edinburgh Clinical Research Facility. This correction has been made to the online version as of Sept 7, 2015, and the printed version is correct.

Introduction Longitudinal surveillance studies using repeated bronchoalveolar lavage in children with cystic fibrosis have reported that 30% of these children have Pseudomonas aeruginosa detected in the first 6 years of life,1 and that infection with significant pathogens occurs in the first 2 years of life in 71% of children.2 Notably, early infection was identified as the major determinant of lung function deterioration by school age, suggesting that it is an important driver of lung inflammation and has a crucial contribution to the development of cystic fibrosis lung disease.2 Young children with cystic fibrosis are generally asymptomatic, cough free, and non-productive of mucus. These children are often incapable of expectorating sputum even if actively coughing during an exacerbation. Effective sampling for lower airway microbiology is therefore problematic, yet remains crucial in this age group if infection is to be effectively treated or prevented, and the potential benefits of newborn screening properly realised.3 Cystic fibrosis standards of care recommend doing regular oropharyngeal cough swabs for bacterial surveillance in young non-expectorating children. However, oropharyngeal cultures are a poor surrogate for cultures from lower airway samples taken at concurrent bronchoalveolar lavage.4 Research in context Evidence before this study

Introduction Longitudinal surveillance studies using repeated bronchoalveolar lavage in children with cystic fibrosis have reported that 30% of these children have Pseudomonas aeruginosa detected in the first 6 years of life,1 and that infection with significant pathogens occurs in the first 2 years of life in 71% of children.2 Notably, early infection was identified as the major determinant of lung function deterioration by school age, suggesting that it is an important driver of lung inflammation and has a crucial contribution to the development of cystic fibrosis lung disease.2 Young children with cystic fibrosis are generally asymptomatic, cough free, and non-productive of mucus. These children are often incapable of expectorating sputum even if actively coughing during an exacerbation. Effective sampling for lower airway microbiology is therefore problematic, yet remains crucial in this age group if infection is to be effectively treated or prevented, and the potential benefits of newborn screening properly realised.3 Cystic fibrosis standards of care recommend doing regular oropharyngeal cough swabs for bacterial surveillance in young non-expectorating children. However, oropharyngeal cultures are a poor surrogate for cultures from lower airway samples taken at concurrent bronchoalveolar lavage.4 Research in context Evidence before this study We initially did a comprehensive review of the use of sputum induction in children with cystic fibrosis on March 1–15, 2011. We searched PubMed for studies published between Jan 1, 1960, and Dec 31, 2010, and updated the search on Dec 6, 2014. We used the following keywords: “induced sputum”, “sputum induction”, “bronchoalveolar lavage”, “cough swab”, “oropharyngeal”, “children”, “child”, “infant” “childhood”, “young”, “cystic fibrosis”, “CF”, and “hypertonic saline”. Few adequately powered studies were found. Six studies assessed sputum induction in children with cystic fibrosis and these were generally in older children who could perform spirometry reliably. Taken together, these studies included 211 patients and reported a 92·5% success rate in obtaining a sputum sample. Four studies compared sputum induction with oropharyngeal samples in children with cystic fibrosis. The two larger studies identified additional organisms on sputum induction in 30% and 42% of cases, but these studies mainly recruited school-age children and teenagers, many of whom could spontaneously expectorate sputum. One small study compared sputum induction with bronchoalveolar lavage in ten children with cystic fibrosis. During the period of recruitment to the present study, one study was published that showed sputum induction to be superior to oropharyngeal sampling in 32 children younger than 5 years, with an approximately two-fold increase in pathogen detection. A further single study compared sputum induction to gold-standard two-lobe bronchoalveolar lavage in children, but paired samples were not necessarily taken at the same visit. That study found sputum induction sensitivity to be 37% and specificity to be 69%, compared with gold-standard two-lobe bronchoalveolar lavage, but discounted any pathogens isolated exclusively on sputum induction as false negatives.

r lavage in children, but paired samples were not necessarily taken at the same visit. That study found sputum induction sensitivity to be 37% and specificity to be 69%, compared with gold-standard two-lobe bronchoalveolar lavage, but discounted any pathogens isolated exclusively on sputum induction as false negatives. Added value of the study To our knowledge, this is the first time a study has compared concomitant cough swab, sputum induction, and the gold-standard two-lobe bronchoalveolar lavage to comprehensive six-lobe bronchoalveolar lavage to systematically define the relative contribution of each approach. Our data establish sputum induction as superior to cough swab and as a credible approach to sampling the lower airway in symptomatic children with cystic fibrosis when compared with bronchoalveolar lavage. This study shows that both sputum induction and six-lobe bronchoalveolar lavage contribute independent, sizeable gains in pathogen detection over and above two-lobe bronchoalveolar lavage, and challenges two-lobe bronchoalveolar lavage as an adequate gold-standard approach to understanding lower airway microbiology. Sputum induction is uniquely placed to sample the large intrathoracic airways, a compartment inadequately considered in the current paradigm of lower airway sampling. In symptomatic patients, doing sputum induction before bronchoalveolar lavage will correctly describe the lower airway pathogen environment in most patients, thereby markedly reducing the number of bronchoscopy procedures required. Implications of all available evidence

To our knowledge, this is the first time a study has compared concomitant cough swab, sputum induction, and the gold-standard two-lobe bronchoalveolar lavage to comprehensive six-lobe bronchoalveolar lavage to systematically define the relative contribution of each approach. Our data establish sputum induction as superior to cough swab and as a credible approach to sampling the lower airway in symptomatic children with cystic fibrosis when compared with bronchoalveolar lavage. This study shows that both sputum induction and six-lobe bronchoalveolar lavage contribute independent, sizeable gains in pathogen detection over and above two-lobe bronchoalveolar lavage, and challenges two-lobe bronchoalveolar lavage as an adequate gold-standard approach to understanding lower airway microbiology. Sputum induction is uniquely placed to sample the large intrathoracic airways, a compartment inadequately considered in the current paradigm of lower airway sampling. In symptomatic patients, doing sputum induction before bronchoalveolar lavage will correctly describe the lower airway pathogen environment in most patients, thereby markedly reducing the number of bronchoscopy procedures required. Implications of all available evidence Our study supports the recommendation that sputum induction and six-lobe lavage should be done together as a new standard of care for comprehensive assessment of the lower airway pathogen environment in children with cystic fibrosis. If sputum induction is done before bronchoalveolar lavage, a substantial number of bronchoscopies could be avoided in symptomatic children with cystic fibrosis. The inclusion of sputum induction as a regular contribution to cystic fibrosis care in children is supported by the tolerability of the procedure in all age groups, the ease of repeatability, the acceptability to parents and patients, the high success rates, the additional pathogens identified, and the clear economic savings.

lusion of sputum induction as a regular contribution to cystic fibrosis care in children is supported by the tolerability of the procedure in all age groups, the ease of repeatability, the acceptability to parents and patients, the high success rates, the additional pathogens identified, and the clear economic savings. Bronchoalveolar lavage is considered to be the gold standard for sampling lower airway microbiology.5 Although the international community is interested in bronchoalveolar lavage-based microbiology-surveillance programmes, little evidence supports recommendation of this invasive approach in routine cystic fibrosis care.6, 7 Bronchoalveolar lavage is generally reserved for children with cystic fibrosis who have not responded to appropriate or empirical antibiotic treatment and in whom oropharyngeal cultures do not offer an explanation for the persistence of symptoms. No consensus exists on methods for bronchoalveolar lavage, and practice varies. Guidelines for children with cystic fibrosis recommend two-lobe bronchoalveolar lavage taken as follows: three-aliquot bronchoalveolar lavage from the right-middle lobe and a single-aliquot bronchoalveolar lavage from the lingular or the most affected lobe.5 A study published in 20118 showed comprehensive six-lobe bronchoalveolar lavage sampling to be safe, well tolerated, and superior to single-lobe9 or two-lobe5 bronchoalveolar lavage, suggesting that bacterial communities might be compartmentalised within the lung.

ar lavage from the lingular or the most affected lobe.5 A study published in 20118 showed comprehensive six-lobe bronchoalveolar lavage sampling to be safe, well tolerated, and superior to single-lobe9 or two-lobe5 bronchoalveolar lavage, suggesting that bacterial communities might be compartmentalised within the lung. Clinically, young children with cystic fibrosis have wet bronchitic-type coughs during infection, suggesting the predominant focus of infection might be the large intrathoracic airways rather than the alveolar bed. The large intrathoracic airway compartment is not routinely sampled because it is largely bypassed by even the most extensive approach to bronchoalveolar lavage. Sputum induction is a safe approach to obtaining lower airway samples from patients who are not spontaneously productive10, 11 and its use in tuberculosis surveillance in children is well established. The role of sputum induction in the care of young children with cystic fibrosis has not been systematically addressed and few studies exist.12, 13, 14, 15, 16 Sample size and patient age have been variable in these studies but for the most part, conclusions have been encouraging. This trial (the Cystic Fibrosis Sputum Induction Trial [CF-SpIT]) takes a systematic approach to comprehensively investigate and compare bacterial sampling techniques in young children with cystic fibrosis.

ize and patient age have been variable in these studies but for the most part, conclusions have been encouraging. This trial (the Cystic Fibrosis Sputum Induction Trial [CF-SpIT]) takes a systematic approach to comprehensively investigate and compare bacterial sampling techniques in young children with cystic fibrosis. We aimed to test sputum induction as an infection diagnostic for bacterial sampling in children with cystic fibrosis compared with concurrent standard cough swab, single-lobe bronchoalveolar lavage, the gold-standard two-lobe bronchoalveolar lavage, and also comprehensive six-lobe bronchoalveolar lavage. Methods Study design and participants Cf-SpIT is a prospective internally controlled interventional single-centre trial done at the Children's Hospital for Wales (Cardiff, UK) in children with cystic fibrosis, comparing sputum induction, as a diagnostic intervention for pathogen detection, with concurrent cough swab, single-lobe bronchoalveolar lavage, gold-standard two-lobe bronchoalveolar lavage, and comprehensive six-lobe bronchoalveolar lavage. The study was subject to Institutional Review by the Cardiff and Vale Research Review Service (CaRRS—Project ID 11-RPM-5216) and approved by the South Wales Research Ethics Committee (11/WA/0334).

e bronchoalveolar lavage, gold-standard two-lobe bronchoalveolar lavage, and comprehensive six-lobe bronchoalveolar lavage. The study was subject to Institutional Review by the Cardiff and Vale Research Review Service (CaRRS—Project ID 11-RPM-5216) and approved by the South Wales Research Ethics Committee (11/WA/0334). We prospectively recruited children with cystic fibrosis aged between 6 months and 18 years, from the South, West and Mid-Wales Children's Cystic Fibrosis Network. All children attending the Children's Hospital for Wales for clinically indicated bronchoscopy, those attending for routine surgery under general anaesthetic, those admitted for treatment of a chest exacerbation, or those attending for annual review in the outpatient clinic were eligible. Children on treatment antibiotics at the time of sampling were excluded, to maximise the chances of successful bacterial culture.

se attending for routine surgery under general anaesthetic, those admitted for treatment of a chest exacerbation, or those attending for annual review in the outpatient clinic were eligible. Children on treatment antibiotics at the time of sampling were excluded, to maximise the chances of successful bacterial culture. The study was structured in two stages, each designed to test different hypotheses. In stage 1, sputum induction as a diagnostic intervention was tested against cough swab. Children were recruited for this stage of the study at any point when they would otherwise have a cough swab taken: in the outpatient clinic or as an inpatient before receiving intravenous antibiotics. In stage 2, sputum induction as a diagnostic intervention was tested against bronchoalveolar lavage. This was done in a subgroup of patients who had been recruited into stage 1, and who were also attending for a clinically indicated bronchoscopy and bronchoalveolar lavage. Specifically, sputum induction was compared to single-lobe bronchoalveolar lavage, two-lobe bronchoalveolar lavage, and six-lobe bronchoalveolar lavage.

was done in a subgroup of patients who had been recruited into stage 1, and who were also attending for a clinically indicated bronchoscopy and bronchoalveolar lavage. Specifically, sputum induction was compared to single-lobe bronchoalveolar lavage, two-lobe bronchoalveolar lavage, and six-lobe bronchoalveolar lavage. Children could volunteer to take part on more than one occasion if the occasions were more than 3 months apart. Sputum induction was done immediately after cough swab and within the 24 h before bronchoscopy if paired with bronchoalveolar lavage. Children were classified as symptomatic at the time of recruitment if they had an increase in respiratory symptoms, defined as a wet or dry cough, wheeze, or coryzal symptoms. The cough was defined as wet if it sounded wet before the procedure. Data on pathogen isolates from the preceding 12 months before the procedure and new treatments commenced because of microbiology results from the procedure were recorded for all children. Informed consent was taken by trained clinicians on the delegation log. Data for recruitment, clinical observations, and primary and secondary outcome measurements were collated by trained clinicians and research staff on the delegation log and managed by the chief investigator. Regular progress and safety reports were submitted to Research and Design and Ethics panels by the chief investigator.

or recruitment, clinical observations, and primary and secondary outcome measurements were collated by trained clinicians and research staff on the delegation log and managed by the chief investigator. Regular progress and safety reports were submitted to Research and Design and Ethics panels by the chief investigator. Procedures Sputum induction was done by a specialist physiotherapist. 8 mL of 7% hypertonic sodium chloride solution was administered through a simple disposable oxygen-driven jet nebuliser set (SideStream disposable kit; Philips Respironics, Murrysville, PA, USA) at 5 L/min for 15 min and physiotherapy was given during and after the nebuliser was completed. A combination of percussion, vibration, positive expiratory pressure, and active cycle of breathing compatible with the patient's normal physiotherapy regimen was used. Oropharyngeal suction using a size 6, 8, or 10F suction catheter was used to obtain a sputum sample in children who could not spontaneously expectorate after the procedure. Duration of the procedure was limited to 30 min. Success of the procedure was defined as the ability to obtain a mucoid sample, per visual inspection. The physiotherapist documented heart rate, respiratory rate, and FEV1 where applicable before and after the procedure as objective measures of tolerance (appendix).

ure. Duration of the procedure was limited to 30 min. Success of the procedure was defined as the ability to obtain a mucoid sample, per visual inspection. The physiotherapist documented heart rate, respiratory rate, and FEV1 where applicable before and after the procedure as objective measures of tolerance (appendix). All bronchoscopy was done under general anaesthetic. Suction of secretions was avoided before bronchoalveolar lavage to preserve localised sampling without contamination. Samples were taken from all six lobes specifically in the following order: right middle lobe (RML), left lingular (LLi), right lower lobe (RLL), right upper lobe (RUL), left lower lobe (LLL), and left upper lobe (LUL). RML bronchoalveolar lavage was done using three aliquots of 1 mL/kg normal saline (maximum 20 mL per aliquot) with low-level suction through the scope used to retrieve the sample between aliquots. A single aliquot of 1 mL/kg bodyweight (maximum 20 mL) was used for all other lobes. This aliquot was retrieved by light suction on the syringe used for instillation before the liquid column was broken. A second instillation of 1 mL/kg (maximum 20 mL) was used for all lobes in which the initial aliquot did not return 40% volume or greater using syringe back-suction.

20 mL) was used for all other lobes. This aliquot was retrieved by light suction on the syringe used for instillation before the liquid column was broken. A second instillation of 1 mL/kg (maximum 20 mL) was used for all lobes in which the initial aliquot did not return 40% volume or greater using syringe back-suction. Bronchoalveolar lavage fluid from each individual was processed as three samples. All three aliquots from the RML were combined to form bronchoalveolar lavage sample 1. The single aliquot from the LLi was used as sample 2. Fluids from the remaining four lobes (RLL, RUL, LLL, and LUL) were combined to form sample 3. These three samples were sent to the microbiology lab where they were processed independently. Because bronchoalveolar lavage samples 1, 2, and 3 were taken in strict sequential order in all bronchoscopy procedures, we were able to combine the pathogens isolated to describe pathogen detection from increasingly extensive bronchoalveolar lavage. Pathogens isolated from sample 1 were used to describe pathogen detection from single-lobe bronchoalveolar lavage; pathogens isolated from samples 1 and 2 were combined to describe pathogen detection from two-lobe bronchoalveolar lavage; and pathogens isolated from samples 1, 2, and 3 were combined to describe pathogen detection from six-lobe bronchoalveolar lavage.

scribe pathogen detection from single-lobe bronchoalveolar lavage; pathogens isolated from samples 1 and 2 were combined to describe pathogen detection from two-lobe bronchoalveolar lavage; and pathogens isolated from samples 1, 2, and 3 were combined to describe pathogen detection from six-lobe bronchoalveolar lavage. Airway samples were processed using standard techniques for bacteria and fungi in the hospital laboratory of the University Hospital of Wales. The laboratory uses standards from the CF Trust Guidelines.17 For all samples, Haemophilus influenzae, Staphylococcus aureus, meticillin-resistant S aureus (MRSA), P aeruginosa, Burkholderia cepacia complex species, non-tuberculous Mycobacterium species, Achromobacter xylosoxidans, Stenotrophomonas maltophilia, and Klebsiella pneumoniae were defined as cystic fibrosis airway pathogens. All airway fluid samples were divided and one aliquot frozen at −80°C within 30 min of collection. Batch DNA extraction was done after a single freeze thaw, in an extraction-dedicated containment level 2 laboratory. Ribosomal intergenic spacer analysis (RISA) was done as described previously (appendix).18, 19

Airway samples were processed using standard techniques for bacteria and fungi in the hospital laboratory of the University Hospital of Wales. The laboratory uses standards from the CF Trust Guidelines.17 For all samples, Haemophilus influenzae, Staphylococcus aureus, meticillin-resistant S aureus (MRSA), P aeruginosa, Burkholderia cepacia complex species, non-tuberculous Mycobacterium species, Achromobacter xylosoxidans, Stenotrophomonas maltophilia, and Klebsiella pneumoniae were defined as cystic fibrosis airway pathogens. All airway fluid samples were divided and one aliquot frozen at −80°C within 30 min of collection. Batch DNA extraction was done after a single freeze thaw, in an extraction-dedicated containment level 2 laboratory. Ribosomal intergenic spacer analysis (RISA) was done as described previously (appendix).18, 19 Outcomes The primary outcome was pathogen detection for all comparisons, measured by the proportion of patients with one or more positive samples (pathogen positive) and the number of pathogens isolated by each sampling approach. Secondary outcomes were success of sputum induction, proportion of cases in which sputum induction resulted in a change of treatment, test-specific detection rates for all approaches against all pathogens isolated, and the sensitivity of each sampling approach to correctly identify all pathogens isolated from the lower airway. Subjective tolerance was assessed using visual analogue Likert-type scales (score 1–10)20 completed by the parent or child, or both, and the physiotherapist who did the procedure.

Outcomes The primary outcome was pathogen detection for all comparisons, measured by the proportion of patients with one or more positive samples (pathogen positive) and the number of pathogens isolated by each sampling approach. Secondary outcomes were success of sputum induction, proportion of cases in which sputum induction resulted in a change of treatment, test-specific detection rates for all approaches against all pathogens isolated, and the sensitivity of each sampling approach to correctly identify all pathogens isolated from the lower airway. Subjective tolerance was assessed using visual analogue Likert-type scales (score 1–10)20 completed by the parent or child, or both, and the physiotherapist who did the procedure. Statistical analysis We used discordant proportions to compute sample size. At the time of study initiation, no data were available on pathogen yield from sputum-induction sampling in children younger than 6 years with cystic fibrosis. Al-Saleh and colleagues13 studied sputum induction and throat swabs in 94 older children (mean age 12·1 years) with cystic fibrosis, most of whom could spontaneously expectorate. Discrepant culture results were seen in 27% of paired samples. For stage 1, assuming sputum induction would be less successful in the younger age group who could not spontaneously expectorate, we powered the study to detect a smaller discrepancy of 20% in culture results between cough swab and sputum induction. We used discordant proportions in keeping with the findings from Al-Saleh and colleagues13 (27% discordance, odds ratio [OR] 8). Using these proportions, we calculated that a sample size of 59 pairs is capable of detecting 20% discordance in culture results with a power of 80% and probability of type I error of 0·05.

. We used discordant proportions in keeping with the findings from Al-Saleh and colleagues13 (27% discordance, odds ratio [OR] 8). Using these proportions, we calculated that a sample size of 59 pairs is capable of detecting 20% discordance in culture results with a power of 80% and probability of type I error of 0·05. We planned subgroup analyses by age (<6 years or ≥6 years) and symptom status (asymptomatic or symptomatic) for the stage 1 comparison of cough swab versus sputum induction. Assuming the same discordant proportions for all age groups, we required a sample size of 59 for each age subgroup. We therefore continued recruiting until at least 59 patients were recruited to each age subgroup, estimating that, taking patient withdrawals and exclusions into account, this would require prospective recruitment of 200 children in total. No data comparing sputum induction with bronchoalveolar lavage were available at the time of study initiation. Therefore, for stage 2, we estimated discordant proportions to calculate the sample size. We estimated that a small proportion of pathogens (5%) would be isolated on sputum induction and not on bronchoalveolar lavage, and estimated discordance in culture results at 34%. A sample size of 44 pairs would be able to detect a 34% discrepancy between sputum induction and bronchoalveolar lavage with a power of 80% and type I error of 0·05.

proportion of pathogens (5%) would be isolated on sputum induction and not on bronchoalveolar lavage, and estimated discordance in culture results at 34%. A sample size of 44 pairs would be able to detect a 34% discrepancy between sputum induction and bronchoalveolar lavage with a power of 80% and type I error of 0·05. We compared all paired proportions between sampling techniques using the two-sided McNemar's test. We used binary logistic regression (BLR) to assess the effect of repeated measurements within the cohort on the success rate of sputum induction and on the rate of pathogen positivity of sputum induction. We used BLR to assess the effect of age, the presence of respiratory symptoms, the ability to expectorate spontaneously before the procedure, and the need for oropharyngeal suction during the procedure, on the number of pathogens detected by sputum induction, using generalised estimating equations (GEE)21 to account for correlation between repeated measurements in the same individual. We used test-specific detection rates when comparing different approaches to sampling, to interrogate the relative pathogen yield and help understand the relative sampling ability of each approach.

We compared all paired proportions between sampling techniques using the two-sided McNemar's test. We used binary logistic regression (BLR) to assess the effect of repeated measurements within the cohort on the success rate of sputum induction and on the rate of pathogen positivity of sputum induction. We used BLR to assess the effect of age, the presence of respiratory symptoms, the ability to expectorate spontaneously before the procedure, and the need for oropharyngeal suction during the procedure, on the number of pathogens detected by sputum induction, using generalised estimating equations (GEE)21 to account for correlation between repeated measurements in the same individual. We used test-specific detection rates when comparing different approaches to sampling, to interrogate the relative pathogen yield and help understand the relative sampling ability of each approach. We generated a sensitivity analysis for each sampling approach against a combined gold standard consisting of all pathogens isolated from sputum induction and six-lobe bronchoalveolar lavage. A positive outcome was defined as the ability to identify all pathogens from the combined gold standard. This enabled us to quantify the ability of any single approach to correctly detect all lower airway pathogens in a given patient. We used BLR to assess the effect of age and the presence of respiratory symptoms on the ability of sputum induction to correctly detect all lower airway pathogens, using GEE21 to account for correlation between repeated measurements in the same individual. Statistical analyses were done using SPSS statistics for Windows, version 23.0. This study is registered with the UK Clinical Research Network (14615) and with the International Standard Randomised Controlled Trial Network Registry (12473810).

ccount for correlation between repeated measurements in the same individual. Statistical analyses were done using SPSS statistics for Windows, version 23.0. This study is registered with the UK Clinical Research Network (14615) and with the International Standard Randomised Controlled Trial Network Registry (12473810). Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. JTF, KR, and J-DT had access to the raw data. The corresponding author had full access to all of the data and the final responsibility to submit for publication. Results Between Jan 23, 2012, and July 4, 2017, 124 patients were prospectively recruited to the trial and had 200 sputum induction procedures (figure 1; table 1). Median time between procedures in patients that contributed more than one sample was 12 months (IQR 5·5–19). 72 (36%) of the 200 procedures were done in children younger than 6 years and 128 (64%) were done in children aged 6 years or older. 128 (64%) of the samples came from children who were symptomatic at the time of sampling.Figure 1 Participant flow diagram Patients could contribute a sample to the trial on more than one occasion if samples were taken at least 3 months apart. Table 1 Patient baseline characteristics

Results Between Jan 23, 2012, and July 4, 2017, 124 patients were prospectively recruited to the trial and had 200 sputum induction procedures (figure 1; table 1). Median time between procedures in patients that contributed more than one sample was 12 months (IQR 5·5–19). 72 (36%) of the 200 procedures were done in children younger than 6 years and 128 (64%) were done in children aged 6 years or older. 128 (64%) of the samples came from children who were symptomatic at the time of sampling.Figure 1 Participant flow diagram Patients could contribute a sample to the trial on more than one occasion if samples were taken at least 3 months apart. Table 1 Patient baseline characteristics Stage 1: sputum induction Stage 2: bronchoalveolar lavage Number of patients recruited 124 35 Number of procedures 200 41 Median age at procedure 8·2 years (4·9–12·6) 8·5 years (6·5–12·6) Number of procedures in children aged <6 years 72 (36%) .. Median age (subgroup <6 years) 3·5 years (1·6–4·9) .. Number of procedures in children aged ≥6 years 128 (64%) .. Median age (subgroup ≥6 years) 11·1 years (8·2–14·3) .. Pseudomonas aeruginosa positive (isolated in preceding 12 months) 24 (12%) 6 (15%) Median FEV1 (where applicable) 89% (76–99) 84% (72–94) Hypertonic saline naive 37 (19%) 3 (7%) Wet cough at time of procedure 66 (33%) 14 (34%) Able to spontaneously expectorate before procedure 22 (11%) 2 (5%) Symptomatic at time of procedure 128 (64%) 32 (82%) Inpatient procedure 80 (40%) 41 (100%) Outpatient procedure 120 (60%) 0 Data are n (%) or median (IQR). Denominators for percentages are number of procedures.

ugh at time of procedure 66 (33%) 14 (34%) Able to spontaneously expectorate before procedure 22 (11%) 2 (5%) Symptomatic at time of procedure 128 (64%) 32 (82%) Inpatient procedure 80 (40%) 41 (100%) Outpatient procedure 120 (60%) 0 Data are n (%) or median (IQR). Denominators for percentages are number of procedures. 167 (84%) of 200 sputum-induction procedures were successful in producing a mucoid sputum sample. Repeated measures in those individuals recruited more than once did not affect the success of sputum induction (p=0·53). 22 (11%) of 200 children were able to expectorate sputum spontaneously before the sputum-induction procedure. 87 (44%) of 200 children could expectorate sputum during the procedure without requiring suction. When analysed by age group, sputum induction was equally as successful in children younger than 6 years (62 [86%] of 72) as in those aged 6 years or older (105 [82%] of 128). Age as a continuous variable did not influence the success of the sputum-induction procedure (p=0·55). However, oral suction was required in 56 (90%) of 62 children younger than 6 years versus 24 (23%) of 105 children aged 6 years or older. The sputum-induction procedure was similarly successful in the inpatient versus outpatient setting (100 [84%] of 120 vs 67 [83%] of 80), in symptomatic versus asymptomatic children (110 [86%] of 128 vs 57 [79%] of 72), in those with a wet versus dry cough (59 [89%] of 66 vs 108 [81%] of 134), and in patients who were naive to hypertonic saline versus those who were not (33 [89%] of 39 vs 134 [82%] of 169).

t setting (100 [84%] of 120 vs 67 [83%] of 80), in symptomatic versus asymptomatic children (110 [86%] of 128 vs 57 [79%] of 72), in those with a wet versus dry cough (59 [89%] of 66 vs 108 [81%] of 134), and in patients who were naive to hypertonic saline versus those who were not (33 [89%] of 39 vs 134 [82%] of 169). Of the 167 paired cough swab and sputum-induction samples, 63 (38%) sputum-induction samples were pathogen positive compared with 24 (14%) cough swabs (p<0·0001; OR 7·5; 95% CI 3·19–17·98). Repeated measures in individuals who gave more than one sample did not affect pathogen positivity in sputum-induction samples (p=0·99). An intention-to-treat analysis (ITT) in which unsuccessful sputum-induction attempts were classified as negative results remained significant (p<0·0001). In subgroup analysis by age, in children younger than 6 years, 18 (29%) of 62 sputum-induction samples were pathogen positive compared with eight (13%) of 62 cough swabs (p=0·021; ITT analysis p=0·049). In children aged 6 years or older, 45 (43%) of 105 sputum-induction samples were pathogen positive compared with 16 (15%) of 105 cough swabs (p<0·0001; ITT analysis p<0·0001). Age as a continuous variable was significant in predicting whether sputum would be pathogen positive (p=0·0028), whereas the ability to expectorate spontaneously before the procedure was not predictive (p=0·86). The ability to expectorate during the procedure did not show an independent effect over age on whether sputum would be pathogen positive (p=0·24).

ficant in predicting whether sputum would be pathogen positive (p=0·0028), whereas the ability to expectorate spontaneously before the procedure was not predictive (p=0·86). The ability to expectorate during the procedure did not show an independent effect over age on whether sputum would be pathogen positive (p=0·24). Sputum induction was more likely to be pathogen positive than cough swab in symptomatic children (46 [42%] of 110 vs 16 [15%] of 110; p<0·0001) and in asymptomatic children (17 [30%] of 57 sputum-induction samples vs eight [14%] of 57 cough swabs; p=0·049). The likelihood of sputum induction being pathogen positive was not significantly affected by whether the child was symptomatic or asymptomatic (p=0·15).

ildren (46 [42%] of 110 vs 16 [15%] of 110; p<0·0001) and in asymptomatic children (17 [30%] of 57 sputum-induction samples vs eight [14%] of 57 cough swabs; p=0·049). The likelihood of sputum induction being pathogen positive was not significantly affected by whether the child was symptomatic or asymptomatic (p=0·15). 86 pathogens were isolated from the 167 paired cough swab and sputum-induction samples (appendix). 79 (92%) were isolated on sputum-induction samples and 27 (31%) were isolated on cough swabs (p<0·0001; figure 2A). Of the 86 pathogens isolated, 59 (69%) were identified by sputum induction only. More than one pathogen was identified on 13 (21%) of 63 positive sputum-induction samples. When analysed by age group, in children younger than 6 years, 20 (83%) of 24 pathogens were isolated on sputum induction versus nine (38%) of 24 pathogens were isolated on cough swabs (p=0·019). In children aged 6 years or older, 59 (95%) of 62 pathogens were isolated on sputum induction whereas 18 (29%) of 62 on cough swab (p<0·0001). Sputum induction identified more of almost all the specific pathogens than cough swab (figure 2B).Figure 2 Pathogen yields from concurrent cough swab and sputum induction in 167 paired samples (A) Total pathogen yield in the whole cohort (n=167) and in subgroups of children younger than 6 years (n=62) and those aged 6 years or older (n=105). (B) Specific pathogen yields in the whole cohort (n=167). Bcc=Burkholderia cepacia complex. MRSA=meticillin-resistant Staphylococcus aureus. nTM=non-tuberculous Mycobacteria.

otal pathogen yield in the whole cohort (n=167) and in subgroups of children younger than 6 years (n=62) and those aged 6 years or older (n=105). (B) Specific pathogen yields in the whole cohort (n=167). Bcc=Burkholderia cepacia complex. MRSA=meticillin-resistant Staphylococcus aureus. nTM=non-tuberculous Mycobacteria. The additional pathogen yield (any pathogen, previously detected or not) from sputum induction compared with paired cough swab resulted in a new treatment in 52 (31%) of 167 cases. When compared with all pathogens isolated on repeated cough swabs from the preceding 12 months (median number of cough swabs six; IQR 5–7), a previously undetected pathogen was isolated in 40 (24%) of 167 sputum samples versus 15 (9%) of 167 cough swab samples (χ2 p=0·00039). Treatment for a previously undetected pathogen isolated exclusively on sputum induction was commenced in 32 (19%) of 167 cases.

12 months (median number of cough swabs six; IQR 5–7), a previously undetected pathogen was isolated in 40 (24%) of 167 sputum samples versus 15 (9%) of 167 cough swab samples (χ2 p=0·00039). Treatment for a previously undetected pathogen isolated exclusively on sputum induction was commenced in 32 (19%) of 167 cases. 41 of the 167 successful sputum induction and cough swab pairs from stage 1 were coupled with bronchoscopy and bronchoalveolar lavage for stage 2 (table 1). 35 patients contributed, of whom six individuals contributed twice (on two separate occasions). Median time between repeat procedures was 38 months (IQR 35–46). 32 (82%) of 41 procedures were done in symptomatic patients. Additionally, eight bronchoscopies were done on asymptomatic patients with unexplained poor spirometry tests and one bronchoscopy was done in an asymptomatic patient at the end of treatment for non-tuberculous Mycobacteria infection. Median age was 8·5 years (IQR 6·5–12·6). Six-lobe bronchoscopy was tolerated well in all patients. At least one pathogen was isolated from at least one of the concurrent samples in 28 (68%) of the 41 pairs. 39 different pathogens were isolated (table 2). Repeated measures in individuals recruited more than once did not affect pathogen positivity in either sputum-induction samples (p=0·94) or bronchoalveolar lavage samples (p=0·75).Table 2 Pathogen isolates from the paired cough swab, sputum induction, and bronchoalveolar samples

erent pathogens were isolated (table 2). Repeated measures in individuals recruited more than once did not affect pathogen positivity in either sputum-induction samples (p=0·94) or bronchoalveolar lavage samples (p=0·75).Table 2 Pathogen isolates from the paired cough swab, sputum induction, and bronchoalveolar samples Cough swab Sputum induction Bronchoalveolar lavage sample 1 (RML) Bronchoalveolar lavage sample 2 (LLi) Bronchoalveolar lavage sample 3 (RLL, RUL, LLL, LUL) 5 .. H influenzae H influenzae H influenzae .. 22* S aureus H influenzae; S aureus; P aeruginosa H influenzae; S aureus H influenzae; S aureus H influenzae; S aureus 45* .. B cenocepacia .. .. .. 57 .. .. H influenzae H influenzae H influenzae 60 .. P aeruginosa .. .. .. 70† .. A xylosoxidans .. H influenzae H influenzae; A xylosoxidans 73† .. S aureus S aureus S aureus S aureus; B multivorans 78 .. P aeruginosa; B multivorans P aeruginosa P aeruginosa; B multivorans P aeruginosa; B multivorans 79 P aeruginosa .. .. .. .. 80† .. S aureus S aureus S aureus S aureus 86‡ .. .. .. M abscessus .. 91 S aureus; S maltophilia S aureus; S maltophilia S aureus S aureus S aureus; S maltophilia 101† .. .. H influenzae; S aureus H influenzae; S aureus H influenzae; S aureus 104† H influenzae; S aureus H influenzae; S aureus; B multivorans H influenzae; S aureus; B multivorans H influenzae; S aureus; B multivorans H influenzae; S aureus; B multivorans 107 .. B cepacia .. .. .. 108* .. S aureus; B multivorans S aureus B multivorans S aureus; B multivorans 115 .. .. MRSA MRSA MRSA 121‡ .. S maltophilia S maltophilia S maltophilia S aureus; S maltophilia 127† .. .. .. S maltophilia S maltophilia 134 .. MRSA .. .. .. 174* P aeruginosa P aeruginosa P aeruginosa P aeruginosa P aeruginosa 178 .. .. .. .. P aeruginosa 179 S aureus S aureus S aureus S aureus S aureus 184 .. H influenzae .. H influenzae H influenzae 196 .. .. S aureus S aureus S aureus 208† .. M abscessus M abscessus .. .. 209 .. P aeruginosa P aeruginosa P aeruginosa P aeruginosa 212 .. H influenzae H influenzae H influenzae H influenzae Of the 13 contributions that were negative with all sampling techniques (not shown), two were from patients who were asymptomatic. Of the six patients who contributed twice, one had no pathogens detected in either contributions. RML=right middle lobe. LLi=left lingular. RLL=right lower lobe. RUL=right upper lobe. LLL=left lower lobe. LUL=left upper lobe. MRSA=meticillin-resistant Staphylococcus aureus.

two were from patients who were asymptomatic. Of the six patients who contributed twice, one had no pathogens detected in either contributions. RML=right middle lobe. LLi=left lingular. RLL=right lower lobe. RUL=right upper lobe. LLL=left lower lobe. LUL=left upper lobe. MRSA=meticillin-resistant Staphylococcus aureus. * Patients who contributed twice; other contribution was negative. † Patients who were asymptomatic. ‡ One patient contributed twice and had different pathogens detected on the repeat procedure. Regarding pathogen yield from the different sampling techniques, sputum induction isolated 27 (69%) of the 39 pathogens compared with 22 (56%; p=0·092; OR 3·3, 95% CI 0·91–12·11) on single-lobe bronchoalveolar lavage, 28 (72%; p=1·0; OR 1·1, 95% CI 0·41–3·15) on two-lobe bronchoalveolar lavage, and 33 (85%; p=0·21; OR 2·2, 95% CI 0·76–6·33) on six-lobe bronchoalveolar lavage (figure 3A). Increasing numbers of pathogens were isolated on sequentially wider bronchoalveolar lavage sampling (χ2 p=0·023).Figure 3 Pathogen yield for concurrent cough swab, sputum induction, and single-lobe, two-lobe, and six-lobe BAL in 41 matched samples (A) Total pathogen yield from each technique. (B) Numbers of unique and overlapping pathogen isolates for the different techniques. (C) Specific pathogen yield. BAL=bronchoalveolar lavage. Bcc=Burkholderia cepacia complex. MRSA=meticillin-resistant Staphylococcus aureus. nTM=non-tuberculous Mycobacteria.

Regarding pathogen yield from the different sampling techniques, sputum induction isolated 27 (69%) of the 39 pathogens compared with 22 (56%; p=0·092; OR 3·3, 95% CI 0·91–12·11) on single-lobe bronchoalveolar lavage, 28 (72%; p=1·0; OR 1·1, 95% CI 0·41–3·15) on two-lobe bronchoalveolar lavage, and 33 (85%; p=0·21; OR 2·2, 95% CI 0·76–6·33) on six-lobe bronchoalveolar lavage (figure 3A). Increasing numbers of pathogens were isolated on sequentially wider bronchoalveolar lavage sampling (χ2 p=0·023).Figure 3 Pathogen yield for concurrent cough swab, sputum induction, and single-lobe, two-lobe, and six-lobe BAL in 41 matched samples (A) Total pathogen yield from each technique. (B) Numbers of unique and overlapping pathogen isolates for the different techniques. (C) Specific pathogen yield. BAL=bronchoalveolar lavage. Bcc=Burkholderia cepacia complex. MRSA=meticillin-resistant Staphylococcus aureus. nTM=non-tuberculous Mycobacteria. The sputum-induction procedure identified 17 (77%) of 22 of the pathogens present on single-lobe bronchoalveolar lavage, 20 (71%) of 28 on two-lobe bronchoalveolar lavage, and 22 (66%) of 33 on six-lobe bronchoalveolar lavage (figure 3B). Conversely, of the 27 pathogens present on sputum induction samples, single-lobe bronchoalveolar lavage isolated 17 (38%), two-lobe bronchoalveolar lavage isolated 20 (70%), and six-lobe bronchoalveolar lavage isolated 22 (80%).

bronchoalveolar lavage, and 22 (66%) of 33 on six-lobe bronchoalveolar lavage (figure 3B). Conversely, of the 27 pathogens present on sputum induction samples, single-lobe bronchoalveolar lavage isolated 17 (38%), two-lobe bronchoalveolar lavage isolated 20 (70%), and six-lobe bronchoalveolar lavage isolated 22 (80%). For some specific pathogens, sputum induction outperformed six-lobe bronchoalveolar lavage (figure 3C). Five important pathogens (13% of total) were identified on sputum induction but not on six-lobe bronchoalveolar lavage (two P aeruginosa and one each of Burkholderia cepacia, Burkholderia cenocepacia, and MRSA). Because these pathogens were not isolated from paired cough swabs either, they are likely to be from the lower airway compartment. Test-specific detection rates were 69% for sputum induction and 72% for two-lobe bronchoalveolar lavage, and for the combination of sputum induction with two-lobe bronchoalveolar lavage it was 90%. The test-specific detection rate for six-lobe bronchoalveolar lavage was 84%, and for the combination of sputum induction with six-lobe bronchoalveolar lavage it was 98%. These data suggest that sputum induction and bronchoalveolar lavage sample closely related, but non-identical, lower airway compartments, and that each therefore has an independent contribution in the identification of lower airway pathogens.

bination of sputum induction with six-lobe bronchoalveolar lavage it was 98%. These data suggest that sputum induction and bronchoalveolar lavage sample closely related, but non-identical, lower airway compartments, and that each therefore has an independent contribution in the identification of lower airway pathogens. To further investigate the sampling ability of sputum induction, we extracted total DNA from cough swab, sputum induction, and bronchoalveolar lavage samples and used RISA profiling to assess the polymicrobial signatures from bacterial DNA present in those samples. In one illustrative example of RISA profiles of concurrent samples by different techniques, the sputum-induction polymicrobial signature is directly related to that of bronchoalveolar lavage, and discrete from cough swab (figure 4A). In another illustrative example, the sputum-induction polymicrobial signature is a combination of contributions from multiple bronchoalveolar lavage sample sites, indicating that sputum induction can be effective in sampling from a very wide lower airway compartment (figure 4B).Figure 4 Two illustrative examples from two individuals of polymicrobial DNA signatures or RISA profiles from concurrent cough swab, sputum induction, and BAL samples (A) Example 1. (B) Example 2. BAL=bronchoalveolar lavage.

To further investigate the sampling ability of sputum induction, we extracted total DNA from cough swab, sputum induction, and bronchoalveolar lavage samples and used RISA profiling to assess the polymicrobial signatures from bacterial DNA present in those samples. In one illustrative example of RISA profiles of concurrent samples by different techniques, the sputum-induction polymicrobial signature is directly related to that of bronchoalveolar lavage, and discrete from cough swab (figure 4A). In another illustrative example, the sputum-induction polymicrobial signature is a combination of contributions from multiple bronchoalveolar lavage sample sites, indicating that sputum induction can be effective in sampling from a very wide lower airway compartment (figure 4B).Figure 4 Two illustrative examples from two individuals of polymicrobial DNA signatures or RISA profiles from concurrent cough swab, sputum induction, and BAL samples (A) Example 1. (B) Example 2. BAL=bronchoalveolar lavage. We did not want to discount those pathogens isolated by sputum induction alone as false positives, but rather, to classify them as additional lower airway pathogens. We therefore generated a sensitivity analysis against a combined gold standard consisting of all pathogens identified by sputum induction and six-lobe bronchoalveolar lavage. We defined a positive outcome as the ability to identify all pathogens in the combined gold standard, so the outcome would enable us to quantify the ability of any single approach to correctly detect all lower airway pathogens in a patient. Sensitivity of sputum induction was 0·63 (95% CI 0·42–0·79), sensitivity of two-lobe bronchoalveolar lavage was 0·59 (0·39–0·77), and sensitivity of six-lobe bronchoalveolar lavage was 0·81 (0·61–0·93). Sensitivity of combined sputum induction and two-lobe bronchoalveolar lavage was 0·93 (0·74–0·99). The ability of the sputum-induction procedure to correctly identify all lower airway pathogens in a given patient was not significantly influenced by age as a continuous variable (p=0·95) or by whether the patient was asymptomatic or symptomatic (p=0·27).

uction and two-lobe bronchoalveolar lavage was 0·93 (0·74–0·99). The ability of the sputum-induction procedure to correctly identify all lower airway pathogens in a given patient was not significantly influenced by age as a continuous variable (p=0·95) or by whether the patient was asymptomatic or symptomatic (p=0·27). Objective tolerance of the sputum induction procedure was good with no significant effects on respiratory rate, heart rate, or FEV1% (figure 5). FEV1 increased by more than 5% in 21 (23%) of 90 people in whom spirometry was done. Subjective tolerance was good. Likert scales rated tolerance high, with mean parent or patient scores of 8·55 (SD 1·65) and physiotherapist scores of 9·09 (1·76).Figure 5 Objective assessment of tolerance to the sputum-induction procedure in 200 attempted procedures Before and after procedure measurements of FEV1 (where applicable), respiratory rate, and heart rate. Children in 27 (14%) of 200 sputum-induction procedures had mild side-effects: 17 (9%) became upset, of which four (2%) could not complete the procedure; six (3%) had mild wheeze, of which two (1%) could not complete the procedure; three (2%) patients vomited during oropharyngeal suction; one (<1%) became transiently dizzy. 108 (96%) of 112 (12 gave no comment) patients or parents were willing to have regular annual sputum-induction procedures.

t complete the procedure; six (3%) had mild wheeze, of which two (1%) could not complete the procedure; three (2%) patients vomited during oropharyngeal suction; one (<1%) became transiently dizzy. 108 (96%) of 112 (12 gave no comment) patients or parents were willing to have regular annual sputum-induction procedures. Discussion In this interventional trial, we took a systematic approach to comprehensively investigate and compare bacterial sampling techniques in young children with cystic fibrosis. In particular, we tested sputum induction as an infection diagnostic technique. We compared pathogen yield from sputum induction with concurrent cough swab, single-lobe bronchoalveolar lavage, two-lobe bronchoalveolar lavage, and comprehensive six-lobe bronchoalveolar lavage to identify the relative contribution of each approach. Most patients recruited to both stages of the study were unable to spontaneously expectorate before the sputum-induction procedure irrespective of whether they had a wet cough or not. We found the sputum-induction procedure to be well tolerated and equally successful in all age groups, in the inpatient or outpatient setting, in those who were asymptomatic or symptomatic, and in children with or without a wet cough.

e the sputum-induction procedure irrespective of whether they had a wet cough or not. We found the sputum-induction procedure to be well tolerated and equally successful in all age groups, in the inpatient or outpatient setting, in those who were asymptomatic or symptomatic, and in children with or without a wet cough. In stage 1 of this study, sputum induction was compared with paired cough swab. Cough swab pathogen positivity in this study was equivalent to that reported in similar populations in other studies.22 Almost three times as many pathogens were identified on sputum induction compared with cough swab, and this benefit was reflected to a broadly similar degree in all age groups. More pathogens were identified on both sputum induction and cough swab in children aged 6 years or older, reflecting the greater pathogen prevalence in the older age group. The benefits of sputum induction over cough swab were seen in symptomatic and asymptomatic children, supporting the use of sputum induction over cough swab in all situations. Sputum induction was positive for a pathogen in 38% of paired samples, positive for a pathogen not identified on paired cough swab in 31% of cases, and positive for a new pathogen not isolated on repeated cough swabs from the preceding 12 months in 24% of cases. Sputum induction had a considerable effect on patient care, with new treatment implemented as a consequence in 31% of cases.

es, positive for a pathogen not identified on paired cough swab in 31% of cases, and positive for a new pathogen not isolated on repeated cough swabs from the preceding 12 months in 24% of cases. Sputum induction had a considerable effect on patient care, with new treatment implemented as a consequence in 31% of cases. In stage 2 of the study, sputum induction was compared with single-lobe, two-lobe, and six-lobe bronchoalveolar lavage in a group of children who were largely symptomatic. Sequentially higher proportions of pathogens were detected by single-lobe, two-lobe, and six-lobe bronchoalveolar lavage. The proportion of pathogens isolated by sputum induction and two-lobe bronchoalveolar lavage were largely equivalent. Six-lobe bronchoalveolar lavage only identified 85% of pathogens isolated from all approaches, since some important pathogens were uniquely isolated on sputum induction. Patient age did not affect the ability for sputum induction to correctly describe all lower airway pathogens.

alveolar lavage were largely equivalent. Six-lobe bronchoalveolar lavage only identified 85% of pathogens isolated from all approaches, since some important pathogens were uniquely isolated on sputum induction. Patient age did not affect the ability for sputum induction to correctly describe all lower airway pathogens. By using multi-approach concurrent sampling from upper and lower airways in the same patient, we estimated whether pathogens identified by sputum induction were upper or lower airway residents. A large proportion of pathogens identified by sputum induction were identified on bronchoalveolar lavage, confirming that sputum induction does effectively sample the lower airway. We found that with sequentially wider bronchoalveolar lavage sampling, more pathogens that were isolated on sputum induction were also identified on bronchoalveolar lavage, indicating that sputum induction is capable of sampling very widely from the lower airway. Using DNA RISA profiling we showed that sputum induction can capture the diversity of bacteria associated with multiple bronchoalveolar compartments.

re isolated on sputum induction were also identified on bronchoalveolar lavage, indicating that sputum induction is capable of sampling very widely from the lower airway. Using DNA RISA profiling we showed that sputum induction can capture the diversity of bacteria associated with multiple bronchoalveolar compartments. A proportion of pathogens (13%) were identified on sputum induction but not on six-lobe bronchoalveolar lavage or cough swabs. With the assumption that concurrent cough swab and bronchoalveolar lavage samples were true negatives,23 this finding suggests that sputum induction can identify pathogens from compartments of the respiratory tract not sampled by the other methods, and raises the question as to where these pathogens reside. Bacterial bronchitis is common in young children with cystic fibrosis who are symptomatic, suggesting the predominant focus of acute infection might often be the large intrathoracic airways rather than the alveolar bed. The large intrathoracic airways together constitute a lower-airway compartment that is most easily sampled by the sputum-induction procedure, is largely bypassed by even the most extensive approach to bronchoalveolar lavage, and is a compartment perhaps inadequately considered in the current paradigm of lower airway sampling. In this study, sputum induction was more successful at isolating the important Gram-negative pathogens, P aeruginosa and B cepacia complex species, compared with concurrent six-lobe lavage (figure 3C). This raises the possibility that for some pathogens the large intrathoracic airways might be the optimal lower airway environment for early infection.

was more successful at isolating the important Gram-negative pathogens, P aeruginosa and B cepacia complex species, compared with concurrent six-lobe lavage (figure 3C). This raises the possibility that for some pathogens the large intrathoracic airways might be the optimal lower airway environment for early infection. We generated a sensitivity analysis against a combined gold standard consisting of all pathogens identified by sputum induction and six-lobe bronchoalveolar lavage. Sputum induction is shown in these data to be marginally more sensitive than the current gold-standard two-lobe bronchoalveolar lavage (0·63 vs 0·59). Six-lobe bronchoalveolar lavage has a higher sensitivity at 0·81. However, sputum induction and two-lobe bronchoalveolar lavage combined have a sensitivity of 0·93. These data highlight independent, sizeable gains in pathogen detection from both sputum induction and extended six-lobe bronchoalveolar lavage, over and above the current gold-standard two-lobe bronchoalveolar lavage. This in turn questions whether two-lobe bronchoalveolar lavage can be considered adequate as a standalone approach to understanding lower airway microbiology. We advocate from the present data that sputum induction and six-lobe lavage should be done together as a new standard of care for comprehensive assessment of the lower airway pathogen environment in children with cystic fibrosis.

e considered adequate as a standalone approach to understanding lower airway microbiology. We advocate from the present data that sputum induction and six-lobe lavage should be done together as a new standard of care for comprehensive assessment of the lower airway pathogen environment in children with cystic fibrosis. The present data also support sputum induction as a non-invasive surrogate for bronchoalveolar lavage. We propose that in symptomatic patients, bronchoscopy and bronchoalveolar lavage should be reserved for those who have not responded to appropriate or empirical antibiotic treatment and whose sputum-induction cultures do not explain the persistence of symptoms. From our data, a successful sputum-induction sample taken before bronchoscopy and six-lobe bronchoalveolar lavage correctly identified all lower airway pathogens in 63% of patients who had one or more pathogen present. The routine use of sputum induction followed by appropriate treatment for those pathogens isolated could therefore substantially reduce the need for bronchoscopy in symptomatic patients with cystic fibrosis. This has notable implications both for quality of care and cost.

patients who had one or more pathogen present. The routine use of sputum induction followed by appropriate treatment for those pathogens isolated could therefore substantially reduce the need for bronchoscopy in symptomatic patients with cystic fibrosis. This has notable implications both for quality of care and cost. Limitations of the study include the fact that study investigators were not blinded to outcome, recruitment was not randomised, and some patients contributed on more than one occasion. However, the study is likely to be representative because 70% of the patients who attend the South, West, and Mid-Wales Children's Cystic Fibrosis Service were recruited into this study. We adjusted statistical outcomes for repeated measures in those patients that were recruited on more than one occasion. The bronchoalveolar lavage data apply to a cohort of patients who were largely symptomatic, because patients were recruited when bronchoscopy was clinically indicated. The outcomes of stage 2 of the study therefore relate to symptomatic patients and might not be directly applicable to surveillance programmes in asymptomatic patients with cystic fibrosis.

y to a cohort of patients who were largely symptomatic, because patients were recruited when bronchoscopy was clinically indicated. The outcomes of stage 2 of the study therefore relate to symptomatic patients and might not be directly applicable to surveillance programmes in asymptomatic patients with cystic fibrosis. In conclusion, we have established sputum induction as superior to cough swab, and as a credible approach to sampling the lower airway in symptomatic children with cystic fibrosis. We showed benefit in patients of all ages, and in those who are unable to spontaneously expectorate. We suggest that the large intrathoracic airways constitute an important lower airway compartment that is inadequately sampled by standard approaches to pathogen surveillance in children with cystic fibrosis. We have shown that both sputum induction and six-lobe bronchoalveolar lavage contribute important independent gains in pathogen detection compared with the current gold-standard two-lobe bronchoalveolar lavage, and advocate sputum induction and six-lobe bronchoalveolar lavage combined as a new standard of care in the assessment of the lower airway pathogen environment in children with cystic fibrosis. In symptomatic patients, doing sputum induction before bronchoalveolar lavage will correctly describe the lower airway pathogen environment in almost two-thirds of patients, and if used routinely, could substantially reduce the number of bronchoscopy procedures required. We recommend implementation of sputum induction as a regular contribution to cystic fibrosis care in children. This recommendation is supported by the tolerability of the procedure in all age groups, the ease of repeatability, the acceptability to parents and patients, the high success of obtaining samples, the high proportion of pathogens identified, the applicability to both the inpatient and outpatient setting, and the clear economic savings compared with bronchoscopy.

ability of the procedure in all age groups, the ease of repeatability, the acceptability to parents and patients, the high success of obtaining samples, the high proportion of pathogens identified, the applicability to both the inpatient and outpatient setting, and the clear economic savings compared with bronchoscopy. Supplementary Material Supplementary appendix Contributors JTF conceived and designed the study, acted as chief investigator, consented and enrolled patients, collected bronchoalveolar lavage samples, collated and managed samples, collected, analysed, and interpreted the data, and wrote the paper. EM and CP collated and managed samples, developed protocols for DNA extraction from the respiratory samples, extracted DNA and generated RISA profiles, and reviewed and approved the paper. LPT and ID contributed to consenting and enrolling patients, to bronchoalveolar lavage collection, and reviewed and approved the paper. KR and J-DT contributed to consenting patients, collected all cough swab and sputum-induction samples, and reviewed and approved the paper. RH managed research samples in the microbiology lab and reviewed and approved the paper. Declaration of interests We declare no competing interests.

urce The funders of the study had no role in study design, data collection, data analysis, data interpretation, writing of the report, or the decision to submit for publication. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results Between Feb 12, 2016, and July 18, 2017, we obtained blood RNA samples from 205 consecutive patients.10 Paired sputum and RNA sequencing data were available in 181 participants included in our analysis (figure 1). Their baseline characteristics are given in table 1; characteristics of participants who were excluded from the analysis are in appendix 1 (p 8). 54 (30%) of 181 patients had pulmonary tuberculosis, confirmed by sputum culture or Xpert, and included all the individuals who received tuberculosis treatment at enrolment, further increasing our confidence in the sensitivity of our standard reference for the diagnosis of tuberculosis (appendix 1, p 15). 44 (24%) of 181 patients were HIV-infected.Figure 1 Study flowchart Table 1 Baseline characteristics of study cohort

Introduction Live attenuated influenza vaccine (LAIV) has been highly efficacious against prepandemic seasonal H1N1 viruses in children, with a meta-analysis1 estimating a pooled efficacy of 85% from several randomised controlled trials. However, concerns have been expressed about protection against pandemic H1N1 (pH1N1) influenza using LAIV. Since 2009, when pH1N1 viruses have circulated as the main seasonal H1N1 strain, vaccine effectiveness of the Ann Arbor-backbone LAIV against pH1N1 in the USA has been low, ranging from −21% in 2015–16 to 17% in 2013–14.2 This reduction in effectiveness resulted in the Advisory Committee on Immunisation Practices removing their recommendation for LAIV use in 2016.3 Research in context Evidence before this study

Introduction Live attenuated influenza vaccine (LAIV) has been highly efficacious against prepandemic seasonal H1N1 viruses in children, with a meta-analysis1 estimating a pooled efficacy of 85% from several randomised controlled trials. However, concerns have been expressed about protection against pandemic H1N1 (pH1N1) influenza using LAIV. Since 2009, when pH1N1 viruses have circulated as the main seasonal H1N1 strain, vaccine effectiveness of the Ann Arbor-backbone LAIV against pH1N1 in the USA has been low, ranging from −21% in 2015–16 to 17% in 2013–14.2 This reduction in effectiveness resulted in the Advisory Committee on Immunisation Practices removing their recommendation for LAIV use in 2016.3 Research in context Evidence before this study We searched Embase, MEDLINE, Global health, and Web of Science databases up to May 9, 2018, with the terms: (“influenza” OR “flu”) AND (“vaccin*” OR “immuni#ation” OR “Influenza Vaccines [Subject heading]”) AND (“effic*” OR “effect*” OR “immune*” OR “respons*” OR “protect*”) AND (“Africa” OR “Africa [Subject heading]” OR each African country [defined by the UN]). This search strategy identified no live attenuated influenza vaccine (LAIV) immunogenicity studies and only two LAIV efficacy studies in African children. The first study was of the Ann-Arbor LAIV in 2001–02, when pre-pandemic H1N1 was circulating. This study was a multicentre randomised placebo-controlled trial that included 277 children aged 6–36 months from South Africa. The efficacy of LAIV in this subset of children was 87% (95% CI 64–95). The second study was a single-centre, randomised, placebo-controlled trial of Russian-backbone LAIV in children aged 2–5 years in Senegal. The study was done in 2013 and, therefore, included pandemic H1N1 (pH1N1). The efficacy against vaccine-matched isolates was −6·1% (95% CI −50·0 to 25·0), and pH1N1 was the predominant circulating strain. In the 2017–18 season, the pH1N1 component in LAIV was updated for the first time in both the Ann Arbor and Russian-backbone LAIVs, with A/Michigan/45/2015-like strains. No studies have been published about whether this change has affected the shedding and immunogenicity of pH1N1 in LAIV.

redominant circulating strain. In the 2017–18 season, the pH1N1 component in LAIV was updated for the first time in both the Ann Arbor and Russian-backbone LAIVs, with A/Michigan/45/2015-like strains. No studies have been published about whether this change has affected the shedding and immunogenicity of pH1N1 in LAIV. Added value of this study Our findings show that A/17/California/2009/38 pH1N1 strain shedding and immunogenicity is less than that of H3N2 and influenza B in the Russian-backbone LAIV, providing an explanation for the lack of efficacy seen in the randomised controlled trial in Senegal. Our data suggest this observation is not attributable to reduced pH1N1 vaccine take secondary to pre-existing immune responses. Our findings show for the first time that updating the Russian-backbone LAIV pH1N1 component has resulted in a vaccine with significantly greater nasopharyngeal shedding, seroconversion, and influenza-specific T-cell induction to pH1N1. We are able to model this difference in replicative ability of old and new pH1N1 LAIV components in vitro. Implications of all the available evidence

Our findings show that A/17/California/2009/38 pH1N1 strain shedding and immunogenicity is less than that of H3N2 and influenza B in the Russian-backbone LAIV, providing an explanation for the lack of efficacy seen in the randomised controlled trial in Senegal. Our data suggest this observation is not attributable to reduced pH1N1 vaccine take secondary to pre-existing immune responses. Our findings show for the first time that updating the Russian-backbone LAIV pH1N1 component has resulted in a vaccine with significantly greater nasopharyngeal shedding, seroconversion, and influenza-specific T-cell induction to pH1N1. We are able to model this difference in replicative ability of old and new pH1N1 LAIV components in vitro. Implications of all the available evidence Impaired replicative ability of pH1N1 components in LAIV might have caused recent low efficacy and effectiveness of LAIV. An improvement in protection against pH1N1 can be expected in the future. These data highlight the importance of assessing viral replicative fitness in addition to antigenicity when selecting vaccine formulations. Studies are needed to ascertain whether improved shedding and immunogenicity translates into improved efficacy and effectiveness. Further research is needed to understand the genetic factors that underlie these phenotypes in vaccine strains to design more rational choices of vaccine antigens for LAIV.

ine formulations. Studies are needed to ascertain whether improved shedding and immunogenicity translates into improved efficacy and effectiveness. Further research is needed to understand the genetic factors that underlie these phenotypes in vaccine strains to design more rational choices of vaccine antigens for LAIV. A randomised controlled trial of Russian-backbone LAIV (Nasovac-S; Serum Institute of India Pvt, Pune, India) among children aged 2–5 years in Senegal did not show efficacy (0·0%, 95% CI −26·4 to 20·9) in 2013, when pH1N1 was the predominant circulating vaccine-matched virus.4 Both LAIV formulations in these studies contained haemagglutinin and neuraminidase from pH1N1 A/California/07/2009-like (Cal09) viruses. It is unclear why protection conferred by the pH1N1 component in LAIV has been suboptimal. Potential reasons include pre-existing immunity, poor viral replicative fitness, or competition from other co-formulated strains—all limiting pH1N1 take and immunogenicity.3

e from pH1N1 A/California/07/2009-like (Cal09) viruses. It is unclear why protection conferred by the pH1N1 component in LAIV has been suboptimal. Potential reasons include pre-existing immunity, poor viral replicative fitness, or competition from other co-formulated strains—all limiting pH1N1 take and immunogenicity.3 The Russian-backbone LAIV (Nasovac-S) was granted a WHO prequalification certificate in 2015, opening up the potential for use in low-income and middle-income countries. The findings in Senegal with this vaccine are especially pertinent because the burden of influenza in Africa is high; influenza-related admissions to hospital in children younger than 5 years are approximately threefold higher than in Europe.5 Superior efficacy of LAIV over inactivated influenza vaccine in young children (predominantly in high-income settings),1 needle-free delivery, and lower manufacturing costs make LAIV an attractive option to tackle this burden in Africa.3 However, few LAIV studies have been done in African cohorts and no published immunogenicity data are available from African children to date.6 In particular, the absence of immunological endpoints from the randomised controlled trial in Senegal makes it difficult to understand the reasons for the lack of efficacy recorded.4

er, few LAIV studies have been done in African cohorts and no published immunogenicity data are available from African children to date.6 In particular, the absence of immunological endpoints from the randomised controlled trial in Senegal makes it difficult to understand the reasons for the lack of efficacy recorded.4 In 2017–18, the pH1N1 Cal09 strain (A/17/California/2009/38) was updated according to WHO recommendations to an A/Michigan/45/2015-like strain (A/17/New York/15/5364 [NY15]), following antigenic drift. This first-ever recommended update to pH1N1 provided a unique opportunity to compare replicative ability and immunogenicity of these two pH1N1 strains. To understand how differences in strain shedding and immunogenicity might account for the findings of the Senegal trial, we compared one cohort of influenza vaccine-naive Gambian children vaccinated with the Russian-backbone Cal09 LAIV formulation from 2016–17 with a second cohort vaccinated with the NY15 LAIV formulation from 2017–18.

differences in strain shedding and immunogenicity might account for the findings of the Senegal trial, we compared one cohort of influenza vaccine-naive Gambian children vaccinated with the Russian-backbone Cal09 LAIV formulation from 2016–17 with a second cohort vaccinated with the NY15 LAIV formulation from 2017–18. Methods Study design and participants We did an open-label, prospective, observational, phase 4 immunogenicity study in Sukuta, a periurban area in The Gambia. Our study is nested within a larger randomised trial comparing microbiome changes in children assigned LAIV with changes in unvaccinated children (NCT02972957; appendix pp 3, 4). Data in our study are from all children enrolled in the randomised trial who were given LAIV. After community sensitisation, parents expressing an interest in the randomised study were invited for consent discussions. Eligible children had to be aged 24–59 months and clinically well, with no history of respiratory illness within the past 14 days (appendix p 2). This study was approved by The Gambia Government and UK Medical Research Council (MRC) joint ethics committee and the Medicines Control Agency of The Gambia, and it was done according to International Conference on Harmonisation Good Clinical Practice standards. A parent provided written or thumbprinted informed consent for their child or children to participate. If parents were not English literate, an impartial witness was present throughout the informed consent discussion undertaken in a local language, who signed to confirm completeness of the consent provided.

standards. A parent provided written or thumbprinted informed consent for their child or children to participate. If parents were not English literate, an impartial witness was present throughout the informed consent discussion undertaken in a local language, who signed to confirm completeness of the consent provided. Procedures When LAIV was updated in 2017–18, haemagglutinin and neuraminidase from pH1N1 Cal09 were replaced with those from NY15, whereas identical H3N2 (A/17/Hong Kong/2014/8296) and B/Vic (B/Texas/02/2013 [Victoria lineage]) strains were used. Vaccine titres per dose (50% egg infectious dose equivalents [EID50eq] per mL) were 1 × 108·0 for pH1N1, 1 × 107·5 for H3N2, 1 × 107·2 for B/Vic in the 2016–17 LAIV and 1 × 107·7 for pH1N1, 1 × 107·6 for H3N2, 1 × 107·3 for B/Vic in the 2017–18 LAIV. The study was done outside the peak influenza transmission season (June to October) based on surveillance data from Senegal and unpublished data from studies in The Gambia.7 Children received one dose of intranasal trivalent Russian-backbone LAIV (Nasovac-S; northern hemisphere formulation) in either 2017 (the Cal09 strain from 2016–17) or 2018 (the NY15 strain from 2017–18 formulation). Children received the vaccine formulation that corresponded with their year of enrolment.

The Gambia.7 Children received one dose of intranasal trivalent Russian-backbone LAIV (Nasovac-S; northern hemisphere formulation) in either 2017 (the Cal09 strain from 2016–17) or 2018 (the NY15 strain from 2017–18 formulation). Children received the vaccine formulation that corresponded with their year of enrolment. Nasopharyngeal swabs were taken before vaccination (day 0), on day 2, and on day 7 using flocked swabs (FLOQSwabs; Copan, Murrieta, CA, USA). We obtained buccal cavity oral fluid with swabs (Oracol Plus; Malvern Medical Development, Worcester, UK) on day 0 and day 21. Whole blood samples were obtained for flow cytometry and serum separation on day 0 and day 21. We chose day 21 to measure vaccine response, in line with previous work.8, 9, 10 Nasopharyngeal swabs, oral fluid, and serum samples were stored at −70°C before further processing.

lopment, Worcester, UK) on day 0 and day 21. Whole blood samples were obtained for flow cytometry and serum separation on day 0 and day 21. We chose day 21 to measure vaccine response, in line with previous work.8, 9, 10 Nasopharyngeal swabs, oral fluid, and serum samples were stored at −70°C before further processing. Haemagglutinin inhibition assays were done according to standard methods,11 using vaccine haemagglutinin-matched and neuraminidase-matched viruses. Seroconversion was defined as a fourfold or greater titre increase (to ≥1:40) from day 0 to day 21. Total and influenza haemagglutinin-specific IgA in oral fluid was detected using a previously described ELISA,12 using recombinant vaccine-matched haemagglutinin. Samples were assayed at dilutions ranging from 1:1000 to 1:20000 for total IgA and from undiluted to 1:16 for influenza-specific IgA, and samples were quantified using an IgA standard curve. Undiluted samples with influenza-specific IgA below the limit of quantitation (LOQ) were assigned LOQ values. We calculated the fold change in the proportion of influenza-specific IgA to total IgA from day 0 to day 21. A twofold increase was considered a significant response.13

e quantified using an IgA standard curve. Undiluted samples with influenza-specific IgA below the limit of quantitation (LOQ) were assigned LOQ values. We calculated the fold change in the proportion of influenza-specific IgA to total IgA from day 0 to day 21. A twofold increase was considered a significant response.13 T-cell responses were quantified by stimulating fresh whole blood (200 μL) on the day of collection for 18 h with overlapping 15–18-mer peptide pools (2 μg/mL) covering vaccine-matched whole haemagglutinin, matrix and nucleoprotein, and co-stimulatory antibodies (antiCD28 and antiCD49; BD Biosciences, Franklin Lakes, NJ, USA). Influenza B responses were measured in 2018 only. We did intracellular cytokine staining for interferon (IFN)γ and interleukin (IL)2 and analysed cells with a flow cytometer (LSR Fortessa; BD Biosciences; appendix pp 5, 6).14 Responses in negative controls (antiCD28 and antiCD49) were subtracted from peptide-stimulated conditions before further analysis; negative values were set to zero. To avoid systematic bias in adjusting for negative values alone, we set a threshold (based on the distribution of negative values; appendix p 7) below which all positive values were also considered a non-response, as described previously.14, 15 In analyses calculating the fold change from day 0 to day 21, null responses were assigned a value halfway between zero and this threshold. A twofold increase after LAIV was considered a significant response.

pendix p 7) below which all positive values were also considered a non-response, as described previously.14, 15 In analyses calculating the fold change from day 0 to day 21, null responses were assigned a value halfway between zero and this threshold. A twofold increase after LAIV was considered a significant response. Vaccine shedding from nasopharyngeal swabs was assessed with monoplex reverse-transcriptase PCR (RT-PCR) using haemagglutinin-specific primers and probes (appendix p 8). In 2018, fully quantitative RT-PCR results were obtained by inclusion of a standard curve with known vaccine titres (log10EID50eq per mL; appendix p 9). RT-PCR assays with primers and probes mapping to internal genes were used to distinguish LAIV from seasonal influenza viruses (appendix p 8).16 Despite optimisation of assay conditions, maximum LAIV dilutions detected by LAIV-specific RT-PCR were at least one log10 lower than those detected by haemagglutinin-specific RT-PCR (appendix p 10). Therefore, only samples with cycle threshold (ct) values of 30 or lower in seasonal influenza assays were tested with LAIV-specific assays, with 100% confirmed as LAIV strains.

dilutions detected by LAIV-specific RT-PCR were at least one log10 lower than those detected by haemagglutinin-specific RT-PCR (appendix p 10). Therefore, only samples with cycle threshold (ct) values of 30 or lower in seasonal influenza assays were tested with LAIV-specific assays, with 100% confirmed as LAIV strains. Primary human nasal epithelial cell cultures (MucilAir; Epithelix Sàrl, Geneva, Switzerland) were used for in vitro viral replication experiments. Madin-Darby Canine Kidney (MDCK) cells (ATCC, Manassas, VA, USA) and MDCK-SIAT cells (WHO Collaborating Centre for Reference and Research on Influenza, London, UK) were maintained at 37°C with 5% CO2 in Dulbecco's modified Eagle's Medium (DMEM; Gibco-Life Technologies, Waltham, MA, USA) supplemented with 10% fetal bovine serum, 1% penicillin–streptomycin, and 1% non-essential amino acids. We also added 1 mg/mL G418 (Gibco-Life Technologies) for MDCK-SIAT cells. Viral stocks of Nasovac-S monovalent forms were titrated by plaque assay at 32°C on MDCK cells (for pH1N1 and influenza B) or MDCK-SIAT cells (for H3N2). Apical surfaces of human nasal epithelial cells were inoculated with each monovalent virus (multiplicity of infection 0·01 plaque-forming units per cell) for 1 h at 32°C and 5% CO2 in triplicate. The inoculum was removed and the apical surface of the human nasal epithelial cells was washed with DMEM before incubation at 32°C. At indicated timepoints (days 1–6 after inoculation), DMEM was added to the human nasal epithelial cells and incubated for 30 min, then it was removed and stored; the stored DMEM—containing virions from the epithelial cell cultures—was titrated by plaque assay. Experiments were done on two separate occasions using cells from different donors.

(days 1–6 after inoculation), DMEM was added to the human nasal epithelial cells and incubated for 30 min, then it was removed and stored; the stored DMEM—containing virions from the epithelial cell cultures—was titrated by plaque assay. Experiments were done on two separate occasions using cells from different donors. To assess the acid stability of pH1N1 strains, Cal09 or NY15 were mixed with pH-adjusted MES (2-[N-morpholino]ethanesulphonic acid) buffer (100 mmol/L MES, 150 mol/L sodium chloride, 0·9 mol/L calcium chloride, 0·5 mol/L magnesium chloride) in triplicate (1:10 dilution) and the mixture was incubated for 15 min at room temperature. The buffer was inactivated with DMEM and infectious virus was titrated by plaque assay. Outcomes Primary shedding and immunogenicity outcomes were the percentage of children with LAIV strain shedding at day 2 and day 7, haemagglutinin inhibition seroconversion, and an increase in influenza haemagglutinin-specific IgA and T-cell responses at day 21 after LAIV.

To assess the acid stability of pH1N1 strains, Cal09 or NY15 were mixed with pH-adjusted MES (2-[N-morpholino]ethanesulphonic acid) buffer (100 mmol/L MES, 150 mol/L sodium chloride, 0·9 mol/L calcium chloride, 0·5 mol/L magnesium chloride) in triplicate (1:10 dilution) and the mixture was incubated for 15 min at room temperature. The buffer was inactivated with DMEM and infectious virus was titrated by plaque assay. Outcomes Primary shedding and immunogenicity outcomes were the percentage of children with LAIV strain shedding at day 2 and day 7, haemagglutinin inhibition seroconversion, and an increase in influenza haemagglutinin-specific IgA and T-cell responses at day 21 after LAIV. Statistical analysis The sample size calculation was based on LAIV microbiome endpoints not presented here (appendix p 3). Differences in unpaired proportions (shedding, seroconversion, IgA responses, and T-cell responses) between years (2017 and 2018) were assessed with either the χ2 test or Fisher's exact test. Differences in continuous variables (ct value, log10EID50eq per mL, and geometric mean fold change in haemagglutinin inhibition) between years (2017 and 2018) or serostatus (positive or negative) were assessed with either the unpaired t test or Mann-Whitney U test. Differences in viral load (log10EID50eq per mL) between strains (NY15, H3N2, and influenza B) within the same visit (paired data) were assessed with the Friedman test (with Dunn's post-test for multiple comparisons). Pairwise viral load correlations were assessed using Spearman's rank-order correlation (rs). Correlation coefficients were interpreted as low (rs=0·30–0·49), moderate (rs=0·50–0·69), high (rs=0·70–0·89), or very high (rs=0·90–1·00). The Wilcoxon signed-rank test was used to compare T-cell responses before and after vaccination. Separate logistic regression analyses were done for dependent variables (shedding, seroconversion, T-cell responses, and IgA responses). Independent variables were selected for multivariable logistic regression models if biologically relevant and the p value from univariable regression was less than 0·2. Each multivariable logistic regression model always included year, age, and sex as potential confounders. Viral loads were ascertained from standard curves using Python version 3.6 (SciPy package). In vitro viral replication was quantified using the area under the curve function in GraphPad Prism 5.0d (GraphPad Software, San Diego, CA, USA) and compared between NY15 and Cal09 strains using an unpaired t test. The proportion of monofunctional and dual-functional T-cell responses were estimated using Boolean gating on FlowJo 10.4 (FlowJo LLC, Ashland, OR, USA) and statistical significance between timepoints tested with the Permutation test in SPICE (version 6.0).15 Proportions are displayed with 95% CIs.

g an unpaired t test. The proportion of monofunctional and dual-functional T-cell responses were estimated using Boolean gating on FlowJo 10.4 (FlowJo LLC, Ashland, OR, USA) and statistical significance between timepoints tested with the Permutation test in SPICE (version 6.0).15 Proportions are displayed with 95% CIs. All tests were two-sided at the 5% significance level and were Bonferroni-adjusted for multiple comparisons within each set of analyses. Statistical analyses were done using R version 3.5.1, Stata release 12 (StataCorp, College Station, TX, USA), and GraphPad Prism 5.0d. Role of the funding source The funder had no role in study design, data collection, data analysis, data interpretation, or writing of the report. BBL, TIdS, EPA, YJJ, NIM, DJ, KH, and AS had access to raw data. The corresponding author had full access to all data in the study and had final responsibility for the decision to submit for publication.

e funder had no role in study design, data collection, data analysis, data interpretation, or writing of the report. BBL, TIdS, EPA, YJJ, NIM, DJ, KH, and AS had access to raw data. The corresponding author had full access to all data in the study and had final responsibility for the decision to submit for publication. Results Between Feb 8, 2017, and April 12, 2017, 118 children were enrolled and received one dose of the 2016–17 northern hemisphere formulation LAIV (Cal09 pH1N1; figure 1A). Between Jan 15, 2018, and March 28, 2018, a separate cohort of 135 children were enrolled and received one dose of the 2017–18 northern hemisphere formulation LAIV (NY15 pH1N1; figure 1B). 118 children in 2017 and 126 children in 2018 completed the study. All study visits were within protocol-defined windows (+1 day for day 2 visit, +7 days for day 7 visits, and +7 days for day 21 visits). In the 2017 cohort, 118 (100%) of 118 day 2 visits were 2 days after LAIV, 115 (97%) of 118 day 7 visits were 7 days after LAIV (three visits were 8 days, 12 days, and 14 days after LAIV), and 112 (95%) of 118 day 21 visits were 21 days after LAIV (five visits were 22 days after LAIV and one was 25 days after LAIV). In the 2018 cohort, 122 (97%) of 126 day 2 visits were 2 days after LAIV (four visits were 3 days after LAIV), 119 (94%) of 126 day 7 visits were 7 days after LAIV (seven visits were 8 days after LAIV), and 117 (93%) of 126 day 21 visits were 21 days after LAIV (eight visits were 22 days after LAIV and one visit was 26 days after LAIV). Baseline demographics did not differ significantly between the two cohorts with the exception of baseline haemagglutinin inhibition titres (table).Figure 1 Study profile

s after LAIV), and 117 (93%) of 126 day 21 visits were 21 days after LAIV (eight visits were 22 days after LAIV and one visit was 26 days after LAIV). Baseline demographics did not differ significantly between the two cohorts with the exception of baseline haemagglutinin inhibition titres (table).Figure 1 Study profile Overview of participants who received (A) the 2016–17 northern hemisphere Russian-backbone LAIV formulation and (B) the 2017–18 northern hemisphere Russian-backbone LAIV formulation. LAIV=live attenuated influenza vaccine. pH1N1=pandemic H1N1. HAI=haemagglutinin inhibition. *The study was nested within a larger randomised controlled trial (NCT02972957; appendix pp 3,4). †Sparse cell populations seen on flow cytometry. ‡Total IgA not detected in sample. §No pH1N1 data for one sample in 2016–17 cohort and no pH1N1 data for four samples and H3N2 data for three samples in 2017–18 cohort because of inadequate sample volume. Table Demographic characteristics and baseline influenza serological data

Overview of participants who received (A) the 2016–17 northern hemisphere Russian-backbone LAIV formulation and (B) the 2017–18 northern hemisphere Russian-backbone LAIV formulation. LAIV=live attenuated influenza vaccine. pH1N1=pandemic H1N1. HAI=haemagglutinin inhibition. *The study was nested within a larger randomised controlled trial (NCT02972957; appendix pp 3,4). †Sparse cell populations seen on flow cytometry. ‡Total IgA not detected in sample. §No pH1N1 data for one sample in 2016–17 cohort and no pH1N1 data for four samples and H3N2 data for three samples in 2017–18 cohort because of inadequate sample volume. Table Demographic characteristics and baseline influenza serological data 2016–17 LAIV (n=118) 2017–18 LAIV (n=126) p value Age (months) 35·1 (28·3–44·9) 35·3 (28·0–40·5) 0·44 Sex ·· ·· 0·61 Female 57 (48%) 56 (44%) ·· Male 61 (52%) 70 (56%) ·· Height (cm) 92·9 (7·4) 91·8 (6·3) 0·23 Weight (kg) 12·9 (2·1) 12·6 (1·7) 0·30 Weight-for-height malnutrition* ·· ·· 0·93 None 76 (64%) 82 (65%) ·· Mild 33 (28%) 36 (29%) ·· Moderate 9 (8%) 8 (6%) ·· Tribe ·· ·· 0·27 Mandinka 96 (81%) 99 (79%) ·· Wolof 5 (4%) 7 (6%) ·· Fula 3 (3%) 5 (4%) ·· Jola 6 (5%) 4 (3%) ·· Serehule 2 (2%) 5 (4%) ·· Serere 5 (4%) 1 (1%) ·· Other 1 (1%) 5 (4%) ·· History of ever being admitted to hospital with a respiratory infection 6 (5%) 3 (2%) 0·32 History of more than one respiratory infection needing medication in the past year 8 (7%) 13 (10%) 0·37 Age when stopped breastfeeding (months) 20 (18–24) 20 (18–24)† 0·80 Baseline seropositive (haemagglutinin inhibition titre ≥1:10) pH1N1‡ 39 (33%) 62 (49%) 0·013 H3N2 90 (76%) 70 (56%) 0·00070 B/Vic 25 (21%) 54 (43%) 0·00040 Haemagglutinin inhibition titre in children seropositive at baseline pH1N1‡ 160 (80–160) 226 (160–320) 0·00050 H3N2 160 (80–160) 160 (80–320) 0·16 B/Vic 160 (80–226·3) 226 (160–320) 0·015 Data are n (%), median (IQR), or mean (SD). pH1N1=pandemic H1N1.

·013 H3N2 90 (76%) 70 (56%) 0·00070 B/Vic 25 (21%) 54 (43%) 0·00040 Haemagglutinin inhibition titre in children seropositive at baseline pH1N1‡ 160 (80–160) 226 (160–320) 0·00050 H3N2 160 (80–160) 160 (80–320) 0·16 B/Vic 160 (80–226·3) 226 (160–320) 0·015 Data are n (%), median (IQR), or mean (SD). pH1N1=pandemic H1N1. * Malnutrition was categorised based on weight-for-height SD (Z score): none (>–1), mild (–2 to <–1), moderate (–3 to <–2). Children with severe malnutrition (weight-for-height SD <–3) were excluded. † Missing data for two children. ‡ pH1N1 virus used for serum haemagglutinin inhibition assays was changed for the cohort given 2017–18 LAIV to reflect the update from Cal09 to NY15.

* Malnutrition was categorised based on weight-for-height SD (Z score): none (>–1), mild (–2 to <–1), moderate (–3 to <–2). Children with severe malnutrition (weight-for-height SD <–3) were excluded. † Missing data for two children. ‡ pH1N1 virus used for serum haemagglutinin inhibition assays was changed for the cohort given 2017–18 LAIV to reflect the update from Cal09 to NY15. No influenza strains were detected from nasopharyngeal swabs taken immediately before vaccination in any children. After administration of the 2016–17 LAIV, pH1N1 Cal09 shedding was seen in significantly fewer children (16 of 118 [14%, 95% CI 8·0–21·1]) at day 2 compared with H3N2 (54 of 118 [46%, 36·6–55·2]; p<0·0001) and B/Vic (95 of 118 [81%, 72·2–87·2]; p<0·0001; figure 2). No pH1N1 Cal09 shedding was recorded at day 7 with the 2016–17 LAIV, with H3N2 shedding noted in 21 of 118 children (18%, 95% CI 11·4–25·9) and B/Vic shedding in 70 of 118 children (59%, 49·9–68·3). Administration of the 2017–18 LAIV resulted in significantly more children shedding pH1N1 NY15 at day 2 (80 of 126 [63%, 95% CI 54·4–71·9]; p<0·0001 compared with pH1N1 Cal09; figure 2A), with shedding of H3N2 seen in 82 of 126 children (65%, 95% CI 56·1–73·4) and shedding of B/Vic reported in 91 of 126 children (72%, 63·5–79·8). Shedding of pH1N1 NY15 was detected at day 7 with the 2017–18 LAIV (65 of 126 children [52%, 42·5–60·6]), with shedding of H3N2 recorded in 40 of 126 children (32%, 95% CI 23·7–40·6) and shedding of B/Vic noted in 60 of 126 children (48%, 38·7–56·7).Figure 2 Shedding of strains in the nasopharynx after vaccination

·8). Shedding of pH1N1 NY15 was detected at day 7 with the 2017–18 LAIV (65 of 126 children [52%, 42·5–60·6]), with shedding of H3N2 recorded in 40 of 126 children (32%, 95% CI 23·7–40·6) and shedding of B/Vic noted in 60 of 126 children (48%, 38·7–56·7).Figure 2 Shedding of strains in the nasopharynx after vaccination (A) Percentage of children shedding vaccine virus with 2016–17 LAIV formulation compared with the 2017–18 LAIV formulation, at day 2 and day 7. Error bars represent the upper 95% CI. (B) Viral load in the nasopharynx is indicated by ct values from RT-PCR. Red bars indicate median ct values. Lower ct values indicate higher viral loads. (C) Quantitative RT-PCR viral load in children from the 2018 cohort for each strain. Red bars indicate median values. p values are Bonferroni-adjusted for multiplicity within each group of analyses. LAIV=live attenuated influenza vaccine. pH1N1=pandemic H1N1. H3N2=A/17/Hong Kong/2014/8296. B/Vic=B/Texas/02/2013 (Victoria lineage). ct=cycle threshold. RT-PCR=reverse-transcriptase PCR. EID50eq=50% egg infectious dose equivalents.

alues. p values are Bonferroni-adjusted for multiplicity within each group of analyses. LAIV=live attenuated influenza vaccine. pH1N1=pandemic H1N1. H3N2=A/17/Hong Kong/2014/8296. B/Vic=B/Texas/02/2013 (Victoria lineage). ct=cycle threshold. RT-PCR=reverse-transcriptase PCR. EID50eq=50% egg infectious dose equivalents. Significantly higher pH1N1 nasopharyngeal viral loads (lower ct values) were also seen with the 2017–18 LAIV compared with the 2016–17 LAIV at day 2 (p=0·0026; figure 2B). Quantitative RT-PCR data showed that pH1N1 NY15 viral loads from the 2017–18 LAIV were significantly higher than H3N2 (p<0·0001) and B/Vic (p=0·0028) at day 2 (figure 2C). To investigate whether the improved replication of pH1N1 with the 2017–18 LAIV resulted in greater competition with H3N2 and B/Vic and, therefore, lower viral loads of these strains, viral loads were compared in co-shedders of each strain. No significant negative effect on H3N2 and B/Vic replication was reported, with a significant positive correlation between pH1N1 and H3N2 shedding noted at day 2 (low correlation, rs=0·40; p=0·0012) and day 7 (moderate correlation, rs=0·51; p=0·0032; appendix p 11).

al loads were compared in co-shedders of each strain. No significant negative effect on H3N2 and B/Vic replication was reported, with a significant positive correlation between pH1N1 and H3N2 shedding noted at day 2 (low correlation, rs=0·40; p=0·0012) and day 7 (moderate correlation, rs=0·51; p=0·0032; appendix p 11). To ascertain whether pre-existing adaptive immunity accounted for poor pH1N1 Cal09 shedding, univariable logistic regression was done to calculate the predicted probability of shedding at each baseline haemagglutinin inhibition titre (figure 3), adjusting for year in the H3N2 and B/Vic models. Although an inverse relation was evident for H3N2 (figure 3C) and B/Vic (figure 3D) between baseline haemagglutinin inhibition titre and shedding, this relation was not evident for Cal09 (figure 3A), for which low shedding was predicted even in seronegative children. By contrast, NY15 shedding was inversely related to the magnitude of the baseline haemagglutinin inhibition titre (figure 3B). Logistic regression also showed no associations between shedding and prevaccination T-cell responses or haemagglutinin-specific mucosal IgA responses for Cal09 (appendix p 12). Similarly, no association was seen between T-cell responses or IgA responses and shedding for H3N2 or B/Vic strains, after adjusting for baseline haemagglutinin inhibition titre (appendix p 12). Significantly lower nasopharyngeal viral loads were recorded at day 2 and day 7 for all three strains in baseline seropositive children compared with baseline seronegative children who received the 2017–18 LAIV formulation (figure 3E), further emphasising the importance of serum antibody in this process.Figure 3 Effect of baseline serum antibody on LAIV strain shedding in the nasopharynx and replicative ability of viruses in primary epithelial cell cultures

red with baseline seronegative children who received the 2017–18 LAIV formulation (figure 3E), further emphasising the importance of serum antibody in this process.Figure 3 Effect of baseline serum antibody on LAIV strain shedding in the nasopharynx and replicative ability of viruses in primary epithelial cell cultures (A–D) Predicted probability from logistic regression of vaccine strain shedding at day 2 after LAIV at a given baseline serum HAI titre to each matched strain. Dots show predicted proportions and shaded areas show 95% CIs. Data shown for Cal09 pH1N1 (A), NY15 pH1N1 (B), H3N2 (C), and B/Vic (D). Upper limit is based on maximum observed HAI titre in the dataset. When data from 2017 and 2018 were combined for H3N2 and B/Vic, results were adjusted for year (appendix p 12). (E) Nasopharyngeal viral load at day 2 and day 7 after 2017–18 LAIV, with participants stratified by baseline serostatus to vaccine haemagglutinin-matched and neuraminidase-matched influenza strains. Red bars indicate median values. (F and G) Replication of pH1N1 (F) or H3N2 and B/Vic (G) vaccine strains in primary nasal epithelium. Dots denote mean values and errors bars the SD. In (F), p<0·0001 comparing area under the curve. (H) Effect of pH on vaccine strain growth in vitro. Dots denote mean values and errors bars the SD. The y axis is a logarithmic scale. LAIV=live attenuated influenza vaccine. HAI=haemagglutinin inhibition. pH1N1=pandemic H1N1. Cal09=A/17/California/2009/38. NY15=A/17/New York/15/5364. H3N2=A/17/Hong Kong/2014/8296. B/Vic=B/Texas/02/2013 (Victoria lineage). EID50eq=50% egg infectious dose equivalents. PFU=plaque-forming units. p values for specific timepoints are *p=0·047, †p=0·0019, ‡p=0·029, §p=0·013, and ¶p<0·0001.

on. pH1N1=pandemic H1N1. Cal09=A/17/California/2009/38. NY15=A/17/New York/15/5364. H3N2=A/17/Hong Kong/2014/8296. B/Vic=B/Texas/02/2013 (Victoria lineage). EID50eq=50% egg infectious dose equivalents. PFU=plaque-forming units. p values for specific timepoints are *p=0·047, †p=0·0019, ‡p=0·029, §p=0·013, and ¶p<0·0001. In the seronegative population, shedding of pH1N1 Cal09 at day 2 was reported in ten of 79 children (13%, 95% CI 7·0–21·8) whereas shedding of pH1N1 NY15 was noted in 58 of 64 children (91%, 81·0–95·6; p<0·0001). Shedding of H3N2 with the 2016–17 LAIV was seen in 21 of 28 seronegative children (75·0%, 95% CI 56·6–87·3) and with the 2017–18 LAIV, shedding of H3N2 was recorded in 46 of 56 seronegative children (82%, 70·2–90·0; p=0·63). In the same seronegative population, shedding of B/Vic with the 2016–17 LAIV was detected in 78 of 93 children (84%, 95% CI 75·1–90·0) and, with the 2017–18 LAIV, shedding of B/Vic was recorded in 57 of 72 children (79%, 68·4–86·9; p=0·57). Comparisons of ct values between 2016–17 and 2017–18 LAIV strains in seronegative children showed a lower ct value (higher viral load) at day 2 with pH1N1 NY15 compared with pH1N1 Cal09 (p<0·0001; appendix p 13).

) and, with the 2017–18 LAIV, shedding of B/Vic was recorded in 57 of 72 children (79%, 68·4–86·9; p=0·57). Comparisons of ct values between 2016–17 and 2017–18 LAIV strains in seronegative children showed a lower ct value (higher viral load) at day 2 with pH1N1 NY15 compared with pH1N1 Cal09 (p<0·0001; appendix p 13). Monovalent vaccine strain replication was tested in primary human nasal epithelial cells cultured at an air–liquid interface to see whether in-vitro kinetics (in the absence of adaptive immune responses) reflected Cal09 and NY15 pH1N1 shedding in children. NY15 replication was greater than Cal09 replication (figure 3F), whereas H3N2 and B/Vic growth was equivalent (figure 3G). Since stability in acidic environments in the upper respiratory tract could be important for replicative ability, Cal09 and NY15 were quantified after exposure to varying pH levels. Greater stability of NY15 was seen in acidic environments compared with Cal09 (figure 3H).

N2 and B/Vic growth was equivalent (figure 3G). Since stability in acidic environments in the upper respiratory tract could be important for replicative ability, Cal09 and NY15 were quantified after exposure to varying pH levels. Greater stability of NY15 was seen in acidic environments compared with Cal09 (figure 3H). The 2016–17 LAIV resulted in significantly fewer children seroconverting to pH1N1 Cal09 (six of 118 [5%, 95% CI 1·9–10·7]) compared with H3N2 (26 of 118 [22%, 14·9–30·6]; p=0·00030) and B/Vic (40 of 118 [34%, 25·4–43·2]; p<0·0001). A significant increase was recorded in pH1N1 NY15 seroconversion with the 2017–18 LAIV compared with Cal09 (24 of 126 [19%, 95% CI 13·2–26·8]; p=0·011; figure 4A), with no difference in H3N2 (35 of 126 [28%, 20·7–36·2]; p=1·00) or B/Vic (30 of 126 [24%, 17·2–32·0]; p=0·11). The improved seroconversion to pH1N1 with NY15 compared with Cal09 was especially evident in seronegative children (24 of 64 [38%, 95% CI 26·7–49·8] vs six of 79 [8%, 2·8–15·8]; p<0·0001; figure 4A), with a significant difference in geometric mean fold change in haemagglutinin inhibition for pH1N1 NY15 in 2017–18 compared with pH1N1 Cal09 in 2016–17 (p<0·0001; figure 4B).Figure 4 Immunogenicity to pH1N1 with the 2016–17 and 2017–18 LAIV formulations

% CI 26·7–49·8] vs six of 79 [8%, 2·8–15·8]; p<0·0001; figure 4A), with a significant difference in geometric mean fold change in haemagglutinin inhibition for pH1N1 NY15 in 2017–18 compared with pH1N1 Cal09 in 2016–17 (p<0·0001; figure 4B).Figure 4 Immunogenicity to pH1N1 with the 2016–17 and 2017–18 LAIV formulations p values are Bonferroni-adjusted for multiplicity within each group of analyses. (A) Percentage of children seroconverting to each LAIV strain, comparing 2016–17 and 2017–18 formulations. Error bars represent the upper 95% CI. (B) Geometric mean fold change in serum haemagglutinin inhibition titre from baseline to day 21, comparing children seronegative at baseline given 2016–17 and 2017–18 LAIVs. Dotted line depicts a fold change of four. y axis is a logarithmic scale. (C) Influenza-specific CD4+ T-cell responses to vaccine strain-matched pH1 haemagglutinin (Cal09 in 2016–17 or NY15 in 2017–18), H3 haemagglutinin, influenza A matrix and nucleoprotein (both matched to LAIV backbone) peptide pools, comparing 2016–17 and 2017–18 LAIVs. Error bars represent the upper 95% CI. (D) Percentage of children with a twofold rise in influenza-specific CD4+ T-cell responses at day 21 after 2016–17 and 2017–18 LAIVs. y axis is a logarithmic scale. (E) Percentage of influenza-specific mucosal IgA responders given the 2016–17 and 2017–18 LAIVs. Error bars represent the upper 95% CI. pH1N1=pandemic H1N1. LAIV=live attenuated influenza vaccine. Cal09=A/17/California/2009/38. NY15=A/17/New York/15/5364. H3N2=A/17/Hong Kong/2014/8296. B/Vic=B/Texas/02/2013 (Victoria lineage). IFNγ=interferon γ.

-specific mucosal IgA responders given the 2016–17 and 2017–18 LAIVs. Error bars represent the upper 95% CI. pH1N1=pandemic H1N1. LAIV=live attenuated influenza vaccine. Cal09=A/17/California/2009/38. NY15=A/17/New York/15/5364. H3N2=A/17/Hong Kong/2014/8296. B/Vic=B/Texas/02/2013 (Victoria lineage). IFNγ=interferon γ. Influenza-specific CD4+ IFNγ-positive, CD4+ IL2-positive, and CD8+ IFNγ-positive T-cell responses were detected at baseline and after vaccination. Although the magnitude of CD8+ responses was generally higher, LAIV-induced responses were predominantly CD4+ (figures 4C and 4D; appendix pp 14, 15). The 2016–17 LAIV did not induce significant pH1 haemagglutinin-specific CD4+ IFNγ-positive or CD4+ IL2-positive responses, whereas H3 haemagglutinin-positive and influenza A matrix and nucleoprotein-specific responses were significantly increased from baseline (figure 4D; appendix p 14). By contrast, the 2017–18 LAIV induced significant pH1 haemagglutinin-specific CD4+ T-cells at day 21. Accordingly, a twofold or greater rise in pH1 haemagglutinin-specific CD4+ T-cell responses was noted in more children given the 2017–18 LAIV than in those given the 2016–17 LAIV (50 of 109 [46%, 95% CI 36·3–55·7] vs 29 of 111 [26%, 18·2–35·3] for CD4+ IFNγ-positive responses; and 57 of 109 [52%, 42·5–61·9] vs 23 of 111 [20·7%, 13·6–29·5] for CD4+ IL2-positive responses; figure 4C). A twofold or greater rise in CD4+ IFNγ-positive and/or CD4+ IL2-positive responses was recorded in 45 of 111 children (41%, 95% CI 31·3–50·3) given the 2016–17 LAIV and in 73 of 111 children (66%, 60·0–75·6) given the 2017–18 LAIV. B/Vic haemagglutinin-specific and influenza B matrix and nucleoprotein-specific CD4+ responses were also induced (appendix p 15). No significant change in the proportion of monofunctional or dual-functional CD4+ T-cell responses was seen after vaccination (appendix p 15). Influenza-specific mucosal IgA responses to pH1N1 did not differ between the 2016–17 LAIV (16 of 117 children [14%, 95% CI 8·0–21·3]) and the 2017–18 LAIV (24 of 121 children [20%, 13·1–28·1]; p=0·27; figure 4E).

tion of monofunctional or dual-functional CD4+ T-cell responses was seen after vaccination (appendix p 15). Influenza-specific mucosal IgA responses to pH1N1 did not differ between the 2016–17 LAIV (16 of 117 children [14%, 95% CI 8·0–21·3]) and the 2017–18 LAIV (24 of 121 children [20%, 13·1–28·1]; p=0·27; figure 4E). The effect of shedding on immunogenicity was investigated using H3N2 data (the largest sample of participants immunised with the same antigen and with T-cell data available). Seroconversion and T-cell responses were highest in children with shedding at both days 2 and 7 (appendix pp 16, 17). Multivariable logistic regression showed a significant effect of this prolonged shedding on the odds of seroconversion (odds ratio [OR] 12·69, 95% CI 4·1–43·6; p<0·0001) and CD4+ T-cell responses (7·83, 2·99–23·5; p<0·0001; appendix pp 16, 17). No such relation was seen with IgA responses (appendix pp 16–18). The odds of seroconversion were also reduced by higher baseline haemagglutinin inhibition titre (OR 0·11, 95% CI 0·04–0·27; p<0·0001) and increased by induction of an H3 haemagglutinin-specific CD4+ IL2-positive response (2·42, 1·05–5·62; p=0·037). Similar findings were seen in B/Vic and NY15 pH1N1 datasets, albeit with smaller sample sizes (appendix pp 18–20).

d by higher baseline haemagglutinin inhibition titre (OR 0·11, 95% CI 0·04–0·27; p<0·0001) and increased by induction of an H3 haemagglutinin-specific CD4+ IL2-positive response (2·42, 1·05–5·62; p=0·037). Similar findings were seen in B/Vic and NY15 pH1N1 datasets, albeit with smaller sample sizes (appendix pp 18–20). Discussion The findings of our study showed limited shedding, in vitro Cal09 replication, and low immunogenicity after administration of the 2016–17 LAIV in Gambian children, providing an explanation for the scant efficacy of this vaccine that was reported in a randomised controlled trial from neighbouring Senegal.4 After the switch to NY15, a significant increase in replication was seen, along with improved serum humoral and cellular immunogenicity. No competitive inhibitory effect of enhanced pH1N1 replication was recorded with H3N2 or B/Vic replication or immunogenicity. Our data also showed that shedding for a longer duration is important for immunogenicity and that viral replicative fitness should be considered alongside antigenicity when selecting vaccine strains. Our findings represent the first reported LAIV immunogenicity data from African children and make a case for further studies of LAIV efficacy in Africa. They are also of relevance to the use of LAIV in other settings.

viral replicative fitness should be considered alongside antigenicity when selecting vaccine strains. Our findings represent the first reported LAIV immunogenicity data from African children and make a case for further studies of LAIV efficacy in Africa. They are also of relevance to the use of LAIV in other settings. In a study of Ann Arbor-backbone LAIV, improved shedding and haemagglutinin inhibition seroconversion was reported with an updated A/Slovenia/2015 pH1N1 strain.17 Parallel findings in two distinct cohorts of children—using two different LAIVs—provide strong support for Cal09 replicative fitness being culpable for the suboptimum pH1N1 LAIV effectiveness seen in recent years.2 Our finding that limited Cal09 shedding is unlikely to be attributable to pre-existing immunity further supports this result and argues against the notion that reduced LAIV effectiveness in the USA might have been due to repeated vaccination in previous years.18

uboptimum pH1N1 LAIV effectiveness seen in recent years.2 Our finding that limited Cal09 shedding is unlikely to be attributable to pre-existing immunity further supports this result and argues against the notion that reduced LAIV effectiveness in the USA might have been due to repeated vaccination in previous years.18 In an earlier study using the Ann Arbor-backbone LAIV, pre-pandemic seasonal H1N1 shedding was found to be higher than for H3N2 or influenza B.19 Why pH1N1 Cal09 replication is impaired is uncertain. Haemagglutinin or neuraminidase residues must be the reason because the remaining six viral gene segments in LAIV are consistent between Cal09, NY15, and H3N2 strains. Differences in Cal09 haemagglutinin thermostability, sialic acid receptor binding, or pH sensitivity are potential explanations for the lower replication noted.3 These properties are important for replication in the human upper respiratory tract. In particular, the pH of the upper respiratory tract in children might be lower than that of adults20 and have a deleterious effect on replication of viruses with labile haemagglutinin. The pH1N1 virus first crossed into human beings in 2009 and has subsequently circulated as a human seasonal virus. During this time, changes in haemagglutinin stability and receptor binding properties might have adapted the virus to replicate better in the human upper respiratory tract.21 Thus, the more recent haemagglutinin from A/Michigan/45/2015-like viruses of 2015 could have conferred enhanced shedding to LAIV pH1N1 components.

uring this time, changes in haemagglutinin stability and receptor binding properties might have adapted the virus to replicate better in the human upper respiratory tract.21 Thus, the more recent haemagglutinin from A/Michigan/45/2015-like viruses of 2015 could have conferred enhanced shedding to LAIV pH1N1 components. We were able to mirror our findings in a primary human nasal epithelial cell model. These cells have a mildy acidic apical surface environment akin to that of the human upper respiratory tract. This strategy could be a practical method for assessing vaccine virus replication before strain choice. Because cell lines traditionally used to culture influenza viruses (eg, MDCK) might not truly reflect replication in the upper respiratory tract, these subtleties were previously underappreciated.22 Ultimately, a greater understanding of the viral genetic determinants of LAIV replicative fitness will be needed to select the best vaccine formulations.

used to culture influenza viruses (eg, MDCK) might not truly reflect replication in the upper respiratory tract, these subtleties were previously underappreciated.22 Ultimately, a greater understanding of the viral genetic determinants of LAIV replicative fitness will be needed to select the best vaccine formulations. Our study also emphasises the multifaceted nature of LAIV-induced immunity. Although seroresponse (the traditional correlate of protection after inactivated influenza vaccine) is modest, LAIV also induces mucosal IgA and T-cell responses. In our cohort, T-cell responses were elicited in a larger proportion of children than were mucosal or serum antibodies, showing the importance of assessing cellular immunity in LAIV studies. Using the 2017–18 LAIV formulation, a CD4+ IFNγ-positive or CD4+ IL2-positive T-cell response was seen in 55–68% of children to the influenza antigens tested, with approximately 80% of children showing a response to haemagglutinin or matrix and nucleoprotein (appendix p 18). LAIV provides protection in the absence of humoral immunity23 and T-cell-mediated immunity is thought to have an important role.24

esponse was seen in 55–68% of children to the influenza antigens tested, with approximately 80% of children showing a response to haemagglutinin or matrix and nucleoprotein (appendix p 18). LAIV provides protection in the absence of humoral immunity23 and T-cell-mediated immunity is thought to have an important role.24 Unlike serum antibody and T-cell responses, we did not see an increase in mucosal IgA responses with NY15. This finding is in keeping with results reported after one dose of the updated Ann Arbor-backbone LAIV,17 although a better response was seen after two doses. In a recent immunogenicity study of Nasovac-S in Bangladesh,25 unlike serum antibody, nasal pH1N1-specific IgA was induced despite scant Cal09 shedding. Furthermore, by contrast to seroconversion and T-cell responses, we noted no association between shedding and IgA responses. Taken together, these data suggest the mechanisms and requirements for serum antibody and mucosal IgA induction by LAIV could be distinct.

1-specific IgA was induced despite scant Cal09 shedding. Furthermore, by contrast to seroconversion and T-cell responses, we noted no association between shedding and IgA responses. Taken together, these data suggest the mechanisms and requirements for serum antibody and mucosal IgA induction by LAIV could be distinct. Our study has several limitations. Although the association between shedding and immunogenicity was a predefined exploratory objective in the larger randomised controlled trial our study was a part of, comparison of formulations containing Cal09 and NY15 was a post-hoc analysis made possible only because of the WHO-recommended update to pH1N1 in the 2017–18 formulation. Because we show an improvement in several shedding and immunogenicity endpoints with NY15, we are confident that our main conclusions are justified. Nevertheless, since our sample size was based on endpoints not reported here, the negative findings reported in some subanalyses should be interpreted with caution. Also, participants were vaccinated with one LAIV dose, in keeping with the prequalification license from WHO and the randomised controlled trials in Senegal4 and Bangladesh.25 Our findings, therefore, might not be generalisable to children in high-income countries who receive booster doses and yearly influenza vaccination. We were also unable to confirm viral shedding with an LAIV-specific RT-PCR in all participants because of lower sensitivity compared with the haemagglutinin-specific RT-PCR, which is important to fully exclude interference from wild-type circulating strains. However, by doing the study outside of the peak influenza season,7 undertaking clinical review at enrolment, and doing baseline RT-PCR screening for influenza virus, it is unlikely that our results were affected by wild-type influenza infections. Our shedding data at day 2 are also similar to those reported from Senegal using Nasovac-S (Cal09 19%, H3N2 48%, and influenza B 66%).4 Because we measured shedding with RT-PCR and not culture, we are unable to confirm to what degree shedding reflected viable viruses. Finally, an important unanswered question from our study is whether NY15 and related pH1N1 strains will result in improved LAIV effectiveness. Data from the 2017–18 UK season estimates the vaccine effectiveness (Ann Arbor-backbone LAIV) to be 90·3% against pH1N1 in children aged 2–17 years.26 However, owing to low-level circulation of pH1N1, the precision around this estimate was low (95% CI 16·4–98·9).

H1N1 strains will result in improved LAIV effectiveness. Data from the 2017–18 UK season estimates the vaccine effectiveness (Ann Arbor-backbone LAIV) to be 90·3% against pH1N1 in children aged 2–17 years.26 However, owing to low-level circulation of pH1N1, the precision around this estimate was low (95% CI 16·4–98·9). Our findings suggest improved effectiveness can indeed be expected with the updated LAIV and, if so, would support wider use of LAIV in the prevention of influenza. Supplementary Material Supplementary appendix

H1N1 strains will result in improved LAIV effectiveness. Data from the 2017–18 UK season estimates the vaccine effectiveness (Ann Arbor-backbone LAIV) to be 90·3% against pH1N1 in children aged 2–17 years.26 However, owing to low-level circulation of pH1N1, the precision around this estimate was low (95% CI 16·4–98·9). Our findings suggest improved effectiveness can indeed be expected with the updated LAIV and, if so, would support wider use of LAIV in the prevention of influenza. Supplementary Material Supplementary appendix Acknowledgments We thank the study participants and their parents who took part in the study; the dedicated team of field and nursing staff led by Janko Camara and Sulayman Bah; Isatou Ndow for clinical trial organisation; the research support and clinical trials support offices at the Medical Research Council (MRC) Unit The Gambia at London School of Hygiene & Tropical Medicine (LSHTM); the Serum Institute of India Pvt for donating the vaccines used in this study; Aminata Ngatou Vilane and Sheikh Jarju for establishing the reverse transcriptase-PCR assays; and Yanchun Peng for help with overlapping peptide pools for influenza T-cell assays. This study was funded by a Wellcome Trust Intermediate Clinical Fellowship award (to TIdS; 110058/Z/15/Z). AS is funded by a Wellcome Trust Clinical Research Training Fellowship (WT105736MA). BK is funded by the UK MRC (grants MR/K007602/1 and MC_UP_A900/1122). Research at the MRC Unit The Gambia at LSHTM is jointly funded by the UK MRC and the UK Department for International Development (DFID) under the MRC/DFID Concordat agreement and is also part of the European & Developing Countries Clinical Trials Partnership 2 programme supported by the EU. TD is supported by the UK MRC and the Chinese Academy of Medical Sciences Innovation Fund for Medical Sciences (grant 2018-I2M-2-002). TIdS, JST, and KH are members of the Human Infection Challenge Network for Vaccine Development, which is funded by the Global Challenge Research Fund Networks in Vaccines Research and Development, which was co-funded by the UK MRC and the Biotechnology and Biological Sciences Research Council. JST, WB, and AS were additionally supported by the National Institute for Health Research Imperial College London Biomedical Research Centre.

he Global Challenge Research Fund Networks in Vaccines Research and Development, which was co-funded by the UK MRC and the Biotechnology and Biological Sciences Research Council. JST, WB, and AS were additionally supported by the National Institute for Health Research Imperial College London Biomedical Research Centre. Contributors TIdS, EC, BK, EPA, and DJ contributed to the clinical study design. TIdS, BBL, YJJ, HJS, KH, JST, AM, TD, AS, and WB contributed to design of laboratory experiments. BBL, EPA, AS, WB, and TIdS contributed to the literature search. BBL, EPA, YJJ, HJS, SD, ES, KH, and AS contributed to data collection. BBL, TIdS, EPA, YJJ, NIM, DJ, KH, and AS contributed to data analysis. TIdS, BBL, EPA, AS, WB, EC, BK, TD, KH, AM, and JST contributed to data interpretation. BBL, TIdS, BK, EC, AS, and WB wrote the report. All authors reviewed the final report. Declaration of interests WB reports personal fees from AstraZeneca for contributing to a virtual advisory board in 2018, outside of the submitted work. BBL, YJJ, EPA, AS, HJS, SD, ES, NIM, DJ, KH, JST, AM, EC, TD, BK, and TIdS declare no competing interests.

Pulmonary arterial hypertension is currently an incurable disease characterised by increased pulmonary vascular resistance and eventual death due to right ventricular failure. Hereditary pulmonary arterial hypertension has been generally thought to represent a small subset of all pulmonary arterial hypertension cases. The most common mutation identified is loss of function in BMPR2, which encodes a cell surface receptor belonging to the transforming growth factor-β (TGFβ) superfamily.1,2 Understanding of the pathogenetic role of BMPR2 haploinsufficiency has uncovered new potential therapeutic targets in pulmonary arterial hypertension, with multiple clinical trials focusing on the TGFβ signalling pathway. Despite the documented contribution of common genetic variants as determinants of complex and multifactorial diseases, previous genome-wide association studies (GWAS) in patients with pulmonary arterial hypertension disease have been scarce. In The Lancet Respiratory Medicine, Christopher Rhodes and colleagues3 report their findings from the largest genetic analysis to date in patients of European ancestry with pulmonary arterial hypertension. By use of two independent GWAS platforms of whole-genome sequencing and genotyping array (discovery phase), and meta-analysis (validation phase) with multiple internationally collaborative cohorts, the study identified two novel loci associated with risk for the development of pulmonary arterial hypertension: an enhancer region in SOX17, and a locus within HLA-DPA1 and HLA-DPB1. The alleles associated with disease were common, with a relatively modest effect size: for example, 59% of patients with pulmonary arterial hypertension compared with 46% of controls were homozygous for the risk allele at both SOX17 SNPs. The results of the discovery and validation phases converged to substantially decrease the possibility of false-positive results.

with a relatively modest effect size: for example, 59% of patients with pulmonary arterial hypertension compared with 46% of controls were homozygous for the risk allele at both SOX17 SNPs. The results of the discovery and validation phases converged to substantially decrease the possibility of false-positive results. The authors reported two independent signals in the transcription factor SOX17 locus significantly associated with pulmonary arterial hypertension (rs10103692, odds ratio 1.80 [95% CI 1.55–2.08], p=5.13 × 10−15, and rs13266183, 1.36 [1.25–1.48], p 1.69 × 10−12). These signals identified enhancer regions leading to modified expression of SOX17, which correlated with the diagnosis of pulmonary arterial hypertension. The functional impact of the novel SOX17 locus on pulmonary arterial hypertension susceptibility was confirmed with Hi-C to determine DNA folding patterns and CRISPR-mediated inhibition in human pulmonary artery endothelial cells. This finding corroborates a previous report describing a rare genetic variant in SOX17 associated with heritable pulmonary arterial hypertension.4 The present study posits that common variation in SOX17 expression is a determinant of pulmonary arterial hypertension and is present more often in patients with pulmonary arterial hypertension than are other rare genetic variants. Conditional deletions of SOX17 in endothelial cells cause abnormal pulmonary vascular morphogenesis5 potentially mediated by Notch pathway signalling to restrict angiogenesis.6 Inactivation of SOX17 in mouse embryos leads to lack of arterial differentiation and vascular remodelling,7 and SOX17 has a major role in the development of haemogenic endothelial cells (the precursor of haemopoietic stem cells).8

orphogenesis5 potentially mediated by Notch pathway signalling to restrict angiogenesis.6 Inactivation of SOX17 in mouse embryos leads to lack of arterial differentiation and vascular remodelling,7 and SOX17 has a major role in the development of haemogenic endothelial cells (the precursor of haemopoietic stem cells).8 The other key genes identified by this study were HLA-DPA1 and HLA-DPB1 (rs2856830, 1.56 [1.42–1.71], p=7.65 × 10−20), which encode the MHC class II DP α and β chains. The specific variant discovered was located in HLA-DPB1 and, surprisingly, correlated with two seemingly opposite outcomes: increased risk for the development of pulmonary arterial hypertension, but improved survival once having developed pulmonary arterial hypertension. The specific alleles associated with the variants were HLA-DPB1*02:01/02:02/16:01, all of which contain a glutamic acid substituted for a lysine residue at position 69. Of note, variants with the same glutamic acid substitution at position 69 have also been described as increasing the risk for developing chronic beryllium disease (a variant of sarcoidosis) among individuals exposed to beryllium.9 In studies using crystallography, the negative charge of the glutamine side chain was found to stabilise the Be2+ ion in the MHC II groove, facilitating a neoantigen that triggers a granulomatous lung disease.10 The finding of the same variants in these patients raises the possibility of an antigenic trigger contributing to the development of pulmonary arterial hypertension. This antigen-triggered disease might have a more benign disease course or improved clinical response to the current clinical armamentarium compared with other forms of pulmonary arterial hypertension, resulting in the improved survival observed.

rigger contributing to the development of pulmonary arterial hypertension. This antigen-triggered disease might have a more benign disease course or improved clinical response to the current clinical armamentarium compared with other forms of pulmonary arterial hypertension, resulting in the improved survival observed. On the basis of the current findings, the next steps include first reproducing and validating these findings with additional and more diverse samples. It will then be important to associate the risk alleles of SOX17 and HLA-DPB1 with clinical characteristics and relevant pathologies in pulmonary arterial hypertension, including inflammatory cytokines, incidence of vasodilator responsiveness, haemodynamics, and right ventricle performance, which will help to contextualise the relevance of the GWAS data to the clinical setting. It remains unanswered whether the significance of these markers generalises to ethnicities, whether these markers drive similar clinical presentations and outcomes in both sexes, whether they can serve as biomarkers to determine predisposition towards and stratification of pulmonary arterial hypertension subtypes, or whether they interact with other genes in a complex disease such as pulmonary arterial hypertension. Transgenic modification of experimental animals with the same mutations will further clarify whether and how these modifications induce disease and provide insights into how they can be pharmacologically targeted. Overall, this study provides hope that the identification of genetic modifiers in pulmonary arterial hypertension will allow more accurate classification of pulmonary arterial hypertension subtypes and more tailored and effective treatments for patients with pulmonary arterial hypertension than are currently available.

Overall, this study provides hope that the identification of genetic modifiers in pulmonary arterial hypertension will allow more accurate classification of pulmonary arterial hypertension subtypes and more tailored and effective treatments for patients with pulmonary arterial hypertension than are currently available. We declare no competing interests. Salary support for the authors was provided by US National Institutes of Health grants T32HL007085 (SG and MHL), P01HL014985 (BBG), R03HL133306 (BBG), and R01HL135872 (BBG); and American Heart Association grant 17POST33670045 (RK).

GBD 2015 Chronic Respiratory Disease Collaborators. Global, regional, and national deaths, prevalence, disability-adjusted life years, and years lived with disability for chronic obstructive pulmonary disease and asthma, 1990–2015: a systematic analysis for the Global Burden of Disease Study 2015. Lancet Respir Med 2017; 5: 691-706—In this Article, Tsegaye T Gebrehiwot and Alan D Lopez should have been listed as authors. The affiliation details for Josep M Antó have been updated. The sixth sentence of the Discussion should read “Age-standardised DALY rates from COPD and asthma declined significantly by 43·7% (39·8–47·0) for COPD and by 42·8% (29·5–52·0) for asthma between 1990 and 2015”. These corrections have been made to the online version as of Sept 14, 2017.

Introduction Identification of people at high risk of developing tuberculosis enables the delivery of preventive treatment for a disease that accounts for more deaths than any other infectious disease worldwide, with an estimated 10 million incident cases and 1·6 million deaths in 2017.1 This approach represents a fundamental component of the WHO End TB strategy, aiming for a 95% reduction in tuberculosis mortality and 90% reduction in tuberculosis incidence by 2035.2 However, these efforts are undermined by the poor positive predictive value of available prognostic tests for development of tuberculosis, which focus on the identification of a T-cell-mediated response to mycobacterial antigen stimulation, as a surrogate for latent tuberculosis infection.3, 4 These tests include the tuberculin skin test and interferon-γ release assays (IGRAs), which have positive predictive values of 1–6% for incident tuberculosis over a 2-year period.4, 5, 6, 7 The poor predictive value of available diagnostics precludes precise delivery of preventive therapy, thus increasing costs and potential adverse effects, attenuating the effectiveness of prevention programmes, and reducing roll-out of preventive treatment in limited-resource settings, where most tuberculosis cases occur. Research in context Evidence before this study

Introduction Identification of people at high risk of developing tuberculosis enables the delivery of preventive treatment for a disease that accounts for more deaths than any other infectious disease worldwide, with an estimated 10 million incident cases and 1·6 million deaths in 2017.1 This approach represents a fundamental component of the WHO End TB strategy, aiming for a 95% reduction in tuberculosis mortality and 90% reduction in tuberculosis incidence by 2035.2 However, these efforts are undermined by the poor positive predictive value of available prognostic tests for development of tuberculosis, which focus on the identification of a T-cell-mediated response to mycobacterial antigen stimulation, as a surrogate for latent tuberculosis infection.3, 4 These tests include the tuberculin skin test and interferon-γ release assays (IGRAs), which have positive predictive values of 1–6% for incident tuberculosis over a 2-year period.4, 5, 6, 7 The poor predictive value of available diagnostics precludes precise delivery of preventive therapy, thus increasing costs and potential adverse effects, attenuating the effectiveness of prevention programmes, and reducing roll-out of preventive treatment in limited-resource settings, where most tuberculosis cases occur. Research in context Evidence before this study We did a systematic review using comprehensive terms for “tuberculosis”, “transcriptome”, “signature” and “blood”, without language or date restrictions, on April 15, 2019. Multiple studies have identified perturbation in the transcriptome that predates clinical diagnosis of tuberculosis and have discovered and assessed performance of one or more signatures for diagnosis of incipient tuberculosis within individual datasets. A head-to-head evaluation of candidate signatures was done, but omitted key signatures, and compared diagnostic accuracy for incipient tuberculosis in only a single dataset over a 0–6-month period. No previous studies have directly compared the diagnostic accuracy of all candidate signatures in a patient-level pooled dataset. It was therefore unknown which signature performs best for diagnosis of incipient tuberculosis, or whether any meets WHO target product profile benchmarks (aiming for sensitivity ≥75% and specificity ≥75% over 2 years).

ctly compared the diagnostic accuracy of all candidate signatures in a patient-level pooled dataset. It was therefore unknown which signature performs best for diagnosis of incipient tuberculosis, or whether any meets WHO target product profile benchmarks (aiming for sensitivity ≥75% and specificity ≥75% over 2 years). Added value of this study To our knowledge, we did the largest direct comparison to date of the performance of whole blood transcriptional signatures for diagnosis of incipient tuberculosis. We tested 17 candidate mRNA signatures, identified through a comprehensive systematic review, in a pooled dataset of 1126 RNA sequencing samples from four countries. We show that a single transcript (BATF2) and seven other multi-transcript signatures, regulated by interferon signalling, perform with equivalent diagnostic accuracy for incipient tuberculosis. The accuracy of all eight signatures declined markedly with increasing intervals to disease. No signature met the minimum WHO target product profile parameters for incipient tuberculosis biomarkers over a 2-year period. In contrast, the eight best performing signatures met or approximated the minimum target product profile parameters over a 0–3-month period. Using a threshold derived from two SDs above the mean of uninfected controls to prioritise specificity, they achieved sensitivities of 47·1–81·0% and specificities of more than 90%, leading to positive-predictive values of 11·2–14·4% and negative-predictive values of more than 98·9%, when assuming 2% pre-test probability. Implications of all the available evidence

To our knowledge, we did the largest direct comparison to date of the performance of whole blood transcriptional signatures for diagnosis of incipient tuberculosis. We tested 17 candidate mRNA signatures, identified through a comprehensive systematic review, in a pooled dataset of 1126 RNA sequencing samples from four countries. We show that a single transcript (BATF2) and seven other multi-transcript signatures, regulated by interferon signalling, perform with equivalent diagnostic accuracy for incipient tuberculosis. The accuracy of all eight signatures declined markedly with increasing intervals to disease. No signature met the minimum WHO target product profile parameters for incipient tuberculosis biomarkers over a 2-year period. In contrast, the eight best performing signatures met or approximated the minimum target product profile parameters over a 0–3-month period. Using a threshold derived from two SDs above the mean of uninfected controls to prioritise specificity, they achieved sensitivities of 47·1–81·0% and specificities of more than 90%, leading to positive-predictive values of 11·2–14·4% and negative-predictive values of more than 98·9%, when assuming 2% pre-test probability. Implications of all the available evidence Multiple transcriptional signatures perform with equivalent diagnostic accuracy for incipient tuberculosis. These biomarkers reflect short-term risk of tuberculosis and only exceed WHO benchmarks if applied to 3–6-month intervals. A screening strategy that incorporates serial testing on a 3–6-monthly basis among carefully selected target groups, such as recent case contacts, might be required for optimal implementation of these biomarkers.

s reflect short-term risk of tuberculosis and only exceed WHO benchmarks if applied to 3–6-month intervals. A screening strategy that incorporates serial testing on a 3–6-monthly basis among carefully selected target groups, such as recent case contacts, might be required for optimal implementation of these biomarkers. Increasing recognition of the continuum of tuberculosis infection and disease has led to renewed interest in the incipient phase of tuberculosis.8, 9, 10 Incipient tuberculosis is defined by WHO as the prolonged asymptomatic phase of early disease before clinical presentation as active disease, during which pathology evolves.11 This definition encompasses the incipient and subclinical phases described elsewhere.12 Tests that identify the incipient phase, between latent infection and active disease, might lead to improved positive predictive value for incident tuberculosis, while still offering an opportunity to prevent tuberculosis-related morbidity and mortality and reduce onward transmission.12 The need for better predictive biomarkers for incident tuberculosis has led to WHO producing a target product profile for incipient tuberculosis diagnostics, stipulating minimum sensitivity and specificity of 75% and optimal sensitivity and specificity of 90% over a 2-year period.11 These minimum criteria are based on achieving a positive predictive value of 5·8%, when assuming 2% pre-test probability, to improve on the predictive ability of existing tests.11

s diagnostics, stipulating minimum sensitivity and specificity of 75% and optimal sensitivity and specificity of 90% over a 2-year period.11 These minimum criteria are based on achieving a positive predictive value of 5·8%, when assuming 2% pre-test probability, to improve on the predictive ability of existing tests.11 Multiple studies have shown changes in the host transcriptome in association with active tuberculosis, when compared with healthy controls or individuals with latent tuberculosis infection or other diseases.13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23 Signatures have become more concise since the initial discovery of a 393-gene signature of active tuberculosis,13 making their translation to near-patient diagnostic tests more achievable. Perturbation in the transcriptome has been found to predate the diagnosis of tuberculosis,17, 24, 25, 26 suggesting that transcriptional signatures might offer an opportunity to diagnose incipient tuberculosis and potentially fulfil the WHO target product profile. However, independent validation of each signature is still limited to a small number of datasets. Which of the multiple candidate transcriptional signatures performs best for the identification of incipient tuberculosis or whether any signatures meet the WHO diagnostic accuracy benchmarks remains unclear. To address these knowledge gaps, we aimed to critically assess the potential value of whole blood transcriptional signatures as biomarkers for incipient tuberculosis in practice.

Multiple studies have shown changes in the host transcriptome in association with active tuberculosis, when compared with healthy controls or individuals with latent tuberculosis infection or other diseases.13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23 Signatures have become more concise since the initial discovery of a 393-gene signature of active tuberculosis,13 making their translation to near-patient diagnostic tests more achievable. Perturbation in the transcriptome has been found to predate the diagnosis of tuberculosis,17, 24, 25, 26 suggesting that transcriptional signatures might offer an opportunity to diagnose incipient tuberculosis and potentially fulfil the WHO target product profile. However, independent validation of each signature is still limited to a small number of datasets. Which of the multiple candidate transcriptional signatures performs best for the identification of incipient tuberculosis or whether any signatures meet the WHO diagnostic accuracy benchmarks remains unclear. To address these knowledge gaps, we aimed to critically assess the potential value of whole blood transcriptional signatures as biomarkers for incipient tuberculosis in practice. Methods Search strategy and selection criteria We hypothesised that any biomarker that distinguishes incipient or active tuberculosis from healthy people might detect incipient disease. We therefore did a systematic review and individual participant data meta-analysis, in accordance with Preferred Reporting Items for a Systematic Review and Meta-analysis of Individual Participant Data standards,27 to identify candidate concise whole blood transcriptional signatures for incipient or active tuberculosis and test their diagnostic accuracy for incipient tuberculosis in published whole blood transcriptomic datasets, in which blood sampling and longitudinal follow-up was done. We searched Medline and Embase on April 15, 2019, without language or date restrictions, using comprehensive terms for “tuberculosis”, “transcriptome”, “signature” and “blood”, with screening of identified titles and abstracts done by two independent reviewers. We included candidate whole blood mRNA signatures discovered with a primary objective of diagnosis of active or incipient tuberculosis compared with controls who were either deemed healthy or had latent tuberculosis infection. We tested the performance of eligible signatures in published whole blood transcriptomic datasets where sampling before tuberculosis diagnosis was done and interval time to disease was available. The full search strategy, eligibility criteria, and screening procedures are outlined in appendix 1 (pp 2–3).

sis infection. We tested the performance of eligible signatures in published whole blood transcriptomic datasets where sampling before tuberculosis diagnosis was done and interval time to disease was available. The full search strategy, eligibility criteria, and screening procedures are outlined in appendix 1 (pp 2–3). In preparation for this meta-analysis, we extended the follow-up of a previously published cohort of London tuberculosis contacts26 by relinking the full cohort to national tuberculosis surveillance records (until Dec 31, 2017; median follow-up increased from 0·9 years [IQR 0·7–1·2] to 1·9 years [1·7–2·2]) held at Public Health England using a validated algorithm.28 National tuberculosis surveillance records include all statutory national tuberculosis notifications. An additional 27 samples and individuals were also available for inclusion in our analysis. The full updated dataset for this study is available in ArrayExpress (accession number E-MTAB-6845). The London contacts study was approved by the UK National Research Ethics Service (reference 14/EM/1208).26 No other ethical approvals were sought for this meta-analysis because all other included patient-level datasets were depersonalised and publicly available.

vailable in ArrayExpress (accession number E-MTAB-6845). The London contacts study was approved by the UK National Research Ethics Service (reference 14/EM/1208).26 No other ethical approvals were sought for this meta-analysis because all other included patient-level datasets were depersonalised and publicly available. Data analysis Individual-level RNA sequencing data were downloaded for eligible studies and processed (including correction of batch effects) as outlined in appendix 1 (p 3). Only samples obtained before the diagnosis of tuberculosis were included. Prevalent tuberculosis was defined as a tuberculosis diagnosis within 21 days of sample collection, as previously.4 Incipient tuberculosis cases were defined as individuals diagnosed with tuberculosis more than 21 days after blood RNA sample collection. Culture-confirmed and clinically or radiologically diagnosed pulmonary or extrapulmonary tuberculosis cases were included in the main analysis. Non-progressors were defined as individuals who remained tuberculosis-free during follow-up. Non-progressor samples with less than 6 months of follow-up from the date of sample collection were excluded owing to risk of outcome misclassification. Participants with prevalent tuberculosis and those who received preventive therapy were excluded. For studies with serial samples from the same individuals, serial samples were included provided that they met these criteria and that they were collected at least 6 months apart, because they were treated as independent samples in the primary analysis.

erculosis and those who received preventive therapy were excluded. For studies with serial samples from the same individuals, serial samples were included provided that they met these criteria and that they were collected at least 6 months apart, because they were treated as independent samples in the primary analysis. Scores were calculated for candidate signatures (using the authors' described methods) for each participant in the pooled dataset. For signatures that required reconstruction of support vector machine or random forest models, we validated the reconstructed model against the original authors' model by comparing receiver operating characteristic curves in their original test dataset when possible. Using a predefined control population (including only participants with negative tests for latent tuberculosis infection among the pooled dataset), batch-corrected signature scores were transformed to Z scores (by subtracting the control mean and dividing by SD) to standardise scaling across signatures.26

hen possible. Using a predefined control population (including only participants with negative tests for latent tuberculosis infection among the pooled dataset), batch-corrected signature scores were transformed to Z scores (by subtracting the control mean and dividing by SD) to standardise scaling across signatures.26 All analyses were done using R (version 3.5.1), unless otherwise specified. Receiver operating characteristic curves for each signature were plotted for a 2-year time horizon. The area under the receiver operating characteristic curve (AUC) and 95% CI were calculated using the DeLong method.29 Any data that was originally used to derive specific signatures were excluded from the pooled dataset used to test the performance of the relevant signature. Receiver operating characteristic curves and AUCs for separate study datasets were initially examined to assess the degree of between study heterogeneity. Because little heterogeneity was observed for all signatures, a one-stage individual participant data meta-analysis, assuming common diagnostic accuracy across studies, was done for the primary analysis. AUCs were directly compared in a pairwise approach using paired DeLong tests.29 The best performing signature available from all samples in the pooled dataset was used as the reference for comparison with all other signatures; signatures with AUCs smaller than the reference and with p values of less than 0·05 were deemed inferior. Correlation between signature scores was assessed by use of Spearman rank correlation. Pairwise Jaccard similarity indices between signatures were calculated using lists of their constituent genes. Clustered cocorrelation and Jaccard index matrices were generated in Morpheus using average Euclidean distance. Upstream analysis of transcriptional regulation was done using Ingenuity Pathway Analysis (version 49932394) and visualised as network diagrams in Gephi (version 0.9.2), depicting all statistically overrepresented molecules predicted to be upstream of more than two target genes for clarity, to highlight the predicted upstream regulators shared by the constituents of the transcriptional signatures.

Analysis (version 49932394) and visualised as network diagrams in Gephi (version 0.9.2), depicting all statistically overrepresented molecules predicted to be upstream of more than two target genes for clarity, to highlight the predicted upstream regulators shared by the constituents of the transcriptional signatures. Receiver operating characteristic curves and AUCs were assessed for the best performing signatures, using prespecified intervals to tuberculosis from sample collection (<3 months, <6months, <1 year, and <2 years). Sensitivity and specificity for each of these time intervals were determined at predefined cutoffs for each signature, defined as a standardised score of two, representing the 97·7th percentile of the IGRA-negative control population assuming a normal distribution, as in previous work.26 These estimates were used to model the estimated predictive values for incident tuberculosis across a range of pre-test probabilities.

each signature, defined as a standardised score of two, representing the 97·7th percentile of the IGRA-negative control population assuming a normal distribution, as in previous work.26 These estimates were used to model the estimated predictive values for incident tuberculosis across a range of pre-test probabilities. We did several sensitivity analyses. First, we restricted inclusion of tuberculosis cases to those with microbiological confirmation. Second, we included only one blood RNA sample per participant from studies that serially sampled by randomly sampling one blood sample per individual. Third, we examined sensitivity and specificity for the best performing signatures using cutoffs defined by the maximal Youden Index30 to achieve the highest accuracy within each time interval. Fourth, we recomputed the receiver operating characteristic curves using mutually exclusive time intervals to tuberculosis of 0–3, 3–6, 6–12, and 12–24 months for each curve excluding participants who had developed tuberculosis in an earlier interval. Finally, we did a two-stage individual participant data meta-analysis to ensure consistency with the primary one-stage analysis, as described in appendix 1 (p 3). This study is registered with PROSPERO, number CRD42019135618. Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

This study is registered with PROSPERO, number CRD42019135618. Role of the funding source The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results 643 unique articles were identified in the systematic review (appendix 1 p 4). Four RNA datasets (table 1) and 17 signatures (table 2) met the criteria for inclusion. The RNA datasets included the Adolescent Cohort Study (ACS) of South African adolescents with latent tuberculosis infection;24 the Bill and Melinda Gates Foundation Grand Challenges 6-74 (GC6-74) household tuberculosis contacts study in South Africa, the Gambia, and Ethiopia;25 a London tuberculosis contacts study;26 and a Leicester tuberculosis contacts study.17 All four eligible datasets were publicly available. The ACS and GC6-74 studies were nested case-control designs within larger prospective cohort studies, whereas the London and Leicester tuberculosis contacts studies were prospective cohort studies, with RNA sequencing done for all participants. All four studies were done in HIV-negative participants. The London tuberculosis contacts study included only baseline samples, whereas the ACS, GC6-74, and Leicester tuberculosis contacts studies included serial sampling. All four studies assessed participants for evidence of prevalent tuberculosis at enrolment through clinical evaluation, and the London and Leicester tuberculosis contacts studies also did chest x-rays. The GC6-74 study excluded participants with tuberculosis diagnosed within 3 months of enrolment, and ACS excluded those diagnosed within 6 months. However, participants who developed tuberculosis within these timeframes following serial sampling events were included. All four studies achieved maximal quality assessment scores (appendix 1 pp 5–6).Table 1 Characteristics of the datasets included in meta-analysis of candidate whole blood transcriptional signatures for incipient tuberculosis

developed tuberculosis within these timeframes following serial sampling events were included. All four studies achieved maximal quality assessment scores (appendix 1 pp 5–6).Table 1 Characteristics of the datasets included in meta-analysis of candidate whole blood transcriptional signatures for incipient tuberculosis Samples included Study design Population Setting HIV status Sampling Follow-up duration and method Tuberculosis case definition RNA sequencing methods Newcastle-Ottawa Scale score Baseline tuberculosis assessment London tuberculosis contacts26 324 (8 tuberculosis; 316 healthy) Cohort Adult tuberculosis contacts London, UK Negative Baseline Median 1·9 (IQR 1·7– 2·2) years, record linkage Culture-confirmed, or clinically diagnosed 15–20 million 41 bp paired-end reads 7/7 Clinical evaluation and chest x-ray Adolescent Cohort Study24 287 (73 tuberculosis; 214 healthy) Nested case-control Adolescents with latent tuberculosis infection South Africa Negative Serial (0, 6, 12, and 24 months) 2 years, active Intrathoracic disease with 2 positive smears, or 1 positive culture 30 million 50 bp paired-end reads 9/9 Clinical evaluation; tuberculosis <6 months from enrolment excluded; chest x-ray not specified Grand Challenges 6-7425 412 (98 tuberculosis; 314 healthy) Nested case-control Adult household pulmonary tuberculosis contacts South Africa, The Gambia, Ethiopia Negative Serial (0, 6, and 18 months) 2 years, active Culture-confirmed or clinically diagnosed 60 million 50 bp paired-end reads 9/9 Clinical evaluation; tuberculosis <3 months from enrolment excluded; chest x-ray not specified Leicester tuberculosis contacts17 103 (4 tuberculosis; 99 healthy) Cohort Adult tuberculosis contacts Leicester, UK Negative Baseline plus serial for a subset* 2 years, active Confirmed by culture or Xpert MTB/RIF 25 million 75 bp paired-end reads 7/7 Clinical evaluation and chest x-ray * Owing to the high frequency of serial sampling (<6-monthly), only baseline samples were included.

Cohort Adult tuberculosis contacts Leicester, UK Negative Baseline plus serial for a subset* 2 years, active Confirmed by culture or Xpert MTB/RIF 25 million 75 bp paired-end reads 7/7 Clinical evaluation and chest x-ray * Owing to the high frequency of serial sampling (<6-monthly), only baseline samples were included. Table 2 Characteristics of candidate whole blood transcriptional signatures for incipient tuberculosis included in systematic review and meta-analysis

Cohort Adult tuberculosis contacts Leicester, UK Negative Baseline plus serial for a subset* 2 years, active Confirmed by culture or Xpert MTB/RIF 25 million 75 bp paired-end reads 7/7 Clinical evaluation and chest x-ray * Owing to the high frequency of serial sampling (<6-monthly), only baseline samples were included. Table 2 Characteristics of candidate whole blood transcriptional signatures for incipient tuberculosis included in systematic review and meta-analysis Original number of genes Model Discovery population Discovery HIV status Discovery setting Discovery approach Intended application Discovery tuberculosis cases Discovery non-tuberculosis controls Eligible signatures discovered* Anderson3819† 42 Disease risk score‡ Children HIV positive and negative South Africa, Malawi Elastic net using genome-wide data Tuberculosis vs latent tuberculosis infection 87 43 1 BATF215 1 NA Adults HIV negative UK SVM using genome-wide data Tuberculosis vs healthy (acute vs convalescent samples) 46 31 1 Gjoen721 7 LASSO regression§ Children HIV negative India LASSO using 198 preselected genes Tuberculosis vs healthy controls and other diseases 47 36 2 Gliddon323 3 Disease risk score‡ Adults HIV positive and negative South Africa, Malawi16 Forward Selection-Partial Least Squares using genome-wide data Tuberculosis vs latent tuberculosis infection 285 (tuberculosis and non-tuberculosis) .. 1 Huang1131† 13 SVM (linear kernel) Adults HIV negative UK22 Common genes from elastic net, L1/2 and LASSO models, using genome-wide data Tuberculosis vs healthy controls and other diseases 16 79 1 Kaforou2516† 27 Disease risk score‡ Adults HIV positive and negative South Africa, Malawi Elastic net using genome-wide data Tuberculosis vs latent tuberculosis infection 285 (tuberculosis and non-tuberculosis) .. 1 Maertzdorf418 4 Random forest¶ Adults HIV negative India Random forest using 360 selected target genes Tuberculosis vs healthy 113 76 2 NPC232 1 NA Adults Not stated Brazil Differential expression using genome-wide data Tuberculosis vs healthy 6 28 3 Qian1733 17 Sum of standardised expression Adults HIV negative UK22 Differential expression of nuclear factor, erythroid 2-like 2-mediated genes Tuberculosis vs healthy controls and other diseases 16 69 1 Rajan520 5 Unsigned sums‡ Adults HIV positive Uganda Differential expression using genome-wide data Tuberculosis vs healthy (active case finding among people living with HIV) 80 total (1:2 cases:controls) ..

of nuclear factor, erythroid 2-like 2-mediated genes Tuberculosis vs healthy controls and other diseases 16 69 1 Rajan520 5 Unsigned sums‡ Adults HIV positive Uganda Differential expression using genome-wide data Tuberculosis vs healthy (active case finding among people living with HIV) 80 total (1:2 cases:controls) .. 1 Roe326 3 SVM (linear kernel) Adults HIV negative UK Stability selection, using genome-wide data Incipient tuberculosis vs healthy 46 31 1 Singhania2017 20 Modified disease risk score‡‖ Adults HIV negative UK, South Africa Random forest using modular approach Tuberculosis vs healthy controls and other diseases Discovery set not explicitly stated .. 1 Suliman27 2 ANKRD22 – OSBPL10 Adults HIV negative Gambia, South Africa, Ethiopia Pair ratios algorithm using genome-wide data Incipient tuberculosis vs healthy 79 328 4 Suliman47** 4 (GAS6 + SEPT4) –(CD1C + BLK) Adults HIV negative Gambia, South Africa Pair ratios algorithm using genome-wide data Incipient tuberculosis vs healthy 45 141 4 Sweeney314 3 (GBP5 + DUSP3) ÷ 2 –KLF2 Adults HIV positive and negative Meta-analysis Significance thresholding and forward search in genome-wide data Tuberculosis vs healthy controls and other diseases 266 931 1 Walter4534† 51 SVM (linear kernel) Adults HIV negative USA SVMs, using genome-wide data Tuberculosis vs latent tuberculosis infection 24 24 1 Zak1624 16 SVM (linear kernel) Adolescents HIV negative South Africa SVM-based gene pair models using genome-wide data Incipient tuberculosis vs healthy 37 77 1 Signatures are referred to by combining the first author's name of the corresponding publication as a prefix, with number of constituent genes as a suffix. For signatures where not all constituent genes were identifiable in the RNA sequencing data (eg, due to records being withdrawn), the suffix indicates the number of identifiable genes included in this analysis. Log2-transformed transcripts per million data used to calculate all signatures, unless otherwise specified. NA=not applicable. SVM=support vector machine. LASSO=least absolute shrinkage and selection operator.

to records being withdrawn), the suffix indicates the number of identifiable genes included in this analysis. Log2-transformed transcripts per million data used to calculate all signatures, unless otherwise specified. NA=not applicable. SVM=support vector machine. LASSO=least absolute shrinkage and selection operator. * Indicates total number of eligible signatures discovered in each study. Where multiple signatures were discovered for the same intended purpose and from the same training dataset, we included the signature with greatest accuracy, as defined by the area under the receiver operating characteristic curve in the validation data. Where accuracy was equivalent, we included the most parsimonious signature. † Anderson38 included 42 genes in the original, Huang11 had 13, Kaforou25 had 27, and Walter45 had 51 (genes not included in current models were either duplicates or not identifiable in RNA sequencing data). ‡ For disease risk scores, the sum of downregulated genes was subtracted from the sum of upregulated genes. For unsigned sums and modified disease risk scores, genes were summed, irrespective of their direction of regulation. § Calculated using non-log-transformed data using model coefficients from original publication. ¶ Required normalisation of the training and test sets. This was done for each gene by subtracting the mean expression across all samples in the dataset and dividing by the SD. ‖ Calculated using non-log-transformed counts per million data with trimmed mean of M-values normalisation, as per original description.

§ Calculated using non-log-transformed data using model coefficients from original publication. ¶ Required normalisation of the training and test sets. This was done for each gene by subtracting the mean expression across all samples in the dataset and dividing by the SD. ‖ Calculated using non-log-transformed counts per million data with trimmed mean of M-values normalisation, as per original description. ** Modelling approach was not clear from the original description. We recreated this using two approaches: as a simple equation of gene pairs ((GAS6+SEPT4)–(CD1C+BLK)) and as an SVM using the four constituent gene pairs, as previously described.35 Because the former approach achieved marginally better performance that was closer to the authors' original description in their test dataset, this was included in the final analysis.

simple equation of gene pairs ((GAS6+SEPT4)–(CD1C+BLK)) and as an SVM using the four constituent gene pairs, as previously described.35 Because the former approach achieved marginally better performance that was closer to the authors' original description in their test dataset, this was included in the final analysis. A total of 1126 samples from 905 patients met our criteria for inclusion (appendix 1 p 7). These included 183 samples from 127 incipient tuberculosis cases, of which 117 (92%) were microbiologically confirmed. Eight (6%) of 127 tuberculosis cases were known to be extra-pulmonary, without pulmonary involvement. Baseline characteristics of the study participants are shown in the appendix 1 (pp 8–9). Of note, a large proportion of participants in the London (112 [35%] of 324) and Leicester (86 [83%] of 103) contact studies were of South Asian ethnicity. Principal component analyses revealed clear separation of samples by dataset when including the entire transcriptome, selected genes comprising only the candidate signatures included in the analysis, and invariant genes, indicative of batch effects in the data due to technical variation in RNA sequencing.36 These batch effects were eliminated after batch correction (appendix 1 p 10).

by dataset when including the entire transcriptome, selected genes comprising only the candidate signatures included in the analysis, and invariant genes, indicative of batch effects in the data due to technical variation in RNA sequencing.36 These batch effects were eliminated after batch correction (appendix 1 p 10). Of the 17 identified signatures (table 2), all were discovered from distinct publications, apart from Suliman4 and Suliman2, which were derived from different discovery populations within the same study. Five studies used existing published datasets for discovery,14, 23, 26, 31, 33 and the remainder used novel data. Two signatures were discovered from paediatric populations.19, 21 Four signature discovery datasets included HIV-infected and HIV-uninfected participants,14, 16, 19, 23 one signature was discovered in an exclusively HIV-infected population for the purpose of active case finding20 and the remainder were discovered in HIV-negative populations. Four signatures were discovered with the intention of diagnosis of incipient tuberculosis.24, 25, 26 The remaining 13 were discovered for diagnosis of active tuberculosis disease, of which five14, 17, 21, 31, 33 targeted discrimination of tuberculosis from other diseases in addition to discriminating people with tuberculosis from people who were healthy or with latent tuberculosis infection. Of the 17 included signatures, only three were not discovered through a genome-wide approach.18, 21, 33 Four signatures required reconstruction of support vector machine models,24, 26, 31, 34 and one required reconstruction of a random forest model.18 Our reconstructed models were validated against the authors' original descriptions by comparing AUCs in common datasets (appendix 1 p 11). The distribution of signature scores, stratified by study, before and after batch correction is shown in appendix 1 (p 12).

d one required reconstruction of a random forest model.18 Our reconstructed models were validated against the authors' original descriptions by comparing AUCs in common datasets (appendix 1 p 11). The distribution of signature scores, stratified by study, before and after batch correction is shown in appendix 1 (p 12). Our analysis initially suggested AUCs for the identification of incipient tuberculosis over a 2-year period were smaller overall in the GC6-74 dataset than in the ACS dataset (appendix 1 pp 13–14). However, the distribution of tuberculosis events during follow-up differed between these studies (appendix 1 pp 8–9). Following stratification by interval to disease, similar AUCs were observed between studies, suggesting that interval to disease confounded the association between source study and AUC. Because little residual between study heterogeneity was observed and principal component analyses after batch correction showed no clustering by study (appendix 1 p 10), we did a pooled data analysis without further adjustment for source study as the primary analysis.

unded the association between source study and AUC. Because little residual between study heterogeneity was observed and principal component analyses after batch correction showed no clustering by study (appendix 1 p 10), we did a pooled data analysis without further adjustment for source study as the primary analysis. We omitted scores for the Suliman2, Suliman4, and Zak16 signatures for samples comprising their corresponding training sets within the GC6-74 and ACS datasets, but included scores for these signatures for all other samples. The signature with the largest AUC for the identification of incipient tuberculosis over a 2-year period tested in pooled data from all 1126 samples was BATF2 (AUC 0·74, 95% CI 0·69–0·78). BATF2 was therefore used as the reference standard for paired comparisons of the other 16 candidate signatures. We found that seven signatures had equivalent AUCs to BATF2: Suliman2 (AUC 0·77 [0·71–0·82]), Kaforou25 (0·73 [0·69–0·78]), Gliddon3 (0·73 [0·68–0·77]), Sweeney3 (0·72 [0·68–0·77]), Roe3 (0·72 [0·67–0·77]), Zak16 (0·7 [0·64–0·76]), and Suliman4 (0·7 [0·64–0·76]). The remaining nine signatures had significantly inferior AUCs (appendix 1 p 15). The distributions of the eight best performing signatures among the IGRA-negative control population followed an approximately normal distribution before Z-score transformation (appendix 1 p 16).

·64–0·76]), and Suliman4 (0·7 [0·64–0·76]). The remaining nine signatures had significantly inferior AUCs (appendix 1 p 15). The distributions of the eight best performing signatures among the IGRA-negative control population followed an approximately normal distribution before Z-score transformation (appendix 1 p 16). The eight signatures identified with equivalent performance showed moderate to high correlation, as defined by Spearman rank correlation (correlation coefficients 0·44–0·84; appendix 1 p 17). In contrast, Singhania20, Anderson38, Huang11, and Walter45 showed little correlation with any other signature. The correlation matrix dendrogram showed the closest associations between signatures identified by the same research group (appendix 1 p 17). Spearman rank correlation and Jaccard Index had a weak positive association, suggesting that overlapping constituent genes might partially account for their correlation (appendix 1 p 17). The 40 genes comprising the eight signatures with equivalent AUCs are shown in figure 1A. Upstream analysis predicted that interferon IFNG, IFNA, STAT1 (the canonical mediator of interferon [IFN] signalling), and tumour necrosis factor (TNF) were the strongest predicted transcriptional regulators of these constituent genes (figure 1B; appendix 2).Figure 1 Genes comprising the eight best performing blood transcriptomic signatures for incipient tuberculosis

IFNA, STAT1 (the canonical mediator of interferon [IFN] signalling), and tumour necrosis factor (TNF) were the strongest predicted transcriptional regulators of these constituent genes (figure 1B; appendix 2).Figure 1 Genes comprising the eight best performing blood transcriptomic signatures for incipient tuberculosis (A) Matrix showing constituent genes for each signature. (B) Network diagram showing statistically enriched (p<0·05) upstream regulators of the 40 genes, identified by Ingenuity Pathway Analysis. Coloured nodes represent the predicted upstream regulators, grouped by function (red=cytokine, blue=transcription factor, green=other). Black nodes represent the transcriptional biomarkers downstream of these regulators. STAT1, represented by a blue node as a predicted upstream regulator of a number of genes, is also gene target for other upstream regulators. The identity of each node is indicated using Human Genome Organisation nomenclature. The size of the nodes is proportional to the number of downstream biomarkers associated with each regulator and the thickness of the edges is proportional to the –log10 p value for enrichment of each of the upstream regulators.

lators. The identity of each node is indicated using Human Genome Organisation nomenclature. The size of the nodes is proportional to the number of downstream biomarkers associated with each regulator and the thickness of the edges is proportional to the –log10 p value for enrichment of each of the upstream regulators. Scores for the eight best performing signatures, stratified by interval to disease, are shown in figure 2 and appendix 1 (p 18). AUCs of these signatures declined with increasing interval to disease (range 0·82–0·91 for 0–3 months vs 0·73–0·82 for 0–12 months; figure 3; appendix 1 p 15).Figure 2 Scatterplots showing scores of eight best performing transcription signatures for incipient tuberculosis, stratified by interval to disease Dashed horizontal lines indicate thresholds set as standardised scores of two for each signature. Number of samples included for each signature, at each timepoint, indicated in the appendix 1 (p 19). Repeated measures analysis of variance with linear trend method showed p<0·0001 for association of categorical interval to disease with decreasing scores for each of the eight signatures. Scatterplots showing scores of these signatures plotted against days to tuberculosis are shown in the appendix 1 (p 18). Figure 3 Receiver operating characteristic curves showing diagnostic accuracy of eight best performing transcriptional signatures for incipient tuberculosis

Dashed horizontal lines indicate thresholds set as standardised scores of two for each signature. Number of samples included for each signature, at each timepoint, indicated in the appendix 1 (p 19). Repeated measures analysis of variance with linear trend method showed p<0·0001 for association of categorical interval to disease with decreasing scores for each of the eight signatures. Scatterplots showing scores of these signatures plotted against days to tuberculosis are shown in the appendix 1 (p 18). Figure 3 Receiver operating characteristic curves showing diagnostic accuracy of eight best performing transcriptional signatures for incipient tuberculosis Receiver operating characteristic curves shown stratified by months from sample collection to disease. Area under the curve estimates and 95% CIs are shown in the appendix 1 (p 15). Number of samples included for each signature, at each timepoint, indicated in the appendix 1 (p 19).

Figure 3 Receiver operating characteristic curves showing diagnostic accuracy of eight best performing transcriptional signatures for incipient tuberculosis Receiver operating characteristic curves shown stratified by months from sample collection to disease. Area under the curve estimates and 95% CIs are shown in the appendix 1 (p 15). Number of samples included for each signature, at each timepoint, indicated in the appendix 1 (p 19). Figure 4 shows the diagnostic accuracy of the eight best performing candidates using prespecified cutoffs of standardised score of two based on the 97·7th percentile of the IGRA-negative control population, stratified by interval to disease and benchmarked against positive-predictive value estimates based on a pre-test probability of 2%. At this threshold, test sensitivities over 0–24 months of the eight best performing signatures ranged from 24·7% (95% CI 16·6–35·1) for the Suliman2 signature to 39·9% (33·0–47·2) for Sweeney3, and corresponding specificities ranged from 92·3% (89·8–94·2) to 95·3% (92·3–96·9). In contrast, over a 0–3-month interval, sensitivities ranged from 47·1% (26·2–69·0) for the Suliman4 signature to 81·0% (60·0–92·3) for the Sweeney3 signature, with corresponding specificities of 90·9% (88·9–92·6) to 94·8% (93·0–96·2). For each of the timepoints, the eight signatures had overlapping confidence intervals, and largely fell in the same positive predictive value plane (5–10% over 0–24 months vs 10–15% over 0–3 months).Figure 4 Diagnostic accuracy of eight best performing transcriptional signatures for incipient tuberculosis shown in receiver operating characteristic space, stratified by months to disease

fidence intervals, and largely fell in the same positive predictive value plane (5–10% over 0–24 months vs 10–15% over 0–3 months).Figure 4 Diagnostic accuracy of eight best performing transcriptional signatures for incipient tuberculosis shown in receiver operating characteristic space, stratified by months to disease Dashed lines represent positive-predictive values of 5%, 10%, and 15%, based on 2% pre-test probability. Grey shading indicates 95% CIs for each signature. Cutoffs derived from two standard scores above the mean of control population. The number of samples included for each signature, at each timepoint, is indicated in the appendix 1 (p 19). Point estimates and 95% CIs are also shown in the appendix 1 (p 20). On the basis of a pre-test probability of 2% at the prespecified cutoffs, all eight best performing signatures achieved a positive predictive value marginally above the WHO benchmark of 5·8% for a 0–24-month period, ranging from 6·8% for Suliman2 to 9·4% for Kaforou25, with corresponding negative-predictive values of 98·4% and 98·6% (appendix 1 p 21). For the 0–3-month period, positive predictive values ranged from 11·2% for Gliddon3 to 14·4% for Zak16, with corresponding negative-predictive values of 99·0% and 99·3% (appendix 1 p 21).

ng from 6·8% for Suliman2 to 9·4% for Kaforou25, with corresponding negative-predictive values of 98·4% and 98·6% (appendix 1 p 21). For the 0–3-month period, positive predictive values ranged from 11·2% for Gliddon3 to 14·4% for Zak16, with corresponding negative-predictive values of 99·0% and 99·3% (appendix 1 p 21). Sensitivities and specificities of the eight equivalent signatures using cutoffs defined by the maximal Youden index for each time interval were smaller than the minimum WHO target product profile criteria for a 0–24-month period but met or approximated the minimum criteria over 0–3 months (appendix 1 pp 22–23). Restricting inclusion of incipient tuberculosis cases to those with documented microbiological confirmation and including only one blood RNA sample per participant (by randomly sampling) produced no significant change to the main results (appendix 1 pp 24–25). Reanalysis of the receiver operating characteristic curves using mutually exclusive periods of 0–3, 3–6, 6–12, and 12–24 months magnified the difference in performance between the intervals, with performance declining more markedly with increasing interval to disease (appendix 1 pp 26–27). AUCs in the 12–24-month interval ranged from 0·60 (95% CI 0·50–0·70) to 0·67 (0·60–0·75) for the eight equivalent signatures. Finally, our two-stage meta-analysis approach showed similar findings to the primary analysis (appendix 1 pp 28–29).

ning more markedly with increasing interval to disease (appendix 1 pp 26–27). AUCs in the 12–24-month interval ranged from 0·60 (95% CI 0·50–0·70) to 0·67 (0·60–0·75) for the eight equivalent signatures. Finally, our two-stage meta-analysis approach showed similar findings to the primary analysis (appendix 1 pp 28–29). Discussion To our knowledge, this is the largest analysis to date of the performance of whole blood transcriptional signatures for incipient tuberculosis. We showed that eight candidate signatures performed with equivalent diagnostic accuracy over a 2-year period. These signatures ranged from a single transcript (BATF2) to 25 genes (Kaforou25). The accuracy of all eight signatures declined markedly with increasing intervals to disease. These signatures only marginally surpassed the WHO target positive predictive value of 5·8% over 2 years, assuming 2% pre-test probability and using a cutoff of two standard scores. However, sensitivity at this cutoff was only 24·7–39·9%, missing most cases. No signature achieved the WHO target sensitivity and specificity of 75% or more over 2 years, even when using the cutoff with maximal accuracy. In contrast, using two standard scores cutoffs over a 0–3-month period, the eight best performing signatures achieved sensitivities of 47·1–81·0% and specificities of more than 90%. This led to positive-predictive values of 11·2–14·4% and negative-predictive values of more than 98·9%, when assuming 2% pre-test probability, suggesting that the WHO target product profile can be achieved over shorter time intervals.

g signatures achieved sensitivities of 47·1–81·0% and specificities of more than 90%. This led to positive-predictive values of 11·2–14·4% and negative-predictive values of more than 98·9%, when assuming 2% pre-test probability, suggesting that the WHO target product profile can be achieved over shorter time intervals. To achieve the WHO target product profile, a screening strategy that incorporates serial testing on a 3–6-monthly basis might therefore be required for transcriptional signatures. Such a strategy, however, is unlikely to be feasible at a population level. Instead, high-risk groups, such as household contacts, could be targeted. However, even this approach will be challenging in high-transmission settings, given the limited global coverage of contact-tracing programmes. In low-transmission, high-resource settings, serial blood transcriptional testing for risk stratification over a defined 1–2-year period might be more achievable, particularly among recent contacts or new entry migrants from high-transmission countries, for whom risk of disease is highest within an initial 2 year interval.4, 26, 37 Integral to scale-up of the use of these biomarkers is translation of transcriptional measurements from genome-wide approaches to the reproducible quantification of selected signature gene transcripts, with appropriately defined cutoffs. Although such targeted transcript quantification has been done for some signatures using PCR-based platforms,23, 24, 25, 38 no signature platforms have been validated for implementation in a near-patient or commercial assay. An additional challenge to implementation is the cost of these assays. This cost is likely to far exceed the US$2 target specified by the WHO target product profile for a nonsputum triage test for tuberculosis disease,39 but might achieve the WHO target price to identify incipient tuberculosis for less than $100, using the price of IGRAs as an initial benchmark.11 The fact that a number of different signatures show equivalent performance enables greater freedom for commercial development of this approach by overcoming restricted access to specific signatures protected by intellectual property rights and encouraging competition to drive down costs.

GRAs as an initial benchmark.11 The fact that a number of different signatures show equivalent performance enables greater freedom for commercial development of this approach by overcoming restricted access to specific signatures protected by intellectual property rights and encouraging competition to drive down costs. The eight signatures that achieved equivalent performance were discovered with the primary intention of diagnosis of incipient tuberculosis,24, 25, 26 or differentiating active tuberculosis from people who are healthy or with latent tuberculosis infection.14, 15, 16, 23 Discovery populations for these eight signatures included adults or adolescents from the UK or sub-Saharan Africa,15, 16, 23, 24, 25, 26 or a meta-analysis of microarray data from multiple studies,14 including a minimum of 37 incipient or active tuberculosis cases. All eight signatures were discovered using genome-wide approaches. In contrast, the nine signatures with inferior performance included two derived from studies in children,19, 21 one from a study that prioritised discrimination of active tuberculosis from other bacterial and viral infections,17 and one from a study that conducted active case-finding for tuberculosis among people living with HIV.20 The differences in primary intended applications, which are reflected in the study populations used for biomarker discovery, might account for their inferior performance when evaluated solely for identification of incipient tuberculosis in a predominantly healthy, HIV-negative adult and adolescent population. The signatures with inferior performance also included three discovered in panels of pre-selected candidate genes, rather than a genome-wide approach,18, 21, 33 and four with only 6–24 tuberculosis cases in the discovery sets.31, 32, 33, 34 These observations suggest that use of a genome-wide approach and inclusion of adequate numbers of diseased cases should be considered during signature discovery to increase the likelihood of identifying generalisable signatures.

18, 21, 33 and four with only 6–24 tuberculosis cases in the discovery sets.31, 32, 33, 34 These observations suggest that use of a genome-wide approach and inclusion of adequate numbers of diseased cases should be considered during signature discovery to increase the likelihood of identifying generalisable signatures. The eight best performing signatures were derived from the application of different computational approaches but showed moderate to high levels of cocorrelation, with the closest associations between signatures identified by the same research group. This finding likely reflects common discovery datasets and modelling approaches used within research groups. Overlapping constituent genes only partially accounted for correlation between signatures, suggesting that they reflect different dimensions of a common host response to infection with Mycobacterium tuberculosis. This hypothesis was strongly supported by the identification IFN and TNF signalling pathways as statistically enriched upstream regulators of the genes across the eight signatures. Although these host response pathways are unlikely to be specific to tuberculosis, the application of these biomarkers for incipient tuberculosis mitigates against the limitations of imperfect specificity by focusing on asymptomatic individuals in whom the probability of other diseases is low. The time-dependent sensitivity of the signatures suggests that the duration of the incipient phase of tuberculosis is typically 3–6 months. However, even within the less than 3-month time interval, the sensitivity of the best performing transcriptional signatures ranged from 47·1–81·0%, indicating that the biomarkers might have imperfect sensitivity for incipient tuberculosis or that the incipient phase can progress very rapidly among a subset of cases. Each signature did exhibit an AUC of more than 0·5 for discriminating incipient tuberculosis from non-progressors even 12–24 months after sampling, suggesting that the incipient phase might be more prolonged in some cases. These slowly progressive cases might reflect those in which the host response initially achieves mycobacterial control in dynamic host–pathogen interactions.40 These findings are generally mirrored in proteomic and metabolomic data from similar cohorts.41, 42

t the incipient phase might be more prolonged in some cases. These slowly progressive cases might reflect those in which the host response initially achieves mycobacterial control in dynamic host–pathogen interactions.40 These findings are generally mirrored in proteomic and metabolomic data from similar cohorts.41, 42 The strengths of this study include the size of the pooled dataset, including 1126 samples from 905 patients and 183 samples from 127 incipient tuberculosis cases. Individual-level data were available for all four eligible studies, all of which achieved maximal quality assessment scores and were done in relevant target populations of either recent tuberculosis contacts or people with latent tuberculosis infection. This facilitated a robust analysis of the diagnostic accuracy of the candidate signatures, stratified by interval to disease. Additionally, we did a comprehensive systematic review and identified 17 candidate signatures. For each of these signatures, gene lists and modelling approaches were extracted and validated by independent reviewers. Moreover, for signatures that required model reconstruction, our models were cross-validated against original models by comparing AUCs using the same dataset wherever possible. This approach facilitated a comprehensive, head-to-head analysis of candidate signatures for incipient tuberculosis for the first time, ensuring that each head-to-head comparison was done on paired data. This approach contrasts with a head-to-head systematic evaluation that included only two of the eight best-performing signatures in our analysis and compared performance for incipient tuberculosis in only one dataset over a 0–6-month period.35 Furthermore, our meta-analytic methods ensured a standardised approach to RNA sequencing data, which included an unbiased approach to batch correction, with unchanged distributions of signature scores within each dataset following correction.

nce for incipient tuberculosis in only one dataset over a 0–6-month period.35 Furthermore, our meta-analytic methods ensured a standardised approach to RNA sequencing data, which included an unbiased approach to batch correction, with unchanged distributions of signature scores within each dataset following correction. A weakness of our analysis is that we were unable to do subgroup analyses by age, ethnicity, or country, because the contributing studies largely defined these strata. There were no clear differences in performance by study, supporting the generalisability of the results. We were also unable to account for previous BCG vaccination status, although we anticipate that BCG coverage is likely to be very high among the study populations included. Additionally, having observed little heterogeneity between studies, we did a pooled analysis, assuming common diagnostic accuracy between studies. The precision of our estimates therefore might be slightly overstated and statistical tests might be anti-conservative. However, sensitivity analysis using a two-stage meta-analysis approach with random effects yielded similar findings, supporting the robustness of our results. Likewise, treating serial samples as independent was anti-conservative, but findings were similar in our sensitivity analysis taking only one sample per individual at random.

itivity analysis using a two-stage meta-analysis approach with random effects yielded similar findings, supporting the robustness of our results. Likewise, treating serial samples as independent was anti-conservative, but findings were similar in our sensitivity analysis taking only one sample per individual at random. All included datasets were from sub-Saharan Africa and the UK, although a substantial proportion of Asian participants were included in the UK studies. No data were available for people living with HIV or children younger than 10 years, among whom different blood transcriptional perturbations might occur in tuberculosis.8, 19 Prospective validation studies in other regions and among these specific target populations are needed and could be used to periodically update this meta-analysis to further increase generalisability. Only eight tuberculosis cases were known to be extra-pulmonary, thus precluding assessment of diagnostic accuracy stratified by tuberculosis disease site. Additionally, most incipient tuberculosis cases were contributed from the African datasets, with 12 cases from the UK studies. Nevertheless, the UK studies were done in appropriate target populations of close contacts of tuberculosis index cases and were done as cohort studies, as opposed to the African case-control designs. High specificity for correctly identifying non-progressors among contacts is a key attribute in improving positive predictive value compared with existing tests. Hence, these UK datasets were valuable additions to the pooled meta-analysis. Furthermore, when multiple signatures were discovered from the same discovery population and for the same purpose, we only included the best performing signature from the original study's validation set in our analysis. We therefore excluded a small number of worse-performing candidate signatures to prioritise a parsimonious list of the most promising candidates. The probability of these excluded signatures performing better than the included signatures is therefore negligible.

from the original study's validation set in our analysis. We therefore excluded a small number of worse-performing candidate signatures to prioritise a parsimonious list of the most promising candidates. The probability of these excluded signatures performing better than the included signatures is therefore negligible. In summary, we show for the first time that eight transcriptional signatures, including a single transcript (BATF2), have equivalent diagnostic accuracy for identification of incipient tuberculosis. Performance appeared similar across studies, including participants from the UK and sub-Saharan Africa. Signature performance was highly time-dependent, with lower accuracy at longer intervals to disease. A screening strategy that incorporates serial testing on a 3–6-monthly basis among selected high-risk groups might be required for these biomarkers to surpass WHO target product profile benchmarks. Supplementary Materials Supplementary appendix 1 Supplementary appendix 2 Acknowledgments The study was funded by National Institute for Health Research (NIHR; DRF-2018-11-ST2-004 to RKG and SRF-2011-04-001 and NF-SI-0616-10037 to IA), by the Wellcome Trust (207511/Z/17/Z to MN), and by NIHR Biomedical Research Funding to University College London and University College London Hospital. This paper presents independent research supported by the NIHR. The views expressed are those of the authors and not necessarily those of the National Health Service, the NIHR, or the Department of Health and Social Care.

R Biomedical Research Funding to University College London and University College London Hospital. This paper presents independent research supported by the NIHR. The views expressed are those of the authors and not necessarily those of the National Health Service, the NIHR, or the Department of Health and Social Care. Contributors RKG, IA, and MN conceived the study. RKG, CTT, and MN wrote the systematic review protocol, did the literature review and extracted signature models. RKG did the analyses and wrote the first draft of the manuscript, supported by CTT and MN. All other authors contributed to the methods or interpretation. All authors have seen and agreed on the final submitted version of the manuscript. Declaration of interests MN has a patent application pending in relation to blood transcriptomic biomarkers of tuberculosis (UK patent application number 1603367.2). All other authors declare no competing interests.

Introduction Delays in diagnosis of active tuberculosis contribute to its high death toll and facilitate onward transmission of infection.1 Current diagnostic tools include smear microscopy, microbiological culture, and molecular detection by Xpert MTB/RIF (Xpert) or Xpert MTB/RIF Ultra (Ultra). These all rely on obtaining sputum or other biological samples from the site of disease. Each approach has additional limitations, such as the poor sensitivity of microscopy, the time delay for culture, the high cost of molecular tests, and false-positive Ultra results arising from detection of non-viable Mycobacterium tuberculosis. WHO has specified an urgent need for a rapid, simple, and low-cost triage test that prioritises sensitivity to confidently rule out tuberculosis, or to identify patients who require further investigation.2 A Delphi process partly informed by cost-effectiveness considerations concluded that such a test required a minimum of 90% sensitivity and 70% specificity.2, 3 As not all patients with tuberculosis produce sputum spontaneously, a nonsputum confirmatory test that prioritises specificity is also advocated.2 Research in context Evidence before this study

Introduction Delays in diagnosis of active tuberculosis contribute to its high death toll and facilitate onward transmission of infection.1 Current diagnostic tools include smear microscopy, microbiological culture, and molecular detection by Xpert MTB/RIF (Xpert) or Xpert MTB/RIF Ultra (Ultra). These all rely on obtaining sputum or other biological samples from the site of disease. Each approach has additional limitations, such as the poor sensitivity of microscopy, the time delay for culture, the high cost of molecular tests, and false-positive Ultra results arising from detection of non-viable Mycobacterium tuberculosis. WHO has specified an urgent need for a rapid, simple, and low-cost triage test that prioritises sensitivity to confidently rule out tuberculosis, or to identify patients who require further investigation.2 A Delphi process partly informed by cost-effectiveness considerations concluded that such a test required a minimum of 90% sensitivity and 70% specificity.2, 3 As not all patients with tuberculosis produce sputum spontaneously, a nonsputum confirmatory test that prioritises specificity is also advocated.2 Research in context Evidence before this study We did a systematic review, using comprehensive terms for “tuberculosis”, “transcriptional”, “signatures”, and “blood”, with no language or date restrictions. Many studies have been done with the aim of discovering whole-blood transcriptional signatures that discriminate individuals with tuberculosis from disease-free controls or from patients with other infectious or respiratory diseases. Several candidate signatures have thus been identified, raising hope of translation into near-patient assays. However, validation of these signatures has been limited, especially in settings where they are needed most and in sick patients undergoing routine investigation for tuberculosis. Only one previous study compared the diagnostic accuracy of candidate signatures in a head-to-head analysis, but key signatures were not included, and validation relied solely on existing case-control datasets. It has therefore been unclear which candidate signature works best for the diagnosis of tuberculosis, or if any signatures meet minimum or optimum benchmarks proposed by WHO in a real-world observational cohort. Addressing these research gaps is crucial to inform whether these biomarkers should be translated into scalable test platforms or considered for adoption by national programmes.

r the diagnosis of tuberculosis, or if any signatures meet minimum or optimum benchmarks proposed by WHO in a real-world observational cohort. Addressing these research gaps is crucial to inform whether these biomarkers should be translated into scalable test platforms or considered for adoption by national programmes. Added value of this study

r the diagnosis of tuberculosis, or if any signatures meet minimum or optimum benchmarks proposed by WHO in a real-world observational cohort. Addressing these research gaps is crucial to inform whether these biomarkers should be translated into scalable test platforms or considered for adoption by national programmes. Added value of this study To our knowledge, we provide the first comprehensive and systematic head-to-head comparison of candidate transcriptional signatures for identification of active tuberculosis in a prospective diagnostic accuracy study. Moreover, we used an unbiased consecutive sampling approach, in contrast to the case-control design of previous studies. Among 181 consecutive patients presenting for investigation of presumptive pulmonary tuberculosis in South Africa, four of 27 candidate transcriptional signatures performed equivalently to each other in discriminating individuals with tuberculosis from those without, irrespective of HIV status and other baseline characteristics. These signatures met or approximated to the minimum WHO target product profile for a triage test (of 90% sensitivity, 70% specificity). However, no signature met the optimum criteria (of 95% sensitivity, 80% specificity) for a tuberculosis triage test, or the minimum criteria for a confirmatory test (65% sensitivity, 98% specificity). The best-performing signatures all improved the specificity of the Xpert MTB/RIF Ultra microbiological molecular test for Mycobacterium tuberculosis DNA, in which the advantages of greater sensitivity have been undermined by a higher rate of false-positive results.

confirmatory test (65% sensitivity, 98% specificity). The best-performing signatures all improved the specificity of the Xpert MTB/RIF Ultra microbiological molecular test for Mycobacterium tuberculosis DNA, in which the advantages of greater sensitivity have been undermined by a higher rate of false-positive results. Implications of all the available evidence Selected blood transcriptional biomarkers show promise as triage tests for patients being investigated for pulmonary tuberculosis in high-incidence settings, exemplified by our study site. The signatures did not achieve the minimum criteria needed for a confirmatory test and should not be used by themselves for this purpose. Nonetheless, they might improve diagnostic accuracy when used in conjunction with highly sensitive molecular tests for M tuberculosis DNA. These data support further development of assays for blood transcriptional biomarkers to enable interventional trials of their potential clinical and health-economic effects in the diagnostic pathway for tuberculosis.

nostic accuracy when used in conjunction with highly sensitive molecular tests for M tuberculosis DNA. These data support further development of assays for blood transcriptional biomarkers to enable interventional trials of their potential clinical and health-economic effects in the diagnostic pathway for tuberculosis. Many host blood transcriptional signatures have been proposed to differentiate patients with pulmonary tuberculosis from healthy controls or patients with other infectious or respiratory diseases,4 raising hopes for translation into near-patient assays. However, validation of these signatures is currently limited to evidence from case-control studies.5 Such studies are prone to overestimate performance because of the spectrum effect arising from differences in disease prevalence and other unmeasured covariates in selected patient subgroups, and biased inclusion of cases at extremes of the distribution of phenotypes that might not be representative of the target population.6 Independent validation in prospective, real-world populations is therefore crucial to assess true test performance, however, there are no comprehensive head-to-head comparisons in such settings for candidate blood transcriptional tuberculosis signatures.

types that might not be representative of the target population.6 Independent validation in prospective, real-world populations is therefore crucial to assess true test performance, however, there are no comprehensive head-to-head comparisons in such settings for candidate blood transcriptional tuberculosis signatures. WHO has endorsed use of Ultra to provide increased sensitivity compared with Xpert for PCR detection of M tuberculosis in sputum specimens.7 However, Ultra returns more false-positive results (culture-negative) than Xpert, particularly within the semi-quantitative trace output category, which detects the lowest bacillary burden of M tuberculosis.8 The large number of false-positives has been attributed to detection of DNA from non-culturable M tuberculosis as a result of past infection, which is more likely in high-burden settings.9 The decreased specificity makes diagnostic interpretation of positive Ultra results challenging, and potentially undermines the value of its greater sensitivity.7, 8, 10 Therefore, in addition to being applied as standalone tests, blood transcriptional biomarkers of tuberculosis could improve the specificity of Ultra by resolving results in which only traces of DNA are detected or those in patients with previous tuberculosis. We undertook a prospective observational study to compare the diagnostic accuracy of candidate transcriptional signatures identified by systematic review. Our primary objective was to benchmark the performance of the signatures against the WHO target product profile (TPP) for a tuberculosis triage test. As secondary objectives, we sought to assess the performance of these signatures against WHO TPP criteria for a blood-based confirmatory tuberculosis test, and to explore their potential use as an add-on confirmatory test to clarify interpretation of positive Ultra results.

profile (TPP) for a tuberculosis triage test. As secondary objectives, we sought to assess the performance of these signatures against WHO TPP criteria for a blood-based confirmatory tuberculosis test, and to explore their potential use as an add-on confirmatory test to clarify interpretation of positive Ultra results. Methods Study design and participants Our study was nested within a diagnostic accuracy study of sputum Xpert and Ultra tests for pulmonary tuberculosis.10 Symptomatic adults (≥18 years) self-presenting for investigation of pulmonary tuberculosis were consecutively recruited in Cape Town, South Africa, from a tuberculosis clinic within a government primary health-care centre (Scottsdene). Patients were screened and investigated according to South African guidelines.11 At recruitment, demographic and clinical metadata were recorded, including a modified tuberculosis symptom score (appendix 1, p1).12 This study was approved by the Stellenbosch University Faculty of Health Sciences Research Ethics Committee (N14/10/136). All participants provided written informed consent.

Methods Study design and participants Our study was nested within a diagnostic accuracy study of sputum Xpert and Ultra tests for pulmonary tuberculosis.10 Symptomatic adults (≥18 years) self-presenting for investigation of pulmonary tuberculosis were consecutively recruited in Cape Town, South Africa, from a tuberculosis clinic within a government primary health-care centre (Scottsdene). Patients were screened and investigated according to South African guidelines.11 At recruitment, demographic and clinical metadata were recorded, including a modified tuberculosis symptom score (appendix 1, p1).12 This study was approved by the Stellenbosch University Faculty of Health Sciences Research Ethics Committee (N14/10/136). All participants provided written informed consent. Specimen microbiology and definitions Blood was collected in Tempus tubes, and patients provided two sputum samples. One was decontaminated by Mycoprep (BD, Johannesburg, South Africa) before double Ziehl-Neelsen smear microscopy and Mycobacteria Growth Indicator Tube 960 liquid culture (appendix 1, p1). The second sputum sample was used for Xpert testing. The next morning, patients provided a third sputum sample for Ultra testing. Sputum samples were either obtained via spontaneous expectoration or induced by nebulising with 5% sodium chloride for 7–10 min. In our primary analysis, patients with tuberculosis were defined as those with either a positive liquid culture or a positive Xpert result, to overcome the limitation of a single culture reference. Patients with missing blood RNA or sputum results were excluded.

Specimen microbiology and definitions Blood was collected in Tempus tubes, and patients provided two sputum samples. One was decontaminated by Mycoprep (BD, Johannesburg, South Africa) before double Ziehl-Neelsen smear microscopy and Mycobacteria Growth Indicator Tube 960 liquid culture (appendix 1, p1). The second sputum sample was used for Xpert testing. The next morning, patients provided a third sputum sample for Ultra testing. Sputum samples were either obtained via spontaneous expectoration or induced by nebulising with 5% sodium chloride for 7–10 min. In our primary analysis, patients with tuberculosis were defined as those with either a positive liquid culture or a positive Xpert result, to overcome the limitation of a single culture reference. Patients with missing blood RNA or sputum results were excluded. Blood RNA sequencing and data processing Extraction and sequencing of blood mRNA was done as previously described,13 resulting in a median of 25 million (range 9–33 million, IQR 21–27 million) 41 bp paired-end reads per sample. Blood samples with an insufficient RNA yield were not processed for sequencing. Data are available on Array Express, accession number E-MTAB-8290. RNA sequencing and data processing were done independently of microbiological test results. RNAseq data were mapped to the reference transcriptome (Ensembl Human GRCh38 release 95) and processed as previously described,14 focusing on protein-coding genes. Unless otherwise specified, log2-transformed transcripts per million values were used for analysis. To account for an observed batch effect that could not be accounted for by any biological or known technical variables (appendix 1, p 13), we tested two batch correction techniques, using the ComBat and sva functions from the sva package in R, respectively (appendix 1, pp 1–2).15 Since surrogate variable analysis preserved specified outcomes of interest (tuberculosis status, HIV status, age, sex, and ethnicity) while correcting any other, unwanted variation, and because samples clustered more tightly after batch correction with surrogate variable analysis (SVA; appendix 1, p 14), we used SVA-adjusted data for the primary analyses.

served specified outcomes of interest (tuberculosis status, HIV status, age, sex, and ethnicity) while correcting any other, unwanted variation, and because samples clustered more tightly after batch correction with surrogate variable analysis (SVA; appendix 1, p 14), we used SVA-adjusted data for the primary analyses. Systematic review of blood transcriptional signatures for tuberculosis We previously did a systematic review14 to identify candidate concise whole-blood transcriptional signatures for incipient or active tuberculosis published before April 15, 2019, including only signatures that were discovered by comparison with asymptomatic controls. In the present study, we extended the inclusion criteria from the previous review to also capture signatures intended to distinguish active tuberculosis from other diseases (appendix 1, p 2). Additionally, following initial peer review, we included two further signatures that met the inclusion criteria but were published after the date limit of our search.16, 17 All screened articles are listed in appendix 2, with reviewed full text articles matched against inclusion criteria.

er diseases (appendix 1, p 2). Additionally, following initial peer review, we included two further signatures that met the inclusion criteria but were published after the date limit of our search.16, 17 All screened articles are listed in appendix 2, with reviewed full text articles matched against inclusion criteria. Signature scores were calculated using the original authors' methods (appendix 1, pp 2–4). Some signatures included genes whose annotations have since been withdrawn, or non-coding RNA and putative pseudogenes that were not present in our protein-coding RNAseq dataset (appendix 3). Where changes to the original model were made, or where a model had to be recreated, we validated the reconstructed model by comparing the area under the receiver operating characteristics curves (AUROCs) in the original dataset where possible (appendix 1, p 7).

resent in our protein-coding RNAseq dataset (appendix 3). Where changes to the original model were made, or where a model had to be recreated, we validated the reconstructed model by comparing the area under the receiver operating characteristics curves (AUROCs) in the original dataset where possible (appendix 1, p 7). Statistical analysis Our sample size was primarily determined by the number of participants in the parent study10 with paired blood RNA and sputum samples. To assess our statistical power, we used published models for estimates of sample size calculations in diagnostic tests (appendix 1, p 12).18, 19 The prevalence of tuberculosis in patients of the parent study was 30% (72/239).10 At this prevalence, a total sample size of more than 135 participants was required to establish whether the blood transcriptional biomarkers could achieve the minimum thresholds of the WHO TPP for a triage test (90% sensitivity and 70% specificity) with a 10% margin of error. Assuming the best-performing test achieved an AUROC of at least 0·9 (as is generally the case in the original reports of each signature), a total sample size of more than 130 participants was required for 80% power to identify a 0·1 difference in AUROCs between paired tests.

70% specificity) with a 10% margin of error. Assuming the best-performing test achieved an AUROC of at least 0·9 (as is generally the case in the original reports of each signature), a total sample size of more than 130 participants was required for 80% power to identify a 0·1 difference in AUROCs between paired tests. This study is reported in accordance with the Standards for Reporting of Diagnostic Accuracy Studies guidelines.20 p values of less than 0·05 were considered statistically significant. Cohort characteristics were compared with χ2 or Mann-Whitney tests. CIs for the differences between proportions were calculated using the Newcombe-Wilson method with continuity correction.21 The pROC package in R was used to construct receiver operating characteristic (ROC) curves to discriminate between patients with and without tuberculosis. CIs for ROC curves' sensitivities were plotted at 1% specificity intervals, using the ci.se function of the pROC package. We compared AUROCs for each candidate signature in a pairwise approach with the DeLong method,22 using the signature with highest AUROC as reference.

between patients with and without tuberculosis. CIs for ROC curves' sensitivities were plotted at 1% specificity intervals, using the ci.se function of the pROC package. We compared AUROCs for each candidate signature in a pairwise approach with the DeLong method,22 using the signature with highest AUROC as reference. To test for differential diagnostic accuracy among predefined population subgroups, we stratified the cohort according to age, sex, ethnicity, HIV infection, previous tuberculosis, and indices of disease severity at presentation (symptom score, body-mass index [BMI], haemoglobin concentration, and sputum smear results). We constructed univariable subgroup-specific ROC curves and compared their AUROCs using DeLong tests. Sensitivity, specificity, and predictive values were reported at the maximum Youden index reflecting the highest test accuracy.23 Additionally, we assessed diagnostic accuracy when fixing sensitivity and specificity at the minimum and optimum thresholds, as defined by the WHO TPP criteria for triage and confirmatory tests of tuberculosis,2 using the coords function in the pROC package. WHO thresholds for a triage test were minimum 90% sensitivity, 70% specificity; optimum 95% sensitivity, 80% specificity. WHO thresholds for a confirmatory test were minimum 65% sensitivity, 98% specificity. CIs for proportions were calculated using the binomial Wilson method,24 implemented in the binconf function of the hmisc R package. We used the upper limit of the CIs for each signature to assess whether they might achieve the required thresholds for sensitivity and specificity. McNemar's tests were used to compare sensitivity and specificity between Ultra analysis alone and a diagnostic algorithm combining sputum Ultra analysis with transcriptional signatures (appendix 1, p 4).

CIs for each signature to assess whether they might achieve the required thresholds for sensitivity and specificity. McNemar's tests were used to compare sensitivity and specificity between Ultra analysis alone and a diagnostic algorithm combining sputum Ultra analysis with transcriptional signatures (appendix 1, p 4). We did three sensitivity analyses to confirm the robustness of our results. First, we restricted the tuberculosis case definition to patients with culture-confirmation, irrespective of Xpert results. Second, we estimated the best possible specificity of the transcriptional signatures by simulating increased sensitivity of the standard reference that might be achieved using four sputum cultures (appendix 1, p 4),25 as previously described.10 Third, ComBat was used as an alternative batch correction method to the surrogate variable analysis used in primary analysis. All statistical analyses were done, and data graphically visualised, in R (version 3.6.0) or GraphPad Prism (version 8.1.1). Role of the funding source The funders of the study had no role in study design, data collection, data analysis, data interpretation, writing of the report, or the decision to submit for publication. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

Results Between Feb 12, 2016, and July 18, 2017, we obtained blood RNA samples from 205 consecutive patients.10 Paired sputum and RNA sequencing data were available in 181 participants included in our analysis (figure 1). Their baseline characteristics are given in table 1; characteristics of participants who were excluded from the analysis are in appendix 1 (p 8). 54 (30%) of 181 patients had pulmonary tuberculosis, confirmed by sputum culture or Xpert, and included all the individuals who received tuberculosis treatment at enrolment, further increasing our confidence in the sensitivity of our standard reference for the diagnosis of tuberculosis (appendix 1, p 15). 44 (24%) of 181 patients were HIV-infected.Figure 1 Study flowchart Table 1 Baseline characteristics of study cohort All (n=181) No tuberculosis (n=127) Positive for pulmonary tuberculosis (n=54) Age, years 35 (27–48) 36 (28–49) 34 (24–43) Sex Male 94 (52%) 66 (52%) 28 (52%) Female 87 (48%) 61 (48%) 26 (48%) Ethnicity Black 28 (15%) 14 (11%) 14 (26%) Mixed ancestry 153 (85%) 113 (89%) 40 (74%) HIV status Unknown* 1 (1%) 1 (1%) 0 Uninfected 136 (75%) 99 (78%) 37 (69%) Infected 44 (24%) 27 (21%) 17 (31%) Antiretroviral therapy† No 24 (55%) 14 (52%) 10 (59%) Yes 15 (34%) 12 (44%) 3 (18%) Unknown* 5 (11%) 1 (4%) 4 (24%) CD4 count†, cells per μL 334 (192–606) 354 (207–707) 326 (128–484) Haemoglobin concentration, g/dL 13·7 (12·4–14·8) 14·2 (13·2–15·4) 12·6 (11·3–13·6) Leucocyte count, × 109 cells per L 8 (6·1–10·2) 7·6 (6–9·8) 9·1 (6·8–11) BMI, kg/m2 19·9 (17·8–22·5) 20·5 (18·4–23·2) 19·1 (16·8–21·5) Tuberculosis symptom score 2 (2–3) 2 (1–3) 3 (2–5) Previous tuberculosis No 115 (64%) 81 (64%) 34 (63%) Yes 66 (36%) 46 (36%) 20 (37%) Liquid culture Positive 53 (29%) NA 53 (98%) Negative 128 (71%) 128 (100%) 1 (2%) Sputum smear Positive 15 (8%) 1 (1%) 14 (26%) Negative 157 (87%) 120 (94%) 37 (69%) Not done 9 (5%) 6 (5%) 3 (6%) Xpert Positive 44 (24%) NA 44 (81%) Negative 134 (74%) 124 (98%) 10 (19%) No result 2 (2%) 2 (2%) NA Not done 1 (1%) 1 (1%) NA Ultra Positive‡ 51 (28%) 10 (8%) 41 (76%) Negative 103 (57%) 94 (74%) 9 (17%) No result 10 (6%) 8 (6%) 2 (4%) Not done 17 (9%) 15 (12%) 2 (4%) Data are n (%) or median (IQR). Individuals positive for pulmonary tuberculosis were defined as those with either a positive liquid culture or a positive Xpert MTB/RIF result, or both. Individuals with missing data: CD4 cell counts (n=1), haemoglobin concentration (n=3), leucocytes (n=3), BMI (n=1), symptom score (n=3). BMI=body-mass index. NA=not applicable.

ls positive for pulmonary tuberculosis were defined as those with either a positive liquid culture or a positive Xpert MTB/RIF result, or both. Individuals with missing data: CD4 cell counts (n=1), haemoglobin concentration (n=3), leucocytes (n=3), BMI (n=1), symptom score (n=3). BMI=body-mass index. NA=not applicable. * Category excluded for χ2 statistical test. † Antiretroviral therapy and CD4 cell counts for HIV-infected patients only. ‡ Positive Ultra results include tests where traces of Mycobacterium tuberculosis were detected. Compared with individuals without tuberculosis, a greater proportion of patients with tuberculosis were black (difference of proportions 0·24 [95% CI 0·04–0·44]). Patients with tuberculosis also had higher symptom scores (difference of means 1·1, [0·6–1·5]), lower haemoglobin concentrations (−1·7 [–2·3 to −1·1]), lower BMI (−1·9 [–3·3 to −0·4]), and increased leucocytes (1·3 [0·2–2·4]). None of these clinical parameters independently discriminated between patients with and without tuberculosis with sufficient diagnostic accuracy for a tuberculosis triage test as defined by WHO TPP (appendix 1, p 16).2

ons (−1·7 [–2·3 to −1·1]), lower BMI (−1·9 [–3·3 to −0·4]), and increased leucocytes (1·3 [0·2–2·4]). None of these clinical parameters independently discriminated between patients with and without tuberculosis with sufficient diagnostic accuracy for a tuberculosis triage test as defined by WHO TPP (appendix 1, p 16).2 27 signatures from 18 of 645 articles identified by our systematic review and expert consultation met the inclusion criteria (appendix 1, p 17; appendix 2; table 2). 14 of these 27 signatures were derived from study populations that included South African participants. Ten signatures were discovered in datasets that included HIV-infected participants. Five signatures were intended for diagnosis of incipient tuberculosis; 22 signatures were discovered with their intended application for diagnosis of active tuberculosis disease. Of these 22 signatures, eight aimed to distinguish tuberculosis from asymptomatic controls (including people who were healthy or with latent tuberculosis infection), five targeted discrimination of tuberculosis from other diseases, and nine aimed to distinguish tuberculosis from a mixed population of patients with other diseases and healthy controls. 24 of the 27 signatures were discovered through a genome-wide approach. Ten signatures required reconstruction of random forest or support vector machine models. We assessed whether each of the models that required reconstruction or had been otherwise altered, achieved the AUROC reported by the authors in the original dataset (appendix 1, p 7). We could not recapitulate the original AUROC for two signatures: Anderson39.OD,26 which had been reduced from 51 transcripts originally to the 39 protein-coding transcripts that were available in our RNAseq dataset, and Duffy10,16 for which our attempt to reconstruct the original model did not achieve the same AUROC as that reported in their validation data. In this case, we used a binary support vector machine model, which did reproduce a similar AUROC in their validation cohort. In addition, this assessment was not possible for two other signatures (Huang11 and Kaforou45) for which the AUROCs were not provided in the original reports.30, 31Table 2 Description of candidate blood transcriptional signatures for tuberculosis

ne model, which did reproduce a similar AUROC in their validation cohort. In addition, this assessment was not possible for two other signatures (Huang11 and Kaforou45) for which the AUROCs were not provided in the original reports.30, 31Table 2 Description of candidate blood transcriptional signatures for tuberculosis Original gene number Model Intended application Discovery datasets Population HIV status Setting Approach Tuberculosis cases Controls Total Anderson39.LTBI26 42 Disease risk score Tuberculosis vs LTBI Children Positive or negative South Africa, Malawi Elastic net using genome-wide data 87 43 130 Anderson39.OD26 51 Disease risk score Tuberculosis vs OD Children Positive or negative South Africa, Malawi Elastic net using genome-wide data 87 134 221 BATF227 1 NA Tuberculosis vs HC (acute vs convalescent) Adults Negative UK SVM using genome-wide data 46 31 77 Duffy1016 10 SVM (linear kernel) Tuberculosis vs LTBI and OD Adults Positive or negative South Africa Multinomial random forest using genome-wide data 93 207 300 Gjoen828 7 LASSO regression Tuberculosis vs HC and OD Children Negative India LASSO using 198 pre-selected genes 47 36 83 Gliddon329 3 (FCGR1A + C1QB) − (ZNF296) Tuberculosis vs LTBI Adults Positive or negative South Africa, Malawi FS-PLS using genome-wide data NS NS 285 Gliddon429 4 (GBP6 + PRDM1) − (TMCC1 + ARG1) Tuberculosis vs OD Adults Positive or negative South Africa, Malawi FS-PLS using genome-wide data NS NS 293 Huang1130 13 SVM (linear kernel) Tuberculosis vs HC and OD Adults Negative UK Common genes from elastic net, L1/2 and LASSO models, using genome-wide data 16 79 95 Kaforou2531 27 Disease risk score Tuberculosis vs LTBI Adults Positive or negative South Africa, Malawi Elastic net using genome-wide data NS NS 285 Kaforou3931 44 Disease risk score Tuberculosis vs OD Adults Positive or negative South Africa, Malawi Elastic net using genome-wide data NS NS 293 Kaforou4531 53 Disease risk score Tuberculosis vs LTBI and OD Adults Positive or negative South Africa, Malawi Elastic net using genome-wide data NS NS NS Maertzdorf432 4 Random forest Tuberculosis vs HC Adults Negative India Random forest using 360 selected target genes 113 76 189 NPC233 1 NA Tuberculosis vs HC and LTBI Adults NS Brazil Differential expression using genome-wide data 6 28 34 Penn-Nicholson617 6 Difference of means Incipient tuberculosis vs HC Adolescents Negative South Africa SVM-based gene pair models using genome-wide data 46 107 153 Qian1734 17 Sum of standardised

113 76 189 NPC233 1 NA Tuberculosis vs HC and LTBI Adults NS Brazil Differential expression using genome-wide data 6 28 34 Penn-Nicholson617 6 Difference of means Incipient tuberculosis vs HC Adolescents Negative South Africa SVM-based gene pair models using genome-wide data 46 107 153 Qian1734 17 Sum of standardised expression Tuberculosis vs HC and OD Adults Negative UK Differential expression of Nrf2-mediated genes 16 69 85 Rajan535 5 Unsigned sums Tuberculosis vs HC (screening among PLHIV) Adults Positive Uganda Differential expression using genome-wide data NS NS 80 (1:2 cases:controls) Roe313 3 SVM (linear kernel) Incipient tuberculosis vs HC Adults Negative UK Stability selection using genome-wide data 46 31 77 Roe427 4 SVM (linear kernel) Tuberculosis vs OD Adults Negative UK SVM using genome-wide data 23 35 58 Roe527 5 SVM (linear kernel) Tuberculosis vs HC and OD Adults Negative UK SVM using genome-wide data 23 50 73 Singhania2036 20 Modified disease risk score Tuberculosis vs HC and OD Adults Negative UK, South Africa Random forest using modular approach NS NS NS Suliman237 2 ANKRD22 −OSBPL10 Incipient tuberculosis vs HC Adults Negative The Gambia, South Africa Pair ratios algorithm using genome-wide data 79 328 407 Suliman437 4 (GAS6 + SEPT4) – (CD1C + BLK) Incipient tuberculosis vs HC Adults Negative The Gambia, South Africa, Ethiopia Pair ratios algorithm using genome-wide data 45 141 186 Sweeney338 3 (GBP5 + DUSP3)/2 −KLF2 Tuberculosis vs LTBI and OD Adults Positive or negative Meta-analysis of South Africa, Malawi, UK, France, USA Significance thresholding and forward search in genome-wide data 296 727 1023 Walter4639 51 SVM (linear kernel) Tuberculosis vs LTBI Adults Negative USA SVM using genome-wide data 24 24 48 Walter3239 47 SVM (linear kernel) Tuberculosis vs OD Adults Negative USA SVM using genome-wide data 24 24 48 Walter10139 119 SVM (linear kernel) Tuberculosis vs LTBI and OD Adults Negative USA SVM using genome-wide data 24 48 72 Zak1640 16 SVM (linear kernel) Incipient tuberculosis vs HC Adolescents Negative South Africa SVM-based gene pair models using genome-wide data 37 77 114 Signatures were identified by systematic literature review and included for analysis.

erculosis vs LTBI and OD Adults Negative USA SVM using genome-wide data 24 48 72 Zak1640 16 SVM (linear kernel) Incipient tuberculosis vs HC Adolescents Negative South Africa SVM-based gene pair models using genome-wide data 37 77 114 Signatures were identified by systematic literature review and included for analysis. Signature names represent the first author's name of the corresponding publication, suffixed with the number of constituent genes that are present in the current RNAseq dataset. Both Anderson signatures resulted in the same number of final genes; these signatures were therefore additionally appended with the comparator control group. Details on how models were recreated are in appendix 1 (pp 2-4). LTBI=latent tuberculosis infection. OD=other diseases. NA=not applicable. HC=healthy controls. SVM=support vector machine. LASSO=least absolute shrinkage and selection operator. FS-PLS=forward selection-partial least squares. NS=not specified. Nrf2=nuclear factor, erythroid 2-like 2. PLHIV=people living with HIV.

2-4). LTBI=latent tuberculosis infection. OD=other diseases. NA=not applicable. HC=healthy controls. SVM=support vector machine. LASSO=least absolute shrinkage and selection operator. FS-PLS=forward selection-partial least squares. NS=not specified. Nrf2=nuclear factor, erythroid 2-like 2. PLHIV=people living with HIV. We ranked the 27 candidate transcriptional signatures by their AUROC for discriminating tuberculosis and no tuberculosis in all 181 patients. The signature with the highest diagnostic accuracy was Sweeney3 (AUROC 90·6% [95% CI 85·6–95·6]), which was derived from an analysis of multiple previously published studies of patients with pulmonary tuberculosis compared with controls comprising both healthy individuals and patients with non-tuberculosis diseases.41 Pairwise comparison of the remaining 26 signatures against Sweeney3 showed that three other signatures had equivalent AUROCs. These were Kaforou25 (86·9% [80·9–92·9]), Roe3 (86·9% [80·3–93·5]), and BATF2 (86·8% [80·6–93·1]; appendix 1, p 9) all derived from individual case-control studies comparing patients with tuberculosis with healthy controls.13, 27, 31 The remaining 23 signatures had inferior AUROCs.

her signatures had equivalent AUROCs. These were Kaforou25 (86·9% [80·9–92·9]), Roe3 (86·9% [80·3–93·5]), and BATF2 (86·8% [80·6–93·1]; appendix 1, p 9) all derived from individual case-control studies comparing patients with tuberculosis with healthy controls.13, 27, 31 The remaining 23 signatures had inferior AUROCs. Test scores of the four signatures with the highest diagnostic accuracy, among all patients and stratified by HIV status, are shown in figure 2. In exploratory subgroup analyses, diagnostic accuracy of these four signatures was not affected by HIV infection (figure 3) or any other patient baseline characteristics, including age, sex, and previous tuberculosis disease (appendix 1, p 10). AUROCs tended to be numerically lower among black patients (compared with those of mixed ancestry), and numerically lower in patients with higher BMI and with tuberculosis symptom scores of less than 3, which might indicate less severe disease. None of these differences was significant for all four signatures. Additionally, there was no systematic effect of sputum smear status or haemoglobin concentration, as other markers of disease severity, on signature performance (appendix 1, p 10). Similarly, signature scores did not correlate with duration of cough, time to culture positivity, or minimum Xpert cycle threshold, as surrogate measures of bacterial load (appendix 1, p 18).Figure 2 Tuberculosis scores of the four transcriptional signatures with the highest diagnostic accuracy overall and stratified by HIV status

ignature scores did not correlate with duration of cough, time to culture positivity, or minimum Xpert cycle threshold, as surrogate measures of bacterial load (appendix 1, p 18).Figure 2 Tuberculosis scores of the four transcriptional signatures with the highest diagnostic accuracy overall and stratified by HIV status Red lines represent the score threshold of the maximal Youden index, identified from analysis of all patients. The score difference between individuals with and without tuberculosis was significant for all four signatures in both the total cohort and in HIV-stratified cohort subsets (Mann-Whitney test p<0·0001). Figure 3 ROC curves for the four transcriptional signatures with the highest diagnostic accuracy in HIV-infected versus HIV-uninfected patients Shaded areas represent the 95% CI of the ROC curve sensitivities, plotted at 1% specificity intervals (red shading represents HIV-infected patients and blue shading represents HIV-uninfected patients). AUROC values are reported with 95% CIs in brackets. p values are derived from pairwise comparison of ROC curves, using DeLong tests. AUROC values and CIs are also in appendix 1 (p 10). ROC=receiver operating characteristic. AUROC=area under the ROC curve.

d patients and blue shading represents HIV-uninfected patients). AUROC values are reported with 95% CIs in brackets. p values are derived from pairwise comparison of ROC curves, using DeLong tests. AUROC values and CIs are also in appendix 1 (p 10). ROC=receiver operating characteristic. AUROC=area under the ROC curve. Table 3 shows the sensitivity and specificity of the BATF2, Kaforou25, Roe3, and Sweeney3 signatures at the maximum Youden index of each in all 181 patients. When ROC curves of these signatures were benchmarked against the WHO TPP criteria for a tuberculosis triage test, point estimates or 95% CIs of all four signatures reached the minimum cutoffs of 90% sensitivity and 70% specificity (figure 4). Similarly, when fixing either sensitivity at 90% or specificity at 70% to enforce minimum WHO TPP diagnostic criteria, all four signatures met or approximated to the required performance thresholds (table 3). However, the optimum target criteria of 95% sensitivity and 80% specificity were beyond the 95% CI of all four signatures, either at the maximum Youden index or when fixing sensitivity or specificity at the required thresholds (figure 4, table 3).Table 3 Performance metrics of the four candidate blood transcriptional signatures with the highest diagnostic accuracy

nsitivity and 80% specificity were beyond the 95% CI of all four signatures, either at the maximum Youden index or when fixing sensitivity or specificity at the required thresholds (figure 4, table 3).Table 3 Performance metrics of the four candidate blood transcriptional signatures with the highest diagnostic accuracy Sensitivity Specificity PPV NPV At maximum Youden index BATF2 87·0% (75·6–93·6) 79·5% (71·7–85·6) 64·4% (52·9–74·4) 93·5% (87·2–96·8) Kaforou25 74·1% (61·1–83·9) 89·8% (83·3–93·9) 75·5% (62·4–85·1) 89·1% (82·5–93·4) Roe3 90·7% (80·1–96·0) 74·0% (65·8–80·9) 59·8% (48·9–69·7) 94·9% (88·7–97·8) Sweeney3 87·0% (75·6–93·6) 85·0% (77·8–90·2) 71·2% (59·4–80·7) 93·9% (88·0–97·0) At minimum sensitivity for a triage test BATF2 90% 59·8% (51·1–68·0) 48·8% (39·2–58·5) 93·4% (85·8–97·0) Kaforou25 .. 62·2% (53·5–70·2) 50·3% (40·5–60·1) 93·6% (86·3–97·2) Roe3 .. 74·0% (65·8–80·9) 59·6% (48·7–69·5) 94·6% (88·2–97·6) Sweeney3 .. 75·6% (67·4–82·2) 61·1% (50·1–71·0) 94·7% (88·5–97·6) At minimum specificity for a triage test BATF2 88·9% (77·8–94·8) 70% 55·7% (45·2–65·8) 93·7% (86·9–97·1) Kaforou25 83·3% (71·3–91·0) .. 54·2% (43·5–64·4) 90·8% (83·4–95·1) Roe3 90·7% (80·1–96·0) .. 56·3% (45·8–66·2) 94·7% (88·1–97·7) Sweeney3 90·7% (80·1–96·0) .. 56·3% (45·8–66·2) 94·7% (88·1–97·7) At optimum sensitivity for a triage test BATF2 95% 25·2% (18·5–33·4) 35·1% (27·8–43·1) 92·2% (78·6–97·5) Kaforou25 .. 28·3% (21·2–36·7) 36·1% (28·6–44·2) 93·0% (80·6–97·7) Roe3 .. 13·4% (8·5–20·4) 31·8% (25·1–39·3) 86·3% (65·3–95·5) Sweeney3 .. 54·3% (45·7–62·7) 46·9% (37·8–56·2) 96·2% (89·0–98·8) At optimum specificity for a triage test BATF2 85·2% (73·4–92·3) 80% 64·4% (52·8–74·5) 92·7% (86·3–96·3) Kaforou25 81·5% (69·2–89·6) .. 63·4% (51·6–73·8) 91·0% (84·3–95·1) Roe3 79·6% (67·1–88·2) .. 62·9% (51·0–73·3) 90·2% (83·4–94·5) Sweeney3 88·9% (77·8–94·8) .. 65·4% (54·0–75·3) 94·4% (88·4–97·4) At minimum sensitivity for a confirmatory test BATF2 65% 85·8% (78·7–90·8) 66·1% (52·7–77·4) 85·2% (78·0–90·3) Kaforou25 .. 92·1% (86·1–95·7) 77·8% (63·8–87·5) 86·1% (79·3–90·9) Roe3 .. 92·1% (86·1–95·7) 77·8% (63·8–87·5) 86·1% (79·3–90·9) Sweeney3 .. 93·7% (88·1–96·8) 81·4% (67·4–90·3) 86·3% (79·6–91·1) At minimum specificity for a confirmatory test BATF2 53·7% (40·6–66·3) 98% 91·9% (77·3–97·4) 83·3% (76·5–88·4) Kaforou25 31·5% (20·7–44·7) .. 87·0% (66·0–95·8) 77·1% (70·0–82·9) Roe3 33·3% (22·2–46·6) .. 87·6% (67·4–96·1) 77·6% (70·5–83·3) Sweeney3 44·4% (32–57·6) ..

3·7% (88·1–96·8) 81·4% (67·4–90·3) 86·3% (79·6–91·1) At minimum specificity for a confirmatory test BATF2 53·7% (40·6–66·3) 98% 91·9% (77·3–97·4) 83·3% (76·5–88·4) Kaforou25 31·5% (20·7–44·7) .. 87·0% (66·0–95·8) 77·1% (70·0–82·9) Roe3 33·3% (22·2–46·6) .. 87·6% (67·4–96·1) 77·6% (70·5–83·3) Sweeney3 44·4% (32–57·6) .. 90·4% (73·7–97·0) 80·6% (73·6–86·0) Data are % (95% CI). WHO defines target product profile criteria for a tuberculosis triage test as minimum 90% sensitivity and 70% specificity, optimum 95% sensitivity and 80% specificity, and for a confirmatory test as minimum 98% specificity and 65% sensitivity. PPV=positive predictive value. NPV=negative predictive value. Figure 4 ROC curves of the four transcriptional signatures with the highest diagnostic accuracy benchmarked against WHO target product profile criteria

90·4% (73·7–97·0) 80·6% (73·6–86·0) Data are % (95% CI). WHO defines target product profile criteria for a tuberculosis triage test as minimum 90% sensitivity and 70% specificity, optimum 95% sensitivity and 80% specificity, and for a confirmatory test as minimum 98% specificity and 65% sensitivity. PPV=positive predictive value. NPV=negative predictive value. Figure 4 ROC curves of the four transcriptional signatures with the highest diagnostic accuracy benchmarked against WHO target product profile criteria (A) Blue shaded areas represent the 95% CIs of the ROC curve sensitivities, plotted at 1% specificity intervals. AUROC values are reported with 95% CIs in brackets. (B) ROC curves are replicated with restricted y axes, and benchmarked against target criteria for a tuberculosis triage test. Minimum criteria (90% sensitivity, 70% specificity) are indicated by the dashed black boxes, optimum criteria (95% sensitivity, 80% specificity) are indicated by the blue boxes. Light blue shaded areas represent the 95% CIs. (C) ROC curves are replicated with restricted x axes and benchmarked against minimum criteria for a confirmatory test (dashed black box: 65% sensitivity, 98% specificity). Light blue shaded areas represent the 95% CIs. ROC=receiver operating characteristic. AUROC=area under the ROC curve.

d areas represent the 95% CIs. (C) ROC curves are replicated with restricted x axes and benchmarked against minimum criteria for a confirmatory test (dashed black box: 65% sensitivity, 98% specificity). Light blue shaded areas represent the 95% CIs. ROC=receiver operating characteristic. AUROC=area under the ROC curve. As a secondary objective, we assessed signature performance as a blood-based confirmatory tuberculosis test, using WHO TPP criteria as a reference (figure 4).2 At the maximum Youden index, all four signatures with the highest diagnostic accuracy failed to reach the required 98% specificity (table 3). Similarly, when setting the test thresholds to enforce either 98% specificity or 65% sensitivity, these four signatures were substantially short of the minimum performance requirements (table 3).

uden index, all four signatures with the highest diagnostic accuracy failed to reach the required 98% specificity (table 3). Similarly, when setting the test thresholds to enforce either 98% specificity or 65% sensitivity, these four signatures were substantially short of the minimum performance requirements (table 3). In view of emerging concerns that the higher sensitivity of the Ultra test might be compromised by false-positive results,7, 8, 10 we also assessed the potential use of blood signatures as an add-on confirmatory test for Ultra-positive patients. Of 51 patients with Ultra-positive results in our cohort, ten (20%) were designated as false-positive by comparison with our standard reference (ie, these individuals were culture-negative and Xpert-negative at enrolment). Six (60%) of the ten Ultra false-positive patients had trace-positive results. Previous tuberculosis disease was more common in patients with Ultra false-positive results compared with patients with Ultra true-positive results (seven [70%] of ten vs 14 [34%] of 41; χ2 test p=0·039; figure 5).Figure 5 Tuberculosis scores of the four transcriptional signatures with the highest diagnostic accuracy in patients with Ultra-positive results

common in patients with Ultra false-positive results compared with patients with Ultra true-positive results (seven [70%] of ten vs 14 [34%] of 41; χ2 test p=0·039; figure 5).Figure 5 Tuberculosis scores of the four transcriptional signatures with the highest diagnostic accuracy in patients with Ultra-positive results Patients with Ultra-positive results were grouped as true-positive tuberculosis (culture-positive or Xpert-positive) and false-positive non-tuberculosis (culture-negative or Xpert-negative). (A) Pie charts representing the proportion of Ultra-positive patients with tuberculosis and individuals without tuberculosis with a history of previous tuberculosis disease. (B) Scores of the four transcriptional signatures with the highest diagnostic accuracy in patients with Ultra-positive results. Red dots indicate patients for whom only traces of Mycobacterium tuberculosis were detected by Ultra analysis. Dashed lines represent the score thresholds of the maximum Youden index, identified from analysis of all patients. The score difference between patients with and without tuberculosis was significant for all four signatures (Mann-Whitney test p<0·0001).

aces of Mycobacterium tuberculosis were detected by Ultra analysis. Dashed lines represent the score thresholds of the maximum Youden index, identified from analysis of all patients. The score difference between patients with and without tuberculosis was significant for all four signatures (Mann-Whitney test p<0·0001). Nine of the ten Ultra false-positive patients scored consistently below the Youden index threshold of all four transcriptional signatures with the highest diagnostic accuracy, correctly classifying them as non-tuberculosis (figure 5). This also included five of the six Ultra trace false-positives. However, two to eight (5–20%) of the 41 true-positive Ultra patients were incorrectly classified as non-tuberculosis at the Youden index threshold of each signature, consistent with the imperfect sensitivity of the transcriptional signatures. A diagnostic algorithm that used the blood transcriptional signature results to re-classify all Ultra-positive patients, or only those with trace results, or those with previous tuberculosis, led to improved specificity compared with Ultra analysis alone, with small associated reductions in sensitivity (table 4). Of note, follow-up of cases that were Ultra-positive but culture-negative in the parent study revealed that three of the ten cases that we designated as Ultra false-positives were diagnosed with tuberculosis at intervals of 295, 432, and 777 days.10Table 4 Sensitivity and specificity of a diagnostic algorithm combining the sputum Ultra test with blood transcriptional signature analysis

ative in the parent study revealed that three of the ten cases that we designated as Ultra false-positives were diagnosed with tuberculosis at intervals of 295, 432, and 777 days.10Table 4 Sensitivity and specificity of a diagnostic algorithm combining the sputum Ultra test with blood transcriptional signature analysis Ultra test alone Addition of signature to: All Ultra-positive individuals Ultra trace-positive individuals Ultra-positive individuals with previous tuberculosis Sensitivity BATF2 82% (69–90) 76% (63–86) 82% (69–90) 80% (67–89) Kaforou25 .. 66% (52–78); p=0·013 82% (69–90) 74% (60–84) Roe3 .. 78% (65–87) 82% (69–90) 80% (67–89) Sweeney3 .. 76% (63–86) 80% (67–89) 78% (65–87) Specificity BATF2 90% (83–95) 99% (95–100); p=0·0077 95% (89–98) 96% (91–98); p=0·041 Kaforou25 .. 100% (96–100); p=0·0044 96% (91–98); p=0·041 97% (92–99); p=0·023 Roe3 .. 99% (95–100); p=0·0077 95% (89–98) 96% (91–98); p=0·041 Sweeney3 .. 100% (96–100); p=0·0044 96% (91–98); p=0·041 97% (92–99); p=0·023 Data are % (95% CI). Only significant p values are shown. Sensitivity and specificity were calculated for 154 patients with Ultra results, with or without reclassification of selected Ultra-positive tests by transcriptional signatures. p value of comparison with Ultra alone using McNemar's test.

41 97% (92–99); p=0·023 Data are % (95% CI). Only significant p values are shown. Sensitivity and specificity were calculated for 154 patients with Ultra results, with or without reclassification of selected Ultra-positive tests by transcriptional signatures. p value of comparison with Ultra alone using McNemar's test. Restricting the tuberculosis case definition to culture-proven patients led to re-assignment of only one culture-negative, Xpert-positive patient as without tuberculosis. Data reanalysis confirmed the finding that the four signatures performed equivalently, independent of HIV status, while meeting or approximating the minimum criteria for a tuberculosis triage, but not confirmatory, test (appendix 1, pp 19–20). The possibility that some patients might have been diagnosed with tuberculosis after enrolment to our study and the absence of multiple sputum cultures might have led to an underestimation of the specificity of the transcriptional signatures. To overcome this limitation, we sought to estimate the best possible specificity that the signatures could achieve if the sensitivity of the standard reference was increased by additional sputum cultures. We reclassified signature false-positive cases (at the Youden index threshold) to true-positive cases by the ratio of the sensitivity expected from four sputum cultures to that of a single culture.25 Even in this analysis, the four signatures with the highest diag-nostic accuracy failed to achieve optimum criteria for a triage test, and minimum criteria for a confirmatory test (appendix 1, p 11). Finally, we repeated our analysis after batch correction with ComBat instead of surrogate variable analysis. Again, our main findings were unchanged, confirming the robustness of our results (appendix 1, pp 21–22).

ieve optimum criteria for a triage test, and minimum criteria for a confirmatory test (appendix 1, p 11). Finally, we repeated our analysis after batch correction with ComBat instead of surrogate variable analysis. Again, our main findings were unchanged, confirming the robustness of our results (appendix 1, pp 21–22). Discussion To our knowledge, this is the first comprehensive head-to-head analysis of candidate blood transcriptional biomarkers of tuberculosis in a prospective validation cohort with a high burden of tuberculosis and HIV. Four signatures (comprising 1–25 genes) had equivalent diagnostic accuracy for differentiating patients with and without tuberculosis, irrespective of HIV status. These signatures met or approximated to the minimum WHO TPP criteria of 90% sensitivity and 70% specificity for a triage test to rule out tuberculosis, but failed to reach the optimum criteria (95% sensitivity and 80% specificity), and at a test threshold that offers the maximum diagnostic accuracy, they missed 9–26% of tuberculosis cases (ie, five to 14 of 54 patients with tuberculosis).

90% sensitivity and 70% specificity for a triage test to rule out tuberculosis, but failed to reach the optimum criteria (95% sensitivity and 80% specificity), and at a test threshold that offers the maximum diagnostic accuracy, they missed 9–26% of tuberculosis cases (ie, five to 14 of 54 patients with tuberculosis). To date, no transcriptional signature has been translated into a point-of-care test, which would require the adaptation and validation of these tests as PCR-based assays. Such studies are underway;17, 29 however, the cost is likely to exceed the target threshold of $2 per sample.2 Taken together with the suboptimal clinical performance observed in our study, the question is raised of whether host transcriptional biomarkers represent a realistic and achievable triage strategy for the resource-limited settings where they are most needed. Of note, the diagnostic accuracy of the best transcriptional biomarkers in the current analysis was similar to that of point-of-care C-reactive protein (CRP) alone for active case-finding among HIV-infected individuals.42 Since CRP testing is likely to be substantially cheaper, prospective assessments of the superiority of transcriptional biomarkers above this benchmark are required if they are to be pursued for this application. We also tested whether transcriptional biomarkers could be used as blood-based confirmatory tests for tuberculosis, for which the WHO-specified maximum target price is higher. However, the transcriptional signatures with the highest diagnostic accuracy in our study had insufficient specificity, making them non-viable for confirmatory tuberculosis diagnostics. A principal advantage of these signatures is the easy accessibility of blood sampling. However, alternative microbiological tests for tuberculosis using non-sputum samples are being developed,43, 44 which might offer greater promise among patient subgroups where obtaining sputum is difficult.

tuberculosis diagnostics. A principal advantage of these signatures is the easy accessibility of blood sampling. However, alternative microbiological tests for tuberculosis using non-sputum samples are being developed,43, 44 which might offer greater promise among patient subgroups where obtaining sputum is difficult. In the current cohort, ten patients had false-positive Ultra results, including six with false-positive Ultra trace results. This finding permitted exploration of alternative clinical applications of host transcriptional signatures. The four signatures with the highest diagnostic accuracy in our study showed promise in correctly classifying Ultra false-positive patients, including those with trace results. Our preliminary results suggest that a diagnostic algorithm combining Ultra sputum analysis with blood transcriptional biomarkers improves Ultra specificity. Large-scale prospective validation studies are required to further assess this potential application, particularly among individuals suspected to be false-positives, such as those with trace results or a history of tuberculosis disease. Of note, Ultra false-positive results have been attributed to non-viable mycobacterial remnants,7, 8, 10 but the fact that three individuals with Ultra false-positive results were diagnosed with tuberculosis after 295–777 days' follow-up raises the possibility that some false-positive results might represent detection of very early paucibacillary or latent infection, undetected by Xpert or culture. In a high-burden setting, we cannot exclude the possibility that these cases were due to acquisition of infection after enrolment. Therefore, whether Ultra-positive results in the absence of prevalent disease predict future incident disease, can only be addressed by randomised trials to test whether tuberculosis treatment in this group will reduce incident disease.

possibility that these cases were due to acquisition of infection after enrolment. Therefore, whether Ultra-positive results in the absence of prevalent disease predict future incident disease, can only be addressed by randomised trials to test whether tuberculosis treatment in this group will reduce incident disease. Among the best-performing signatures, BATF2, Kaforou25, and Roe3 were originally discovered by comparing patients with active tuberculosis with asymptomatic individuals.13, 27, 31 Nonetheless, their performance in this observational cohort of almost exclusively symptomatic patients suggests that these signatures can discriminate between tuberculosis and the casemix of other symptomatic illness in this context. Assessing the extent to which these findings are generalisable will require similar observational studies in settings that might have a different casemix. Additionally, whether existing signatures have reached the maximum possible diagnostic accuracy using blood transcriptomics, or whether novel signatures, derived on even larger discovery datasets, might lead to further improvements in diagnostic accuracy, remains to be seen. Likewise, whether integration of clinical metadata with biomarkers will generate models with greater diagnostic accuracy also needs to be tested using independent training and validation cohorts.

s, derived on even larger discovery datasets, might lead to further improvements in diagnostic accuracy, remains to be seen. Likewise, whether integration of clinical metadata with biomarkers will generate models with greater diagnostic accuracy also needs to be tested using independent training and validation cohorts. Within the limitations of the statistical power in our cohort, signature performance was independent of age, sex, HIV coinfection, or previous tuberculosis disease, and preserved in subgroup analyses of patients stratified by sputum smear status or haemoglobin concentrations as surrogate measures of disease severity. The point estimates for test performance among black patients, and patients with higher BMI and lower tuberculosis symptom scores were lower, but our study had insufficient power to assess the significance of these observations for all four signatures with the highest diagnostic accuracy.

measures of disease severity. The point estimates for test performance among black patients, and patients with higher BMI and lower tuberculosis symptom scores were lower, but our study had insufficient power to assess the significance of these observations for all four signatures with the highest diagnostic accuracy. An important strength of our study was the clinically relevant, real-life population of patients who were evaluated for tuberculosis in a high-burden setting, with both HIV-infected and HIV-uninfected individuals, and patients with varying severity of tuberculosis disease. We induced sputum, ensuring that we did not include only patients who could expectorate, for whom there is less need for non-sputum tests. Importantly, the non-tuberculosis group was not pre-selected to be homo-genous, thus likely encompassing a casemix of people with latent tuberculosis infection and other diseases. Second, we used a robust standard of culture or Xpert positivity as a diagnostic reference for our primary analysis, and confirmed that the most optimistic estimates of additional cultures in the standard reference would not significantly improve signature performance. Third, we did a systematic review to identify 27 candidate transcriptional signatures for tuberculosis to undertake a comprehensive head-to-head analysis. Finally, our dataset was exclusively used for validation rather than discovery, making it a truly independent diagnostic accuracy study.

prove signature performance. Third, we did a systematic review to identify 27 candidate transcriptional signatures for tuberculosis to undertake a comprehensive head-to-head analysis. Finally, our dataset was exclusively used for validation rather than discovery, making it a truly independent diagnostic accuracy study. A limitation of our study was the observed batch effect in RNA sequencing data, which appeared to result from a mixture of technical batch factors. We addressed this effect with two different data adjustment approaches, and found in both analyses that the same four signatures performed equivalently, irrespective of HIV status, and met or approximated the minimum criteria for a tuberculosis triage but not a confirmatory test. A second limitation of our study was that our cohort was restricted to adults with possible pulmonary tuberculosis. Similar independent validation studies are needed for children and patients with extrapulmonary tuberculosis. Since inclusion criteria for our systematic review were not limited by age or site of disease, the 27 candidate signatures identified could be tested in such a study. Third, no alternative diagnoses were available for patients without tuberculosis; thus, we were not able to establish whether false-positive results were related to particular non-tuberculosis diseases. Finally, this study was limited to transcriptional biomarkers. Prospective head-to-head studies comparing performance of transcriptional signatures with other candidate triage test biomarkers, such as point-of-care CRP42 and automated chest radiograph interpretation tools,45 or with strategies that integrate biomarkers with clinical metadata, are needed.

ranscriptional biomarkers. Prospective head-to-head studies comparing performance of transcriptional signatures with other candidate triage test biomarkers, such as point-of-care CRP42 and automated chest radiograph interpretation tools,45 or with strategies that integrate biomarkers with clinical metadata, are needed. In conclusion, we showed that four blood transcriptional signatures have equivalent diagnostic accuracy for active tuberculosis, independent of HIV status. These biomarkers achieved the WHO minimum diagnostic accuracy parameters required for a tuberculosis triage test but failed to meet the criteria for a confirmatory test in the present setting. Notwithstanding the challenge of achieving the desired target price for such tests, further validation studies are needed to assess their application in different settings alongside head-to-head comparisons with other candidate triage biomarkers, with a view to interventional trials to assess their clinical and health economic effects. Supplementary Materials Supplementary appendix 1 Supplementary appendix 2 Supplementary appendix 3

In conclusion, we showed that four blood transcriptional signatures have equivalent diagnostic accuracy for active tuberculosis, independent of HIV status. These biomarkers achieved the WHO minimum diagnostic accuracy parameters required for a tuberculosis triage test but failed to meet the criteria for a confirmatory test in the present setting. Notwithstanding the challenge of achieving the desired target price for such tests, further validation studies are needed to assess their application in different settings alongside head-to-head comparisons with other candidate triage biomarkers, with a view to interventional trials to assess their clinical and health economic effects. Supplementary Materials Supplementary appendix 1 Supplementary appendix 2 Supplementary appendix 3 Acknowledgments This study was funded by a Medical Research Council Fellowship (MR/L001756/1 to JKR), a Royal Society Newton Advanced fellowship (NA-150-202 to GT), the Wellcome Trust (207511/Z/17/Z to MN), and by National Institute of Health Research (NIHR) Biomedical Research Funding to UCL and UCLH. Additionally, Cepheid gave in-kind cartridges and equipment donations to the clinical study (BAR-TB) from which the specimens in this manuscript were derived. This paper presents independent research supported by the NIHR. The views expressed are those of the author(s) and not necessarily those of the UK National Health Service, the NIHR or the UK Department of Health and Social Care.

ns to the clinical study (BAR-TB) from which the specimens in this manuscript were derived. This paper presents independent research supported by the NIHR. The views expressed are those of the author(s) and not necessarily those of the UK National Health Service, the NIHR or the UK Department of Health and Social Care. Contributors GT and MN conceived the study. ZP and BWPR were responsible for sample and metadata collection. JKR, PM, GRN and ET processed the samples. CTT and RKG analysed the data with input from RFM, GT, and MN. CTT, RKG, and MN wrote the manuscript with input from all other authors. Declaration of interests RFM reports personal fees from Gilead. MN reports grants from the Wellcome Trust and the UK National Institute for Health Research (NIHR). BWPR reports non-financial support from Cepheid. RKG reports grants from NIHR. JKR reports grants from the UK Medical Research Council. GT reports grants from the Royal Society Newton Advanced fellowship. JKR and MN have a UK patent application pending (1603367.2) in relation to blood transcriptomic biomarkers of tuberculosis. GRN, ZP, ET, CTT, and PM declare no competing interests.

Introduction Cystic fibrosis is the most common life-limiting inherited disease in white populations, with most patients dying prematurely from respiratory failure. Children with cystic fibrosis in the UK and in other high-income countries are usually diagnosed in the first year of their life,1 and subsequently need intensive support from family and health-care services.

limiting inherited disease in white populations, with most patients dying prematurely from respiratory failure. Children with cystic fibrosis in the UK and in other high-income countries are usually diagnosed in the first year of their life,1 and subsequently need intensive support from family and health-care services. Cystic fibrosis is of particular interest in the study of health inequalities, because it is a genetic disease and there is no social gradient in incidence of the disorder—it affects all socioeconomic groups equally (appendix). Inequalities can develop, however, in the outcomes experienced by people with the disease. People with cystic fibrosis from socioeconomically disadvantaged backgrounds, for example, die younger than do those in more advantaged social positions in the UK and the USA.2–5 Between 1986 to 1994, the adjusted risk of death was 3·65 times higher in patients with cystic fibrosis in the USA with Medicaid cover (taken as an indicator of poverty) than it was in those without Medicaid cover.2 In England and Wales, between 1959 and 2008, Barr and Fogarty recorded an increased risk of dying later, at an age above the median of all deaths due to cystic fibrosis, in more advantaged social classes, a pattern that has persisted for more than four decades.5 As with other chronic diseases, this social patterning of survival in cystic fibrosis implies that social and environmental factors affect outcomes.6,7 Inequalities in access to specialist health care might also be important, because in many health-care systems provision and use of services decreases with patients' income,8,9 the so-called inverse care law.10

al patterning of survival in cystic fibrosis implies that social and environmental factors affect outcomes.6,7 Inequalities in access to specialist health care might also be important, because in many health-care systems provision and use of services decreases with patients' income,8,9 the so-called inverse care law.10 To gain a better understanding of when and how inequalities in outcomes develop in cystic fibrosis, we undertook a longitudinal registry study to explore the effect of deprivation on growth, nutrition, lung function, risk of Pseudomonas aeruginosa colonisation, and the use of major cystic fibrosis treatment modalities in a UK-wide population cohort, in the context of a universal health-care system, free at the point of use. Methods Study design and data sources We undertook a longitudinal retrospective cohort study of individuals in the UK cystic fibrosis registry who were younger than 40 years at last follow-up, with at least one outcome measurement and a valid postcode between Jan 1, 1996, and Dec 31, 2009. The UK cystic fibrosis registry is supported and coordinated by the UK Cystic Fibrosis Trust.11,12 The UK cystic fibrosis registry is maintained to a high standard of data quality, and is estimated to include nearly all people thought to have cystic fibrosis in the UK population13 and is therefore ideally suited to the study of outcomes and treatments across the whole socioeconomic spectrum in the UK (appendix).

,12 The UK cystic fibrosis registry is maintained to a high standard of data quality, and is estimated to include nearly all people thought to have cystic fibrosis in the UK population13 and is therefore ideally suited to the study of outcomes and treatments across the whole socioeconomic spectrum in the UK (appendix). NHS research ethics approval (Huntingdon Research Ethics Committee 07/Q0104/2) was granted for the collection of data into the UK database. The Cystic Fibrosis Trust database committee approved the use of anonymised data in this study. Primary outcome and covariates The primary clinical outcomes were weight, height, body-mass index (BMI), percent predicted forced expiratory volume in 1 s (%FEV1), and prevalence of P aeruginosa colonisation. Anthropometric values were converted into standard deviation [SD] scores using the UK reference population.14 The primary health-care outcomes were use of treatments in the previous year (yes or no): intravenous antibiotics, supplemental nutritional support, DNase, or inhaled antibiotic treatment. Conditional on the use of intravenous treatment, we also used the log total number of days on intravenous treatment as a secondary outcome.

ary health-care outcomes were use of treatments in the previous year (yes or no): intravenous antibiotics, supplemental nutritional support, DNase, or inhaled antibiotic treatment. Conditional on the use of intravenous treatment, we also used the log total number of days on intravenous treatment as a secondary outcome. The primary exposure measure was a small-area-based measure of deprivation of area of residence. Postcodes were used to derive Index of Multiple Deprivation scores for the constituent UK countries15 and each person was allocated to a deprivation score on the basis of the first recorded postcode on entry to the dataset. Other baseline covariates in the analysis were: sex, genotype coded as the number of delta F508 alleles (0, 1, or 2), year of birth, screening status (diagnosis by neonatal screening or otherwise), and ethnic origin (white or other). Time-varying covariates were age, presence of cystic fibrosis related diabetes (CFRD), and presence of pancreatic insufficiency (ie, whether or not an individual used pancreatic enzyme supplementation). In our health-care use analyses, we adjusted for disease severity on the basis of current %FEV1, P aeruginosa status, and BMI SD score.

iates were age, presence of cystic fibrosis related diabetes (CFRD), and presence of pancreatic insufficiency (ie, whether or not an individual used pancreatic enzyme supplementation). In our health-care use analyses, we adjusted for disease severity on the basis of current %FEV1, P aeruginosa status, and BMI SD score. Statistical analysis Full details are provided in the supplementary appendix. Briefly, we fitted separate longitudinal models in the paediatric (<18 years) and adult (18–40 years) age ranges. We then approximated time-trends using linear functions (eg, for %FEV1), piecewise or broken-stick functions (weight, BMI), or quadratics (eg, any intravenous treatment), as appropriate. For instance, population weight SD score increased to about age 3 years, and then decreased subsequently (appendix). The modelling approach involved first fitting a model adjusted for age and the baseline covariates defined above, and then testing for the significance of adding deprivation. Finally, the time-varying covariates were added to the model, to assess whether the deprivation coefficient was modified. We estimated all model parameters by maximum likelihood, using linear or generalised linear mixed effects models.16 We used generalised likelihood ratio statistics to compare nested models, and Wald statistics to test hypotheses about model parameters. We used R (version 2.9.2) for all statistical analyses.

s modified. We estimated all model parameters by maximum likelihood, using linear or generalised linear mixed effects models.16 We used generalised likelihood ratio statistics to compare nested models, and Wald statistics to test hypotheses about model parameters. We used R (version 2.9.2) for all statistical analyses. Role of the funding source The study sponsor had no role in the design, collection, analysis, or interpretation of the data, in the writing of the report, or in the decision to submit the article for publication. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

y sponsor had no role in the design, collection, analysis, or interpretation of the data, in the writing of the report, or in the decision to submit the article for publication. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results The final dataset for weight SD scores, the most commonly collected outcome, contained information collected at 49 337 annual reviews for 8055 patients between Jan 1, 1996 and Dec 31, 2009, in the UK (table 1 and appendix). 5324 (66%) individuals had five or more follow-up measures (appendix), with a total of 48 425 person-years of follow-up. We recorded no relation between sex ratios, birth cohort, neonatal screening and deprivation status (table 1), or number of incident cases, and age at diagnosis and deprivation status (appendix). We recorded a slight trend towards fewer heterozygote delta F508 carriers (p=0·0022), more people with no delta F508 genes (p<0·0001), and a greater proportion of non-white patients with increasing level of deprivation (p<0·0001). Compared with the UK reference population, the population of patients with cystic fibrosis weighed less (SD score −0·37, 95% CI −0·43 to −0·35 [35th centile]), were shorter (–0·50, −0·53 to −0·47 [30th centile]), and had a lower BMI (–0·08, −0·11 to −0·06 [46th centile]; in models ignoring time trends).

ion (p<0·0001). Compared with the UK reference population, the population of patients with cystic fibrosis weighed less (SD score −0·37, 95% CI −0·43 to −0·35 [35th centile]), were shorter (–0·50, −0·53 to −0·47 [30th centile]), and had a lower BMI (–0·08, −0·11 to −0·06 [46th centile]; in models ignoring time trends). Weight SD scores increased from diagnosis up to about the age of 3 years, decreasing thereafter (appendix). After adjustment for baseline factors, at diagnosis, the weight of children in the most deprived quintile was lower than that of children in the least deprived quintile (weight SD score −0·54, 95% CI −0·73 to −0·34). The deprivation gap diminished with increasing age up to age 3 years, and from then on remained constant (table 2 and appendix). A higher weight SD score was associated with male sex, screened patients, heterozygotes for delta F508, and white patients (appendix). In adults, adjusted weight-for-age was lower in more deprived groups (table 2). The average height of individuals in the most deprived quintile compared with the least deprived quintile was also about a third of an SD score shorter in the adjusted analysis, a difference that remained constant across all ages (table 2 and appendix). In patients younger than 18 years, a bigger height SD score was statistically significantly associated with male sex and screened patients, and statistically significantly increased in white patients with age (appendix).

the adjusted analysis, a difference that remained constant across all ages (table 2 and appendix). In patients younger than 18 years, a bigger height SD score was statistically significantly associated with male sex and screened patients, and statistically significantly increased in white patients with age (appendix). We modelled BMI SD score much like we modelled weight SD score, with a split-line at age 3 years. In the paediatric age range, there was a deprivation gap (with lower scores in the most derived groups) of −0·13 (–0·22 to −0·04; table 2). Higher BMI was associated with male sex in the paediatric age range (individuals ages 0–18 years), and had a steeper rate of decline in delta F508 homozygotes after the age of 3 years (appendix). In the adult age range, we recorded no association between BMI SD score and deprivation status (–0·12, −0·25 to 0·01; figure 1 and table 2). Addition of the time-varying covariates did not substantially alter the deprivation effects for growth outcomes, and the estimates were consistent with a monotonic dose-response relation between deprivation and both weight and height.

We modelled BMI SD score much like we modelled weight SD score, with a split-line at age 3 years. In the paediatric age range, there was a deprivation gap (with lower scores in the most derived groups) of −0·13 (–0·22 to −0·04; table 2). Higher BMI was associated with male sex in the paediatric age range (individuals ages 0–18 years), and had a steeper rate of decline in delta F508 homozygotes after the age of 3 years (appendix). In the adult age range, we recorded no association between BMI SD score and deprivation status (–0·12, −0·25 to 0·01; figure 1 and table 2). Addition of the time-varying covariates did not substantially alter the deprivation effects for growth outcomes, and the estimates were consistent with a monotonic dose-response relation between deprivation and both weight and height. In the final model for %FEV1, we detected a difference of −4·1 percentage points (–5·0 to −3·1) when comparing children (<18 years) in the most deprived quintile with those in the least deprived quintile (a difference that was apparent from as soon as %FEV1 can be measured at about 5 years of age), but there was no evidence of an increased rate of decline in children from more deprived quintiles. Higher %FEV1 was associated with male sex, screened patients, heterozygote delta F508 status, white patients, no CFRD, no P aeruginosa colonisation, and higher BMI (figure 2 and appendix). Further adjustment for Burkholderia cenocepacia status and care centre did not change the deprivation effect on %FEV1 (appendix). The addition of BMI SD score to the model reduced the %FEV1 deprivation gap to −3·5 percentage points (–5·2 to −1·8). There was no statistically significant association between %FEV1 and social deprivation in the adult age range (table 2). The cross-sectional proportion of people with chronic P aeruginosa infection increased steadily with age to about 60% by the age of 20 years, and was more common in the most deprived quintile, with an odds ratio (OR) of 1·9 (95% CI 1·3 to 2·7) in the adjusted paediatric analysis for the most deprived quintile (table 2 and figure 2). An increased likelihood of P aeruginosa colonisation was associated with female sex, homozygote delta F508 status, CFRD, pancreatic insufficiency, and lower %FEV1, but adjustment for these factors did not substantially alter the deprivation effect (data not shown). The estimates were consistent with a monotonic dose-response relation between deprivation and %FEV1 (appendix) and risk of P aeruginosa colonisation (data not shown).

s, CFRD, pancreatic insufficiency, and lower %FEV1, but adjustment for these factors did not substantially alter the deprivation effect (data not shown). The estimates were consistent with a monotonic dose-response relation between deprivation and %FEV1 (appendix) and risk of P aeruginosa colonisation (data not shown). The use of any intravenous treatment, after adjustment for disease severity, was more than twice as common in the most deprived children cohort compared with the least deprived children cohort (table 2), and this deprivation difference was also present in adults (table 2 and figure 3). Further adjustment for care centre did not change this effect. Conditional on receipt of intravenous treatment, and after adjustment for disease severity, people in the most deprived quintile had more days of intravenous treatment in both the paediatric and adult age range (table 2). We analysed the receipt of hospital and at-home intravenous treatment separately and noted that the higher prevalence of any intravenous treatment seen in the most deprived quintile was almost entirely due to delivery of such treatment in hospital rather than home (figure 3). For intravenous treatment at home, the association with social deprivation was much less strong, and, in the cross-sectional analysis, at-home intravenous treatment was more common in the least deprived quintile compared with the most deprived quintile in patients between the ages of 10 years and 27 years (figure 3). Prevalence of any supplemental feeding therapy in the previous year was more common in the most deprived quintile, compared to the least, across the entire age range from age 0 years to age 40 years (OR 1·78, 95% CI 1·42 to 2·2, adjusted for baseline variables, P aeruginosa infection status, and BMI, in the 5–18 age group, figure 3).

ny supplemental feeding therapy in the previous year was more common in the most deprived quintile, compared to the least, across the entire age range from age 0 years to age 40 years (OR 1·78, 95% CI 1·42 to 2·2, adjusted for baseline variables, P aeruginosa infection status, and BMI, in the 5–18 age group, figure 3). We detected no statistically significant association between DNase use and deprivation in the paediatric age range before we adjusted for disease severity. After adjustment for disease severity, treatment was less likely in the most deprived quintile, in both children and adults, although the association with deprivation was stronger in adults. We saw a similar pattern for inhaled antibiotic treatment (table 2 and figure 3). The estimates were consistent with a monotonic dose-response relation between socioeconomic status and treatment outcomes (appendix). Discussion Our findings show that children with cystic fibrosis from the most disadvantaged areas in the UK have lower weight, height, and BMI in the first years of life after diagnosis, are more likely to have chronic P aeruginosa infection, and have a lower %FEV1 than do children in the least disadvantaged areas. These social inequalities persist into adulthood but do not widen.

from the most disadvantaged areas in the UK have lower weight, height, and BMI in the first years of life after diagnosis, are more likely to have chronic P aeruginosa infection, and have a lower %FEV1 than do children in the least disadvantaged areas. These social inequalities persist into adulthood but do not widen. Our findings suggest evidence of positive discrimination, or so-called pro-poor bias, in the provision of some key treatments, on the basis of socioeconomic circumstances. We show that in the NHS, compared with children with cystic fibrosis in the least disadvantaged areas, children with cystic fibrosis from the most disadvantaged areas are about twice as likely, after adjustment for disease severity, to receive intravenous antibiotics (specifically in hospital) and nutritional support. Our findings also show some apparent bias in favour of wealthier populations, a so-called pro-rich bias, in two other treatments, DNase and inhaled antibiotics, with patients from the most affluent areas being more likely to receive these treatments after adjustment for disease severity.

pital) and nutritional support. Our findings also show some apparent bias in favour of wealthier populations, a so-called pro-rich bias, in two other treatments, DNase and inhaled antibiotics, with patients from the most affluent areas being more likely to receive these treatments after adjustment for disease severity. Key strengths of this study include the population-wide coverage of the UK cystic fibrosis registry, the high quality of the data, and the longitudinal analysis. However, our study does have limitations. First, it relies on retrospective, routinely collected data and we used a standard measure of deprivation of area of residence. Each small area contains about 1500 people, and, in this respect, the Index of Multiple Deprivation scores allowed much finer resolution than US analyses3,17,18 that have used ZIP-code-linked income data, because every ZIP code contains about 30 000 people (panel).19 There is always the possibility of ecological fallacy (whereby inferences made at the group level do not apply to the individual), but this possibility is unlikely in view of the fact that similar associations have been seen in the US studies that use both area and individual measures of socioeconomic status.3,17,18 Second, we had valid postcodes for only 90% of the sample, although our sample size was large, with no pronounced gradient in the proportion of patients by deprivation quintile. The excluded population—those with no valid postcode—were largely older birth cohorts, owing to the improved collection of postcodes by clinical staff over time, but we do not believe that this has biased the associations detected in our analysis (appendix). Third, there is a strong cohort effect in cystic fibrosis and, with datasets of this type, age and cohort effects confound one another, and cannot be completely separated.20 We have adjusted for both in our analysis, to estimate the adjusted effect on deprivation.

d the associations detected in our analysis (appendix). Third, there is a strong cohort effect in cystic fibrosis and, with datasets of this type, age and cohort effects confound one another, and cannot be completely separated.20 We have adjusted for both in our analysis, to estimate the adjusted effect on deprivation. Overall, the UK cystic fibrosis population is underweight and shorter compared with the UK reference population, by about a third of an SD score. Deprivation roughly doubles this effect, lowering the SD score by another third. How much of the effect of socioeconomic status on growth outcomes is specific to cystic fibrosis, and how much is attributable to socioeconomic status effects in the general population is unclear. Comparable data in contemporary representative cohorts in the UK is absent, but the age-related changes in growth in the general population are characterised by increasing obesity in childhood from the age of 4 years onwards, with higher BMI in the more deprived populations,21,22 findings which contrast with the patterns seen in our study. The projected weight difference at intercept in our study (–0·54) by socioeconomic status is also larger than those in other recent studies,23,24 but direct comparison between these studies and ours is complicated by the use of different socioeconomic status measures. We speculate that having cystic fibrosis is likely to amplify the effects of socioeconomic status on nutritional status at birth and in the first few years of an individual's life.

nt studies,23,24 but direct comparison between these studies and ours is complicated by the use of different socioeconomic status measures. We speculate that having cystic fibrosis is likely to amplify the effects of socioeconomic status on nutritional status at birth and in the first few years of an individual's life. The inequality in weight is greatest at around the time of diagnosis, and becomes narrower over the first 3 years of life. This is an important finding, because a widening of inequalities over time is often the norm.7,25,26 These findings suggest that extending the period of differential weight gain for as long as possible might reduce inequalities, further lending support to neonatal screening programmes to enable early diagnosis and treatment.27 We speculate that by extending this period of catch-up for as long as possible by early diagnosis (ie, screening) we might see an attenuation of the deprivation effect over time. In this study, we detected no difference in the age at diagnosis by deprivation, but screening was associated with increased weight and height and improved lung function in children. Furthermore, our finding that the prevalence of supplemental feeding treatment was higher, after adjusting for disease severity, in the most disadvantaged patients suggests that NHS professionals are actively engaged in trying to boost the nutrition of poorer patients, recognising their health disadvantage.

n children. Furthermore, our finding that the prevalence of supplemental feeding treatment was higher, after adjusting for disease severity, in the most disadvantaged patients suggests that NHS professionals are actively engaged in trying to boost the nutrition of poorer patients, recognising their health disadvantage. The socioeconomic gradient in lung function, evident as soon as it can be routinely measured at the age of 5 years, points to the crucial role of environmental and health-care factors operating in the early years of life to produce inequalities. It further reinforces the need for early diagnosis and action to prevent adverse consequences for children with cystic fibrosis living in disadvantaged circumstances. In Schechter and colleagues' cross-sectional study of US data,2 inequalities in %FEV1 by Medicaid status widened slightly from age 5 years to age 20 years. The magnitude of the inequalities in lung function at the age of 5 years seen in Schechter's study was larger (about a 9% difference) than in our UK study (4%), as was the magnitude of inequalities in lung function seen in O'Connor and colleagues' US study,3 which showed a difference of 5·5% between the most and least deprived quintiles. Methodological differences between the studies, however, make a direct comparison between these UK and US findings inappropriate. This study is the first to examine the relation between deprivation and %FEV1 in a population-level, adult cohort. We did not detect an association, despite the higher prevalence of P aeruginosa. We speculate that this finding might relate to the complication of progressive drop-out in older patients, and the insensitivity of %FEV1 as an outcome measure in adults.20

between deprivation and %FEV1 in a population-level, adult cohort. We did not detect an association, despite the higher prevalence of P aeruginosa. We speculate that this finding might relate to the complication of progressive drop-out in older patients, and the insensitivity of %FEV1 as an outcome measure in adults.20 The increased prevalence of chronic P aeruginosa infection in patients from more deprived areas, after adjusting for %FEV1, is a new finding in a population-level cohort. In Schechter and colleagues' study,2 Medicaid-insured patients were more likely to have P aeruginosa infection than were patients who were not eligible for Medicaid insurance, but when adjusted for %FEV1 there was no statistically significant difference—another US cohort study did not detect an association either.28 Previously identified risk factors for P aeruginosa acquisition, which is associated with worse lung function, include female sex and genotype (both associations shown in this study), and exposure to other patients with P aeruginosa colonisation.29 Our finding that more deprived groups are more likely to receive intravenous treatment in hospital might result in more deprived patients having greater exposure to other patients with chronic P aeruginosa, therefore increasing their risk of infection.

), and exposure to other patients with P aeruginosa colonisation.29 Our finding that more deprived groups are more likely to receive intravenous treatment in hospital might result in more deprived patients having greater exposure to other patients with chronic P aeruginosa, therefore increasing their risk of infection. We saw substantial socioeconomic differences in the reported use of key cystic fibrosis treatments in two contrasting ways. First, children from the most deprived quintile were about twice as likely to receive hospital intravenous antibiotic treatment and nutritional support, after adjustment for disease severity, compared with those from the least deprived quintile. We can speculate, from our knowledge of UK cystic fibrosis services, that clinicians in the NHS are more likely to bring children from more deprived areas into hospital for intravenous treatment because of concerns about the difficulties in delivering treatments in their homes. Conversely, children living in more affluent areas might receive intravenous treatment at home because of judgments about the adequacy of support and adherence to treatment in their home or because of their families' wish to avoid disruption to schooling and family life. This equitable model of care, with positive discrimination for socially disadvantaged children and adults with cystic fibrosis, is an uncommon finding in health systems, when access, particularly to secondary care for adults, often decreases with increasing deprivation, after adjusting for differential need.8,30 While several studies have seen use of health services by level of deprivation to be more equal in relation to children than adults,31 we have detected evidence in children with cystic fibrosis that goes even further with a pro-poor bias in the NHS for specific treatments. Coupled with our findings of inequalities in outcomes by deprivation, which do not widen over time, we speculate that the treatment decisions being made by clinicians might mitigate some effects of social disadvantage. This provides encouragement that there are interventions that health services can make to reduce the adverse effects of deprivation on chronic disorders such as cystic fibrosis.

do not widen over time, we speculate that the treatment decisions being made by clinicians might mitigate some effects of social disadvantage. This provides encouragement that there are interventions that health services can make to reduce the adverse effects of deprivation on chronic disorders such as cystic fibrosis. In the USA, with use of ZIP-code-linked income of an area as the socioeconomic indicator, there was no gradient in intravenous treatment use in children younger than 12 years, but in young people aged 13–18 years, those living in more affluent areas were more likely to be treated (13·8% in the lowest income category compared with 19·2% in the highest).18

ked income of an area as the socioeconomic indicator, there was no gradient in intravenous treatment use in children younger than 12 years, but in young people aged 13–18 years, those living in more affluent areas were more likely to be treated (13·8% in the lowest income category compared with 19·2% in the highest).18 Our second, and contrasting, set of findings on cystic fibrosis treatments, however, point to an apparent pro-rich bias in two other treatments, which were more evident in adults than in children: more affluent adults in the UK were more likely to receive DNase and inhaled antibiotics than were their more disadvantaged counterparts. DNase is an expensive treatment to reduce viscosity of sputum and to aid sputum expectoration, and some evidence exists that it prevents decrease in %FEV1.32 These treatments, although expensive, are free of charge to all patients in the NHS. One possibility for the social disparity in access to them is that they are both home-based treatments, requiring regular and long-term administration. Socially disadvantaged patients with cystic fibrosis are less likely to adhere to treatments,29 and if they report poor adherence, clinicians might be less likely to prescribe these drugs because they are unlikely to be as effective if taken inconsistently. Evidence from the USA shows no difference in use of DNase in children by area income quintile, but Medicaid-insured children (ie, those receiving free or subsidised care) were more likely to receive DNase than were children who were not eligible for Medicaid insurance.17

kely to be as effective if taken inconsistently. Evidence from the USA shows no difference in use of DNase in children by area income quintile, but Medicaid-insured children (ie, those receiving free or subsidised care) were more likely to receive DNase than were children who were not eligible for Medicaid insurance.17 Further research is needed to clarify which elements of the cystic fibrosis care model might contribute to a reduction in the adverse outcomes associated with deprivation. A cause for concern is the fact that the most disadvantaged families have a higher burden of treatment, in terms of time spent in hospital, which increases disruption to school and family life. Furthermore, the link with P aeruginosa colonisation requires further investigation. Higher socioeconomic status, as measured by parental education status, is associated with improved adherence to treatment in cystic fibrosis,29 and further research is needed to investigate the processes that lead to these differences. Systems to support the provision of intravenous treatment at home for more deprived groups in the UK should be explored.

ured by parental education status, is associated with improved adherence to treatment in cystic fibrosis,29 and further research is needed to investigate the processes that lead to these differences. Systems to support the provision of intravenous treatment at home for more deprived groups in the UK should be explored. Differences in access to health care cannot explain the differences in weight and height, by socioeconomic status, that are evident at the time of diagnosis, and are unlikely to explain the gradient in lung function evident at around the age of 5 years. The UK cystic fibrosis registry does not capture data about smoking in the home and these early effects might be associated with the known differences in smoking prevalence by socioeconomic status in the UK.33 The effect of socioeconomic status on growth in utero and in the early years of life in people with cystic fibrosis, might be mediated, at least in part, by maternal smoking, thus affecting subsequent outcomes and ultimately survival. Future studies should focus on the assessment of interventions, such as the reduction of exposure to environmental tobacco smoke,34 which might mitigate the effects of deprivation during the critical early years of life, and on the identification of aspects of health-care provision in cystic fibrosis that would help overcome the extra burden of adverse consequences of cystic fibrosis faced by patients living in economically-disadvantaged circumstances. Supplementary Material Supplementary appendix

Future studies should focus on the assessment of interventions, such as the reduction of exposure to environmental tobacco smoke,34 which might mitigate the effects of deprivation during the critical early years of life, and on the identification of aspects of health-care provision in cystic fibrosis that would help overcome the extra burden of adverse consequences of cystic fibrosis faced by patients living in economically-disadvantaged circumstances. Supplementary Material Supplementary appendix Acknowledgments DCT-R is supported by an MRC Population Health Scientist Fellowship (G0802448). We thank the UK Cystic Fibrosis Trust for access to the UK cystic fibrosis registry and all the centre directors for the input of data to the registry. Elaine Gunn assisted with access to the data. Contributors DCT-R, PJD, MW, and RLS had the idea for and designed the study and were named on the original MRC Fellowship application. DCT-R undertook the analysis and PJD supervised analysis. DCT-R, MW, RLS, and PJD interpreted the results and drafted the paper. All authors contributed to and approved the final draft for publication. Conflicts of interest We declare that we have no conflicts of interest. Figure 1 Comparison of anthropometric outcomes, by age and socioeconomic status Mean cross-sectional (A) weight, (B) height, and (C) body-mass index (BMI). Figure 2 Comparison of respiratory outcomes, by age and socioeconomic status Mean cross-sectional (A) FEV1 and (B) Pseudomonas aeruginosa colonisation prevalence. Figure 3 Comparison of treatment methods, by age and socioeconomic status

Figure 1 Comparison of anthropometric outcomes, by age and socioeconomic status Mean cross-sectional (A) weight, (B) height, and (C) body-mass index (BMI). Figure 2 Comparison of respiratory outcomes, by age and socioeconomic status Mean cross-sectional (A) FEV1 and (B) Pseudomonas aeruginosa colonisation prevalence. Figure 3 Comparison of treatment methods, by age and socioeconomic status Proportion of patients who received (A) any intravenous antibiotic treatment, (B) home intravenous antibiotic treatment, (C) hospital intravenous antibiotic treatment, (D) supplemental feeding, (E) DNase, and (F) inhaled antibiotics. Table 1 Unadjusted characteristics of study population by deprivation quintile (UK cystic fibrosis registry 1996 to 2009)

Proportion of patients who received (A) any intravenous antibiotic treatment, (B) home intravenous antibiotic treatment, (C) hospital intravenous antibiotic treatment, (D) supplemental feeding, (E) DNase, and (F) inhaled antibiotics. Table 1 Unadjusted characteristics of study population by deprivation quintile (UK cystic fibrosis registry 1996 to 2009) 1 (least deprived) 2 3 4 5 (most deprived) All p value Number of patients 1537 (19%) 1563 (19%) 1591 (20%) 1736 (22%) 1628 (20%) 8055 0·0018 Observations (for weight SD score) 9500 (19%) 9706 (20%) 9708 (20%) 10 550 (21%) 9873 (20%) 49 337 <0·0001 Female sex 712 (46%) 726 (46%) 728 (46%) 825 (48%) 773 (48%) 3764 (47%) 0·38 Age in days at diagnosis (IQR) 121 (30–731) 121 (30–670) 113 (30–730) 109 (30–728) 120 (30–730) 120 (30–730) 0·39 Number of delta F508 alleles 2 824 (54%) 827 (53%) 822 (52%) 907 (52%) 779 (48%) 4159 (52%) 0·0022 1 543 (35%) 556 (36%) 560 (35%) 609 (35%) 594 (37%) 2862 (36%) 0·63 0 170 (11%) 180 (12%) 209 (13%) 220 (13%) 255 (16%) 1034 (13%) <0·0001 Non-white 31 (2%) 31 (2%) 52 (3%) 73 (4%) 120 (7%) 307 (4%) <0·0001 Screened 233 (15%) 272 (17%) 245 (15%) 282 (16%) 277 (17%) 1309 (16%) 0·39 Birth cohort 1957 to 1966 62 (4%) 49 (3%) 64 (4%) 51 (3%) 35 (2%) 261 (3%) <0·0045 1967 to 1976 157 (10%) 172 (11%) 182 (11%) 171 (10%) 153 (9%) 835 (10%) 0·23 1977 to 1986 329 (21%) 384 (25%) 369 (23%) 426 (25%) 396 (24%) 1904 (24%) 0·09 1987 to 1996 496 (32%) 478 (31%) 489 (31%) 535 (31%) 530 (33%) 2528 (31%) 0·82 1997 to 2006 396 (26%) 393 (25%) 396 (25%) 427 (25%) 410 (25%) 2022 (25%) 0·62 2007 to <2010 97 (6%) 87 (6%) 91 (6%) 126 (7%) 104 (6%) 505 (6%) 0·32 Data are n (%) unless otherwise stated.

(25%) 369 (23%) 426 (25%) 396 (24%) 1904 (24%) 0·09 1987 to 1996 496 (32%) 478 (31%) 489 (31%) 535 (31%) 530 (33%) 2528 (31%) 0·82 1997 to 2006 396 (26%) 393 (25%) 396 (25%) 427 (25%) 410 (25%) 2022 (25%) 0·62 2007 to <2010 97 (6%) 87 (6%) 91 (6%) 126 (7%) 104 (6%) 505 (6%) 0·32 Data are n (%) unless otherwise stated. Table 2 Summary of adjusted effects of deprivation on clinical outcomes and use of treatments in patients with cystic fibrosis in the UK Patients younger than 18 years Patients aged 18 years to <40 years Clinical outcomes* FEV1 (percentage points [95% CI]) −4·12 (−5·01 to −3·19) −1·6 (−4·41 to 1·25) Weight-for-age (SD score [95% CI]) −0·28 (−0·38 to −0·18) −0·31 (−0·46 to −0·16) Height-for-age (SD score [95% CI]) −0·31 (−0·40 to −0·21) −0·31 (−0·43 to −0·19) BMI-for-age (SD score [95% CI]) −0·13 (−0·22 to −0·04) −0·12 (−0·25 to 0·01) Pseudomonas aeruginosa colonisation (OR [95% CI]) 1·89 (1·34 to 2·66) 1·78 (1·26 to 2·51) Treatments Any intravenous treatment (OR [95% CI])† 2·52 (1·92 to 3·17) 1·89 (1·51 to 2·38) Total intravenous days per year (% change [95% CI])† 15·9 (8·2 to 24) 10·6 (2·5 to 19·2) Supplemental feeding (OR [95% CI])‡ 1·78 (1·42 to 2·2) 2·38 (1·69 to 3·36) DNase treatment (OR [95% CI])† 0·40 (0·21 to 0·72) 0·37 (0·26 to 0·52) Use of inhaled antibiotics (OR [95% CI])† 0·66 (0·47 to 0·93) 0·40 (0·31 to 0·5) All estimates compare the most deprived quintile to the least deprived (reference) quintile. * The outcomes are from separate longitudinal models adjusted for time trends, sex, genotype, screening status, and ethnic origin.

Patients younger than 18 years Patients aged 18 years to <40 years Clinical outcomes* FEV1 (percentage points [95% CI]) −4·12 (−5·01 to −3·19) −1·6 (−4·41 to 1·25) Weight-for-age (SD score [95% CI]) −0·28 (−0·38 to −0·18) −0·31 (−0·46 to −0·16) Height-for-age (SD score [95% CI]) −0·31 (−0·40 to −0·21) −0·31 (−0·43 to −0·19) BMI-for-age (SD score [95% CI]) −0·13 (−0·22 to −0·04) −0·12 (−0·25 to 0·01) Pseudomonas aeruginosa colonisation (OR [95% CI]) 1·89 (1·34 to 2·66) 1·78 (1·26 to 2·51) Treatments Any intravenous treatment (OR [95% CI])† 2·52 (1·92 to 3·17) 1·89 (1·51 to 2·38) Total intravenous days per year (% change [95% CI])† 15·9 (8·2 to 24) 10·6 (2·5 to 19·2) Supplemental feeding (OR [95% CI])‡ 1·78 (1·42 to 2·2) 2·38 (1·69 to 3·36) DNase treatment (OR [95% CI])† 0·40 (0·21 to 0·72) 0·37 (0·26 to 0·52) Use of inhaled antibiotics (OR [95% CI])† 0·66 (0·47 to 0·93) 0·40 (0·31 to 0·5) All estimates compare the most deprived quintile to the least deprived (reference) quintile. * The outcomes are from separate longitudinal models adjusted for time trends, sex, genotype, screening status, and ethnic origin. † Adjusted for time trends, sex, genotype, screening status, (FEV1), and Pseudomonas aeruginosa colonisation status. ‡ Adjusted for time trends, sex, genotype, screening status, and body mass index (BMI) SD score. Panel Research in context Systematic review

* The outcomes are from separate longitudinal models adjusted for time trends, sex, genotype, screening status, and ethnic origin. † Adjusted for time trends, sex, genotype, screening status, (FEV1), and Pseudomonas aeruginosa colonisation status. ‡ Adjusted for time trends, sex, genotype, screening status, and body mass index (BMI) SD score. Panel Research in context Systematic review We searched PubMed with the terms “(cystic fibrosis) and (inequality OR equity OR inequity OR socioeconomic OR disadvantage OR vulnerable OR poverty OR social class OR disparity)” to identify relevant studies on the effect of socioeconomic status on outcomes and treatment in people with cystic fibrosis. We applied no date or language restrictions. We identified a review that summarises all studies,29 much of which were done in the USA, where the health-care system is different to that in the UK. People with cystic fibrosis from socioeconomically disadvantaged backgrounds die younger than do those in more advantaged social positions in the UK5 and the USA.2 The key challenge is to understand how and when these inequalities develop, and to understand how the health-care system in the UK can mitigate or perpetuate these effects to identify promising options for intervention. Interpretation

We searched PubMed with the terms “(cystic fibrosis) and (inequality OR equity OR inequity OR socioeconomic OR disadvantage OR vulnerable OR poverty OR social class OR disparity)” to identify relevant studies on the effect of socioeconomic status on outcomes and treatment in people with cystic fibrosis. We applied no date or language restrictions. We identified a review that summarises all studies,29 much of which were done in the USA, where the health-care system is different to that in the UK. People with cystic fibrosis from socioeconomically disadvantaged backgrounds die younger than do those in more advantaged social positions in the UK5 and the USA.2 The key challenge is to understand how and when these inequalities develop, and to understand how the health-care system in the UK can mitigate or perpetuate these effects to identify promising options for intervention. Interpretation This study has identified important longitudinal differences in weight, height, body-mass index, forced expiratory volume in 1 s, and risk of Pseudomonas aeruginosa colonisation by deprivation in people with cystic fibrosis in the UK, which start early in life, but do not increase over time. We detected socioeconomic differences in the reported use of key treatments in the UK. People from more deprived areas are about twice as likely to receive in-hospital intravenous antibiotic treatment and nutritional support, but less likely to receive DNase and inhaled antibiotics. Interventions to reduce inequalities in outcomes in cystic fibrosis need to be focused in the antenatal period and the early years of life. Such interventions include smoking prevention and public health initiatives to address inequalities in maternal and child health. Further research is needed to clarify which elements of the cystic fibrosis care model in the UK might contribute to a reduction in the adverse outcomes associated with deprivation, and to investigate identified differences in access to inhaled treatments.

Introduction Wheeze in children aged 10 months to 5 years is characterised by recurrent episodes that are frequently triggered by viral colds.1 Episodes of wheeze in young children might be clinically severe and can result in parents seeking medical attention.2 Indeed, an audit of UK paediatric hospital admissions for acute wheeze from 1998 to 2005 showed that most admissions were of children younger than 5 years.3 Because wheeze in young children is characterised by long asymptomatic periods interspersed with short intense episodes,1 intermittent treatment strategies have been assessed. We previously reported that a short course of oral corticosteroids initiated by parents at the onset of a wheeze episode is not effective for reducing the severity of wheeze in children aged 1–5 years.4 By contrast, intermittent high-dose inhaled corticosteroids reduce the risk of clinically severe wheeze episodes by 30% in that age group.5 However, this strategy is associated with clinically relevant growth suppression.5 Because montelukast (a cysteinyl leukotriene receptor blocker) does not suppress growth,6 the effectiveness of intermittent montelukast for wheeze in young children is of clinical interest.

vere wheeze episodes by 30% in that age group.5 However, this strategy is associated with clinically relevant growth suppression.5 Because montelukast (a cysteinyl leukotriene receptor blocker) does not suppress growth,6 the effectiveness of intermittent montelukast for wheeze in young children is of clinical interest. To date, trials of intermittent montelukast have reported conflicting results: findings from a subgroup analysis in Robertson and colleagues' trial7 of children aged 2–14 years showed that intermittent montelukast given over 12 months reduced unscheduled use of acute health-care resources by 38%; Bacharier and colleagues8 reported that intermittent montelukast therapy over 12 months does not decrease wheeze severity in young children or need for oral corticosteroid therapy; and Valovirta and colleagues9 reported no beneficial effect of a 12 month course of intermittent montelukast on wheeze attacks in young children. Reasons for these inconsistent results could be the substantial heterogeneity in treatment effect in young children with wheeze,10 and that the response to montelukast is limited to a subgroup of children.

eported no beneficial effect of a 12 month course of intermittent montelukast on wheeze attacks in young children. Reasons for these inconsistent results could be the substantial heterogeneity in treatment effect in young children with wheeze,10 and that the response to montelukast is limited to a subgroup of children. Studies of adults with asthma suggest that heterogeneity in response to montelukast is partly determined by a polymorphism in the arachidonate 5-lipoxygenase (ALOX5) gene promoter. The ALOX5 gene encodes 5-lipoxygenase—the rate-limiting enzyme in the cysteinyl leukotriene biosynthetic pathway.11,12 This polymorphism results from variable numbers of copies of the Sp1 binding motif GGGCGG, whereby five Sp1 repeats are the major allele.13 Thus individuals are classified as either 5/5, or 5/x (in which x is not equal to 5), or x/x.14 To date, the ALOX5 promoter genotype grouping that best defines montelukast responsiveness in adults is unclear. For example, Telleria and colleagues15 reported increased montelukast responsiveness in adults with the 5/5 and the 5/x genotype (compared with x/x), whereas Lima and colleagues14 reported that both the 5/x and x/x genotypes were responsive to montelukast. We did the Wheeze And Intermittent Treatment (WAIT) trial to assess the efficacy of intermittent montelukast for wheeze in young children at increased risk of clinically severe episodes of wheeze.

Studies of adults with asthma suggest that heterogeneity in response to montelukast is partly determined by a polymorphism in the arachidonate 5-lipoxygenase (ALOX5) gene promoter. The ALOX5 gene encodes 5-lipoxygenase—the rate-limiting enzyme in the cysteinyl leukotriene biosynthetic pathway.11,12 This polymorphism results from variable numbers of copies of the Sp1 binding motif GGGCGG, whereby five Sp1 repeats are the major allele.13 Thus individuals are classified as either 5/5, or 5/x (in which x is not equal to 5), or x/x.14 To date, the ALOX5 promoter genotype grouping that best defines montelukast responsiveness in adults is unclear. For example, Telleria and colleagues15 reported increased montelukast responsiveness in adults with the 5/5 and the 5/x genotype (compared with x/x), whereas Lima and colleagues14 reported that both the 5/x and x/x genotypes were responsive to montelukast. We did the Wheeze And Intermittent Treatment (WAIT) trial to assess the efficacy of intermittent montelukast for wheeze in young children at increased risk of clinically severe episodes of wheeze. Methods Study design and participants We did this multicentre, parallel-group, randomised, placebo-controlled trial between Oct 1, 2010, and Dec 20, 2013, at 21 primary care sites and 41 secondary care sites in England and Scotland. Eligible children were aged between 10 months and 5 years and had had two or more previous episodes of wheeze, at least one of which was physician-confirmed, and at least one of which had happened in the preceding 3 months. We excluded children if they had a pre-existing respiratory vulnerability such as cystic fibrosis, sickle-cell disease, severe developmental delay with feeding difficulty, history of neonatal chronic lung disease, or structural airways disease. Children were also excluded if they had been enrolled in a therapeutic trial in the previous 3 months or were taking continuous oral montelukast at the time of enrolment. To represent the overall population of young children with wheeze, and in line with the population in our previous trials,4,16 we did not exclude children receiving continuous inhaled corticosteroid therapy. The study was approved by the UK National Health Service Multicenter Research Ethics Committee (reference number 09/H1102/110), by local institutional review boards, and by the UK Medicines and Healthcare Products Regulatory Agency (21313/0024/01-0001); the UK Medicines for Children Research Network supported the study. An independent data and safety monitoring committee not involved with patient enrolment reviewed adverse events. Written informed consent was obtained from the parent or guardian of each child enrolled in the study.

ulatory Agency (21313/0024/01-0001); the UK Medicines for Children Research Network supported the study. An independent data and safety monitoring committee not involved with patient enrolment reviewed adverse events. Written informed consent was obtained from the parent or guardian of each child enrolled in the study. Randomisation and masking Participants were allocated to either a 5/5 or 5/x+x/x ALOX5 promoter genotype stratum, then randomly assigned (1:1), via a permuted block schedule (size ten) developed by the manufacturer (Novalabs, Leicester, UK), to receive montelukast or placebo (appendix). Clinical investigators and parents were masked to treatment group and genotype strata. Placebo and montelukast were packaged as identical granules in identical sachets labelled with participant number only. Emergency code break was allowed in cases of a suspected severe adverse reaction when knowledge of patient allocation could have affected clinical management of a study participant, in the case of a suspected unexpected severe adverse reaction, and in any other circumstance in which the principal investigator considered that an emergency code break was indicated.

a suspected severe adverse reaction when knowledge of patient allocation could have affected clinical management of a study participant, in the case of a suspected unexpected severe adverse reaction, and in any other circumstance in which the principal investigator considered that an emergency code break was indicated. Procedures At enrolment, parents completed a structured questionnaire administered by research study personnel, which asked about previous wheeze, present treatment, and risk factors (appendix). Saliva from each child was collected with the Oragene OG-250 collection kit in combination with the CS-1 saliva collection kit for young children (both manufactured by DNA Genotek, Ottawa, ON, Canada) and transferred to Queen Mary University of London (London, UK) for analysis. The simple sequence-length polymorphism in the promoter region of ALOX5 (rs59439148) was genotyped as described previously.17 Alleles were classified according to the number of simple repeats (appendix), and children were identified as belonging to either 5/5 or 5/x+x/x strata.

sity of London (London, UK) for analysis. The simple sequence-length polymorphism in the promoter region of ALOX5 (rs59439148) was genotyped as described previously.17 Alleles were classified according to the number of simple repeats (appendix), and children were identified as belonging to either 5/5 or 5/x+x/x strata. Parents were advised to commence the trial drug at the onset of each viral cold or wheezing episode over the 12-month study period. Parents continued all other drugs prescribed by their managing clinician (including bronchodilators and inhaled corticosteroids), and completed a diary of symptoms, medicine use, adverse events, and medical attendance for each day the trial drug was given (appendix). Investigators asked parents by telephone survey about usage of trial drug, use of oral corticosteroid rescue therapy, and medical attendances at two-monthly intervals during the 12-month study period (appendix). Parents who could not be contacted received a maximum of two letters offering continued involvement in the study. When parents could not be contacted for two successive phone calls, parent and child were regarded as withdrawn from the study. Medical attendances for wheeze were independently verified by study investigators by contact with the managing clinician.

a maximum of two letters offering continued involvement in the study. When parents could not be contacted for two successive phone calls, parent and child were regarded as withdrawn from the study. Medical attendances for wheeze were independently verified by study investigators by contact with the managing clinician. Urine was obtained from asymptomatic children at baseline. Urine was transported on ice, and stored at −80°C within 1 h of collection. We analysed urinary leukotriene E4—the final urinary metabolite of cysteinyl leukotriene production—by high-performance liquid chromatography–tandem mass spectrometry (ABI SCIEX 4000 QTRAP, Framingham, MA, USA), as previously described (appendix).18 Concentrations were expressed in proportion to urinary creatinine. We excluded samples with a urinary creatinine concentration of less than 0·1 mg/mL because correction is inaccurate in very dilute samples. We monitored children for adverse events with a diary card report and telephone follow-up. Hospital admission for exacerbation of wheeze, acute lower-respiratory-tract infection, acute febrile illness, febrile convulsion, gastroenteritis, and exacerbation of eczema were not classed as serious adverse events in the trial protocol.

d children for adverse events with a diary card report and telephone follow-up. Hospital admission for exacerbation of wheeze, acute lower-respiratory-tract infection, acute febrile illness, febrile convulsion, gastroenteritis, and exacerbation of eczema were not classed as serious adverse events in the trial protocol. Outcomes Our primary outcome was the number of unscheduled medical attendances for wheezing episodes. Such attendances were defined as those to a family doctor, an asthma nurse or similarly trained health-care professional, an accident and emergency department, hospital via accident and emergency (hospital admission), or any combination of these. Secondary outcomes were duration of hospital admission, number of wheeze episodes, duration of wheeze episodes, number of courses of oral steroids per year, proportion of children receiving oral corticosteroids, use of trial drug, time to first unscheduled medical attendance, and time to first unscheduled attendance by site of medical attendance. We did a prespecified subgroup analysis that assessed unscheduled medical attendances for wheeze episodes by ALOX5 promoter genotype strata (5/5 and 5/x+x/x). Other prespecified subgroups for analysis were multitrigger and episodic wheeze at baseline, use of either continuous inhaled corticosteroids or no inhaled corticosteroids at baseline, and the alternative genotype grouping of 5/5+5/x and x/x.

nces for wheeze episodes by ALOX5 promoter genotype strata (5/5 and 5/x+x/x). Other prespecified subgroups for analysis were multitrigger and episodic wheeze at baseline, use of either continuous inhaled corticosteroids or no inhaled corticosteroids at baseline, and the alternative genotype grouping of 5/5+5/x and x/x. Statistical analysis The trial was powered to detect a difference in the number of unscheduled medical attendances for wheeze episodes between participants in the intervention and control groups, and to detect differential responsiveness to montelukast in the 5/5 stratum compared with the 5/x+x/x stratum, with the assumption that montelukast leads to a 60% reduction in attendances in the 5/x+x/x stratum, and a 20% reduction in the 5/5 stratum. With use of data from the UK General Practitioner Research Database, with courses of oral steroids as a proxy for unscheduled medical attendances for wheeze episodes, we estimated a mean of 0·76 [SD 1·22] such attendances per year. Because data follow an overdispersed Poisson distribution, we used Markov chain Monte Carlo simulation in WinBUGs (version 1.4) to estimate required sample sizes. 1050 children were needed to detect a 33% drop in unscheduled medical attendances for wheeze episodes, with a power of 90% at a significance level of 5%, with a 6% loss to follow up. A 33% drop in attendances equates to an attack rate of 0·51 for the treatment group. The clinical significance of these changes is that roughly four children would need to be treated to prevent one unscheduled medical attendance. Because a sample size of 1200 provides just more than 80% power at the 5% significance level to detect an interaction between treatment and ALOX5 genotype, 1300 children needed to be recruited, assuming a 6% dropout. Interim safety analyses were done at 6-monthly intervals. Efficacy analyses were done at the end of the trial.

a sample size of 1200 provides just more than 80% power at the 5% significance level to detect an interaction between treatment and ALOX5 genotype, 1300 children needed to be recruited, assuming a 6% dropout. Interim safety analyses were done at 6-monthly intervals. Efficacy analyses were done at the end of the trial. For each child, we analysed unscheduled medical attendances for wheeze episodes and episodes of viral cold with a Poisson regression model. For each episode of wheeze and viral cold, duration of hospital admission, and number of symptom days were also analysed with Poisson regression models. We included follow-up time for each child as an exposure variable and a random effect fitted for each child to account for overdispersion or when episode was the unit of analysis. Follow-up time was based on time from randomisation until either the 12 month end of trial date or date of last phone call.

gression models. We included follow-up time for each child as an exposure variable and a random effect fitted for each child to account for overdispersion or when episode was the unit of analysis. Follow-up time was based on time from randomisation until either the 12 month end of trial date or date of last phone call. For unscheduled medical attendances for wheeze episodes, we assessed the differential effect of treatment in predefined subgroups by inclusion of an interaction term. Proportions of patients who had any unscheduled medical attendance, or those receiving oral corticosteroid rescue therapy, were analysed with logistic regression. We analysed time to first unscheduled medical attendance with Cox regression models. All models were fitted on the available case population with modified intention-to-treat principles and included fixed effects for stratification factor and treatment. We did a per-protocol analysis that excluded any children randomised not according to schedule and that corrected for those randomised under the incorrect stratum. Parents who withdrew their children from the study and provided permission to use their data were included in the analysis to the point of withdrawal. Parents who withdrew their children and did not provide permission for their data to be used were excluded from the analysis. Because we anticipated few missing data, no imputation of missing data was done. All analyses were two-sided with a 5% significance level. Results are presented as incidence rate ratios (IRRs), odds ratios (ORs), or hazard ratios (HRs) as appropriate, with corresponding 95% CIs. To assess the effect of ALOX5 genotype on urinary leukotriene E4, data were first log10 transformed to normalise distribution. Groups were compared with either ANOVA and Dunnett's multiple comparisons test, or with t test using GraphPad Prism version 6.00 for Windows (GraphPad Software, La Jolla, CA, USA). Analyses were done with STATA Statistical Software: release 12.1. This trial is registered with ClinicalTrials.gov, number NCT01142505.

on. Groups were compared with either ANOVA and Dunnett's multiple comparisons test, or with t test using GraphPad Prism version 6.00 for Windows (GraphPad Software, La Jolla, CA, USA). Analyses were done with STATA Statistical Software: release 12.1. This trial is registered with ClinicalTrials.gov, number NCT01142505. Role of funding source The sponsor of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. All authors had full access to all raw data and the corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

sponsor of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. All authors had full access to all raw data and the corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. Results Figure 1 shows the trial profile. Parents of 1366 children provided consent to enter the study, of whom eight withdrew children before randomisation. The remaining 1358 children were randomly assigned to receive montelukast (n=669) or placebo (n=677; figure 1). Data for the primary outcome were obtained from 1308 (96%) children whose parents responded to at least one follow-up phone call (figure 1). Baseline demographic characteristics were similar between treatment groups. The per-protocol analysis included 1297 children. 11 children were excluded who had been incorrectly randomised and the strata was corrected for two children who were randomised with incorrect strata. There were no major differences in baseline variables between children in the placebo and montelukast groups or between the two genetic strata (table 1). The dominant allele was five repeats (table 1), and consistent with previous reports,14,19 black children had a greater frequency of x alleles (75% vs 31% in white children; appendix).

r differences in baseline variables between children in the placebo and montelukast groups or between the two genetic strata (table 1). The dominant allele was five repeats (table 1), and consistent with previous reports,14,19 black children had a greater frequency of x alleles (75% vs 31% in white children; appendix). Overall, we recorded 1310 unscheduled medical attendances for wheeze episodes in the montelukast group and 1480 such attendances in the placebo group. There was no difference in mean medical attendances between the montelukast and placebo groups (table 2). These conclusions remained the same when the analysis was repeated in the per-protocol population. Compared with placebo, children in the 5/5 ALOX5 promoter stratum had reductions in unscheduled medical attendances for wheeze episodes (table 2). By contrast, there was no difference in medical attendances between children in the montelukast and placebo groups in the 5/x+x/x stratum (table 2).

per-protocol population. Compared with placebo, children in the 5/5 ALOX5 promoter stratum had reductions in unscheduled medical attendances for wheeze episodes (table 2). By contrast, there was no difference in medical attendances between children in the montelukast and placebo groups in the 5/x+x/x stratum (table 2). No difference was recorded between the montelukast and placebo groups for the number of children who had at least one unscheduled medical attendance for wheeze episodes, the number of wheeze episodes, or the duration of wheeze episodes (table 3). There was also no difference between treatment groups for time to first unscheduled medical attendance (table 3). Time to first hospital admission was increased in the montelukast group (p=0·04; appendix). There was no difference between the montelukast and placebo groups for attendances to accident and emergency (appendix). Mean number of courses of rescue oral corticosteroids were lower in children given montelukast than in those given placebo (table 3), but there was no difference in the proportion of children receiving at least one course of rescue oral corticosteroids (appendix). In the montelukast group, study drugs were reported to be effective by 323 (56%) of 579 parents at the 12-month timepoint; 58 (10%) parents were unsure, and 69 (12%) did not respond. In the placebo group, study drugs were reported to be effective by 299 (52%) of 575 parents; 58 (10%) parents were unsure, and 78 (14%) did not respond.

ast group, study drugs were reported to be effective by 323 (56%) of 579 parents at the 12-month timepoint; 58 (10%) parents were unsure, and 69 (12%) did not respond. In the placebo group, study drugs were reported to be effective by 299 (52%) of 575 parents; 58 (10%) parents were unsure, and 78 (14%) did not respond. There was no significant interaction for pattern of wheeze at baseline (multitrigger vs episodic wheeze), use of regular inhaled corticosteroids, or a different grouping of ALOX5 promoter genotype 5/5+5/x and x/x (appendix). Of the 940 adverse events reported in the study, 657 (70%) were classified as definitely not related to study drug, 179 (19%) as probably not related, 93 (10%) as possibly related, 11 (1%) as probably related, and no adverse event was definitely related (appendix). We recorded one serious adverse event, which was a skin reaction in a child allocated to placebo (appendix). The distribution of adverse events was similar between groups (table 4). Urine was obtained from 975 asymptomatic children at recruitment. We excluded children with concentrations of urinary creatinine of less than 0·1 mg/mL (n=26), resulting in analysis of 597 (63%) children with the 5/5 genotype, 312 (33%) with the 5/x genotype, and 40 (4%) with the x/x genotype. Urinary leukotriene E4 (log10 transformed) was higher in children with the x/x genotype than in those with the 5/5 genotype (figure 2). There was no significant difference in urinary leukotriene E4 between the 5/5 and 5/x genotypes, or the 5/5 and 5/x+x/x genotypes (data not shown).

, and 40 (4%) with the x/x genotype. Urinary leukotriene E4 (log10 transformed) was higher in children with the x/x genotype than in those with the 5/5 genotype (figure 2). There was no significant difference in urinary leukotriene E4 between the 5/5 and 5/x genotypes, or the 5/5 and 5/x+x/x genotypes (data not shown). Discussion Our findings show that intermittent montelukast treatment, although not associated with side-effects, did not reduce unscheduled medical attendances for wheeze episodes in children younger than 5 years. These results are in line with those of Bacharier and colleagues,8 who reported that intermittent montelukast in young children with wheeze does not increase the proportion of episode-free days or decrease the proportion of children who need urgent medical care, and with those of Valovirta and colleagues9 who noted that intermittent montelukast does not reduce the number of wheeze episodes culminating in need for unscheduled medical care or rescue oral corticosteroids. Use of oral steroid rescue therapy in our study was much lower than unscheduled medical attendances for wheeze episodes. We postulate that this finding shows a change in UK prescribing practice in view of studies reporting oral steroids to be ineffective in acute wheeze in young children.4,16 We recorded a reduction in use of oral corticosteroid in children given montelukast, but in the context of present UK prescribing practice, the clinical significance of a change in this indirect marker of wheeze severity is unclear. Our results differ to those from Robertson and colleagues7 who, in a subgroup analysis, showed that intermittent montelukast is effective in reducing unscheduled use of health-care resources in children aged 2–5 years. To resolve these contradictory findings, we did a meta-analysis of trials of intermittent montelukast for unscheduled medical attendances for wheeze episodes (appendix). Findings of this meta-analysis showed no benefit of a 12 month period of intermittent montelukast therapy on unscheduled medical attendances for wheeze (appendix). This outcome suggests that intermittent montelukast is not an effective treatment strategy for treatment of young children with a history of two or more episodes of wheeze (panel).

alysis showed no benefit of a 12 month period of intermittent montelukast therapy on unscheduled medical attendances for wheeze (appendix). This outcome suggests that intermittent montelukast is not an effective treatment strategy for treatment of young children with a history of two or more episodes of wheeze (panel). In the present study, the 95% CI of the IRR for unscheduled medical attendances for wheeze excluded a 33% reduction in such attendances. However, the fewer unscheduled medical attendances in the montelukast group, albeit non-significant, suggests heterogeneity of treatment response—a characteristic of previous studies in young children with wheeze. For example, response to continuous inhaled corticosteroids is most favourable in the subgroup of white males with an unscheduled medical attendance for wheeze in the previous 12 months and aeroallergen sensitisation.21 Furthermore, in Bacharier and colleagues' study,8 intermittent montelukast, despite having no overall benefit, reduced the area under the curve for wheezing score in children with a positive asthma predictive index, defined as four or more wheezing episodes with at least one diagnosed by a doctor, and one or more major criteria of parental asthma, doctor-diagnosed dermatitis, allergic sensitisation to one or more aeroallergen, or at least two minor criteria of allergic sensitisation to milk, egg, or peanuts; wheeze unrelated to colds; and blood eosinophils greater than 4%.22 We did not stratify by asthma predictive index because Meyer and colleagues22 reported that no clinical variable predicts response to continuous montelukast in wheeze in young children, and blood sampling, in our experience, greatly reduces the willingness of parents to enter their infants into a therapeutic trial. Furthermore, use of parental-reported diagnosis for disorders such as eczema overestimates physician-diagnosed disease.23 As such, we cannot exclude montelukast responsiveness in children with a positive asthma predictive index. However, our prespecified subgroup analyses showed that neither the pattern of wheeze nor use of inhaled corticosteroids was associated with montelukast response, although our study was not powered for these interactions.

we cannot exclude montelukast responsiveness in children with a positive asthma predictive index. However, our prespecified subgroup analyses showed that neither the pattern of wheeze nor use of inhaled corticosteroids was associated with montelukast response, although our study was not powered for these interactions. In adults with asthma, heterogeneity in response to montelukast is associated with a polymorphism in the ALOX5 promoter.14,15 In line with these studies in adults, we recorded a 20% reduction in unscheduled medical attendances for wheeze in children in the montelukast group with the 5/5 ALOX5 promoter genotype, and no effect of intermittent montelukast in those with the 5/x+x/x genotype. The montelukast-responsive genotype (5/5) in the present study is, however, different from our a-priori hypothesis, as suggested by the 5/x+x/x grouping from Lima and colleagues' study.14 But other studies in adults report montelukast responsiveness of the 5/5 genotype. For example, Telleria and colleagues15 reported decreased asthma exacerbations and improved lung function in adults with the 5/5 genotype who were given montelukast, and Drazen and colleagues24 showed that ABT-761 (a 5-lipoxygenase inhibitor) improved lung function in adults with the 5/5 genotype, but not in those with the x/x genotype. We sought support for a differential response to montelukast between genotypes by measurement of urinary leukotriene E4.25 In the only study in children to date, Mougey and colleagues19 measured urinary leukotriene E4 and identified ALOX5 polymorphism status in 270 6–17-year-old children with poorly controlled asthma enrolled into a 6 month (negative) trial of acid-reflux treatment. Children with the x/x genotype (73% of whom were receiving montelukast) had significantly higher concentrations of urinary leukotriene E4, worse forced expiratory volume in 1 s, and a trend for poorer asthma control than those with the 5/5+5/x genotypes.19 Similarly, we recorded increased urinary leukotriene E4 in children with the x/x genotype compared with those with the 5/5 or 5/5+5/x genotypes. These data provide support for a differential response to montelukast between 5/5 and x/x genotypes; however, they do not explain a differential response between the 5/x and 5/5 genotypes. We postulate that differences in production of cysteinyl leukotriene between 5/x and 5/5 genotypes might be shown during children's wheeze episodes when cysteinyl leukotriene production is increased.26

between 5/5 and x/x genotypes; however, they do not explain a differential response between the 5/x and 5/5 genotypes. We postulate that differences in production of cysteinyl leukotriene between 5/x and 5/5 genotypes might be shown during children's wheeze episodes when cysteinyl leukotriene production is increased.26 These data do not support the routine use of intermittent montelukast for wheeze in children aged 10 months to 5 years. Further stratified trials should be done to confirm the presence of a responsive subgroup. Supplementary Material Supplementary appendix Acknowledgments This study was funded by the Medical Research Council (UK), in partnership with the National Institute for Health Research (reference number 08/43/03). We thank independent members of the trial steering committee (listed in appendix), centres responsible for primary care recruitment (appendix), and the National Institute for Health Research (NIHR) Medicines for Children Network for helping with recruitment.

al Institute for Health Research (reference number 08/43/03). We thank independent members of the trial steering committee (listed in appendix), centres responsible for primary care recruitment (appendix), and the National Institute for Health Research (NIHR) Medicines for Children Network for helping with recruitment. Contributors JG was the chief investigator, planned and provided overall supervision of the study, wrote with CN the first and final drafts of the report, and vouches for these data. CN supervised the study, and wrote with JG the first and final drafts of the report. HP, ST, DP, and CJG contributed to study planning and to the final manuscript. TV contributed to study planning, supervised genotype analysis, and contributed to the final manuscript. ID contributed to study planning, genotype analysis, and the final manuscript. JWH contributed to study planning, advised on genotype analysis, and contributed to the final manuscript. MS did the urinary leukotriene analysis and contributed to the final manuscript. RB supervised the study, did the combined analysis, and contributed to the final manuscript. LK did genotyping and was responsible for audit of genotype data, CR supported the data monitoring committee, wrote the final statistical analysis plan, and did the statistical analysis. SE contributed to study planning and supervised the statistical analysis.

combined analysis, and contributed to the final manuscript. LK did genotyping and was responsible for audit of genotype data, CR supported the data monitoring committee, wrote the final statistical analysis plan, and did the statistical analysis. SE contributed to study planning and supervised the statistical analysis. Declaration of interests JG received personal fees for Advisory Board membership for new asthma treatments in children from GlaxoSmithKine, Boehringer Ingelheim, and Novartis while the study was being done. DP has received fees paid to Research in Real Life for lecture and speaking engagements from Almirall, AstraZeneca, Boehringer Ingelheim, Chiesi, Cipla, GlaxoSmithKline, Kyorin, Meda, Merck, Mundipharma, Novartis, Pfizer, SkyePharma, Takeda, and Teva; for manuscript preparation from Mundipharma and Teva; for travel, accommodation, and meeting expenses from Aerocrine, Boehringer Ingelheim, Mundipharma, Napp, Novartis, and Teva; for patient enrolment or completion of research from Almirall, Chiesi, Teva, and Zentiva; for contract research from Aerocrine, AKL, Almirall, Boehringer Ingelheim, Chiesi, Meda, Mundipharma, Napp, Novartis, Orion, Takeda, and Zentiva; has an AKL patent pending; and has shares in AKL, which produces phytopharmaceuticals and owns 80% of Research in Real Life and its subsidiary social enterprise Optimum Patient Care. All other authors have no competing interests. Figure 1 Trial profile

Declaration of interests JG received personal fees for Advisory Board membership for new asthma treatments in children from GlaxoSmithKine, Boehringer Ingelheim, and Novartis while the study was being done. DP has received fees paid to Research in Real Life for lecture and speaking engagements from Almirall, AstraZeneca, Boehringer Ingelheim, Chiesi, Cipla, GlaxoSmithKline, Kyorin, Meda, Merck, Mundipharma, Novartis, Pfizer, SkyePharma, Takeda, and Teva; for manuscript preparation from Mundipharma and Teva; for travel, accommodation, and meeting expenses from Aerocrine, Boehringer Ingelheim, Mundipharma, Napp, Novartis, and Teva; for patient enrolment or completion of research from Almirall, Chiesi, Teva, and Zentiva; for contract research from Aerocrine, AKL, Almirall, Boehringer Ingelheim, Chiesi, Meda, Mundipharma, Napp, Novartis, Orion, Takeda, and Zentiva; has an AKL patent pending; and has shares in AKL, which produces phytopharmaceuticals and owns 80% of Research in Real Life and its subsidiary social enterprise Optimum Patient Care. All other authors have no competing interests. Figure 1 Trial profile *Perceived inefficacy is on the side of patient. †Data for the primary outcome were obtained from children whose parents responded to at least one follow-up phone call. Figure 2 Dot plot of urinary LTE4 by variable numbers of copies of the Sp1-binding motif (either 5/5, 5/x, or x/x, in which x does not equal 5) in the ALOX5 promoter region

*Perceived inefficacy is on the side of patient. †Data for the primary outcome were obtained from children whose parents responded to at least one follow-up phone call. Figure 2 Dot plot of urinary LTE4 by variable numbers of copies of the Sp1-binding motif (either 5/5, 5/x, or x/x, in which x does not equal 5) in the ALOX5 promoter region 11 datapoints were outside the axis and are not shown for convenience. Horizontal bars within plots represent mean. LTE4=leukotriene E4. ALOX5=arachidonate 5-lipoxygenase. Table 1 Baseline characteristics

Figure 2 Dot plot of urinary LTE4 by variable numbers of copies of the Sp1-binding motif (either 5/5, 5/x, or x/x, in which x does not equal 5) in the ALOX5 promoter region 11 datapoints were outside the axis and are not shown for convenience. Horizontal bars within plots represent mean. LTE4=leukotriene E4. ALOX5=arachidonate 5-lipoxygenase. Table 1 Baseline characteristics Montelukast group (n=669) Placebo group (n=677) 5/5 5/x+x/x Total 5/5 5/x+x/x Total n (%) 416 (62%) 253 (38%) 669 (100%) 426 (63%) 251 (37%) 677 (100%) Height (cm) 90·0 (10·3) 89·8 (10·5) 89·9 (10·4) 89·9 (10·5) 91·8 (11·7) 90·6 (11·0) Weight (kg) 14·0 (3·0) 13·9 (3·7) 14·0 (3·3) 14·0 (3·3) 14·6 (3·8) 14·2 (3·5) Age (years) 2·6 (1·1) 2·5 (1·1) 2·6 (1·1) 2·6 (1·1) 2·8 (1·2) 2·7 (1·1) Male sex 262 (63%) 164 (65%) 426 (64%) 276 (65%) 161 (64%) 437 (65%) Ethnic origin White 335 (81%) 179 (71%) 514 (77%) 338 (79%) 174 (69%) 512 (76%) Black 5 (1%) 14 (6%) 19 (3%) 4 (1%) 14 (6%) 18 (3%) Asian 55 (13%) 37 (15%) 92 (14%) 58 (14%) 46 (18%) 104 (15%) Other 21 (5%) 23 (9%) 44 (7%) 26 (6%) 17 (7%) 43 (6%) Preterm birth (<37 weeks) 58 (14%) 40 (16%) 98 (14%) 56 (13%) 42 (17%) 98 (15%) Birthweight (<2500g) 51 (12%) 28 (11%) 79 (12%) 42 (10%) 28 (11%) 70 (10%) Food allergy 64 (15%) 44 (18%) 108 (16%) 64 (15%) 47 (19%) 111 (17%) Drug allergy 26 (6%) 12 (5%) 38 (6%) 23 (6%) 19 (8%) 42 (6%) Itchy rash (>6 months, ever)* 98 (23%) 64 (25%) 162 (24%) 104 (25%) 60 (24%) 164 (25%) Eczema (ever)† 207 (49%) 121 (48%) 328 (48%) 215 (52%) 134 (53%) 349 (52%) History of asthma in mother 156 (37%) 95 (38%) 251 (37%) 141 (34%) 89 (35%) 230 (34%) History of asthma in father 126 (30%) 73 (29%) 199 (29%) 126 (30%) 81 (32%) 207 (31%) Age at first wheeze (months) 12·4 (9·8) 13·5 (10·5) 12·8 (10·1) 12·4 (10·4) 13·6 (11·5) 12·9 (10·8) Children with episodic viral wheeze 296 (71%) 181 (72%) 477 (71%) 295 (69%) 191 (76%) 486 (72%) Children with multitrigger wheeze 120 (29%) 72 (28%) 192 (29%) 131 (31%) 60 (24%) 191 (28%) Interval between onset of URTI and wheezing (h)‡ 31·6 (27·4) 28·8 (25·2) 30·5 (26·6) 27·3 (23·4) 28·2 (26·0) 27·7 (24·4) Children with more than one hospital admission for wheeze in the past year 363 (87%) 216 (85%) 579 (87%) 351 (82%) 203 (81%) 554 (82%) Courses of oral corticosteroids in past year 2·0 (1·9) 1·8 (1·8) 1·9 (1·8) 1·9 (1·9) 1·8 (2·0) 1·9 (2·0) USMA in previous year 5·5 (4·3) 5·4 (4·1) 5·4 (4·2) 5·7 (5·3) 5·6 (4·6) 5·6 (5·1) Continuous inhaled corticosteroids 118 (28%) 66 (26%) 184 (28%) 144 (34%) 69 (27%) 213 (31%) Data are mean (SD) or n (%), unless otherwise indicated

corticosteroids in past year 2·0 (1·9) 1·8 (1·8) 1·9 (1·8) 1·9 (1·9) 1·8 (2·0) 1·9 (2·0) USMA in previous year 5·5 (4·3) 5·4 (4·1) 5·4 (4·2) 5·7 (5·3) 5·6 (4·6) 5·6 (5·1) Continuous inhaled corticosteroids 118 (28%) 66 (26%) 184 (28%) 144 (34%) 69 (27%) 213 (31%) Data are mean (SD) or n (%), unless otherwise indicated . USMA=unscheduled medial attendance for wheeze. URTI=upper-respiratory-tract infection. * A question to parents from the International Study of Asthma and Allergies in Childhood questionnaire was used to identify symptoms suggestive of eczema. † Eczema from birth was based on parental report to recruiting investigator at enrolment. ‡ Based on parental report of the usual interval between URTI and onset of wheezing. Table 2 Treatment response in the primary analysis, and by 5/5 and 5/x+x/x strata Montelukast group (n=652) Placebo group (n=656) Adjusted incidence rate ratio (95% CI) p value pinteracttion Primary analysis USMA episodes 2·0 (2·6) 2·3 (2·7) 0·88 (0·77–1·01) 0·06 .. Subgroup analysis USMA in 5/5 stratum 2·0 (2·7) 2·4 (3·0) 0·80 (0·68–0·95) 0·01 .. USMA in 5/x+x/x stratum 2·0 (2·5) 2·0 (2·3) 1·03 (0·83–1·29) 0·79 0·08 Data are mean (SD), unless otherwise indicated. We obtained primary outcome data from the phone call that took place every 2 months. USMA=unscheduled medial attendance for wheeze. Table 3 Secondary outcomes

Montelukast group (n=652) Placebo group (n=656) Adjusted incidence rate ratio (95% CI) p value pinteracttion Primary analysis USMA episodes 2·0 (2·6) 2·3 (2·7) 0·88 (0·77–1·01) 0·06 .. Subgroup analysis USMA in 5/5 stratum 2·0 (2·7) 2·4 (3·0) 0·80 (0·68–0·95) 0·01 .. USMA in 5/x+x/x stratum 2·0 (2·5) 2·0 (2·3) 1·03 (0·83–1·29) 0·79 0·08 Data are mean (SD), unless otherwise indicated. We obtained primary outcome data from the phone call that took place every 2 months. USMA=unscheduled medial attendance for wheeze. Table 3 Secondary outcomes Montelukast group (n=652) Placebo group (n=656) Point estimate (95% CI) p value Children with one or more USMA 426 (65%) 456 (70%) OR 0·83 (0·66–1·04) 0·10 Time to first USMA (days)* 147 (50–365) 130 (38–)† HR 0·89 (0·78–1·02) 0·09 Need for rescue oral corticosteroids (courses per child)‡ 0·26 (0·7) 0·33 (0·9) IRR 0·75 (0·58–0·98) 0·03 Wheeze episodes‡ 2·7 (2·9) 2·6 (3·0) IRR 1·02 (0·91–1·16) 0·68 Duration of wheeze episodes (days) 5·2 (4·0) 5·4 (3·8) IRR 0·97 (0·89–1·06) 0·53 Duration of hospital admission (days per admission) 1·8 (1·3) 1·7 (1·1) IRR 1·05 (0·94–1·18) 0·40 Symptomatic days per wheeze episode 4·9 (3·5) 4·8 (3·8) IRR 0·96 (0·88–1·05) 0·36 Data are n (%), median (IQR), or mean (SD), unless otherwise indicated. USMA=unscheduled medical attendance for wheeze episodes. OR=odds ratio. HR=hazard ratio. IRR=incidence rate ratio. * Seven participants were missing dates for USMA and seven participants had their first medical attendance on the day of randomisation and were hence excluded.

Montelukast group (n=652) Placebo group (n=656) Point estimate (95% CI) p value Children with one or more USMA 426 (65%) 456 (70%) OR 0·83 (0·66–1·04) 0·10 Time to first USMA (days)* 147 (50–365) 130 (38–)† HR 0·89 (0·78–1·02) 0·09 Need for rescue oral corticosteroids (courses per child)‡ 0·26 (0·7) 0·33 (0·9) IRR 0·75 (0·58–0·98) 0·03 Wheeze episodes‡ 2·7 (2·9) 2·6 (3·0) IRR 1·02 (0·91–1·16) 0·68 Duration of wheeze episodes (days) 5·2 (4·0) 5·4 (3·8) IRR 0·97 (0·89–1·06) 0·53 Duration of hospital admission (days per admission) 1·8 (1·3) 1·7 (1·1) IRR 1·05 (0·94–1·18) 0·40 Symptomatic days per wheeze episode 4·9 (3·5) 4·8 (3·8) IRR 0·96 (0·88–1·05) 0·36 Data are n (%), median (IQR), or mean (SD), unless otherwise indicated. USMA=unscheduled medical attendance for wheeze episodes. OR=odds ratio. HR=hazard ratio. IRR=incidence rate ratio. * Seven participants were missing dates for USMA and seven participants had their first medical attendance on the day of randomisation and were hence excluded. † The 75th percentile could not be calculated for this IQR because more than 25% of children never had a USMA. ‡ Analysis included all children. 446 children had no diary data and these participants were considered to have no wheeze and cold episodes. When the analysis was repeated with these patients treated as missing, there was no difference in the IRR between treatment and placebo. Table 4 Non-serious adverse events

† The 75th percentile could not be calculated for this IQR because more than 25% of children never had a USMA. ‡ Analysis included all children. 446 children had no diary data and these participants were considered to have no wheeze and cold episodes. When the analysis was repeated with these patients treated as missing, there was no difference in the IRR between treatment and placebo. Table 4 Non-serious adverse events Montelukast (n=669) Placebo (n=677) Number of events* 397 543 Participants with events 197 (29%) 235 (35%) Intensity Mild 314 (79%) 426 (78%) Moderate 77 (19%) 108 (20%) Severe 6 (2%) 9 (2%) Minor injury 27 (7%) 22 (4%) Gastrointestinal 86 (22%) 122 (22%) Upper-respiratory-tract infection 73 (18%) 103 (19%) CNS 25 (6%) 46 (8%) Minor infection 87 (22%) 107 (20%) Allergy 16 (4%) 20 (4%) Cutaneous 32 (8%) 54 (10%) Respiratory 34 (9%) 54 (10%) Haematological 5 (1%) 7 (1%) Genitourinary 10 (3%) 6 (1%) Major injury 2 (1%) 1 (<1%) Musculoskeletal 0 1 (<1%) Data are n (%), unless otherwise indicated. See appendix for full details of adverse events. * No adverse events were definitely treatment-related Panel Research in Context Systematic review

Montelukast (n=669) Placebo (n=677) Number of events* 397 543 Participants with events 197 (29%) 235 (35%) Intensity Mild 314 (79%) 426 (78%) Moderate 77 (19%) 108 (20%) Severe 6 (2%) 9 (2%) Minor injury 27 (7%) 22 (4%) Gastrointestinal 86 (22%) 122 (22%) Upper-respiratory-tract infection 73 (18%) 103 (19%) CNS 25 (6%) 46 (8%) Minor infection 87 (22%) 107 (20%) Allergy 16 (4%) 20 (4%) Cutaneous 32 (8%) 54 (10%) Respiratory 34 (9%) 54 (10%) Haematological 5 (1%) 7 (1%) Genitourinary 10 (3%) 6 (1%) Major injury 2 (1%) 1 (<1%) Musculoskeletal 0 1 (<1%) Data are n (%), unless otherwise indicated. See appendix for full details of adverse events. * No adverse events were definitely treatment-related Panel Research in Context Systematic review We did a search between June 30, and July 10, 2014, using the research strategy reported by Ducharme and colleagues.20 We searched Embase, Scopus, Medline, and the Cochrane Airways Group trials register for additional studies between Jan 1, and July 30, 2014, with search terms “wheez* or asthm*”, “preschool* or preschool**”, “randomised or randomized or randomly or trial”, “leukotriene* or anti-leukotriene or antileukotriene or montelukast”. We also included “viralwheeze or viral-wheeze”, “young children and infant”, “intermittent, pre-emptive, and preemptive”. Our search retrieved no additional trials to those previously identified.7–9 Interpretation

We did a search between June 30, and July 10, 2014, using the research strategy reported by Ducharme and colleagues.20 We searched Embase, Scopus, Medline, and the Cochrane Airways Group trials register for additional studies between Jan 1, and July 30, 2014, with search terms “wheez* or asthm*”, “preschool* or preschool**”, “randomised or randomized or randomly or trial”, “leukotriene* or anti-leukotriene or antileukotriene or montelukast”. We also included “viralwheeze or viral-wheeze”, “young children and infant”, “intermittent, pre-emptive, and preemptive”. Our search retrieved no additional trials to those previously identified.7–9 Interpretation Whether intermittent treatment with montelukast is effective for treatment of wheeze in children aged 10 months to 5 years is unclear: one randomised trial7 showed that intermittent montelukast is effective for wheeze in that population, whereas two other trials8,9 reported no benefit. We therefore sought to establish the efficacy of intermittent montelukast in young children with wheeze. Because young children with wheeze exhibit marked heterogeneity in response to montelukast, and in adults, copy numbers of the GGGCGG Sp1 binding motif in the arachidonate 5-lipoxygenase (ALOX5) gene promoter (either 5/5, 5/x, or x/x, in which x does not equal 5) are associated with heterogeneity in montelukast response,14,15 we stratified the trial by 5/5 and 5/x+x/x genotypes. Our findings show that intermittent montelukast is no better than placebo for reducing the need for unscheduled medical attention in young children with a history of clinically severe wheeze. Evidence suggested that children with the 5/5 genotype might be responsive to intermittent montelukast treatment. For clinicians, these data suggest that intermittent montelukast should not be routinely used to treat wheeze in young children. Further data from stratified trials are needed before treatment is targeted to a responsive subgroup.

Introduction Despite advances in the treatment and prevention of infectious diseases, the incidence of sepsis is rising.1–3 Mortality rates for sepsis remain unacceptably high at around 20–30%,2–5 and the effect on health-care expenditure and resource use has been substantial.6,7 Moreover, for those who survive the acute illness, the risk of death is increased for up to 5 years after the septic episode8,9 and quality of life is significantly impaired.10 Attempts to reduce mortality in patients with severe sepsis by modulating the host response have proved disappointing, partly because of poor understanding of the complex mechanisms that regulate innate immunity and the inflammatory cascade.11 Furthermore, such interventions are often delayed, and have usually been applied unselectively to heterogeneous groups of patients, without considering the potential influence of host genetic diversity on response to treatment. Genomics has the potential to substantially advance our understanding of the key biological pathways implicated in human disease, and to suggest new targets for treatment or prevention.12 Additionally, characterisation of genetic variants associated with outcome from sepsis could enable us to identify those at high risk who might benefit from more aggressive interventions or from specific, individually targeted, early, or pre-emptive measures.