Browse the corpus

e paediatric admission record and documentation of a wider range of clinical and demographic data (n=49 demographic, symptom and sign characteristics). To promote informal benchmarking, the adequacy of documentation for the core 16 clinical variables was also summarised and presented in reports that span all hospitals. To illustrate the overall effect, we created an index of missing data based on the 49 required core admission variables (demographic, symptom and sign characteristics) for each case. A similar index was created for the subset of 16 core clinical characteristics specifically included in the feedback reports. We show in figures 2 and 3 below how clinical documentation has improved (missing data have declined) for each hospital over time, including in these figures an indication of the timing of major CIN meetings. With the improvement in data, fuller descriptions of patient populations are now possible and are presented elsewhere.28 Figure 2 Trends of rate of missing data for all core signs and symptoms documented during admission. CIN, Clinical Information Network. Figure 3 Trends of rate of missing data for signs and symptoms documented during admission included in feedback reports to hospitals. CIN, Clinical Information Network.

Key questions What is already known about this topic? Collaborative health information networks have helped improve outcomes of care, accelerated knowledge discovery and advanced cross-domain development of digital architecture to support research in high-income settings. Central to such networks is the collection of standardised data across hospital sites that can be used for tracking or benchmarking performance while promoting the sharing of experiences and innovations to improve care. What are the new findings? Establishing health information networks in low-resource settings has multiple unique challenges that new research needs to address. These challenges include the development of new data collection procedures and new methods to implement the provision of accurate reporting to hospitals. Recommendations for policy This study provides evidence that operationalising clinical information networks in low-income countries can be achieved by addressing: – Technical rules for improving the data quality collected in a resource-limited setting using open source and non-commercial standardised patient data collection tools. – Behavioural rules of collaborative health networks to improve organisational culture to enable new systems for gathering and using information for improving care delivery. Introduction The need to improve healthcare delivery has been highlighted in a number of reports from low and middle-income countries (LMIC),1 2 including Kenya.3–5

– Behavioural rules of collaborative health networks to improve organisational culture to enable new systems for gathering and using information for improving care delivery. Introduction The need to improve healthcare delivery has been highlighted in a number of reports from low and middle-income countries (LMIC),1 2 including Kenya.3–5 The Kenya Medical Research Institute (KEMRI)-Wellcome Trust Research Programme's (KWTRP) Health Services Unit has collaborated with the Kenyan Ministry of Health since 2002 to develop national evidence-based clinical guidelines for paediatric care,6 to conduct implementation research and pragmatic clinical trials,7 8 and to conduct surveys of the quality of care within hospitals.9 On the basis of these experiences and a review of the wider literature,10 a new programme of work was developed to focus on improving the delivery of essential interventions during inpatient paediatric care. Kenya is similar to many LMIC in that hospitals often have no electronic systems for recording the care they provide. This means that in order to improve the delivery of essential interventions, we first need to establish a new method for collecting data on paediatric admissions to Kenyan hospitals. A new partnership between researchers, the Ministry of Health, The Kenyan Paediatric Associated and 14 country (district) level hospitals was formed to create a Clinical Information Network (CIN) to provide an accurate picture of healthcare provision to paediatric inpatients in the participating hospitals.

n hospitals. A new partnership between researchers, the Ministry of Health, The Kenyan Paediatric Associated and 14 country (district) level hospitals was formed to create a Clinical Information Network (CIN) to provide an accurate picture of healthcare provision to paediatric inpatients in the participating hospitals. The CIN follows the approach of other clinical networks that have been a feature of efforts to improve care in high-income (eg, the Northern Neonatal Network,11 the Vermont Oxford Network12) and middle-income (eg, the Child Healthcare Problem Identification Programme13) countries. A network has been described as ‘a grouping that aims to improve clinical care and service delivery using a collegial approach to identify and implement a range of [improvement] strategies’,14 and the CIN follows this approach. More recently, clinical information networks have helped improve outcomes of care,15 accelerated knowledge discovery,16 and advanced cross-domain development of digital architecture to support research.17 Central to such networks is the collection of standardised data across sites that can be used for tracking or benchmarking performance while promoting the sharing of experiences and innovations to improve care. However, there are few published reports of attempts to develop collaborative information networks in LMIC.

7 Central to such networks is the collection of standardised data across sites that can be used for tracking or benchmarking performance while promoting the sharing of experiences and innovations to improve care. However, there are few published reports of attempts to develop collaborative information networks in LMIC. In this paper, we describe the challenges faced by Kenya and other low-income countries with the collection of data on routine care and provide an overview of the approach used to address these challenges in the area of paediatric admissions, the focus of our CIN. We describe how hospitals were provided with routine reports to help improve clinical documentation, and then consider the potential future value of such a network.

f data on routine care and provide an overview of the approach used to address these challenges in the area of paediatric admissions, the focus of our CIN. We describe how hospitals were provided with routine reports to help improve clinical documentation, and then consider the potential future value of such a network. Background Quality is multidimensional and often described as comprising structure (inputs), process (activities) and outcomes.18–20 In recent years, increasing attention has been devoted to assessing the process aspect of delivering quality in healthcare. Optimal processes can be defined by clinical practice standards or summarised as guidelines. These can provide an explicit link between research evidence and practice. It therefore follows that the gap between these standards and the care that is actually delivered provides one measure of quality care: it indicates how successfully (new) interventions are adopted in practice and also whether any benefits from research are realised. Central to many strategies to improve process quality is therefore the ability to measure adherence to guidelines and tracking the progress of such indicators as part of ‘Plan, Do, Study Act’ cycles. However, in low-income settings, routine health information systems often provide data of poor quality,4 21 which preclude their use in such improvement exercises. Specific challenges are listed in the following section.

to guidelines and tracking the progress of such indicators as part of ‘Plan, Do, Study Act’ cycles. However, in low-income settings, routine health information systems often provide data of poor quality,4 21 which preclude their use in such improvement exercises. Specific challenges are listed in the following section. The challenges Poor clinical documentation Inpatient clerking in public district hospitals in LMIC is predominantly paper-based and patients' clinical features are often poorly documented.22 This often makes the subsequent medical records an inadequate source of accurate patient data. Information on patient assessment, investigations carried out and treatment prescribed are also often only partially documented. However, in prior work, the CIN team has been able to develop and implement a medical record tool that enables clinicians to document patient admissions in a standardised fashion, and data on treatments can also be improved through the use of routine treatment charts.7 23

ribed are also often only partially documented. However, in prior work, the CIN team has been able to develop and implement a medical record tool that enables clinicians to document patient admissions in a standardised fashion, and data on treatments can also be improved through the use of routine treatment charts.7 23 Limitations of National Health Information Systems Kenya has an electronic national health data collection system, called DHIS2, that is now in use in many low-income countries.24 Summary data from hospitals are usually collated from paper medical records (which suffer from the issues described above) and entered through a web-portal onto the national DHIS2 system. In current practice, each disease episode is assigned an International Classification of Diseases 10th Edition code and DHIS summary reports are based on these codes rather than on patient counts.25 As a result, a patient with more than one diagnosis contributes more than one disease episode and this makes it hard to disambiguate prevalence rates from DHIS reports by patient count rather than disease episode count. Use of these limited data for basic tasks (eg, tracking patient outcomes) is further hampered by poor standardisation of coding and gaps in reporting data such as whether the patient lived or died.5 22 The lack of information on patients' key symptoms or signs, any investigations used and their results and of how treatments are used makes exploring the process aspects of quality impossible using data collected through the current national Health Information System.

data such as whether the patient lived or died.5 22 The lack of information on patients' key symptoms or signs, any investigations used and their results and of how treatments are used makes exploring the process aspects of quality impossible using data collected through the current national Health Information System. Information culture in hospitals Kenyan hospitals often do not have a culture of using information to systematically improve patient care as the lack of longitudinal data (as described above) means that information is not available to inform efforts at quality improvement audit cycles.26 27 Some sporadic information-gathering exercises are conducted, such as mortality audits, and most health institutions have a process for delivering Continuous Medical Education (CME) to physicians. However, these exercises rarely feed back into process improvement due to the insufficiency and poor quality of the available information, and a lack of subsequent monitoring or evaluation of any possible change in care.22

alth institutions have a process for delivering Continuous Medical Education (CME) to physicians. However, these exercises rarely feed back into process improvement due to the insufficiency and poor quality of the available information, and a lack of subsequent monitoring or evaluation of any possible change in care.22 The CIN therefore initially set out to overcome these challenges and produce high-quality process and outcome data from individual admissions to paediatric wards in Kenyan hospitals as a prelude to using these data to inform improvement strategies. Our initial focus was on improving information on the most common childhood illnesses in Kenya, which account for up to 80% of all admission episodes in many African countries and the CIN.28 Quality of care indicators for these common illnesses have previously been identified through an international and national Delphi exercise linked to standards encompassed in the WHO and Kenyan paediatric guidelines.29 30 These indicators have been successfully used in previous assessments of the quality of paediatric inpatient care.3–5 9 31 Our strategies for tackling the challenges of enabling routine measurement of such quality indicators are outlined in the next section.

encompassed in the WHO and Kenyan paediatric guidelines.29 30 These indicators have been successfully used in previous assessments of the quality of paediatric inpatient care.3–5 9 31 Our strategies for tackling the challenges of enabling routine measurement of such quality indicators are outlined in the next section. Data quality improvement strategies Improving routine clinical documentation To facilitate improved clinical documentation, hospitals were encouraged to promote good prescribing practices and to implement both more formal discharge forms and a standard paediatric admission record.23 Much of the focus of initial data use was to provide feedback to hospitals on the quality of their clinical documentation. This anticipates improvements from network activities, which have included feedback and mentorship through telephone calls and 4 monthly face-to-face meetings.32 33

d paediatric admission record.23 Much of the focus of initial data use was to provide feedback to hospitals on the quality of their clinical documentation. This anticipates improvements from network activities, which have included feedback and mentorship through telephone calls and 4 monthly face-to-face meetings.32 33 The informatics framework Data capture in CIN hospitals happens at the point of patient discharge where data from the paediatric inpatient paper records are abstracted directly into a non-commercial electronic tool, REDCap.34 A minimum data set required for the national reporting system (DHISv224) is collected on all patients admitted to the paediatric wards for all sites. Comprehensive data for all admissions aged 1 month or more without burns or a surgical diagnosis to the paediatric ward(s) are entered in 12 hospitals and, because of the high workload, on a random selection of records in 2 hospitals (35% and 70% records). The comprehensive data comprise clinical, investigation and treatment data focused on admission events and then discharge data with up to 350 variables per patient encounter. As is summarised in figure 1, data are collected by trained clerks35 and preprogrammed field validation rules in the REDCap tool are used to check data quality as it is entered. All data subsequently shared with the central network analysis team are de-identified. R (R Core Team, R: A Language and Environment for Statistical Computing. 2014: Vienna, Austria) statistical software has been installed on hospital sites' computers and, through a process of meta-programming (writing code that writes itself during runtime based on predefined clinical guidelines22 23 29), R software autogenerates code that is used for running on-site checks daily. It then also cleans and recodes data to enable indicator measurement and reporting. These R resources are available for reuse in other projects.36 36 A detailed report of CIN's data management framework is described elsewhere.35

ines22 23 29), R software autogenerates code that is used for running on-site checks daily. It then also cleans and recodes data to enable indicator measurement and reporting. These R resources are available for reuse in other projects.36 36 A detailed report of CIN's data management framework is described elsewhere.35 Figure 1 Informatics infrastructure framework to support data use. KEMRI, Kenya Medical Research Institute. Results of assessment Data use The auto-generated R scripts are used to prepare reports for each hospital on a 2–3 monthly basis. Additionally, a combined hospital report is generated for the Ministry of Health in Kenya using cleaned datasets from all hospitals. For each hospital, data were initially used to provide feedback on the completeness of documentation of a set of 16 core symptoms and signs on admission. Over the first 2 years of operation, reports have been delivered to CIN hospitals on eight occasions. There have also been three face-to-face CIN meetings with paediatricians that included, on two occasions, senior nurses and health record offices. These reports and meetings were supplemented with telephone discussions with paediatricians every 2–4 weeks that promoted better use of the paediatric admission record and documentation of a wider range of clinical and demographic data (n=49 demographic, symptom and sign characteristics). To promote informal benchmarking, the adequacy of documentation for the core 16 clinical variables was also summarised and presented in reports that span all hospitals.

To illustrate the overall effect, we created an index of missing data based on the 49 required core admission variables (demographic, symptom and sign characteristics) for each case. A similar index was created for the subset of 16 core clinical characteristics specifically included in the feedback reports. We show in figures 2 and 3 below how clinical documentation has improved (missing data have declined) for each hospital over time, including in these figures an indication of the timing of major CIN meetings. With the improvement in data, fuller descriptions of patient populations are now possible and are presented elsewhere.28 Figure 2 Trends of rate of missing data for all core signs and symptoms documented during admission. CIN, Clinical Information Network. Figure 3 Trends of rate of missing data for signs and symptoms documented during admission included in feedback reports to hospitals. CIN, Clinical Information Network. Such feedback reports and participation in the network have prompted greater adoption and use of the standard paediatric admission record form and, consequently, overall improvements in documentation of clinical characteristics. Plotting the median value of the missing data index for each case record for the broad set of demographic and clinical characteristics and the core set of clinical features suggests that those items that are directly the subject of feedback have shown greater improvement, although there is improvement for all aspects of documentation (figure 4). In a specific example, the recording of the presence or absence of the Alert, Verbal response, Pain, Unresponsive (AVPU) danger signs and ability to drink has improved from 64% in all admissions in the first 3 months each hospital joined the network to 95% in the most recent 3-month period.

of documentation (figure 4). In a specific example, the recording of the presence or absence of the Alert, Verbal response, Pain, Unresponsive (AVPU) danger signs and ability to drink has improved from 64% in all admissions in the first 3 months each hospital joined the network to 95% in the most recent 3-month period. Figure 4 Median rate per month of missing data comparing documentation of items included in feedback reports versus all signs and symptoms collected at admission. CIN, Clinical Information Network. Discussion A community of practice The CIN has been relatively successful in creating an opportunity for frontline caregivers, health researchers and informatics specialists to learn as a community to improve availability of clinical data and begin to promote their use. The hospitals in the network have begun supplying and promoting the use of more structured medical records. This has been helped, we believe, by slowly changing the hospital culture through sustained engagement and by providing peer support by linking hospitals within the network.9 37 In this way, new staff quickly become familiar with the clinical forms and are integrated into thinking about data-informed quality improvement efforts at the hospital level, something that is not routine.37 38 This is especially important in low-income countries as clinical staff in training programmes (who often are the ones admitting patients) rotate through different hospitals on a 3-monthly basis.

ated into thinking about data-informed quality improvement efforts at the hospital level, something that is not routine.37 38 This is especially important in low-income countries as clinical staff in training programmes (who often are the ones admitting patients) rotate through different hospitals on a 3-monthly basis. Clinical teams may feel criticised if key indicators show poorer performance than they had been anticipating. The efforts made to adopt an inclusive, facilitative and supportive way of using data have resulted in refinements to indicators that better reflect practice and have built trust in the results. A supportive rather than regulatory approach that appreciates challenges to improvement efforts (eg, lack of Mid-Upper Arm Circumference (MUAC) tapes or pulse oximeters), coupled with face-to-face meetings, is encouraging growing ownership of the data by the clinical teams.

practice and have built trust in the results. A supportive rather than regulatory approach that appreciates challenges to improvement efforts (eg, lack of Mid-Upper Arm Circumference (MUAC) tapes or pulse oximeters), coupled with face-to-face meetings, is encouraging growing ownership of the data by the clinical teams. Digital architecture and links to quality improvement The CIN collated anonymised data on over 65 000 admissions in its first 2 years of operations. It is producing comprehensive clinical paediatric data, which are of moderately good quality and are trackable. This provides opportunities for exploring the use and value of these data as part of CIN's longer term aims to improve care. A full account of CIN's data management framework is provided elsewhere,35 but the focus on using non-commercial or open-source software provides future opportunities for sharing all tools, standard operating procedures and approaches to analysis. At each hospital, only one personal computer, an internet link and a clerk are required, supported by a centralised data management and analysis team working with paediatricians.

ercial or open-source software provides future opportunities for sharing all tools, standard operating procedures and approaches to analysis. At each hospital, only one personal computer, an internet link and a clerk are required, supported by a centralised data management and analysis team working with paediatricians. The data sharing approach and work to automate production of CIN routine reports means they can be fed back to the hospital management and clinical teams in CIN hospitals as documents and presentations with discussion facilitated by telephone, social networks and occasional face-to-face meetings of network partners. The focus can be put on key indicators that show poor performance in the hospitals and possible interventions suggested, implemented and tested to try to improve clinical performance. For example, the continuing poor documentation of ‘ability to drink’ prompted an exploration of why this occurred in some hospitals when in other hospitals it had improved. A lack of recognition of the value of this sign and limited local supervisory attention were identified as contributory factors.

nical performance. For example, the continuing poor documentation of ‘ability to drink’ prompted an exploration of why this occurred in some hospitals when in other hospitals it had improved. A lack of recognition of the value of this sign and limited local supervisory attention were identified as contributory factors. Promoting learning The aim of the CIN is to evaluate common clinical practices and to support the local team take on the responsibility of developing strategies for tackling any deficiencies based on an understanding of the specific hospital context.5 The approach thus draws on principles that underlie successful improvement collaboratives. Such collaboratives require data, the primary focus of our initial work. However, the CIN could also support broader learning aims outside the immediate network if a common data framework was adopted across hospitals. This would allow variability in and associations with mortality to be examined and more detailed audit approaches to be added as have been successful in South Africa.13 Potentially, such data might be used to track adoption of interventions and their effects over time at scale. One example would be examining diarrhoea/dehydration admissions after introduction of rotavirus vaccination. More specifically, organised networks may contribute to the more efficient conduct of pragmatic trials.39 This could help reduce the duration and costs and help enable more rapid translation of research into practice. In other areas, work within the CIN could explore different theory-driven feedback approaches to determine which might be best used to change behaviour. All such learning can feed in at policy level to help develop wider monitoring and evaluation linked to efforts to improve quality and health information systems.

ice. In other areas, work within the CIN could explore different theory-driven feedback approaches to determine which might be best used to change behaviour. All such learning can feed in at policy level to help develop wider monitoring and evaluation linked to efforts to improve quality and health information systems. Conclusions The work undertaken to date within the CIN has demonstrated that although electronic medical records spanning inpatient care are yet to be deployed in Kenyan wards, it is possible to produce standardised data from multiple sites and improve their quality through partnerships with hospital teams. This has been achieved using low-cost software and innovative adaptations by a local but centralised informatics team working closely with clinicians. These data are used to create timely reports for hospitals that have traditionally had no access to routine information that includes process and outcomes for their patients. Having established this platform, the CIN is now able to begin work with all partners to improve the quality of care and to develop an appreciation of the importance of good information and longer term learning strategies. Correction notice: This article has been corrected since it was first published. The abstract has been included. Handling editor: Soumitra Bhuyan.

Conclusions The work undertaken to date within the CIN has demonstrated that although electronic medical records spanning inpatient care are yet to be deployed in Kenyan wards, it is possible to produce standardised data from multiple sites and improve their quality through partnerships with hospital teams. This has been achieved using low-cost software and innovative adaptations by a local but centralised informatics team working closely with clinicians. These data are used to create timely reports for hospitals that have traditionally had no access to routine information that includes process and outcomes for their patients. Having established this platform, the CIN is now able to begin work with all partners to improve the quality of care and to develop an appreciation of the importance of good information and longer term learning strategies. Correction notice: This article has been corrected since it was first published. The abstract has been included. Handling editor: Soumitra Bhuyan. The authors would like to thank the Ministry of Health who gave permission for this work to be developed and have supported the implementation of the CIN together with the county health executives and all hospital management teams. Collaboration with officers from the Ministry of Health's national Health Management-Information System, the Monitoring and Evaluation Unit and the Maternal, Neonatal, Child and Adolescent Health Unit has been important to the initiation of the CIN. The authors are grateful to the Kenya Paediatric Association for promoting the aims of the CIN and for the support they provide through their officers and membership. They also thank the hospital paediatricians and clinical teams on all the paediatric wards who provide care to the children for whom this project is designed. This work is also published with the permission of the Director of KEMRI.

aims of the CIN and for the support they provide through their officers and membership. They also thank the hospital paediatricians and clinical teams on all the paediatric wards who provide care to the children for whom this project is designed. This work is also published with the permission of the Director of KEMRI. Collaborators: TT, MB, LM, NM, DG, SG, GM, GI, WN, MO, TJ and ME developed the CIN approach, database tools and analytical approaches and supported hospital feedback. The CIN authors who contributed to the conduct of the work, collection of data, data quality assurance and development of reporting frameworks include: Rachel Nyamai (Ministry of Health), Fred Were (University of Nairobi),Philip Ayieko (KWTRP), John Chengondu (Vihiga County Hospital), Ernest Namayi (Mbale Rural Health and Demonstration Centre), Josephat Shikokoti (Kakamega Provincial General Hospital), Joseph Nganga (Mbagathi District Hospital), Edward Ngugi (Kiambu County Hospital), Esther Mwangi (Mama Lucy Kibaki Hospital), Caroline Mwari, Esther Muthiani (Machakos Level 5 Hospital), Julian M. Ndungu (Nyeri Level 5 hospital), Rinnie Juma (Kisumu East District Hospital), Pauline W. Njeru (Embu Provincial General Hospital), Consolata Kinyua (Karatina District Hospital), Mary Nguri (Kerugoya District Hospital) and Jane Mokua (Kitale District Hospital). TT, MB, CP and ME helped draft the initial report and all authors contributed to its development and approved the final version.

spital), Pauline W. Njeru (Embu Provincial General Hospital), Consolata Kinyua (Karatina District Hospital), Mary Nguri (Kerugoya District Hospital) and Jane Mokua (Kitale District Hospital). TT, MB, CP and ME helped draft the initial report and all authors contributed to its development and approved the final version. Funding: Funds from The Wellcome Trust (#097170) awarded to ME as a Senior Fellowship and (#106823) awarded to TT as a Master's Fellowship together with additional funds from a Wellcome Trust core grant awarded to the KEMRI-Wellcome Trust Research Programme (#092654) supported this work. Competing interests: None declared. Ethics approval: KEMRI—Ethical Review Committee. Provenance and peer review: Not commissioned; externally peer reviewed. Data sharing statement: No additional data are available.

Key questions What is already known about this topic? There is a global shortage of health professionals, particularly in sub-Saharan Africa; associate clinicians (ACs; clinical officers, in Malawi; previously non-physician clinicians) are in the frontline of healthcare in countries like Malawi. Countries, like Malawi, while making some advances, are struggling to achieve the United Nations Millennium Developmental Goals for child mortality and maternal health (MDG 4 and 5). In 2007, the Malawian Ministry of Health, recognising some of these issues, set out a National Road Map for accelerating the reduction of maternal and child mortality, and achieving MGD 5. Up-skilling ACs may provide a solution to some of these issues. What are the new findings? This is one of the first trials taking an in-depth look at the impact on health outcomes in districts across central and northern Malawi with a programme of knowledge, skills and clinical leadership training for ACs. The trial's aim is to see if the up-skilling of this important cadre of health workers can impact on district maternal mortality rates, perinatal mortality rates, and key obstetric and birth complications. The results show some very positive trends. It is our belief that as the trainees share their new skills and knowledge, this positive impact will grow.

What are the new findings? This is one of the first trials taking an in-depth look at the impact on health outcomes in districts across central and northern Malawi with a programme of knowledge, skills and clinical leadership training for ACs. The trial's aim is to see if the up-skilling of this important cadre of health workers can impact on district maternal mortality rates, perinatal mortality rates, and key obstetric and birth complications. The results show some very positive trends. It is our belief that as the trainees share their new skills and knowledge, this positive impact will grow. Recommendations for policy This cadre is an important component in helping to relieve the chronic shortages of trained medical professionals in sub-Saharan Africa, and for helping countries move towards realisation of the millennium development goals. Further evaluations of the up-skilling of this cadre are needed.

ing from a lower baseline, the maternal mortality ratio increases throughout the trial period. Indeed, the absolute ratio in the intervention and control districts by the final quarter of 2013 were 90.23, a reduction of 61.01 points from intervention start and 139.03, an increase of 33.6 points from intervention start. Table 4 shows the actual mortality rates/ratios over the lifetime of the project. In PNM rates, there was an improvement of 32% in the intervention districts compared with only 30% in the control, with intervention districts having consistently lower PNM rates than control. The rate of early neonatal deaths, a key component of PNM, decreases over the 3 years of the project in the intervention districts (from 5.8 in 2011 to 4 in 2013, per 1000 births) by 31%; in the control districts, the rate is higher overall and has increased by 2% in 2013 when compared with 2011 (see table 4). Fresh stillbirth rates reduce by nearly 46% in the control districts and by only 32% in the intervention districts. Maternal mortality in the intervention districts reduces by 40%; however, in the control districts, the ratio increases by 31% by the end of the project. Table 4 District facility-based mortality ratios/rates per year for 2011–2013

It is our belief that as the trainees share their new skills and knowledge, this positive impact will grow. Recommendations for policy This cadre is an important component in helping to relieve the chronic shortages of trained medical professionals in sub-Saharan Africa, and for helping countries move towards realisation of the millennium development goals. Further evaluations of the up-skilling of this cadre are needed. Background Many African countries, like Malawi, have a cadre of health workers called associate clinicians (ACs), previously non-physician clinicians, who are trained in basic medical diagnosis and treatment.1 ACs are often the most experienced health worker in hospitals across the country. Many of these ACs specialise in emergency obstetric and neonatal care (EmONC), and are in the frontline providing care for mothers and babies. The value of ACs cannot be understated; it will take many more years before countries like Malawi and most countries in sub-Saharan Africa have enough doctors. ACs, known as clinical officers in Malawi, are providing essential, valuable, safe and effective EmONC services across sub-Saharan African countries.1–6 Recent work has shown that while they are a valuable resource, they often feel undervalued and undersupported, and this has an impact on their performance and retention.7 8 The WHO recognises the important role this cadre of health workers can play in maternal and newborn healthcare, and has made recommendations so that their role can be elevated.9 Major surveys consistently show that extra training and support can improve task shifting, and reduce maternal and neonatal mortality and morbidity in the areas where extra training and support have been piloted.3 4 6 Training skilled attendants to prevent, detect and manage major obstetric complications, including undertaking emergency caesarean surgery in complicated deliveries, is arguably the single most important factor in preventing maternal deaths and protecting the human rights of women.2–4 6 To be effective, ACs need the appropriate knowledge, skills, equipment, drugs and the technology essential for managing obstetric complications in rural or deprived communities.

plicated deliveries, is arguably the single most important factor in preventing maternal deaths and protecting the human rights of women.2–4 6 To be effective, ACs need the appropriate knowledge, skills, equipment, drugs and the technology essential for managing obstetric complications in rural or deprived communities. In 2011, the European Commission FP7 funded the enhancing of human resources and use of appropriate technologies for maternal and perinatal survival in sub-Saharan Africa (ETATMBA, Enhancing Training And Technology for Mothers and Babies in Africa) project. The project set out to implement and evaluate a programme of locally based clinical service improvement in Tanzania and Malawi. A cohort of EmONC ACs (in both countries) were provided with a programme of evidenced-based skills, knowledge and clinical leadership training delivered by European and African specialist clinicians and academicians. Trainees had links to specialist support and in Malawi, two UK-based obstetricians, at specialist registrar level with 5 years of clinical experience, worked alongside the ACs for 2 weeks in their districts, providing peer support and sharing of skills and knowledge.10 Here we report on the evaluation of the impact of the training in Malawi. The main aim is to evaluate the impact of the ETATMBA training on health outcomes (maternal and perinatal morbidity and mortality) by comparing districts where trainees were based with districts where there were no ETATMBA trainees.

In 2011, the European Commission FP7 funded the enhancing of human resources and use of appropriate technologies for maternal and perinatal survival in sub-Saharan Africa (ETATMBA, Enhancing Training And Technology for Mothers and Babies in Africa) project. The project set out to implement and evaluate a programme of locally based clinical service improvement in Tanzania and Malawi. A cohort of EmONC ACs (in both countries) were provided with a programme of evidenced-based skills, knowledge and clinical leadership training delivered by European and African specialist clinicians and academicians. Trainees had links to specialist support and in Malawi, two UK-based obstetricians, at specialist registrar level with 5 years of clinical experience, worked alongside the ACs for 2 weeks in their districts, providing peer support and sharing of skills and knowledge.10 Here we report on the evaluation of the impact of the training in Malawi. The main aim is to evaluate the impact of the ETATMBA training on health outcomes (maternal and perinatal morbidity and mortality) by comparing districts where trainees were based with districts where there were no ETATMBA trainees. Methods Trial design A cluster randomised controlled trial (RCT) with a 1:1 allocation. However, for the purpose of analyses, we adopted a quasiexperimental research design, as we were using district population-level data, in order to capture the longitudinal effects of the intervention through regression modelling. The main advantage of this approach is that it makes full use of the longitudinal nature of the data, and accounts for preintervention trends.11 12

xperimental research design, as we were using district population-level data, in order to capture the longitudinal effects of the intervention through regression modelling. The main advantage of this approach is that it makes full use of the longitudinal nature of the data, and accounts for preintervention trends.11 12 Study setting The study was conducted in districts within the central and northern regions of Malawi. There were a total of 14 districts (clusters) in these regions which were randomised to either intervention districts (where ETATMBA trainees were based) or control districts. Participants There are no ‘participants’ per se in this study; the trainees who were recruited in the district are in reality the intervention group.

Study setting The study was conducted in districts within the central and northern regions of Malawi. There were a total of 14 districts (clusters) in these regions which were randomised to either intervention districts (where ETATMBA trainees were based) or control districts. Participants There are no ‘participants’ per se in this study; the trainees who were recruited in the district are in reality the intervention group. Intervention Following randomisation, ACs within the intervention districts were invited to enrol on the ETATMBA training programme. The training was delivered between late 2011 and early 2014, and consisted of eight modules with mentoring and support offered between modules. The training was accredited at undergraduate degree level, and was specifically aimed at improving skills and knowledge related to EmONC and neonatal care, and had a strong clinical leadership component. Figure 1 outlines the training modules and their timeline. More detailed information on recruitment and modules is available elsewhere,10 13 in the online supplementary appendix. Fifty-four trainees were included in the study, and they were based in the eight intervention districts. In these districts, a minimum of two ETATMBA trainees were included. The control districts received no ETATMBA trainees. 10.1136/bmjgh-2015-000020.supp1supplementary appendix Figure 1 An overview of the ETATMBA training modules and when these were delivered. ETATMBA, Enhancing Training And Technology for Mothers and Babies in Africa; NPC, non-physician clinician.

Intervention Following randomisation, ACs within the intervention districts were invited to enrol on the ETATMBA training programme. The training was delivered between late 2011 and early 2014, and consisted of eight modules with mentoring and support offered between modules. The training was accredited at undergraduate degree level, and was specifically aimed at improving skills and knowledge related to EmONC and neonatal care, and had a strong clinical leadership component. Figure 1 outlines the training modules and their timeline. More detailed information on recruitment and modules is available elsewhere,10 13 in the online supplementary appendix. Fifty-four trainees were included in the study, and they were based in the eight intervention districts. In these districts, a minimum of two ETATMBA trainees were included. The control districts received no ETATMBA trainees. 10.1136/bmjgh-2015-000020.supp1supplementary appendix Figure 1 An overview of the ETATMBA training modules and when these were delivered. ETATMBA, Enhancing Training And Technology for Mothers and Babies in Africa; NPC, non-physician clinician. Outcomes Facility-based perinatal mortality (PNM; pragmatically defined in this study as fresh stillbirths and neonatal deaths before discharge from the healthcare facility) is our primary outcome, and our secondary outcomes were facility-based maternal mortality, neonatal mortality, stillbirths, postpartum haemorrhage, caesarean section, eclampsia and sepsis. All of these outcomes were the numbers of events as recorded in maternity records by clinical staff within the health facilities. Each facility within a district records birth events in a maternity record book. This book is collated monthly and summarised, presenting a sum total of each of the variables from the book (eg, number of births, number of women with pre-eclampsia, number of stillbirths (macerated and fresh), etc) and sent to a central point within the district for reporting to the Ministry of Health. It is from these summary records that we gathered our data from each of the districts.

each of the variables from the book (eg, number of births, number of women with pre-eclampsia, number of stillbirths (macerated and fresh), etc) and sent to a central point within the district for reporting to the Ministry of Health. It is from these summary records that we gathered our data from each of the districts. Power calculation In this study, we based our power calculation on a neonatal mortality rate of 30 per 1000 live births.14 The study was powered to detect a 20% difference between intervention and control districts in PNM rate (neonate survival until discharge from facility). The study has 0.80 power to detect the 20% difference with an α of 0.05. This was based on each of the trainees being exposed to 700 birth events in each of the eight intervention districts. Randomisation Using data from a 2011 Republic of Malawi Ministry of Health report,15 the study statistician looked at each of the 14 districts in terms of maternal deaths, stillbirths and neonatal deaths per 1000 (population), ranking each district for each variable. Summing these three ranks gave a score for each of the 14 districts. The 14 districts were then placed in two strata based on a median split of the ranked score (of which there were 7 in each). The strata represent high and low ranked districts. Using a random number generator in STATA software, four districts were randomised to the intervention from each strata.

ch of the 14 districts. The 14 districts were then placed in two strata based on a median split of the ranked score (of which there were 7 in each). The strata represent high and low ranked districts. Using a random number generator in STATA software, four districts were randomised to the intervention from each strata. Data analysis Descriptive statistics were generated for all variables, and PNM rates (per 1000 live births) and maternal mortality ratios (per 100 000 live births) were calculated (MS Excel and SPSS V.22) based on the number of birth events.

ch of the 14 districts. The 14 districts were then placed in two strata based on a median split of the ranked score (of which there were 7 in each). The strata represent high and low ranked districts. Using a random number generator in STATA software, four districts were randomised to the intervention from each strata. Data analysis Descriptive statistics were generated for all variables, and PNM rates (per 1000 live births) and maternal mortality ratios (per 100 000 live births) were calculated (MS Excel and SPSS V.22) based on the number of birth events. The primary outcome, PNM rates and maternal mortality ratios were examined using an interrupted time series (ITS). The time series looked at quarterly periods across the 3 years with the intervention (interruption) being introduced in the first quarter of 2012 (giving a preinterruption slope of 12 months prior to any training/exposure). The ITS, in our case, is a robust technique having the ability to evaluate both intended and unintended consequences of interventions, such as significant challenges, not least the confounding influence of training programmes in control districts that might have affected the intervention. The ITS (statistical comparison of time trends preintervention and postintervention) was carried out using SPSS (V.22). Autoregressive integrated moving average models were generated for the primary variable and maternal mortality ratios for intervention and control districts. Effects are reported from the slope of the regression line preintervention and postintervention (overall), and at 3-month intervals for 21 months.11 12 CIs (95%) are calculated for all effects. In addition, to aid the comparison between intervention and control districts, percentages of the absolute effects are calculated as ‘relative effects’, that is, 100×(actual effect)/(predicted effect)−(actual effect).

tion (overall), and at 3-month intervals for 21 months.11 12 CIs (95%) are calculated for all effects. In addition, to aid the comparison between intervention and control districts, percentages of the absolute effects are calculated as ‘relative effects’, that is, 100×(actual effect)/(predicted effect)−(actual effect). Data are presented as tables, figures or charts, as appropriate. Data were transcribed from records held at the district hospitals. Data represented the quarterly figures for a particular district and in total, we collected data for three whole years (2011–2013 inclusive). Quarterly data from January 2011 to January 2012 (five quarters) represents the pre-ETATMBA training period. The remaining seven quarters up to December 2013 was the follow-up period. The outcomes chosen were linked to elements of the training provided to the ACs, and are data that are routinely recorded and stored. Deviations from original protocol Owing to unforeseen circumstances, we needed to adjust our protocol slightly once we started the study. We have, within this paper, clarified our power calculation and randomisation as there was some confusion with the original protocol. While our pragmatic decision was to split Lilongwe into two because of its size, in reality this proved impossible; hence, we ended up with eight intervention districts (Lilongwe, Nkhotakota, Ntcheu, Chitipa, Karonga, Mzimba/Mzuzu, Kasungu and Rumphi) and six control districts (Dedza, Dowa, Mchinji, Ntchisi, Salima and Nkhata Bay).

gmatic decision was to split Lilongwe into two because of its size, in reality this proved impossible; hence, we ended up with eight intervention districts (Lilongwe, Nkhotakota, Ntcheu, Chitipa, Karonga, Mzimba/Mzuzu, Kasungu and Rumphi) and six control districts (Dedza, Dowa, Mchinji, Ntchisi, Salima and Nkhata Bay). We stated in our protocol that primary data would be extracted from the maternity logs (Malawi Ministry of Health Maternity Register, V.2 (July 2008)) at the district hospital and also the summary data for all other facilities within the district, which were also held there. Data were collected at three points in time from all districts.16 Local (project-related) and national (Government-related) resource issues early in the project period forced us to change a number of things. First, it was impractical within our resource constraints to extract data directly from the registers; therefore, we collected the monthly or quarterly summary data for the whole district from the district hospital. Second, our data collection was carried out in two rather than the three visits that was originally planned. Ethics The study was reviewed and approved by the Biomedical Research Ethics Committee (BREC) at the University of Warwick, UK (143/09/2011), and the College of Medicine Research Ethics Committee (COMREC), Malawi (P.07/11/1102). It had also the approval and support of the Malawi Ministry of Health.

We stated in our protocol that primary data would be extracted from the maternity logs (Malawi Ministry of Health Maternity Register, V.2 (July 2008)) at the district hospital and also the summary data for all other facilities within the district, which were also held there. Data were collected at three points in time from all districts.16 Local (project-related) and national (Government-related) resource issues early in the project period forced us to change a number of things. First, it was impractical within our resource constraints to extract data directly from the registers; therefore, we collected the monthly or quarterly summary data for the whole district from the district hospital. Second, our data collection was carried out in two rather than the three visits that was originally planned. Ethics The study was reviewed and approved by the Biomedical Research Ethics Committee (BREC) at the University of Warwick, UK (143/09/2011), and the College of Medicine Research Ethics Committee (COMREC), Malawi (P.07/11/1102). It had also the approval and support of the Malawi Ministry of Health. Role of the funding source The funders of this study had no input into the design and delivery of the programme and were not involved in any way with the studies data and its analysis.

Ethics The study was reviewed and approved by the Biomedical Research Ethics Committee (BREC) at the University of Warwick, UK (143/09/2011), and the College of Medicine Research Ethics Committee (COMREC), Malawi (P.07/11/1102). It had also the approval and support of the Malawi Ministry of Health. Role of the funding source The funders of this study had no input into the design and delivery of the programme and were not involved in any way with the studies data and its analysis. Results Table 1 shows the total birth events for each district, and overall for the intervention and control districts. Here we see that the intervention districts had almost twice as many birth events when compared with the control over the study period (eg, in 2013, 155 425 compared with 79 437). There were missing data from one district and for reasons unknown, in 2011, Dowa appears to have 8000 more births (these data were checked and verified). Table 1 Total birth events within health facilities over the study period 2011–2013 by district

Results Table 1 shows the total birth events for each district, and overall for the intervention and control districts. Here we see that the intervention districts had almost twice as many birth events when compared with the control over the study period (eg, in 2013, 155 425 compared with 79 437). There were missing data from one district and for reasons unknown, in 2011, Dowa appears to have 8000 more births (these data were checked and verified). Table 1 Total birth events within health facilities over the study period 2011–2013 by district Districts Number of health facilities* I/C† Total births (n) 2011 2012 2013 Chitipa 9 I 7186 8308 8173 Karonga 15 I 7018 6088 8240 Kasungu 21 I 14 190 14 480 17 761 Mzimba 40 I 28 095 29 198 28 800 Ntcheu 22 I 18 732 17 290 18 245 Rumphi 15 I 8316 7732 8191 Nkhotakota 17 I 9940 9966 10 031 Lilongwe 54 I 53 810 55 922 55 984 Dedza 27 C 21 627 22 685 23 501 Dowa 20 C 20 417 12 647 12 986 Mchinji 13 C 19 373 20 486 17 512 Nkhata bay 17 C 6193 5893 6223 Ntchisi 10 C 5855 5986 6920 Salima 14 C MD 11 148 12 295 Intervention (I) 147 287 148 984 155 425 Control (C) 73 465 78 845 79 437 *This is the number of health facilities that are included in the data for each of the districts. †C, control districts; I, intervention districts. MD, missing data. Below we present the ITS analyses which explores the two primary mortality figures in more detail followed by the actual mortality figures, and the key obstetric and birth variables. A full data set, of all variables, broken down to individual districts is provided as an online supplementary appendix.

†C, control districts; I, intervention districts. MD, missing data. Below we present the ITS analyses which explores the two primary mortality figures in more detail followed by the actual mortality figures, and the key obstetric and birth variables. A full data set, of all variables, broken down to individual districts is provided as an online supplementary appendix. Facility-based PNM (ITS) Figure 2 and table 2 show the results for the ITS of PNM rates, our primary outcome. For the first quarter of 2011 (first data point) in the intervention and control districts, rates were 21.12 and 27.35 (per 1000 births), respectively. The rates reduce by 0.407 and 0.0966 points, respectively, per quarter prior to the intervention (at the end of the fourth quarter of 2011). When this trend is taken into account in the ITS analyses, it is uncertain what, if any, impact the ETATMBA training has had on the districts PNM rate. There is a consistent downward trend in the intervention and control districts throughout the follow-up period. However, the decline in the rate in the control district, which starts from a higher point than the intervention districts, is significantly better at p=0.05 at all points. Table 2 Effects from the ITS models for district facility-based perinatal mortality rates, comparing intervention with control

Facility-based PNM (ITS) Figure 2 and table 2 show the results for the ITS of PNM rates, our primary outcome. For the first quarter of 2011 (first data point) in the intervention and control districts, rates were 21.12 and 27.35 (per 1000 births), respectively. The rates reduce by 0.407 and 0.0966 points, respectively, per quarter prior to the intervention (at the end of the fourth quarter of 2011). When this trend is taken into account in the ITS analyses, it is uncertain what, if any, impact the ETATMBA training has had on the districts PNM rate. There is a consistent downward trend in the intervention and control districts throughout the follow-up period. However, the decline in the rate in the control district, which starts from a higher point than the intervention districts, is significantly better at p=0.05 at all points. Table 2 Effects from the ITS models for district facility-based perinatal mortality rates, comparing intervention with control Intervention Control Effect SE CI 95% p Value Relative effect (%) Effect SE CI 95% p Value Relative effect (%) 3 months −1.36 1.18 (−4.16 to 1.43) 0.286 −7 −9.14 2.98 (−16.18 to −2.09) 0.018 −27 6 months −1.73 1.36 (−4.94 to 148) 0.243 −9 −10.58 3.45 (−18.73 to −2.43) 0.018 −34 9 months −2.10 1.58 (−5.83 to 1.63) 0.224 −12 −12.02 4.03 (−21.54 to −2.50) 0.020 −36 12 months −2.47 1.83 (−6.79 to 1.84) 0.218 −14 −13.46 4.68 (−24.52 to −2.41) 0.024 −38 15 months −2.84 2.09 (−7.78 to 2.10) 0.216 −16 −14.91 5.37 (−27.60 to −2.21) 0.027 −42 18 months −3.21 2.37 (−8.81 to 2.38) 0.217 −19 −16.35 6.09 (−30.75 to −1.95) 0.031 −46 21 months −3.58 2.65 (−9.85 to 2.69) 0.219 −21 −17.79 6.83 (−33.95 to −1.64) 0.035 −47 Effect—estimate of effect from ARIMA ITS model.

) 0.024 −38 15 months −2.84 2.09 (−7.78 to 2.10) 0.216 −16 −14.91 5.37 (−27.60 to −2.21) 0.027 −42 18 months −3.21 2.37 (−8.81 to 2.38) 0.217 −19 −16.35 6.09 (−30.75 to −1.95) 0.031 −46 21 months −3.58 2.65 (−9.85 to 2.69) 0.219 −21 −17.79 6.83 (−33.95 to −1.64) 0.035 −47 Effect—estimate of effect from ARIMA ITS model. Relative effect, percentage change (compared with preslope trend). ITS, interrupted time series. Figure 2 Interrupted time series: district health facility perinatal mortality rate (per 1000 births) comparing intervention districts (Int) with control districts (Cont). Comparison of the standardised relative effects illustrates the greater decline in the control district rates compared with the intervention (eg, at 18 months, 46% decline in control compared with 19% in intervention; table 2). Figure 2 illustrates the PNM rates across the lifetime of the trial showing the downward trends with slight increases at the last data point (the last quarter of 2013), but rates at this time are clearly much lower in the intervention districts compared with the control districts (13.28 and 22.24, respectively).

le 2). Figure 2 illustrates the PNM rates across the lifetime of the trial showing the downward trends with slight increases at the last data point (the last quarter of 2013), but rates at this time are clearly much lower in the intervention districts compared with the control districts (13.28 and 22.24, respectively). Facility-based maternal mortality (ITS) Figure 3 and table 3 show the results for the ITS of maternal mortality rates. Maternal mortality ratios for the first quarter of 2011 (first data point) in the intervention and control districts were 171.58 and 83.43 (per 100 000 live births), respectively. It is clear that the intervention districts are reporting a higher ratio than the control at the time of the intervention start, the first quarter of 2012, with ratios of 151.24 and 105.44, respectively. The ratios do reduce by 6.879 and 7.304 points, respectively, per quarter prior to the intervention (at the end of the fourth quarter of 2011). When we take this trend into account, there is a consistent reduction in maternal mortality ratio in the intervention districts while in the control districts there is a steady increase. The CIs around the effect estimates are wide and make us cautious in overinterpreting this result (table 3). Table 3 Effects from the ITS models for district facility-based maternal mortality ratios, comparing intervention with control

Facility-based maternal mortality (ITS) Figure 3 and table 3 show the results for the ITS of maternal mortality rates. Maternal mortality ratios for the first quarter of 2011 (first data point) in the intervention and control districts were 171.58 and 83.43 (per 100 000 live births), respectively. It is clear that the intervention districts are reporting a higher ratio than the control at the time of the intervention start, the first quarter of 2012, with ratios of 151.24 and 105.44, respectively. The ratios do reduce by 6.879 and 7.304 points, respectively, per quarter prior to the intervention (at the end of the fourth quarter of 2011). When we take this trend into account, there is a consistent reduction in maternal mortality ratio in the intervention districts while in the control districts there is a steady increase. The CIs around the effect estimates are wide and make us cautious in overinterpreting this result (table 3). Table 3 Effects from the ITS models for district facility-based maternal mortality ratios, comparing intervention with control Intervention Control Effect SE CI 95% p Value Relative effect (%) Effect SE CI 95% p Value Relative effect (%) 3 months −14.87 22.78 (−68.73 to 38.99) 0.535 −9 4.85 36.88 (−82.37 to 92.06) 0.899 4 6 months −18.68 26.03 (−80.24 to 42.88) 0.496 −12 5.97 42.49 (−94.50 to 106.43) 0.892 5 9 months −22.49 30.13 (−93.74 to 48.77) 0.480 −16 7.08 49.60 (−110.19 to 124.36) 0.890 6 12 months −26.30 34.78 (−108.54 to 55.94) 0.474 −18 8.20 57.66 (−128.13 to 144.53) 0.891 8 15 months −30.11 39.79 (−124.19 to 63.96) 0.474 −21 9.32 66.32 (−147.51 to 166.14) 0.892 8 18 months −33.93 45.03 (−140.40 to 72.54) 0.476 −24 10.44 75.39 (−167.82 to 188.70) 0.894 8 21 months −38.11 50.30 (−157.06 to 80.84) 0.473 −29 11.55 87.72 (−195.87 to 218.98) 0.895 9 Effect—estimate of effect from ARIMA ITS model.

months −30.11 39.79 (−124.19 to 63.96) 0.474 −21 9.32 66.32 (−147.51 to 166.14) 0.892 8 18 months −33.93 45.03 (−140.40 to 72.54) 0.476 −24 10.44 75.39 (−167.82 to 188.70) 0.894 8 21 months −38.11 50.30 (−157.06 to 80.84) 0.473 −29 11.55 87.72 (−195.87 to 218.98) 0.895 9 Effect—estimate of effect from ARIMA ITS model. Relative effect, percentage change (compared with preslope trend). ITS, interrupted time series. Figure 3 Interrupted time series: district health facility maternal mortality ratio (per 100 000 births), comparing intervention districts (Int) with control districts (Cont). However, comparison of the standardised relative effects shows that there is a consistent reduction in the intervention districts compared with a consistent increase in the control districts (table 3). Figure 3 illustrates the maternal mortality ratios across the lifetime of the trial and shows that starting from a higher baseline, the intervention districts maternal mortality ratio reduces. For the control districts, starting from a lower baseline, the maternal mortality ratio increases throughout the trial period. Indeed, the absolute ratio in the intervention and control districts by the final quarter of 2013 were 90.23, a reduction of 61.01 points from intervention start and 139.03, an increase of 33.6 points from intervention start.

Table 4 shows the actual mortality rates/ratios over the lifetime of the project. In PNM rates, there was an improvement of 32% in the intervention districts compared with only 30% in the control, with intervention districts having consistently lower PNM rates than control. The rate of early neonatal deaths, a key component of PNM, decreases over the 3 years of the project in the intervention districts (from 5.8 in 2011 to 4 in 2013, per 1000 births) by 31%; in the control districts, the rate is higher overall and has increased by 2% in 2013 when compared with 2011 (see table 4). Fresh stillbirth rates reduce by nearly 46% in the control districts and by only 32% in the intervention districts. Maternal mortality in the intervention districts reduces by 40%; however, in the control districts, the ratio increases by 31% by the end of the project. Table 4 District facility-based mortality ratios/rates per year for 2011–2013 Intervention districts N=8 Control districts N=6 Variable and year Births n= Events n= Rates*/ratio† Births n= Events n= Rates*/ratio† Perinatal mortality (PNM) and PNM rate per 1000 births 2011‡ 147 287 3057 20.8 73 465 2130 29.0 2012 148 984 2599 17.4 78 845 1825 23.2 2013 155 425 2840 14.2 79 437 1612 20.3 Early neonatal death (ND) and ND rate per 1000 births 2011‡ 147 287 859 5.8 73 465 703 9.6 2012 148 984 788 5.3 78 845 694 8.8 2013 155 425 625 4.0 79 437 778 9.8 Stillbirths fresh (SBF) and SBF rate per 1000 births 2011‡ 147 287 2198 14.9 73 465 1427 19.4 2012 148 984 1811 12.2 78 845 1131 14.3 2013 155 425 1574 10.1 79 437 834 10.5 Stillbirths macerated (SBM) and SBM rate per 1000 births 2011‡ 147 287 880 6 73 465 569 7.7 2012 148 984 856 5.7 78 845 641 8.1 2013 155 425 1063 6.2 79 437 633 8 Maternal mortality (MM) and MM ratio per 100.000 births 2011‡ 147 287 262 177.9 73 465 72 98.0 2012 148 984 211 141.6 78 845 94 119.2 2013 155 425 185 107.5 79 437 102 128.4 *Rates are calculated as number of events divided by total births, multiplied by 1000.

4 856 5.7 78 845 641 8.1 2013 155 425 1063 6.2 79 437 633 8 Maternal mortality (MM) and MM ratio per 100.000 births 2011‡ 147 287 262 177.9 73 465 72 98.0 2012 148 984 211 141.6 78 845 94 119.2 2013 155 425 185 107.5 79 437 102 128.4 *Rates are calculated as number of events divided by total births, multiplied by 1000. †Ratios for MM are calculated as number of events divided by total births, multiplied by 100 000. ‡Missing data from one control district for the whole of 2011. Obstetric complications and caesarean sections Table 5 shows the number and rate of key obstetric complications over the trials lifetime. There are increases in the cases/rates of prolonged labour across both intervention and control districts. Cases of (pre)-eclampsia remain similar throughout as do many of the other complications. Table 5 Comparison of key obstetric complications (facility based) by year

Obstetric complications and caesarean sections Table 5 shows the number and rate of key obstetric complications over the trials lifetime. There are increases in the cases/rates of prolonged labour across both intervention and control districts. Cases of (pre)-eclampsia remain similar throughout as do many of the other complications. Table 5 Comparison of key obstetric complications (facility based) by year Intervention districts N=8 Control districts N=6* Variable and year Births n= Events n= Rate† Births n= Events n= rate† Prolonged labour (PL) and PL rate per 1000 births 2011* 147 287 2864 19.5 73 465 2627 35.8 2012 148 984 2931 19.7 78 845 2690 34.2 2013 155 425 5098 32.8 79 437 3132 39.4 (Pre-)eclampsia (pE) and pE rate per 1000 births 2011* 147 287 593 4.0 73 465 406 5.5 2012 148 984 947 6.4 78 845 462 5.9 2013 155 425 895 5.8 79 437 484 6.1 Sepsis (maternal) (Sm) and Sm rate per 1000 births 2011* 147 287 153 1.0 73 465 91 1.2 2012 148 984 209 1.4 78 845 92 1.2 2013 155 425 198 1.27 79 437 69 0.9 Ruptured uterus (RU) and RU rate per 1000 births 2011* 147 287 143 1.0 73 465 97 1.3 2012 148 984 127 0.9 78 845 97 1.2 2013 155 425 216 1.4 79 437 75 1.0 Haemorrhage (H) and H rate per 1000 births 2011* 147 287 1705 11.6 73 465 1125 15.3 2012 148 984 2821 19.0 78 845 1518 19.3 2013 155 425 2291 15.8 79 437 1404 18.7 *Missing data from one control district for the whole of 2011. †Rates are calculated as number of events divided by total births, multiplied by 1000.

Intervention districts N=8 Control districts N=6* Variable and year Births n= Events n= Rate† Births n= Events n= rate† Prolonged labour (PL) and PL rate per 1000 births 2011* 147 287 2864 19.5 73 465 2627 35.8 2012 148 984 2931 19.7 78 845 2690 34.2 2013 155 425 5098 32.8 79 437 3132 39.4 (Pre-)eclampsia (pE) and pE rate per 1000 births 2011* 147 287 593 4.0 73 465 406 5.5 2012 148 984 947 6.4 78 845 462 5.9 2013 155 425 895 5.8 79 437 484 6.1 Sepsis (maternal) (Sm) and Sm rate per 1000 births 2011* 147 287 153 1.0 73 465 91 1.2 2012 148 984 209 1.4 78 845 92 1.2 2013 155 425 198 1.27 79 437 69 0.9 Ruptured uterus (RU) and RU rate per 1000 births 2011* 147 287 143 1.0 73 465 97 1.3 2012 148 984 127 0.9 78 845 97 1.2 2013 155 425 216 1.4 79 437 75 1.0 Haemorrhage (H) and H rate per 1000 births 2011* 147 287 1705 11.6 73 465 1125 15.3 2012 148 984 2821 19.0 78 845 1518 19.3 2013 155 425 2291 15.8 79 437 1404 18.7 *Missing data from one control district for the whole of 2011. †Rates are calculated as number of events divided by total births, multiplied by 1000. Birth complications Table 6 shows the number and rate of key birth complications and caesarean sections over the lifetime of the trial. Of note, here is an increase in the reported rates of neonatal asphyxia, which increases in the intervention and control districts. Caesarean sections (cases/percentages) increase in the intervention district while remaining fairly constant in the control districts. Table 6 Comparison of key birth complications and caesarean sections (facility based) by year

Birth complications Table 6 shows the number and rate of key birth complications and caesarean sections over the lifetime of the trial. Of note, here is an increase in the reported rates of neonatal asphyxia, which increases in the intervention and control districts. Caesarean sections (cases/percentages) increase in the intervention district while remaining fairly constant in the control districts. Table 6 Comparison of key birth complications and caesarean sections (facility based) by year Intervention districts N=8 Control districts N=6 Variable and year Births n= n Rate* Births n= n Rate* Premature birth (PB) and PB rate per 1000 births 2011† 147 287 3025 20.6 73 465 2095 28.6 2012 148 984 2424 16.3 78 845 1996 25.3 2013 155 425 3036 19.5 79 437 2065 26 Low birthweight (LBW) and LBW rate per 1000 births‡ 2011† 147 287 2949 20.0 73 465 3723 50.7 2012 148 984 2968 20.0 78 845 3272 41.5 2013 155 425 4517 29.1 79 437 3042 38.3 Neonatal asphyxia (NA) and NA rate per 1000 births 2011† 147 287 2301 15.6 73 465 1425 19.4 2012 148 984 2704 18.2 78 845 2010 25.5 2013 155 425 4104 26.4 79 437 2710 34.1 Neonatal sepsis (NS) and NS rate per 1000 births 2011† 147 287 454 3.1 73 465 587 8.0 2012 148 984 707 4.8 78 845 415 5.3 2013 155 425 892 5.7 79 437 543 6.8 Vacuum extraction (VE) and VE rate per 1000 births 2011† 147 287 2056 14.0 73 465 691 9.4 2012 148 984 2165 10.2 78 845 1052 13.3 2013 155 425 5601 13.9 79 437 1235 15.6 Breech delivery (BD) and BD rate per 1000 births 2011† 147 287 2165 14.7 73 465 2166 29.5 2012 148 984 1603 10.8 78 845 1943 24.6 2013 155 425 2035 13.1 79 437 1618 20.4 Caesarean sections (CS) and CS percentage§ n %§ n %§ 2011† 147 287 5601 3.8 73 465 3923 5.3 2012 148 984 6319 4.2 78 845 3942 5.0 2013 155 425 9368 6.0 79 437 4163 5.2 *Rates are calculated as number of events divided by total births, multiplied by 1000.

603 10.8 78 845 1943 24.6 2013 155 425 2035 13.1 79 437 1618 20.4 Caesarean sections (CS) and CS percentage§ n %§ n %§ 2011† 147 287 5601 3.8 73 465 3923 5.3 2012 148 984 6319 4.2 78 845 3942 5.0 2013 155 425 9368 6.0 79 437 4163 5.2 *Rates are calculated as number of events divided by total births, multiplied by 1000. †Missing data from one control district for the whole of 2011. ‡Birthweight <2500 g. §Percentage=number of caesarean section/number births×100. ETATMBA trainee outcomes Fifty-four trainees were recruited, representing 67% (54/81) of the ACs working in emergency obstetric and neonatal care (EmONC) in the intervention districts. Of those recruited, 46 (85%) remained in the training programme till the end, 25 from the central region of Malawi drawn from nine hospitals (district and central hospitals) and 21 from the northern region drawn from six hospitals (district and central hospitals). One of the smaller districts in the northern region had one ETATMBA trainee working in its district hospital. Nearly all the trainees were male, with only two females. All of the 46 trainees completed the training and were awarded their degree in late 2014.

ion drawn from six hospitals (district and central hospitals). One of the smaller districts in the northern region had one ETATMBA trainee working in its district hospital. Nearly all the trainees were male, with only two females. All of the 46 trainees completed the training and were awarded their degree in late 2014. Discussion The main aims of this study were to evaluate the impact on health outcomes (perinatal and maternal morbidity and mortality) of the ETATMBA training programme. We are pleased to see an overall reduction in PNM rates (control and intervention). Attributing this reduction to our training is complex as reductions are statistically better in control districts. However, on closer examination of the early neonatal death rates (a key component of PNM), we find a fall by 31% (per 1000 births) in the intervention districts and a 2% increase in the control districts over the 3 years of the project. There were twice as many birth events within the intervention districts compared with control over the lifetime of the project implying that our trainees were exposed to more birth events. Our original assumption was that training would reduce perinatal deaths in the intervention district more than control through early intervention (eg, effective resuscitation) and enhanced care of the neonate. We are cautious in interpreting these results as there are confounding factors, but evidence from our qualitative studies in Malawi and Tanzania do support the notion that our training has been effective.13

rict more than control through early intervention (eg, effective resuscitation) and enhanced care of the neonate. We are cautious in interpreting these results as there are confounding factors, but evidence from our qualitative studies in Malawi and Tanzania do support the notion that our training has been effective.13 A problem with interpreting these data is that when resuscitating babies born in very poor condition, some cases end in reclassifying fresh stillbirths as neonatal deaths. The Helping Babies Breath initiative was occurring at the same time as the ETATMBA training programme across Malawi, and this effect of moving some fresh stillbirths to neonatal deaths occurred in the intervention and control districts and explains the drop in fresh stillbirths without a drop in neonatal deaths in the control districts. Importantly, in the intervention districts, both stillbirths and neonatal deaths decreased suggesting that our training (which included resuscitation and early neonatal care) ended in more successful resuscitations in the intervention districts.

sh stillbirths without a drop in neonatal deaths in the control districts. Importantly, in the intervention districts, both stillbirths and neonatal deaths decreased suggesting that our training (which included resuscitation and early neonatal care) ended in more successful resuscitations in the intervention districts. For maternal mortality, we see a consistent reduction in ratio over the 3 years in the intervention districts compared with a gradual increase within control districts. This does suggest that the training was starting to have an impact. Our qualitative work gave strong indications that at the local level (in the facilities where a trainee worked), maternal mortality had reduced. It takes effective teams to prevent maternal deaths, and the combination of knowledge, practical and leadership training was effective.13 We again are cautious in interpreting this result, but we do feel that there is evidence that women's lives were saved as a result of our training.13

maternal mortality had reduced. It takes effective teams to prevent maternal deaths, and the combination of knowledge, practical and leadership training was effective.13 We again are cautious in interpreting this result, but we do feel that there is evidence that women's lives were saved as a result of our training.13 Caesarean sections increased in the intervention districts while remaining at a similar level in control districts. There are many factors that can contribute to this, but our process evaluation has shown us that trainees were more confident to intervene and work with the local team to provide a better outcome for mothers.13 In high-income countries, the increasing caesarean section rates are associated with an increase in maternal morbidity.17 In contrast, in low-income countries like Malawi where there are low caesarean section rates (∼5%), increasing the rates is associated with an increase in maternal and perinatal survival.17 Furthermore, our training included improving decision-making for indications of caesarean section, improving skills to avoid caesarean section (vaginal breech and vacuum delivery) and reducing complications from caesarean section (improved surgical technique, transverse skin incisions, antibiotic prophylaxis, use of WHO checklist, better management of intraoperative haemorrhage, improved communications between clinical team). Thus, the training may be responsible for the increased rate of caesarean section with a decrease in neonatal and maternal mortality.

gical technique, transverse skin incisions, antibiotic prophylaxis, use of WHO checklist, better management of intraoperative haemorrhage, improved communications between clinical team). Thus, the training may be responsible for the increased rate of caesarean section with a decrease in neonatal and maternal mortality. As expected, carrying out an RCT in a sub-Saharan African setting has a number of limitations. Two major initiatives from US aid organisations were active across Malawi at the time of our study: Helping Babies Breathe and Kangaroo Mother Care. The control districts could not ethically be deprived of these initiatives, and this may have contributed as a major confounder to our results.

as a number of limitations. Two major initiatives from US aid organisations were active across Malawi at the time of our study: Helping Babies Breathe and Kangaroo Mother Care. The control districts could not ethically be deprived of these initiatives, and this may have contributed as a major confounder to our results. This was not a traditional cluster RCT design. There are no actual participants; the clusters are whole districts and intervention districts had a numbers of ACs who received the ETATMBA training, which did not happen in the control districts. Along with this, our reliance on locally recorded data in the districts health facilities is also a limitation. Indeed, we are cautious about overinterpreting results. Our researchers visited the districts at least twice, collecting the data and transcribing it from data pooled at the district centres/hospitals. While they were able to check some data against register entries, to do this for all facilities within a district would have required a huge number of research staff. While we do have missing data (and report this), we are confident that the data collected are an accurate reflection of events in all of the districts included in the trial. Our design using district data was pragmatic, and a limitation is that it will not give the true picture that would be obtained from population mortality surveillance; however, as all the trainees were practising at the district hospitals, we believe it provides a comparable measure of change.

ts included in the trial. Our design using district data was pragmatic, and a limitation is that it will not give the true picture that would be obtained from population mortality surveillance; however, as all the trainees were practising at the district hospitals, we believe it provides a comparable measure of change. Our original pragmatic plan to split Lilongwe into two was thwarted in reality, another limitation of this study. This is the largest district in the study with the largest concentration of ETATMBA trainees (nine) whose influence was potentially district-wide. This has somewhat unbalanced the study with many more birth events in the intervention districts overall (see online supplementary appendix). It may be considered that our randomisation failed because of the imbalance but, in good faith, we based our randomisation on previously published data. We believe that we ensured that all necessary steps in randomisation were taken care of by using the appropriate statistical programme and randomisation technique, which cover the control of variability, levels of randomisation, size of intervention arms and power to detect causal effects, as well as the many problems that commonly lead to postintervention bias. We present an accurate reflection of the reality.

ng the appropriate statistical programme and randomisation technique, which cover the control of variability, levels of randomisation, size of intervention arms and power to detect causal effects, as well as the many problems that commonly lead to postintervention bias. We present an accurate reflection of the reality. We report elsewhere the high value the trainees placed on the mentoring and support they received from the visiting obstetricians and the ETATMBA team.13 A key part, and indeed a unique part, of the ETATMBA training was the integration of clinical leadership with clinical skills and knowledge teaching. Trainees were actively encouraged to take leadership roles, and cascade their new skills and knowledge within their districts, which included travelling out to other facilities. Our qualitative work provides evidence that cascading took place with more effective team work and commitment to improve facilities.13 However, countrywide political and infrastructure problems (eg, fuel shortages and electricity outages) early in the project did place restrictions on the trainees’ ability to cascade their skills (eg, travelling to other facilities). Our hopes are that the ripples from the training are far reaching and ongoing; hence, our objective is to look at district-wide outcomes. Indeed, results from the Tanzanian arm of the ETATMBA show similar outcomes.18 19

restrictions on the trainees’ ability to cascade their skills (eg, travelling to other facilities). Our hopes are that the ripples from the training are far reaching and ongoing; hence, our objective is to look at district-wide outcomes. Indeed, results from the Tanzanian arm of the ETATMBA show similar outcomes.18 19 A recent review looking at obstetric and newborn care capacity building in rural sub-Saharan Africa concluded that the millennium development goals will not be met, but suggests that simple packaged emergency obstetric interventions could have an impact in the future.20 Given the critical shortages of qualified obstetricians in countries like Malawi, some have started training cadres like ACs in obstetric surgery, particularly in rural/remote areas, with the hope of alleviating the problem.21–26 In Malawi, the health facilities have chronic shortages of essential equipment and drugs.25 This coupled with the health provider crisis demonstrate the challenges faced in trying to make an impact on health outcomes.24–26 Training in obstetric emergencies that includes high fidelity simulations, leadership training and clinical teaching can improve obstetric and neonatal outcomes; features included in the ETATMBA training.27–29

e health provider crisis demonstrate the challenges faced in trying to make an impact on health outcomes.24–26 Training in obstetric emergencies that includes high fidelity simulations, leadership training and clinical teaching can improve obstetric and neonatal outcomes; features included in the ETATMBA training.27–29 Very few RCTs of training with the outcome measure of obstetric complications have been reported and there is an urgent need for these.29 30 Reported improvements in obstetric complications have been demonstrated in ‘before and after’ studies.27 29 In low-income countries, the ability to control other variables (eg, confounders) is challenging and in judging the outcomes, one must take into account both qualitative and quantitative data over longer time frames and in context.

ts in obstetric complications have been demonstrated in ‘before and after’ studies.27 29 In low-income countries, the ability to control other variables (eg, confounders) is challenging and in judging the outcomes, one must take into account both qualitative and quantitative data over longer time frames and in context. In conclusion, it is heartening to see the reductions in PNM, maternal and neonatal mortality rates/ratios presented here. We are cautious in our interpretation of these results. Attributing the changes to the ETATMBA training is complex. Although there have been a large number of challenges, we have successfully trained 46 ACs with advanced skills and knowledge in obstetric, neonatal care and clinical leadership. This training has had an excellent retention rate, and was well received by the trainees and those around them in the districts (all were awarded a BSc in International Obstetrics by the University of Warwick in October 2014). We feel that these results, supported by the qualitative evidence, show the training has changed practice, and as a result may have contributed to the positive downward trends in maternal and neonatal mortality rates, and an increase in numbers and quality of lifesaving obstetric interventions such as caesarean sections. Providing this cadre with the leadership and practical skills and knowledge, based on best evidence and tailored to be delivered in a low-resource setting, could be a practical solution to the doctor shortages in African countries. Our hope is that the ETATMBA trainees will have an enduring influence that will impact positively on future practice in Malawi and Tanzania, and will help in realisation of MDGs 4 and 5.

e and tailored to be delivered in a low-resource setting, could be a practical solution to the doctor shortages in African countries. Our hope is that the ETATMBA trainees will have an enduring influence that will impact positively on future practice in Malawi and Tanzania, and will help in realisation of MDGs 4 and 5. Enhancing Human Resources and Use of Appropriate Technologies for Maternal and Perinatal Survival in sub-Saharan Africa (ETATMBA) is a collaborative project funded by the European Commission, Seventh Framework Programme (THEME (HEALTH.2010.3.4-2) (Project no. 266290)). This trial was embedded within this programme of work. All authors are part of the ETATMBA team. The ETATMBA would like to thank all of the clinical officers, the district medical and nursing officers for their hard work and support. They would also like to thank the staff at the Ministry of Health in Malawi. This project benefited from facilities funded through Birmingham Science City Translational Medicine Clinical Research and Infrastructure Trials Platform, with support from Advantage West Midlands. Handling editor: Seye Abimbola

Enhancing Human Resources and Use of Appropriate Technologies for Maternal and Perinatal Survival in sub-Saharan Africa (ETATMBA) is a collaborative project funded by the European Commission, Seventh Framework Programme (THEME (HEALTH.2010.3.4-2) (Project no. 266290)). This trial was embedded within this programme of work. All authors are part of the ETATMBA team. The ETATMBA would like to thank all of the clinical officers, the district medical and nursing officers for their hard work and support. They would also like to thank the staff at the Ministry of Health in Malawi. This project benefited from facilities funded through Birmingham Science City Translational Medicine Clinical Research and Infrastructure Trials Platform, with support from Advantage West Midlands. Handling editor: Seye Abimbola Collaborators: Authors are acting on behalf of the ETATMBA study group below. The ETATMBA Study Group: Malawi: University of Malawi College of Medicine: FK, CM, WC, Chikayiko Chiwandira, Queen Dube. Tanzania: Ministry of Health, Malawi: Fannie Kachale, Chimwemwe Mvula. Tanzania: Ifakara Health Institute, Tanzania: Godfrey Mbaruku, Paul Kihaile, Sidney Ndeki, Hamed Mohamed, Senga Pemba, Aloisia Shemdoe, Festo Mazuguni, Angelo Nyamtema. Sweden: Karolinska Institutet, Sweden: Staffan Bergström. UK: GE Healthcare: Alan Davies; The University of Warwick, UK: JPO, SQ, DS, DD, DRE, Frances Griffiths, Ngianga-bakwin Kandala, Anne-Marie Brennan, Edward Peile, Anne-Marie Slowther, Saliya Chipwete, Paul Beeby, Gregory Eloundou, Harry Gee, Vinod Patel.

yamtema. Sweden: Karolinska Institutet, Sweden: Staffan Bergström. UK: GE Healthcare: Alan Davies; The University of Warwick, UK: JPO, SQ, DS, DD, DRE, Frances Griffiths, Ngianga-bakwin Kandala, Anne-Marie Brennan, Edward Peile, Anne-Marie Slowther, Saliya Chipwete, Paul Beeby, Gregory Eloundou, Harry Gee, Vinod Patel. Contributors: DRE, JPO, WC, DS, CM, FK and SQ were involved in conception and design of the study. DRE drafted the manuscript supported by all authors. JPO, FK, CM, SQ, DS and DD were responsible for the design, management and delivery of the training. N-bK is the study statistician. The corresponding author (DRE) confirms that he had full access to all the data in the study and had final responsibility for the decision to submit for publication. Funding: Seventh Framework Programme (Project no. 266290).

Contributors: DRE, JPO, WC, DS, CM, FK and SQ were involved in conception and design of the study. DRE drafted the manuscript supported by all authors. JPO, FK, CM, SQ, DS and DD were responsible for the design, management and delivery of the training. N-bK is the study statistician. The corresponding author (DRE) confirms that he had full access to all the data in the study and had final responsibility for the decision to submit for publication. Funding: Seventh Framework Programme (Project no. 266290). Competing interests: JPO is the principal investigator for the trial and is Associate Clinical Professor at Warwick Medical School (UK) in the department of Metabolic and Vascular Health and Translational Medicine. DRE is a Principal Research Fellow in the Warwick Clinical Trials unit (UK) and has expertise in research design, implementation and evaluation. FK is a Consultant Obstetrician/Gynaecologist and is principal investigator for the trial at College of Medicine, Malawi. CM is Senior Lecturer and Consultant Obstetrician/Gynaecologist, College of Medicine, University of Malawi. DS is Associate Professor in Child Health with a research interest in international child health. SQ is Professor of Obstetrics and Honorary Consultant Obstetrician at University Hospitals Coventry and Warwickshire (UK) with research interests being translational research in recurrent miscarriage, implantation, preterm and dysfunctional labour, and obesity in pregnancy. DD is an Associate Professor (Reader) in the Warwick Medical School Educational Development and Research Team. His research interests are primarily in global health education and educational technology, and e-learning in medical education. N-bK is a Professor of statistics at the Department of Mathematics and Information sciences, Faculty of Engineering and Environment, Northumbria University, Newcastle upon Tyne, UK, and Head of Unit of Health Economics and Evidence Synthesis Research Unit, Department of Population Health, Luxembourg Institute of Health, Luxembourg. WC is a researcher and PhD student at the College of Medicine, Malawi.

ciences, Faculty of Engineering and Environment, Northumbria University, Newcastle upon Tyne, UK, and Head of Unit of Health Economics and Evidence Synthesis Research Unit, Department of Population Health, Luxembourg Institute of Health, Luxembourg. WC is a researcher and PhD student at the College of Medicine, Malawi. Ethics approval: Research Ethics Committee (BREC) at the University of Warwick, UK (143/09/2011), and the College of Medicine Research Ethics Committee (COMREC), Malawi (P.07/11/1102). It had also the approval and support of the Malawi Ministry of Health. Provenance and peer review: Not commissioned; externally peer reviewed. Data sharing statement: No additional data are available.

Key questions What is already known about this topic? Poverty-related diseases are a devastating health and socioeconomic problem in many low-income countries, particularly in sub-Saharan Africa. Traditional education and training in biomedical sciences usually do not offer perspectives on poverty and its related health conditions. What are the new findings? We report our experience and evaluation of a unique international biomedical training programme that created a South-South and South-North network of young researchers, solely focused on poverty-related diseases. Poverty-Related Diseases College trained young biomedical scientists to become future trainers of the next generation of scientists in their own countries, addressing the needs of the poor in Africa. Recommendations for policy Networks connecting researchers in disease-endemic countries and high-income countries are needed to build scientific capacity for poverty-related diseases and to conduct research across disciplines and diseases. Development of infrastructure, high-level education programmes and career promotion plans, supported by African home institutions and supervisors, are required for this. The sustainability of such networks largely depends on long-term funding.

Recommendations for policy Networks connecting researchers in disease-endemic countries and high-income countries are needed to build scientific capacity for poverty-related diseases and to conduct research across disciplines and diseases. Development of infrastructure, high-level education programmes and career promotion plans, supported by African home institutions and supervisors, are required for this. The sustainability of such networks largely depends on long-term funding. Training the future generation of African and European scientists in poverty-related diseases Poverty is such a dominant factor affecting global health1–3 that the objectives of poverty alleviation programmes are usually strongly linked to health indicators.4 5 This multidimensional concept of poverty is, for example, clearly expressed in the overlapping targets, goals and indicators of the recently proposed Sustainable Development Goals, defining the post-2015 health agenda.6–8 Poverty-related diseases (PRDs, see box 1) remain a devastating set of health and socioeconomic problems in many low-income countries, particularly in sub-Saharan Africa.9 Poor people do not only suffer from the greatest disease burden, but this burden is also given the least medical research attention.10 11 Traditional education and training in biomedical sciences do not usually offer perspectives on poverty and its related health consequences.12–14 Training young researchers in the many aspects of PRD is therefore a pre-requisite for sustainable healthcare and development in developing countries. An additional major challenge is the poor career opportunities for researchers in Africa.15–17 As a consequence, young researchers often emigrate early in their scientific career, with little contribution to development of their native country. Strengthening health research capacity in Africa has been recognised as a key factor to address this, by funders and policymakers alike.18–20 Box 1 Definition of poverty-related diseases (PRD) Many of the diseases contributing to the disease burden in low-income countries are tightly linked to the debilitating conditions of poverty, such as a lack of access to proper sanitation, health education and safe drinking-water, and poor nutrition and indoor air pollution. Diseases of poverty are often easily avoidable, preventable or treatable with existing medical interventions.

come countries are tightly linked to the debilitating conditions of poverty, such as a lack of access to proper sanitation, health education and safe drinking-water, and poor nutrition and indoor air pollution. Diseases of poverty are often easily avoidable, preventable or treatable with existing medical interventions. According to the most recent Global Burden of Disease report by the WHO, 52% of the total disease burden in low-income countries is caused by PRDs.21 PRDs include the so called ‘neglected tropical diseases’, but also extend to a much wider spectrum of diseases and conditions causing high, though preventable, morbidity and mortality worldwide in low-income countries. They include, among others: HIV/AIDS Malaria Tuberculosis Parasitic diseases (eg, leishmaniasis, schistosomiasis, filariasis, trypanosomiasis) Other tropical diseases (eg, dengue, yellow fever, Buruli ulcer, leptospirosis) Treatable childhood diseases (eg, polio, measles, pertussis) Respiratory infections (eg, pneumonia) Diarrhoeal diseases Nutritional deficiencies Other perinatal and maternal conditions

Parasitic diseases (eg, leishmaniasis, schistosomiasis, filariasis, trypanosomiasis) Other tropical diseases (eg, dengue, yellow fever, Buruli ulcer, leptospirosis) Treatable childhood diseases (eg, polio, measles, pertussis) Respiratory infections (eg, pneumonia) Diarrhoeal diseases Nutritional deficiencies Other perinatal and maternal conditions The Poverty-Related Diseases College (PRDC) was a virtual African-European college created to give supplementary training to the doctors, health scientists and policymakers of tomorrow's developing world. PRDC aimed at enhancing exchange, and increasing knowledge and experience in PRD in order to build sustainable career opportunities for young biomedical scientists in Africa. PRDC merged theoretical teaching with hands-on experience in new skills and technologies, and provided this as a training programme to young researchers with different scientific backgrounds in Africa and Europe. The connection between basic and applied sciences was achieved by programming research electives within the context of the disease burden in Africa. The forging of South-North and South-South networks was considered an essential part of career development. The final and long-term goal was to build scientific capacity in poverty-related biomedical research in Africa, and to contribute to the development of highly motivated scientists embedded in and sustained by international networks. These scientists will act as the future trainers of the next generation of scientists in their own countries or contribute to health policies by their thorough understanding of fundamental and applied biomedical sciences in an African context.

ly motivated scientists embedded in and sustained by international networks. These scientists will act as the future trainers of the next generation of scientists in their own countries or contribute to health policies by their thorough understanding of fundamental and applied biomedical sciences in an African context. The objective of this analysis paper is to present our experiences as an example of scientists and funders jointly inspired to develop sustainable scientific capacity and networks that address the medical needs of the poor in Africa. The Poverty-Related Diseases College Scope, structure and aims The PRDC consortium was a virtual institute comprising five African and five European institutions providing a comprehensive training programme in scientific, technical and soft skills to a selected group of African and European biomedical research fellows. The structure of PRDC is shown in figure 1. Figure 1 Structure of organisation of PRDC and origin of fellows and faculty. FP7, Seventh Framework Programme. PRDC, The Poverty-Related Diseases College.

The Poverty-Related Diseases College Scope, structure and aims The PRDC consortium was a virtual institute comprising five African and five European institutions providing a comprehensive training programme in scientific, technical and soft skills to a selected group of African and European biomedical research fellows. The structure of PRDC is shown in figure 1. Figure 1 Structure of organisation of PRDC and origin of fellows and faculty. FP7, Seventh Framework Programme. PRDC, The Poverty-Related Diseases College. The training programme consisted of a modular training curriculum (figure 2), preparing the fellows for a ‘reality check’ in Africa, followed by a science exchange programme at research laboratories in Europe and the USA and finalised by ‘outlook’ activities. The programme was guided by the following aims: (1) training and mentoring of future students by the fellows, (2) building sustainable research careers, (3) building networks around PRD, (4) specific training in advanced scientific technologies useful in PRD research and (5) improving cultural understanding. Figure 2 Overview of the various modules of the PRDC programme. PRDC, The Poverty-Related Diseases College. The funder The programme was funded by the European Commission's FP7 within the funding scheme, Coordination and Support Action (call topic: ‘HEALTH-2007-2.3.2-14—Next generation of researchers for HIV/AIDS, malaria, tuberculosis and neglected infectious diseases’). The programme started in July 2009 and was completed in March 2013.

programme was funded by the European Commission's FP7 within the funding scheme, Coordination and Support Action (call topic: ‘HEALTH-2007-2.3.2-14—Next generation of researchers for HIV/AIDS, malaria, tuberculosis and neglected infectious diseases’). The programme started in July 2009 and was completed in March 2013. The faculty The faculty of PRDC was composed by an African coordinator, the University of Yaoundé I, Cameroon, and supported by principal investigators (PI) originating from five African research centres and four European institutions, all involved in different areas of scientific research and medical aspects of infectious-related and PRDs (see figure 1). The PIs undertook the task of organising and implementing the different joint courses, developing the topics and selecting the advisory group and the fellows. In addition, they facilitated the reality check, science exchange and outlook activities. Six senior scientists with broad experience in research, teaching and administration, composed the advisory group from African and European institutions (figure 1). The advisors participated in the teaching and continuous contact with the fellows as well as with the PIs, supporting and advising on all modules of the programme.

or scientists with broad experience in research, teaching and administration, composed the advisory group from African and European institutions (figure 1). The advisors participated in the teaching and continuous contact with the fellows as well as with the PIs, supporting and advising on all modules of the programme. The fellows In early 2010, following a competitive international open call for application, 24 fellowships were awarded to 16 African and 8 European young scientists, working in the area of PRD and affiliated to an academic institution as a student or as an employee. One additional African fellowship was awarded due to a drop-out during the programme. Selection criteria focused on personal talents and achievements (eg, publication record) but also included institutional backing of the potential fellow and their institutional supervisor's interests in the activities of PRDC, and required reference letters. The fellows' countries of affiliation are shown in figure 1. Twenty-four per cent of African fellows were female, in contrast to 75% of the European fellows (total 40%). Among the African fellows, 11 were Anglophone versus 6 Francophone. The average age (range) was 28 (26–30) years and 35 (24–42) years for European and African fellows, respectively. Seventy-six per cent of the fellows had a MD or equivalent master's degree as highest degree at the time of enrolment, 5% a bachelor's degree and 19% (all of whom were African) had already obtained their PhD degree.

ge age (range) was 28 (26–30) years and 35 (24–42) years for European and African fellows, respectively. Seventy-six per cent of the fellows had a MD or equivalent master's degree as highest degree at the time of enrolment, 5% a bachelor's degree and 19% (all of whom were African) had already obtained their PhD degree. Programme management The programme was managed in accordance with all requirements of FP7 grant agreements. An unexpected and erroneous administrative interruption of the programme occurred during the organisation of the science exchange. This effectively resulted in a temporary arrest of forwarding funds from the funder and halted all activities for more than a year. Training activities in PRDC Courses The PRD College offered three educational packages: (1) a ‘Basic Educational Package’, concerning theoretical biological, clinical, epidemiological and public health aspects of PRD; (2) an ‘Advanced Educational Package’, concerning practical technology mainly focused on modern immunological laboratory assays; and (3) a ‘Soft Skills Package’, focusing on research support, communication, administration, patents, grant applications and management, collaborations, gender balance, ethics, dissemination, public communication, philosophy and others (figure 2). The three educational programmes were specifically designed to synchronise the educational level of the fellows so as to make them able to follow the longer and more demanding practical programmes.

Training activities in PRDC Courses The PRD College offered three educational packages: (1) a ‘Basic Educational Package’, concerning theoretical biological, clinical, epidemiological and public health aspects of PRD; (2) an ‘Advanced Educational Package’, concerning practical technology mainly focused on modern immunological laboratory assays; and (3) a ‘Soft Skills Package’, focusing on research support, communication, administration, patents, grant applications and management, collaborations, gender balance, ethics, dissemination, public communication, philosophy and others (figure 2). The three educational programmes were specifically designed to synchronise the educational level of the fellows so as to make them able to follow the longer and more demanding practical programmes. The reality check An important pillar of the PRDC training programme was the ‘Reality Check’ module. In this module, hosted by the African institutions, the fellows were exposed to the reality of health and disease in low-income regions. The main aims were to train fellows in identifying determinants of health and health seeking in poor communities, and the responses of the care providers, regular and traditional, to existing health needs. To optimise the module, the fellows were grouped in teams mixing African and European fellows, which were operational at seven locations for a period of 1–2 months: Langa township, Cape Town, South Africa; Kati district, Mali; Mbeya, Tanzania; Chipulukusu, Ndola, Zambia; Malawi; Mbengwi, Cameroon; and Kasangati, Uganda. Fellows carried out 21 assignments ranging from ‘handshaking’ with local authorities, to collecting available local health data, conducting interviews with officials at various levels of health services, and organising focus group discussions to find out perceptions of diseases and opinions about the available healthcare systems among the population.

ts ranging from ‘handshaking’ with local authorities, to collecting available local health data, conducting interviews with officials at various levels of health services, and organising focus group discussions to find out perceptions of diseases and opinions about the available healthcare systems among the population. The science exchange The ‘Science Exchange’ activity consisted of 3-month training periods of the African fellows at European laboratories, in order to generate research data useful for their PhD projects, or for learning new skills and techniques, and experiencing different working and cultural environments. Owing to unforeseen administrative problems, this module was interrupted for an extensive period, demanding rescheduling of the training activities. Three fellows managed to conduct their training before the interruption, another eight fellows were able to conduct their training after the interruption and six fellows did not accept this opportunity, because they had continued their career in other directions. The overall output of the Science Exchange, despite the long interruption, was considered highly valuable, and turned out to be an important experience for all fellows involved.

training after the interruption and six fellows did not accept this opportunity, because they had continued their career in other directions. The overall output of the Science Exchange, despite the long interruption, was considered highly valuable, and turned out to be an important experience for all fellows involved. Outlook activities The PRDC activities were concluded with a 2-day Outlook Retreat in Italy, where fellows, PIs and advisors reported their evaluation of the whole programme. Fellows were additionally asked to prepare joint grant proposals. These proposals were reviewed by the faculty and, based on their quality, six were allocated a small sum of money to develop a full grant proposal. On follow-up, three proposals were not developed due to individual problems of the fellows, and two were merged, given their common goals and participants. Thus, two grant proposals were finalised and submitted for funding. Neither was successful. One of them has been revised and resubmitted recently. Expectations and evaluations of PRDC Fellows' expectations and evaluations Following the first course module, a semistructured questionnaire was developed to collect the fellows' viewpoints regarding their future career tracks and expectations from their enrolment in PRDC. The response rate to this questionnaire was 100%, and a descriptive and qualitative content analysis was performed on clustered answers.

irst course module, a semistructured questionnaire was developed to collect the fellows' viewpoints regarding their future career tracks and expectations from their enrolment in PRDC. The response rate to this questionnaire was 100%, and a descriptive and qualitative content analysis was performed on clustered answers. At the time of enrolment, almost 90% of the African fellows had the expectation that PRDC would increase their chances on the job market compared to 50% of the European fellows. For the Africans, the main reason for joining PRDC was the scientific and technological content of the programme, while for the Europeans it was the possibility to expand their research collaborative network. Likewise, 80% of all fellows expected PRDC to change their view on PRD, mainly by linking a multitude of disciplines and backgrounds, and by increasing basic knowledge on, for example, pathophysiology and epidemiology of PRD. Most of the African fellows wanted to continue their research career after PRDC in their own country (50%) or to get limited training abroad and return to their country afterwards (12.5%), while most European fellows desired to use PRDC as preparation to network in an African context to advance their career in PRD (75%).

st of the African fellows wanted to continue their research career after PRDC in their own country (50%) or to get limited training abroad and return to their country afterwards (12.5%), while most European fellows desired to use PRDC as preparation to network in an African context to advance their career in PRD (75%). The majority of fellows found it more difficult to conduct research on PRD than to do so on other diseases, mainly due to limited funds. Possibly related to this, 50% and 37.5% of the African and European fellows, respectively, did not earn enough salary with their current research employment to make ends meet. Nevertheless, financial incentives to choose a professional direction were more apparent among African fellows. Two years after the end of PRDC, in August 2015, another semistructured questionnaire was developed, to assess the impact of PRDC on personal and career development of the fellows. The response rate to this questionnaire was 88%, and a descriptive and qualitative content analysis was performed on clustered answers (the following percentages are based on the total number of respondents).

ed questionnaire was developed, to assess the impact of PRDC on personal and career development of the fellows. The response rate to this questionnaire was 88%, and a descriptive and qualitative content analysis was performed on clustered answers (the following percentages are based on the total number of respondents). At the time of evaluation, 90% of the fellows had obtained their PhD degree. Most fellows remained in a research environment: 90% were employed by a university or university hospital, spending an average of 35% of their time in research and science, and 37% were in teaching. Most importantly, 81% of all respondents (93% of the African fellows) indicated that they were still involved in and actively pursuing research on PRD, in accordance with the PRDC's objectives. Their main research topics are summarised in table 1. In the 2 years after the end of PRDC, the fellows had published an average number of 4.5 journal papers per fellow (range 0–17). An average of 0.9 (range 0–6) papers per fellow was directly related to PRDC activities.22 The majority of fellows (67%) reported having transferred their PRDC training to their own students; for example, six fellows were supervising a total of 13 PhD students at the time of follow-up. Seven fellows had obtained grants during the follow-up period with a median value of €37 650 (range €10 000–346 623). When examining the outcomes versus the initial aims of PRDC (see section 2.1 above), ≥70% of the fellows found that PRDC indeed had a positive impact on all of those components (see figure 2). In general, the African fellows were more positive than the Europeans about the impact of PRDC on their careers. This might be explained by the fact that many aspects, such as teaching and knowledge on scientific technologies, were already appropriately covered in the curricula of the European graduate programmes. Nevertheless, almost all fellows (95%) indicated that PRDC had contributed to their mutual cultural understanding. Close interactions between all participants during all of the modules of PRDC and the frank scheduled discussions about encountered cultural differences contributed to the appreciation and acceptance of these differences. Many fellows explicitly acknowledged the positive impact of PRDC on their network and personal links (figure 3), but admitted that they had made little use of it in practice.

C and the frank scheduled discussions about encountered cultural differences contributed to the appreciation and acceptance of these differences. Many fellows explicitly acknowledged the positive impact of PRDC on their network and personal links (figure 3), but admitted that they had made little use of it in practice. While quantification remains difficult, the created networks did result in tangible results already within the (short) follow-up period: a South-South-North network established through the Science Exchange recently published the results of their collaboration in the Malaria Journal,22 and various PRDC fellows had been offered postdoctoral positions through connections made within PRDC. Table 1 Current main poverty-related disease research interests of the PRDC fellows Poverty-related disease Number HIV 4 Leishmaniasis 1 Hepatitis B/C virus 1 Lymphatic filariasis 1 Dengue 1 Malaria/helminth coinfection during pregnancy 1 Malaria 5 Vector-borne diseases 1 TB/HIV 1 PRDC, The Poverty-Related Diseases College; TB, tuberculosis. Figure 3 Impact of PRDC on various aspects of the fellows' career and personal development. The percentages indicate the proportion of positive impact evaluations. PRDC, The Poverty-Related Diseases College.

Poverty-related disease Number HIV 4 Leishmaniasis 1 Hepatitis B/C virus 1 Lymphatic filariasis 1 Dengue 1 Malaria/helminth coinfection during pregnancy 1 Malaria 5 Vector-borne diseases 1 TB/HIV 1 PRDC, The Poverty-Related Diseases College; TB, tuberculosis. Figure 3 Impact of PRDC on various aspects of the fellows' career and personal development. The percentages indicate the proportion of positive impact evaluations. PRDC, The Poverty-Related Diseases College. The group dynamics among the different groups of fellows working and living closely together, showed rather normal and predictable features. Interpersonal dynamics sometimes adopted cultural issues, which were subsequently construed through focused group discussions. Although not explicitly measured, the focus on soft skills probably helped the fellows to use their enhanced communication skills to solve encountered problems.

normal and predictable features. Interpersonal dynamics sometimes adopted cultural issues, which were subsequently construed through focused group discussions. Although not explicitly measured, the focus on soft skills probably helped the fellows to use their enhanced communication skills to solve encountered problems. Advisors' evaluation The advisors were asked to evaluate the project and to provide their opinion and personal experience of it. All declared that, in spite of certain administrative problems, the PRDC achieved its goals; and that the project was a very positive experience on a personal level. In addition to the main positive aspects, good spirit, motivation and dedication of both, fellows and faculty contributed to effective exchange of ideas and solutions to problems. The advisors encouraged the faculty to continue following and supporting the career of the fellows, and suggested creating an alumni association to provide means and incentives for maintaining contact between fellows. It is worth mentioning that the University of Yaoundé I has converted the PRDC curriculum into a full graduate programme.

aged the faculty to continue following and supporting the career of the fellows, and suggested creating an alumni association to provide means and incentives for maintaining contact between fellows. It is worth mentioning that the University of Yaoundé I has converted the PRDC curriculum into a full graduate programme. Conclusion and discussion General conclusion The PRDC programme shows that it is possible to establish highly productive research and training networks that are based on international collaboration and inspiration. The programme was unique in its category, which warrants an analysis of its strengths and weaknesses. PRDC was also an impressive personal experience to fellows, faculty and advisors. It brought together, for almost 4 years, young and senior scientists coming from a number of different countries in Africa and Europe, with diverse scientific and cultural backgrounds, but who were invariably inspired to conduct research on PRD. The aims of PRDC in relation to the personal ambitions and expectations, assessed in the presented analysis, were largely fulfilled, with generally positive evaluations by the fellows (figure 3). The programme partially failed, however, to create a sustainable formal continuation of the created scientific network.

PRD. The aims of PRDC in relation to the personal ambitions and expectations, assessed in the presented analysis, were largely fulfilled, with generally positive evaluations by the fellows (figure 3). The programme partially failed, however, to create a sustainable formal continuation of the created scientific network. Weaknesses, threats and suggested solutions It proved difficult to implement a dynamic and efficient training programme within the context of ongoing and highly variable PhD tracks. In fact, the programme spanned 3 years (prolonged to four), during which time the students followed their individual PhD or postdoctoral paths at their home institutions. This sometimes gave rise to problems with their supervisors, because of the interruptions in their PhD projects while attending the PRDC courses and initiatives. This problem seems intrinsic to every ‘interfaculty’ initiative and may possibly be overcome by allocating more resources to the programme, and to its capacities to synchronise with existing PhD programmes. A more comprehensive research period in Africa, better integrated and focused on the PhD topic/programme of the fellow by involving the supervisors and departments at early planning stages of the fellows' activities, may have helped to make both programmes more compatible.

s to synchronise with existing PhD programmes. A more comprehensive research period in Africa, better integrated and focused on the PhD topic/programme of the fellow by involving the supervisors and departments at early planning stages of the fellows' activities, may have helped to make both programmes more compatible. On the African side, a major issue was the limited infrastructure, especially with regard to laboratory equipment and reagents. It is imperative to persuade funders to develop grants that specifically include infrastructure development in African institutions. Another issue was the encountered gender imbalance, which could have been addressed by a quota for female fellows in the selection procedure. The unforeseen and erroneous arrest in forwarding funds resulted in the need to suspend the entire programme for almost 1 year. This interruption resulted in the loss of involvement of some of the fellows, who had to continue their careers outside PRDC and were much less available once the programme resumed. Such unforeseen calamities can affect all programmes depending on external funding but the conditions of this error were typical of programmes that are carried out in and with developing countries. To prevent damage, commitment of the funder is needed to quickly repair administrative errors that threaten the continuation of the programme.

mities can affect all programmes depending on external funding but the conditions of this error were typical of programmes that are carried out in and with developing countries. To prevent damage, commitment of the funder is needed to quickly repair administrative errors that threaten the continuation of the programme. Future perspectives The future of PRDC has two main aspects: the individual future career development of its fellows, and the development of the PRDC concept into new initiatives. To maintain the network and increase its efficacy, the advisors suggested establishing an alumni programme. The sustainability of such a programme would be increased by the effort in developing the PRDC concept into long-term training initiatives that could take up the PRDC training modules and intercultural exchange plans. The University of Yaoundé I has started a self-sustaining Masters graduate programme that reproduces the PRDC curriculum since 2013. A similar programme in Public Health Biotechnology was started at the University of Ibadan, Nigeria, in 2015. The ‘Reality Check’ module has been adopted in the Masters programme in Health Economics at the Catholic University of Cameroon in Bamenda. Such programmes could be linked and connected through existing networks, such as The Global Health Network.23 The aspects of South-South and North-South collaboration and exchange are considered the major positive outputs of PRDC, and these need to be maintained in future dedicated programmes.

y of Cameroon in Bamenda. Such programmes could be linked and connected through existing networks, such as The Global Health Network.23 The aspects of South-South and North-South collaboration and exchange are considered the major positive outputs of PRDC, and these need to be maintained in future dedicated programmes. PRDC has shown that careful forging of bonds between fellows and PIs is probably key to success in intercultural and interdisciplinary training programmes. By emphasising respect for transcultural and other differences, group dynamics become very positive and productive. Group dynamics are rarely used as an instrument for scientific training. The experience of PRDC shows its potential. Recommendations to policymakers and funders There is an urgent need for effective and sustainable actions to build scientific capacity to address poverty-related topics in Africa. To achieve this, it will be essential to put in place a network between researchers in disease-endemic countries and researchers in the developed parts of the world, who together will implement research efforts across disciplines and diseases. This will require promotion of high-level training of African researchers and their career development in African institutions.

ce a network between researchers in disease-endemic countries and researchers in the developed parts of the world, who together will implement research efforts across disciplines and diseases. This will require promotion of high-level training of African researchers and their career development in African institutions. The promotion activities should be three-tiered: Development of infrastructure. Targeted equipment and infrastructure grants are needed to upgrade African institutions to become internationally more competitive. Funders should have a clear and transparent plan of research support that will allow maintenance of the infrastructures of African research institutions, universities and hospitals, at a competitive level. Such efforts could be coordinated by, for example, the African Academy of Sciences. Development of high-level educational programmes. The PRDC experience highlights the feasibility of implementing successful educational programmes that, in addition to technical training, aim at cross-continental, cross-cultural and cross-disciplinary networking and collaboration. This kind of experience is of key importance for future scientists from both, low-income and high-income countries. Development of a career promotion plan, especially for African researchers, supported by African home institutions and supervisors with the aim to attract back researchers from the large pool of highly skilled African scientists living outside Africa, as well as to reduce future brain-drain.

Development of high-level educational programmes. The PRDC experience highlights the feasibility of implementing successful educational programmes that, in addition to technical training, aim at cross-continental, cross-cultural and cross-disciplinary networking and collaboration. This kind of experience is of key importance for future scientists from both, low-income and high-income countries. Development of a career promotion plan, especially for African researchers, supported by African home institutions and supervisors with the aim to attract back researchers from the large pool of highly skilled African scientists living outside Africa, as well as to reduce future brain-drain. Importantly, the sustainability of such a network, as it was developed within PRDC, is largely dependent on the continuation of funding and interest of the funding agencies, subject to their critical evaluation. In our opinion, this long-term funding perspective would be pivotal to maintaining the momentum that was created by PRDC among the fellows, the involved institutions and the global health community, to consolidate medical science and training on PRD in the development of Africa. Handling editor: Seye Abimbola Twitter: Follow Thomas Dorlo at @thomasdorlo The authors are grateful to all fellows and advisors of the PRDC consortium for answering the questionnaires. The authors are especially indebted to A Conesa-Botella and J Fokam, for their help in revising the questionnaires.

Importantly, the sustainability of such a network, as it was developed within PRDC, is largely dependent on the continuation of funding and interest of the funding agencies, subject to their critical evaluation. In our opinion, this long-term funding perspective would be pivotal to maintaining the momentum that was created by PRDC among the fellows, the involved institutions and the global health community, to consolidate medical science and training on PRD in the development of Africa. Handling editor: Seye Abimbola Twitter: Follow Thomas Dorlo at @thomasdorlo The authors are grateful to all fellows and advisors of the PRDC consortium for answering the questionnaires. The authors are especially indebted to A Conesa-Botella and J Fokam, for their help in revising the questionnaires. Contributors: TPCD, CF, MT-B and DB conceived the outline of the manuscript. TPCD wrote the first draft and coordinated the preparation of the manuscript. CF, MT-B, DB and PJdV participated in the writing and revising of the manuscript. WFM coordinated the college and revised the manuscript. All the authors have approved the final version of the manuscript. Funding: This training programme was funded by the European Commission's 7th Framework Programme (FP7-HEALTH); under project reference number 223581. Competing interests: None declared. Provenance and peer review: Not commissioned; externally peer reviewed. Data sharing statement: No additional data are available.

Key questions What is already known about this topic? In developing countries, underuse of maternal healthcare services is attributed to an array of supply and demand factors, social structure and health beliefs. Among various socioeconomic factors, maternal education level and economic status are the most important determinants of usage of health services. What are the new findings? Wealth status, education and area of residence (urban vs rural) have a significant impact on uptake of all three types of maternal health services in the study population. Compared with the women in the poorest wealth quintile, those in the higher quintile have significantly higher odds of receiving at least four antenatal care visits, skilled birth attendance and postnatal care. Comparatively, women with relatively higher education had higher odds of receiving at least four antenatal care visits, skilled birth attendance and postnatal care. Women in rural areas were less likely to receive four antenatal visits, skilled birth attendance and postnatal care. Antenatal care usage is particularly lower in Malawi and more focus should be directed at improving antenatal visits as per the WHO standards. Recommendations for policy Provision of quality healthcare by increasing education and reducing area (urban vs rural) and wealth inequality should be a top public health priority in the Sustainable Development Goals (SDGs).

Antenatal care usage is particularly lower in Malawi and more focus should be directed at improving antenatal visits as per the WHO standards. Recommendations for policy Provision of quality healthcare by increasing education and reducing area (urban vs rural) and wealth inequality should be a top public health priority in the Sustainable Development Goals (SDGs). Since the healthcare system is fraught with a range of funding and logistical issues, more nuanced cooperation between local and international development organisations is needed to successfully achieve the maternal health-related targets in the country. The Government of Malawi should invest more in education and infrastructural development to begin removing the structural causes of non-use of maternal healthcare services. Introduction Despite constant efforts by the global community to reduce the burden of death arising from pregnancy and childbirth, maternal mortality (maternal deaths/100 000 live births) still remains a serious issue, affecting about 800 lives a day.1 Pregnancy/childbirth is arguably the most anticipated event in the life of a woman, yet complications during this period (from pregnancy to delivery) constitute the leading cause of death among women aged between 15 and 49 years.2 The Millennium Development Goal 5 (MDG 5) was dedicated to the goal of reducing, between 1990 and 2015, the maternal mortality ratio (MMR) by three-quarter.3

ife of a woman, yet complications during this period (from pregnancy to delivery) constitute the leading cause of death among women aged between 15 and 49 years.2 The Millennium Development Goal 5 (MDG 5) was dedicated to the goal of reducing, between 1990 and 2015, the maternal mortality ratio (MMR) by three-quarter.3 Although some countries have shown promising outcomes in terms of reduction in MMR, many are still behind track and continue to face maternal mortality as a major population health challenge. Statistics reveal that practically all-maternal deaths (99%) occur in the low and middle income countries (LMICs), among which Africa alone accounts for over 50% of the deaths.1 According to the reports by the United Nations Population Fund (UNDP), UNICEF and WHO, maternal mortality is as high as 1 in 11 in Eastern Africa compared with 1 in 3500 in North America and 1 in 4000 in Western Europe.4 The most important causes of maternal mortality in developing countries are unsafe abortion, haemorrhage, eclampsia and obstructed labour as they together account for nearly two-third of total maternal mortality globally.1 3 5 A growing consensus suggests that a vast majority of these deaths are actually preventable by adopting the necessary precautions provisioned through basic maternal healthcare services (MHS).6

rrhage, eclampsia and obstructed labour as they together account for nearly two-third of total maternal mortality globally.1 3 5 A growing consensus suggests that a vast majority of these deaths are actually preventable by adopting the necessary precautions provisioned through basic maternal healthcare services (MHS).6 The basic components of MHS: (1) during pregnancy (antenatal care), (2) during the intrapartum period (labour and delivery) and (3) during the postpartum period, postdelivery has been proposed as a key strategy to combat maternal mortality in resource-poor countries such as Malawi. In developed countries, nearly all women (98%) receive antenatal care (ANC) and deliver under the supervision of skilled health professionals (94%).7 In LMICs, on the other hand, about half of the women are deprived of ANC services8 and more than half of all births take place, outside institutional settings, mostly in unhygienic and unsafe conditions.9 10 The WHO recommends at least four antenatal visits for all pregnant women. However, almost half of the pregnant women worldwide do not receive this level of care, which is more common in LMICs. Poor attendance of ANC is associated with increased rates of low birthweight babies and neonatal deaths. Again, postpartum care is a crucial need for the survival of the mother and the newborns, as most maternal and infant deaths occur during this period. The WHO recommends that mothers and newborns receive initial postnatal care (PNC) within the first 24 hours after delivery and a minimum of three additional PNC visits within 48–72 hours, 7–14 days and 6 weeks after delivery.11 Despite the strong recommendation of the WHO in this context, this is the most neglected period for the provision of quality care.12 Unfortunately, a great majority of the women in LMICs remain deprived of the basic MHS due to various socioeconomic13 and cultural constraints.14

, 7–14 days and 6 weeks after delivery.11 Despite the strong recommendation of the WHO in this context, this is the most neglected period for the provision of quality care.12 Unfortunately, a great majority of the women in LMICs remain deprived of the basic MHS due to various socioeconomic13 and cultural constraints.14 Again, at the advocacy level, these issues are treated with a reductionist approach, although with a purpose only to relegate the complexities of implementation, given the diverse and intricate case-specific realities.15 In one of the studies in Tanzania, for example, despite higher antenatal care coverage and a positive notion about antenatal care on the part of women, there were significant gaps in antenatal care quality, based on factors such as avoiding repeated visits to the clinic, lack of money or concerns about caesarean deliveries.16 This catches the moot point of approaching indicators of maternal health like skilled birth attendance; antenatal care coverage and postnatal care with caution. Reasonable indicators may not necessarily mean reasonable maternal health. For example, the fuller potential of antenatal care strategy in reducing maternal mortality can be accomplished by providing what is called a ‘focused antenatal package’, including identifying possible obstetric complications and planning in advance for the emergency obstetric care possibilities and realities in terms of geographical location and service availabilities among others. A study in Uganda found that emergency obstetric care is a priority in reducing maternal mortality.17 Yet these aspects of MHS remain neglected.

c complications and planning in advance for the emergency obstetric care possibilities and realities in terms of geographical location and service availabilities among others. A study in Uganda found that emergency obstetric care is a priority in reducing maternal mortality.17 Yet these aspects of MHS remain neglected. The topic of socioeconomic inequality has received growing research attention in the domain of population health. Numerous studies have shown that the economically disadvantaged sections of the society are also the ones that suffer the worst health conditions. However, the association between economic inequality and MHS usage is less widely studied. Intuitively, economic constraints are a strong limiting factor for the accessibility and affordability of healthcare services for mothers from poor households. Women from well-off families are more likely to be able to pay for the costs associated with healthcare visits, medications and transportation.

widely studied. Intuitively, economic constraints are a strong limiting factor for the accessibility and affordability of healthcare services for mothers from poor households. Women from well-off families are more likely to be able to pay for the costs associated with healthcare visits, medications and transportation. Evidence from a Multiple Indicator Cluster Survey (MICS; 2007) study in Vanuatu found that mothers in the highest wealth status were, respectively, 5.50 and 2.12 times more likely to be assisted by skilled birth assistance (SBA) and have institutional deliveries.14 One Malawian study based on Demographic and Health Surveys (DHS 1992, 2000 and 2004) concluded that non-poor who suffered less frequently from selected diseases (including selected MHS) received more of the treatment/interventions, compared with the poorer counterparts, who with a greater proportion of disease burden get less of the interventions.18 Experience from other sub-Saharan nations also reveal similar situations.19 20 Inequality in MHS can greatly thwart progress towards maternal mortality-related MDGs and thus impede national progress owing to the direct and indirect losses arising from poor maternal and child health. It should also be noted that wealth inequality cannot be isolated from the compounding effects of other factors like education and the place of residence, that is, urban versus rural. With an aim to understand maternal healthcare seeking behaviour in relation to wealth inequality, education level and differences between urban and rural residence in Malawi, we conducted this study by analysing the most recent MICS data and estimated the rate of MHS usage and how usage status varies across selected variables.

aim to understand maternal healthcare seeking behaviour in relation to wealth inequality, education level and differences between urban and rural residence in Malawi, we conducted this study by analysing the most recent MICS data and estimated the rate of MHS usage and how usage status varies across selected variables. Methods About the survey and study population The MICS programme has the recognition of being the most comprehensive and reliable source of data on maternal and child health issues in developing countries. Operating in technical cooperation with UNICEF, the programme encompasses 108 countries and has completed over 280 surveys since its inception in 1995. The data sets serve crucial tools for monitoring progress towards MDGs and has become a vital component for evidence-based public health and social policymaking across countries. Data for the present study on Malawi was obtained from the fifth round of the survey (MICS 5), which was conducted in Malawi, from November 2013 to April 2014. The survey (Malawi MDG Endline Survey 2014) included information on various indicators of MDGs and other key socioeconomic and demographic variables and was carried out with an aim to measure progress towards MDGs and other development programmes in the country.21 The survey employed a multistage cluster-sampling strategy to select a sample population in 27 districts. In total, 24 230 women aged between 15 and 49 years were interviewed, with a response rate of 95.3%. However, an inclusion criterion for this study was the birth of a child in the past 2 years.

es in the country.21 The survey employed a multistage cluster-sampling strategy to select a sample population in 27 districts. In total, 24 230 women aged between 15 and 49 years were interviewed, with a response rate of 95.3%. However, an inclusion criterion for this study was the birth of a child in the past 2 years. Selection of variables The outcome variables of interest were antenatal care, skilled birth attendance and postnatal care. Wealth status, type of residence, that is, urban versus rural and education were the independent variables of interest. MHS included three basic components and was categorised as yes/no in SPSS, the statistical software package used for analysing the data. The component of MHS included the following: During pregnancy (ANC): As per the recommendation by World Health Organization, ANC was defined as having at least four visits to a qualified healthcare provider in their pregnancy. During the intrapartum period (SBA): This was determined by the usage of SBA during delivery. During the postpartum period (PNC): Whether or not mothers underwent a health check-up after delivery. The study considered a postnatal check-up within 48 hours after birth as a potential maternal healthcare indicator as per the WHO.

During the intrapartum period (SBA): This was determined by the usage of SBA during delivery. During the postpartum period (PNC): Whether or not mothers underwent a health check-up after delivery. The study considered a postnatal check-up within 48 hours after birth as a potential maternal healthcare indicator as per the WHO. Wealth index: Household wealth status is representative of an individual's affordability of expenses arising from healthcare needs. MICS programmes employ wealth index as a proxy indicator for household wealth status. The process involves assigning wealth scores, which is performed by principal components analysis, based on a selected range of household assets, for example, number of household members, floor, wall and roof material; type of cooking fuel; access to potable water and sanitation, ownership of radio, TV, refrigerator, motorcycle and others. Based on their weighted wealth scores, households fall into five wealth quintiles ranging from poorest to richest (poorest, poorer, middle, richer, richest). Measurement of wealth index is explained in detail elsewhere.21 Education: Education of the mother was categorised as those having education up to primary school and those having education of secondary school or higher. Type of residence: This variable included the categories of urban and rural residences.

Wealth index: Household wealth status is representative of an individual's affordability of expenses arising from healthcare needs. MICS programmes employ wealth index as a proxy indicator for household wealth status. The process involves assigning wealth scores, which is performed by principal components analysis, based on a selected range of household assets, for example, number of household members, floor, wall and roof material; type of cooking fuel; access to potable water and sanitation, ownership of radio, TV, refrigerator, motorcycle and others. Based on their weighted wealth scores, households fall into five wealth quintiles ranging from poorest to richest (poorest, poorer, middle, richer, richest). Measurement of wealth index is explained in detail elsewhere.21 Education: Education of the mother was categorised as those having education up to primary school and those having education of secondary school or higher. Type of residence: This variable included the categories of urban and rural residences. Covariates Defining explanatory variables: Selected socioeconomic and demographic variables such as mother's age at birth, attended school: yes/no; religious faith; wealth quintile (poorest, poorer, middle, richer, richest); type of residence, that is, urban versus rural and geographical region of residence, that is, Northern, Central or Southern, were used as covariates. Three outcome variables for MHS included in the study were antenatal care, skilled birth attendance and PNC for mothers.

ntile (poorest, poorer, middle, richer, richest); type of residence, that is, urban versus rural and geographical region of residence, that is, Northern, Central or Southern, were used as covariates. Three outcome variables for MHS included in the study were antenatal care, skilled birth attendance and PNC for mothers. Data analysis The sociodemographic characteristics of participants were analysed by descriptive statistics. Cross-tabulation was performed, and χ2 bivariate tests were used to check for statistical significance with MHS usage status and as a guide to the explanatory variables which are to be included in the multivariate analysis. All the covariates were entered as categorical variables. Variables that were found to have significant association from the χ2 test in cross-tabulation were entered into the regression model. Given the clustered nature of the survey, we used generalised estimating equations (GEE) for regression analysis. The aim of the final analysis was to adjust for potential confounders and calculate the ORs to assess the likelihood of using MHS. A p Value of <0.05 (two-tailed) was considered statistically significant for all associations. Data analysis was performed using SPSS V.24 for Windows (SPSS, Chicago, Illinois, USA). Ethical approval This research was based on secondary data available in the public domain by the courtesy of the MICS programme of UNICEF and hence was not subject to ethical approval.

Data analysis The sociodemographic characteristics of participants were analysed by descriptive statistics. Cross-tabulation was performed, and χ2 bivariate tests were used to check for statistical significance with MHS usage status and as a guide to the explanatory variables which are to be included in the multivariate analysis. All the covariates were entered as categorical variables. Variables that were found to have significant association from the χ2 test in cross-tabulation were entered into the regression model. Given the clustered nature of the survey, we used generalised estimating equations (GEE) for regression analysis. The aim of the final analysis was to adjust for potential confounders and calculate the ORs to assess the likelihood of using MHS. A p Value of <0.05 (two-tailed) was considered statistically significant for all associations. Data analysis was performed using SPSS V.24 for Windows (SPSS, Chicago, Illinois, USA). Ethical approval This research was based on secondary data available in the public domain by the courtesy of the MICS programme of UNICEF and hence was not subject to ethical approval. Results Sociodemographic characteristics Table 1 shows the results of the descriptive analysis (frequencies and percentages) on the sociodemographic characteristics of the sample population. In total, 7572 women were included in the study with an average age of 26.88 (±6.68) years. A high number of pregnancies (46.5%) were noted in the high-risk age range, either between 15 and 19 years or above 30 years. It should be noted as per Donoso et al22 that 20–29 years is the age range with a lesser general reproductive risk. About 1 in 10 reported ‘yes’ to the question ‘ever attended school’ (88.8%). Nearly one-fifth of the population belonged to the Christian faith. The population with Muslim faith was 14.2%, and 3.8% and 0.5% belonged to either no religion or other religion, respectively. In terms of education, only 20.1% had secondary education or higher and 79.9% had education up to primary school. Wealth status quintiles were distributed in the study sample, with 23.6%, 22.7% and 21.3% belonging to the poorest, poor and middle quintiles, respectively, whereas 17.3% and 15% belonged to the rich and richest quintiles of the study sample, respectively. The majority (88.8%) of the study sample belonged to the rural areas, whereas geographical distribution of the study sample closely matched the census 2010 data, with 17.2% belonging to the Northern region, 33.9% belonging to the Central region and 49% belonging to the Southern region. Rates of use of MHS by type, usage of MHS at different education levels, wealth groups and areas (urban vs rural) are presented in figures 1–4 below. As seen in figure 1, only 44.70% of women use antenatal care (ANC) services as per the WHO defined criteria of at least four ANC visits during pregnancy. The rates of use of PNC and SBA are above 80%. Similarly, figure 2 shows the usage of MHS, ANC, PNC and SBA at different education levels. Figure 3 shows usage of MHS in different wealth quintiles and figure 4 shows the usage of MHS in different areas.

WHO defined criteria of at least four ANC visits during pregnancy. The rates of use of PNC and SBA are above 80%. Similarly, figure 2 shows the usage of MHS, ANC, PNC and SBA at different education levels. Figure 3 shows usage of MHS in different wealth quintiles and figure 4 shows the usage of MHS in different areas. Table 1 Cross-tabulation results with covariates

WHO defined criteria of at least four ANC visits during pregnancy. The rates of use of PNC and SBA are above 80%. Similarly, figure 2 shows the usage of MHS, ANC, PNC and SBA at different education levels. Figure 3 shows usage of MHS in different wealth quintiles and figure 4 shows the usage of MHS in different areas. Table 1 Cross-tabulation results with covariates Covariates; n=7572 n% ANC>4 times SBA PNC Age Low risk group 4052 (53.50%) 1735 (43.90%) 3566 (89%) 3282 (82.50%) High risk group 3520 (46.50%) 1558 (47.30%) 3011 (86.40%) 2832 (81.80%) p Value NS p<0.001 NS Education Primary/less 5372 (79.90%) 5219 (43.30%) 4635(87.10%) 4288 (81.30%) Secondary/higher 1351 (20.10%) 700 (52.60%) 1254 (93.60%) 1183 (88.50%) p Value p<0.0001 p<0.0001 p<0.0001 Religion Christian 6176 (81.60%) 2714 (45.10%) 5371 (87.80%) 5031 (82.80%) Muslim 1072 (14.20%) 450 (43.60%) 935 (88.30%) 838 (79.60%) No religion 285 (3.80%) 115 (41.10%) 235 (83.60%) 215 (78.50%) Other 39 (0.50%) 14 (36.80%) 36 (92.30%) 30 (78.90%) p Value NS NS p=0.012 Wealth index Poorest 1789 (23.60%) 721 (41.40%) 1506 (84.80%) 1379 (78.21) Second 1720 (22.70%) 711 (42.60%) 1468 (86.30%) 1355 (80.60%) Middle 1616 (21.30%) 684 (43.3%) 1408 (87.90%) 1310 (82.30%) Fourth 1312 (17.30%) 587 (46%) 1148 (88.70%) 1082 (84.10%) Richest 1135 (15%) 590 (53.60%) 1047 (93.40%) 988 (88.50%) p Value p<0.0001 p<0.0001 p<0.0001 Region Northern 1299 (17.20%) 542 (42.30%) 1154 (89.50%) 1118 (87.50%) Central 2566 (33.9%) 1117 (44.6%) 2204 (86.80%) 2110 (83.60%) Southern 3707 (49%) 1634 (45.70%) 3219 (87.90%) 2886 (79.40%) p Value NS p=0.025 p<0.0001 Area Urban 866 (11.40%) 458 (54.10%) 800 (93.20%) 758 (88.60%) Rural 6706 (88.60%) 2835 (43.50%) 5277 (87.10%) 5356 (81.30%) p Value p<0.0001 p<0.0001 p<0.0001 The p denotes level of significance estimated from the χ2 test.

3.60%) Southern 3707 (49%) 1634 (45.70%) 3219 (87.90%) 2886 (79.40%) p Value NS p=0.025 p<0.0001 Area Urban 866 (11.40%) 458 (54.10%) 800 (93.20%) 758 (88.60%) Rural 6706 (88.60%) 2835 (43.50%) 5277 (87.10%) 5356 (81.30%) p Value p<0.0001 p<0.0001 p<0.0001 The p denotes level of significance estimated from the χ2 test. ANC, antenatal care; NS, not significant; PNC, postnatal care; SBA, skilled birth assistance. Figure 1 Use of MHS by type in Malawi. MICS 2013–2014. ANC, antenatal care; MHS, maternal healthcare service; MICS, Multiple Indicator Cluster Survey; PNC, postnatal care; SBA, skilled birth assistance. Figure 2 Utilisation rate of MHS at different education levels in Malawi. MICS 2013–2014. ANC, antenatal care; MHS, maternal healthcare service; MICS, Multiple Indicator Cluster Survey; PNC, postnatal care; SBA, skilled birth assistance. Figure 3 Utilisation rate of MHS in different wealth groups in Malawi. MICS 2013–2014. ANC, antenatal care; MHS, maternal healthcare service; MICS, Multiple Indicator Cluster Survey; PNC, postnatal care; SBA, skilled birth assistance. Figure 3 Continued Figure 3 Continued Figure 4 Utilisation rate of MHS in different areas in Malawi. MICS 2013–2014. ANC, antenatal care; MHS, maternal healthcare service; MICS, Multiple Indicator Cluster Survey; PNC, postnatal care; SBA, skilled birth assistance. As per the χ2 tests, education, wealth index quintiles and area (urban vs rural) were significantly associated with the usage status of all the three types of MHS. Detailed cross-tabulation results are as displayed in table 1 below.

Figure 4 Utilisation rate of MHS in different areas in Malawi. MICS 2013–2014. ANC, antenatal care; MHS, maternal healthcare service; MICS, Multiple Indicator Cluster Survey; PNC, postnatal care; SBA, skilled birth assistance. As per the χ2 tests, education, wealth index quintiles and area (urban vs rural) were significantly associated with the usage status of all the three types of MHS. Detailed cross-tabulation results are as displayed in table 1 below. Table 2 shows the usage of additional health services generally prescribed for pregnant mothers. Well over four-fifth of the women had their blood pressure measured and a little less than one-third had their urine sample tested. A majority of the women (93.7%) had their blood sample taken during pregnancy. Table 2 Additional health services during pregnancy

Table 2 shows the usage of additional health services generally prescribed for pregnant mothers. Well over four-fifth of the women had their blood pressure measured and a little less than one-third had their urine sample tested. A majority of the women (93.7%) had their blood sample taken during pregnancy. Table 2 Additional health services during pregnancy Tests Percentage of women who reported yes Any tetanus toxoid injection during pregnancy 83.40 Blood pressure measured during pregnancy 87.70 Blood sample taken during pregnancy 93.70 Urine sample taken during pregnancy 31.30 Multivariate analysis Table 3 shows the ORs with 95% CIs obtained from the multiple logistic regression for usage status of MHS. Logistic regression was performed to ascertain the effects of wealth index, education, area of residence, that is, urban versus rural, religion and region on the likelihood that participants have used ANC. The logistic regression model was statistically significant, χ2 (5)=68.204, p<0.0001. The model explained 1.4% (Nagelkerke R2) of the variance in ANC and correctly classified 56.3% of cases. Sensitivity was 20.3%, specificity was 85.9%, positive predictive value was 54.3% and negative predictive value was 56.63%. Of the five predictor variables, only four were statistically significant: wealth index, education, area and region. Moving from poorest to richest, a one-unit increase in the wealth index quintile had 1.063 times higher odds of having used ANC services as per the WHO standards. Similarly, moving from primary or lower education to secondary or higher education had 1.3 times higher odds of having used ANC services at the given standards. A unit increase in area, that is, moving from urban to rural, decreased the likelihood of using ANC services and, in terms of region, moving from the Northern to Southern region increased the likelihood of using ANC services. The results of logistic regression are as shown in table 3 below.

ices at the given standards. A unit increase in area, that is, moving from urban to rural, decreased the likelihood of using ANC services and, in terms of region, moving from the Northern to Southern region increased the likelihood of using ANC services. The results of logistic regression are as shown in table 3 below. Table 3 Results of logistic regression model showing OR and CI, sensitivity, specificity and PPV and NPV for respective models, for usage of different types of MHS among women in Malawi, 2013–2014 Variable OR Sensitivity (%) Specificity (%) PPV (%) NPV (%) SBA Wealth index 1.10 (1.03 to 1.17) 100 0 88.40 0 Education 1.79 (1.40 to 2.29) Area 0.68 (0.50 to 0.93) Religion 1.00 (0.87 to 1.16) Region 1.02 (0.92 to 1.13) PNC Wealth index 1.08 (1.02 to 1.14) 100 0 82.70 0 Education 1.44 (1.19 to 1.75) Area 0.71 (0.55 to 0.91) Religion 0.94 (0.83 to 1.06) Region 1.02 (0.92 to 1.13) ANC Wealth index 1.06 (1.02 to 1.10) 20 86 54.30 56.60 Education 1.30 (1.14 to 1.48) Area 0.81 (0.69 to 0.96) Religion 0.99 (0.89 to 1.09) Region 1.10 (1.03 to 1.18) ANC, antenatal care; NPV, negative predictive value; PNC, postnatal care; PPV, positive predictive value; SBA, skilled birth assistance.

02 (0.92 to 1.13) ANC Wealth index 1.06 (1.02 to 1.10) 20 86 54.30 56.60 Education 1.30 (1.14 to 1.48) Area 0.81 (0.69 to 0.96) Religion 0.99 (0.89 to 1.09) Region 1.10 (1.03 to 1.18) ANC, antenatal care; NPV, negative predictive value; PNC, postnatal care; PPV, positive predictive value; SBA, skilled birth assistance. Discussion On the basis of the MICS wave 5 data, this study attempts to demonstrate the impact of wealth inequality, education and area of residence on selected indicators of MHS usage in Malawi. Despite a considerable drop in MMR at the global stage during the past few decades (44% between 1990 and 2015), progress has been lowest in sub-Saharan Arica (SSA) as the countries continue to share a disproportionate burden of maternal and neonatal mortalities. According to the 2015 MDG progress report, SSA accounts for about two-third of maternal and half of neonatal mortality globally.23 The same report classified Malawi as having a very high MMR in 2015 with about 634 deaths per 100 000 live births. Several studies have attempted to explore the root causes of high MMR in Malawi. While a growing body of literature is documenting the impact of inequality, education and area of residence, that is, urban versus rural, in maternal health, quality evidence is lacking for countries in SSA. Previous experience from developing regions in Asia and Africa reveals a positive association between MHS uptake and wealth inequality.19 24 25 Studies have also indicated the impact of education and the area of residence, that is, urban versus rural, on maternal health services usage.26

s lacking for countries in SSA. Previous experience from developing regions in Asia and Africa reveals a positive association between MHS uptake and wealth inequality.19 24 25 Studies have also indicated the impact of education and the area of residence, that is, urban versus rural, on maternal health services usage.26 In this study, we sought to investigate how household wealth inequality, area (urban vs rural) and education affect the usage status of MHS among Malawian women. Our finding showed that wealth status had a significant impact on uptake of all three types of MHS in the study population. Compared with the women in the poorest wealth quintile, those in the higher quintile have significantly higher odds of receiving at least four ANC visits, skilled birth attendance and attending PNC. Our results are consistent with past findings. A previous study by analysing DHS data for the years between 1990 and 1998 in 45 developing countries showed that the use of skilled assistance at delivery and antenatal care is 80% or higher for the richest quintile.27 Another DHS study in 56 countries during 1990–2002 found that women in the richest quintile were nearly five times more likely to experience skilled assistance at delivery than the poorest.28 This finding indicates that wealth inequality is a limiting factor for MHS usage in Malawi. Past evidences suggest that addressing wealth inequalities in MHS usage is essential for achieving the maternal health-related MDGs.11

early five times more likely to experience skilled assistance at delivery than the poorest.28 This finding indicates that wealth inequality is a limiting factor for MHS usage in Malawi. Past evidences suggest that addressing wealth inequalities in MHS usage is essential for achieving the maternal health-related MDGs.11 Financial barriers to usage of facility-based care are prohibitive among the poor, even where the actual care is free of charge. Some countries in the SSA are implementing policies to lower/exempt direct out-of-pocket (OOP) costs29 to promote maternal health in the region. Direct OOP costs associated with maternity care include all formal, official fees charged for delivery care, bed stay, and required drugs and supplies. In addition to direct financial expenditures, there may be additional indirect costs of care seeking, such as lost wages or earnings. Such costs are difficult to measure as they vary according to income and employment status, and may be subject to seasonal variation as well. Indirect costs of care seeking can exceed direct OOP costs. Owing to system inefficiency and poor accountability and transparency, unofficial fees were on average 12 times higher than official fees.30 National health policymaking should take into account the direct as well as indirect expenditures to promote MHS uptake among the disadvantaged sections of the society.

OOP costs. Owing to system inefficiency and poor accountability and transparency, unofficial fees were on average 12 times higher than official fees.30 National health policymaking should take into account the direct as well as indirect expenditures to promote MHS uptake among the disadvantaged sections of the society. In Malawi, healthcare financing is faced with serious constraints and is highly dependent on external sources of financing. Budgetary failure (Abuja Declaration of 15% of the national budget), decreasing share of private sources in healthcare expenditure and rising health expenditure (US$12 in 1998/1999 to US$25 in 2005/2006) are concerns for healthcare financing in the country.31 More than three quarters of the population live below poverty line. Hence despite reduction in out-of-pocket payments, this still remains worth consideration. Poor households usually spend a large share of income for food and any amount of spending can be competitive for household food availability and education of children. Thus, the burden of maternal healthcare is unlikely to be affordable especially for poor households.

t payments, this still remains worth consideration. Poor households usually spend a large share of income for food and any amount of spending can be competitive for household food availability and education of children. Thus, the burden of maternal healthcare is unlikely to be affordable especially for poor households. Similarly, those residing in rural areas are significantly less likely to use maternal health services as seen by logistic regression results. Education was another variable, which had a significant impact on utilisation of all three types of services. Those with no education or up to primary school level had lower odds for using all the three types of MHS compared with those with secondary or higher education. Given the relatively higher rates of usage of all three types of MHS, the impact of area and education become even more relevant. It raises a moot point about the quality of MHS, which might be subject to the impact of area and education. For example, a study from Tanzania noted that those in a rural set-up face barriers of transportation and reaching the health facility to receive appropriate antenatal care.32 Area of residence impacts the quality of MHS services through standards of care for antenatal visits, timing of postpartum care and identification of intra-partum risk factors, as found from one of the studies in rural India.33

of transportation and reaching the health facility to receive appropriate antenatal care.32 Area of residence impacts the quality of MHS services through standards of care for antenatal visits, timing of postpartum care and identification of intra-partum risk factors, as found from one of the studies in rural India.33 Another qualitative study from Malawi has shown that there are factors apart from maternal health services usage, like lack of appropriate resources for maternal health services, overloaded staff, etc. A study from Sudan also revealed a substantial impact of education and area of residence on the quality of maternal health services received as well as the rates of maternal health services.34 Another study from rural Tanzania revealed the overarching influence of the rural set-up on the perception of postpartum complication and quality of health services received.16 It should be noted, however, that this study is related to the impact of area of residence, education and wealth inequity on the usage of maternal health services and the point about quality of maternal health services calls for further research. The discussion does, however, underscore the influence of factors beyond mere utilisation rates of maternal health services and emphasises the import of focusing on reducing maternal mortality versus improving indicators of maternal health.

the point about quality of maternal health services calls for further research. The discussion does, however, underscore the influence of factors beyond mere utilisation rates of maternal health services and emphasises the import of focusing on reducing maternal mortality versus improving indicators of maternal health. Conclusion The finding of this study reveals reasonable rates of usage of MHS. The rates of usage are significantly impacted by the differences in education, area of residence and wealth inequalities. Barriers to maternal health care due to socioeconomic and cultural factors are well recognised in the country, which necessitates special intervention programs that directly benefit the poor, particularly in most underdeveloped areas. The focus should also be on increasing women's education above secondary/higher levels. Despite reasonable rates of MHS usage, the maternal mortality rates continue to remain high. Therefore, provision of quality healthcare by increasing education and reducing wealth inequality as well as reducing the urban–rural divide should be a top public health priority in the Sustainable Development Goals (SDGs). Since the healthcare system is fraught with a range of funding and logistical issues, more nuanced cooperation between local and international development organisations is needed to successfully achieve the maternal health-related targets in the country.

blic health priority in the Sustainable Development Goals (SDGs). Since the healthcare system is fraught with a range of funding and logistical issues, more nuanced cooperation between local and international development organisations is needed to successfully achieve the maternal health-related targets in the country. Limitations of the study As an observational study, the findings do not indicate a cause–effect relationship between wealth inequality, education and area of residence with usage of MHS. The survey also relied on participants' ability to correctly recall the timing and frequency of the services they availed. So there is a strong possibility of recall error and under-reporting by the participants. Handling editor: Valery Ridde Competing interests: None declared. Provenance and peer review: Not commissioned; externally peer reviewed. Data sharing statement: No additional data are available.

Key questions What is already known about this topic? Skilled birth attendance and emergency obstetric care are potentially life-saving care packages which need to be available 24/7 at health facility level to reduce maternal and newborn mortality. There is a lack of information regarding how routine health services are affected when large-scale severe epidemics occur. What are the new findings? Across all districts in Sierra Leone, during the Ebola virus epidemic, there was an 18% decrease in the number of women attending for antenatal care, a 22% decrease in women seeking postnatal care and a 11% decrease in the number of women attending for birth at a healthcare facility. During the Ebola virus epidemic, there was a 34% increase in the facility maternal mortality ratio and 24% increase in the stillbirth rate. Recommendations for policy In the post-Ebola phase, ‘readiness’ (or not) for future epidemics has been the focus of much debate. This ‘readiness’ is particularly important in fragile states where the impact of epidemics may be greater. Emergency preparedness plans need to be in place that take into account the capacity of healthcare facilities to provide both routine and emergency care as well as the need for early community mobilisation and involvement.

diness’ is particularly important in fragile states where the impact of epidemics may be greater. Emergency preparedness plans need to be in place that take into account the capacity of healthcare facilities to provide both routine and emergency care as well as the need for early community mobilisation and involvement. Introduction In May 2014, Sierra Leone, along with Guinea and Liberia, was hit by the biggest Ebola virus disease (EVD) epidemic ever recorded. The EVD epidemic was officially declared over in November 2015 by the Government of Sierra Leone and the WHO; there had been 8704 cases and 3589 deaths.1 2 However, there was a flare up of EVD reported in Sierra Leone in January 2016 and declaration that the country was again free of EVD on 17 March 2016.3 It is considered likely that the full impact of the EVD outbreak is not just related to the disease itself but also to its effect on other aspects of healthcare provision. This might be particularly so for maternity services as these are expected to be available 24/7 at health facility level. At least 80% of all maternal deaths globally result from five complications which are well understood and can be prevented or managed by experienced healthcare providers. Key services that need to be in place include antenatal (ANC) and postnatal care (PNC), skilled birth attendance (SBA) and, for the 15% of women who are expected to have a complication, access to and availability of, either basic (BEmOC) or comprehensive emergency obstetric care (CEmOC; table 1).4 Table 1 Signal functions of EmOC17

Introduction In May 2014, Sierra Leone, along with Guinea and Liberia, was hit by the biggest Ebola virus disease (EVD) epidemic ever recorded. The EVD epidemic was officially declared over in November 2015 by the Government of Sierra Leone and the WHO; there had been 8704 cases and 3589 deaths.1 2 However, there was a flare up of EVD reported in Sierra Leone in January 2016 and declaration that the country was again free of EVD on 17 March 2016.3 It is considered likely that the full impact of the EVD outbreak is not just related to the disease itself but also to its effect on other aspects of healthcare provision. This might be particularly so for maternity services as these are expected to be available 24/7 at health facility level. At least 80% of all maternal deaths globally result from five complications which are well understood and can be prevented or managed by experienced healthcare providers. Key services that need to be in place include antenatal (ANC) and postnatal care (PNC), skilled birth attendance (SBA) and, for the 15% of women who are expected to have a complication, access to and availability of, either basic (BEmOC) or comprehensive emergency obstetric care (CEmOC; table 1).4 Table 1 Signal functions of EmOC17 BEmOC CEmOC 1. iv/im antibiotics All included in BEmOC (1–7) plus 2. iv/im oxytocic drugs 8. Blood transfusion 3. iv/im anticonvulsants 9. Caesarean section 4. Manual removal of retained placenta 5. Removal of retained products of conception (by manual vacuum aspiration) 6. Assisted vaginal delivery (ventouse delivery) 7. Resuscitation of the newborn (using a bag and mask) BEmOC, basic emergency obstetric care; CEmOC, comprehensive emergency obstetric care; EmOC, emergency obstetric care; im, intramuscular; iv, intravenous.

retained products of conception (by manual vacuum aspiration) 6. Assisted vaginal delivery (ventouse delivery) 7. Resuscitation of the newborn (using a bag and mask) BEmOC, basic emergency obstetric care; CEmOC, comprehensive emergency obstetric care; EmOC, emergency obstetric care; im, intramuscular; iv, intravenous. It can be difficult to differentiate between complications of pregnancy such as obstetric haemorrhage (the leading cause of maternal death in low resource settings) and the case definition of EVD, which could affect patient management.5 Women suspected of having EVD would normally be isolated until their Ebola status is confirmed. However, such isolation may also prevent women from receiving timely obstetric care.6 Obstetric care is considered one of the most high-risk areas for exposure to body fluids, through which EVD is spread, and as a consequence healthcare providers face increased risks of exposure to Ebola and so may be reluctant to assist at the time of birth or carry out invasive procedures if adequate protection is not in place.

care is considered one of the most high-risk areas for exposure to body fluids, through which EVD is spread, and as a consequence healthcare providers face increased risks of exposure to Ebola and so may be reluctant to assist at the time of birth or carry out invasive procedures if adequate protection is not in place. In October 2014, the United Nations Population Fund (UNFPA) estimated that 800 000 women were due to give birth over the following 12 months in the three EVD affected countries and that 12 000 women and babies would require some level of emergency obstetric care (EmOC).7 Before the onset of the EVD epidemic, Sierra Leone had made excellent progress towards achieving Millennium Development Goal 5 with an estimated 52% reduction in the maternal mortality ratio between 1990 and 2013 and 97% of women attending for one or more ANC visits.8 However, with a still fragile health system and an estimated maternal mortality ratio of 1100/100 000 live births, Sierra Leone could ill afford to lose the gains it had made.9 Although there has been significant interest in, and critique of, the international response to the EVD epidemic,10 much less is known about ‘readiness’ and functioning of the existing health system for non-EVD-related and more routine health service provision which, it could be argued, is at least as important during an epidemic. This study aimed to look at the impact of the EVD epidemic on the availability, uptake and outcomes of maternal and newborn health services in Sierra Leone.

of the existing health system for non-EVD-related and more routine health service provision which, it could be argued, is at least as important during an epidemic. This study aimed to look at the impact of the EVD epidemic on the availability, uptake and outcomes of maternal and newborn health services in Sierra Leone. Methods Selection of healthcare facilities Sierra Leone is divided into 13 districts, each district has one healthcare facility designated to provide CEmOC and five or six designated to provide BEmOC. During the epidemic, there was concern about the impact of Ebola on pregnant women, in particular the difficultly in differential diagnosis between obstetric emergencies and infection with the Ebola virus6 and the increased risk to healthcare workers because of the high exposure to body fluids during obstetric care. The decision to focus on provision and uptake of care in BEmOC and CEmOC facilities was based on their key role in providing EmOC. Data were unavailable from two BEmOC facilities (one each in Bo and Kenema districts) as contact with these facilities could not be established. Thus, we included all 13 facilities designated to provide CEmOC across Sierra Leone and 65 of 67 facilities designated to provide BEmOC. All included healthcare facilities are designated to provide ANC, SBA and PNC.

ilities (one each in Bo and Kenema districts) as contact with these facilities could not be established. Thus, we included all 13 facilities designated to provide CEmOC across Sierra Leone and 65 of 67 facilities designated to provide BEmOC. All included healthcare facilities are designated to provide ANC, SBA and PNC. Data collection The numbers of EVD cases and peak incidence of the disease varied over time across the 13 districts of Sierra Leone, with some districts reporting higher overall numbers than others and a corresponding difference in facility impact. Therefore, we collected data on the number of EVD cases per week per district from confirmed patient databases and situation reports from the National Ebola Response Centre (http://nerc.sl/).

ierra Leone, with some districts reporting higher overall numbers than others and a corresponding difference in facility impact. Therefore, we collected data on the number of EVD cases per week per district from confirmed patient databases and situation reports from the National Ebola Response Centre (http://nerc.sl/). An electronic data collection tool developed at the Centre for Maternal and Newborn Health—Liverpool School of Tropical Medicine (LSTM) was used to aid data collection and possibly reduce the risk of any cross-infection. Data were collected by experienced LSTM Sierra Leonean technical officers based in Freetown who had worked within the health system and understood the impact Ebola was having on the country. All data collectors were given instructions on how to maintain their own safety when visiting facilities and where allowed to restrict their visits if they felt their own safety was at risk. Data collectors reported any difficulties in data collection due to the epidemic and a collective decision was made whether to suspend or delay data collection dependent on conditions within the each facility. Data were able to be gathered from all but two of the targeted health facilities.

heir own safety was at risk. Data collectors reported any difficulties in data collection due to the epidemic and a collective decision was made whether to suspend or delay data collection dependent on conditions within the each facility. Data were able to be gathered from all but two of the targeted health facilities. Data were obtained on the availability of healthcare providers, availability and provision of EmOC signal functions, drugs, equipment, number of ANC and PNC visits, number of births at the facility, reported number of emergency complications, maternal deaths, and stillbirths.11 Signal functions were classified as being available if the equipment, drugs and appropriately trained staff were available to perform the signal function. Data for the availability of healthcare providers was obtained from facility attendance registers. Data regarding availability (or not) of equipment and drugs was obtained from facility registers. Information on each indicator of interest was obtained retrospectively from facility registers for each month (for each of the 12 months before and 10 months during the EVD epidemic) during a visit made to each health facility for this purpose by trained staff based in Freetown. Data analysis For each facility, data for each month was available and analysed for four groups of outcomes: (1) number of staff available by cadre; (2) availability of each EmOC signal function; (3) number of antenatal visits, postnatal visits and births; (4) maternal deaths and stillbirths.

Information on each indicator of interest was obtained retrospectively from facility registers for each month (for each of the 12 months before and 10 months during the EVD epidemic) during a visit made to each health facility for this purpose by trained staff based in Freetown. Data analysis For each facility, data for each month was available and analysed for four groups of outcomes: (1) number of staff available by cadre; (2) availability of each EmOC signal function; (3) number of antenatal visits, postnatal visits and births; (4) maternal deaths and stillbirths. To examine the possible effects of the EVD epidemic on each of these outcomes, mixed-effects models were used because of the longitudinal nature of the data. To assess the impact of the EVD epidemic on each of the outcomes of interest, an indicator variable was defined (occurrence of one or more EVD cases in district in the immediately preceding month). Relevant risk factors were: type of facility (BEmOC or CEmOC), district and month of the year; each of these was defined as a categorical variable. For all analyses, facilities were treated as random effects. An alternative approach to account for the EVD epidemic used the occurrence of EVD cases in any preceding month. This approach did not yield different results unless indicated in the Results section. A further alternative approach would be to use the first occurrence of EVD in the country. This approach was not considered as it would be less sensitive to the likely variation within districts and between healthcare facilities.

th. This approach did not yield different results unless indicated in the Results section. A further alternative approach would be to use the first occurrence of EVD in the country. This approach was not considered as it would be less sensitive to the likely variation within districts and between healthcare facilities. p Values were obtained using likelihood ratio tests, and when this was not possible (because fitting of the submodel with Ebola excluded was not possible), Wald tests are reported. A p value <0.05 was considered to be statistically significant. Tests of interaction between facility type and EVD were performed using models which involved only the explanatory variables facility type and EVD presence. For availability of healthcare providers Mixed-effects Poisson regression models were used, to model separately the effect of the EVD epidemic on the total number of staff and the number of staff in each cadre. Poisson models were used because of the count nature of the data. Type of facility and district were both included. Separate district-level analyses were not performed since most healthcare providers are drawn from a central pool and a population of limited size. Means and ratios are reported for occurrences of Ebola and type of facility.

used because of the count nature of the data. Type of facility and district were both included. Separate district-level analyses were not performed since most healthcare providers are drawn from a central pool and a population of limited size. Means and ratios are reported for occurrences of Ebola and type of facility. For availability of EmOC For availability of signal functions (a binary outcome, available or not available), mixed-effects logistic regression models were used. For each month, the outcome was whether the signal function was in principle available or not. ORs are reported for the association of availability of each signal function with onset of the EVD epidemic after accounting for variables within the analysis. District was included for signal functions 1, 2 and 3; type of facility for signal functions 1, 2, 3 and 7 (otherwise they were omitted to ensure plausibility of the model); month was included for all signal functions. For each facility assessment, availability (or not) of the required cadre of healthcare provider, equipment and/or drugs needed for each of the signal functions of EmOC was assessed. For uptake of services For numbers of events (ANC visits, PNC visits and births) mixed-effects Poisson regression models were used. Type of facility, month and district were each included in analysis of data for all districts. For analysis of each district, type of facility and month were included. Ratios of mean number of events per facility are reported for onset of EVD epidemic and type of facility.

d-effects Poisson regression models were used. Type of facility, month and district were each included in analysis of data for all districts. For analysis of each district, type of facility and month were included. Ratios of mean number of events per facility are reported for onset of EVD epidemic and type of facility. Maternal deaths and stillbirths The stillbirth rate was calculated for each facility as number of stillbirths recorded per 1000 live births. The maternal mortality ratio was calculated as number of maternal deaths recorded per live 100 000 births. Mixed-effects Poisson regression models were used, with number of live births used to define exposure, and thus derive incidence rates. Type of facility, district and month were included in the analysis of data for all districts. For analysis of each district, facility was included when there were deaths for both types of facility (CEmOC and BEmOC), otherwise only data for the type of facility at which deaths occurred were used. Incidence rate ratios (IRRs) are reported for EVD onset (or not) and for type of facility (when included). The statistical package Stata V.12.1 was used for all analyses. Ethical approval Ethical approval for the study was obtained from the Sierra Leone Research and Scientific Committee and from the LSTM Ethics Committee (reference number 15.004RS). Results Data for a total 78 facilities were available for 22 months, giving a total of 1716 month–facility combinations of which 474 with EVD and 1242 with no EVD present. All districts had at least 4 months in which EVD cases were reported (figure 1).

Ethical approval Ethical approval for the study was obtained from the Sierra Leone Research and Scientific Committee and from the LSTM Ethics Committee (reference number 15.004RS). Results Data for a total 78 facilities were available for 22 months, giving a total of 1716 month–facility combinations of which 474 with EVD and 1242 with no EVD present. All districts had at least 4 months in which EVD cases were reported (figure 1). Figure 1 Number of Ebola Virus cases in Sierra Leone, number of antenatal visits (ANC), postnatal care (PNC) visits and births at health facility level (institutional delivery). Availability of healthcare providers Overall, there was a small 3% (IRR 1.03, 95% CI 1.00 to 1.07; p=0.09), but not statistically significant increase in the total number of healthcare providers deployed and working at both CEmOC-level and BEmOC-level health facilities (excluding students and traditional birth attendants) following the onset of the EVD epidemic (table 2). However, the number of student healthcare providers decreased by two-thirds (IRR 0.33, 95% CI 0.29 to 0.37; p<0.001). When combined, the total (including trainees) number of healthcare providers in place reduced by 8% (IRR 0.92, 95% CI 0.90 to 0.95). Table 2 Association between the EVD epidemic and numbers and cadres of healthcare providers in post in Sierra Leone at healthcare facilities designated to provide BEmOC or CEmOC

Availability of healthcare providers Overall, there was a small 3% (IRR 1.03, 95% CI 1.00 to 1.07; p=0.09), but not statistically significant increase in the total number of healthcare providers deployed and working at both CEmOC-level and BEmOC-level health facilities (excluding students and traditional birth attendants) following the onset of the EVD epidemic (table 2). However, the number of student healthcare providers decreased by two-thirds (IRR 0.33, 95% CI 0.29 to 0.37; p<0.001). When combined, the total (including trainees) number of healthcare providers in place reduced by 8% (IRR 0.92, 95% CI 0.90 to 0.95). Table 2 Association between the EVD epidemic and numbers and cadres of healthcare providers in post in Sierra Leone at healthcare facilities designated to provide BEmOC or CEmOC Mean number of healthcare providers per healthcare facility EVD epidemic* Facility providing BEmOC Facility providing CEmOC Cadre of staff No Ebola (n=1035) Ebola (n=395) No Ebola (n=207) Ebola (n=79) Ratio of means (95% CI) p Value Specialist doctor 0 0 0.34 0.39 1.00 (0.67 to 1.51) 0.99 Medical doctor‡ 0.002 0.018 1.61 1.90 1.12 (0.93 to 1.36) 0.24 CHO 1.05 1.10 1.29 1.46 1.03 (0.93 to 1.14) 0.54 Registered midwives 1.04 1.02 2.63 2.71 0.97 (0.88 to 1.06) 0.52 Registered nurses 0 0 0.55 0.60 0.98 (0.71 to 1.37) 0.92 Nurse anaesthetist 2.10 2.40 1.01 (0.86 to 1.22) 0.87 State enrolled community nurses 1.56 1.85 9.96 11.33 1.06 (1.00 to 1.12) 0.06 Maternal and child health aides 1.82 2.04 1.52 1.51 1.02 (0.94 to 1.10) 0.65 Nurse aide 0.35 0.42 2.57 2.69 1.03 (0.91 to 1.16) 0.62 Total of cadre above 5.82 6.46 24.0 26.2 1.03 (1.00 to 1.07) 0.09 Traditional birth attendant 3.85 4.35 1.19 0.98 1.03 (0.97 to 1.09) 0.33 Student§ 1.86 0.34 5.28 3.39 0.33 (0.29 to 0.37) <0.001 Total 11.49 11.11 29.2 29.9 0.92 (0.90 to 0.95) <0.001 *Ebola cases were confirmed in the district in the previous month.

.62 Total of cadre above 5.82 6.46 24.0 26.2 1.03 (1.00 to 1.07) 0.09 Traditional birth attendant 3.85 4.35 1.19 0.98 1.03 (0.97 to 1.09) 0.33 Student§ 1.86 0.34 5.28 3.39 0.33 (0.29 to 0.37) <0.001 Total 11.49 11.11 29.2 29.9 0.92 (0.90 to 0.95) <0.001 *Ebola cases were confirmed in the district in the previous month. ‡Data missing at one CEmOC facility from March 2014, treated as 0 in derivation of totals, but missing in analysis of this cadre. §Data missing throughout at five facilities, and on four other occasions in April and May 2013, treated as 0 in derivation of totals, but missing in analysis of this cadre. BEmOC, basic emergency obstetric care; CEmOC, comprehensive emergency obstetric care; CHO, Community Health Worker; EVD, Ebola virus disease. Availability of signal functions of EmOC Overall, for all districts combined, there is no evidence of an association between the onset of EVD epidemic in that district and the ability (or not) to provide the components (signal functions) of the EmOC care package (table 3). This includes intravenous or intramuscular antibiotics, oxytocics, anticonvulsants, manual removal of a retained placenta, removal of retained products of conception (signal functions 1–5), blood transfusion and caesarean section (signal functions 8 and 9). Table 3 Association between EVD epidemic and availability of signal functions at healthcare facilities designated to provide BEmOC or CEmOC

Availability of signal functions of EmOC Overall, for all districts combined, there is no evidence of an association between the onset of EVD epidemic in that district and the ability (or not) to provide the components (signal functions) of the EmOC care package (table 3). This includes intravenous or intramuscular antibiotics, oxytocics, anticonvulsants, manual removal of a retained placenta, removal of retained products of conception (signal functions 1–5), blood transfusion and caesarean section (signal functions 8 and 9). Table 3 Association between EVD epidemic and availability of signal functions at healthcare facilities designated to provide BEmOC or CEmOC Availability of EmOC signal function (% of month–facility occasions when signal function available) Effect of EVD epidemic Facility providing BEmOC Facility providing CEmOC All facilities combined EmOC signal function No Ebola (n=1035) Ebola (n=395) No Ebola (n=207) Ebola (n=79) No Ebola (n=1242) Ebola (n=474) OR (95% CI) p Value 1. im/iv antibiotics 95.7 96.5 95.7 93.7 95.7 96.0 1.01 (0.48 to 2.12) 0.99 2. im/iv oxytocics 94.1 91.7 95.2 92.4 94.3 91.8 0.67 (0.39 to 1.16) 0.59 3. im/iv anticonvulsants 97.6 98.0 95.7 96.2 97.3 97.7 1.99 (0.71 to 5.57) 0.18 4. Manual removal of placenta 100.0 100.0 100.0 100.0 100.0 100.0 NV 5. Removal of retained products of conception 79.7 84.3 100.0 100.0 83.1 86.9 1.65 (0.71 to 3.83) 0.23 6. Assisted vaginal delivery 84.4 81.3 100.0 100.0 87.0 84.4 0.45 (0.20 to 1.03) 0.056 7. Neonatal resuscitation 79.8 88.4 77.8 81.0 79.5 87.1 NR 8. Blood transfusion NA NA 91.3 86.1 NA NA NE 9. Caesarean section NA NA 91.6 91.6 NA NA NV BEmOC, basic emergency obstetric care; CEmOC, comprehensive emergency obstetric care; EmOC, emergency obstetric care; EVD, Ebola virus disease; im, intramuscular; iv, intravenous; n, number of facility–month combinations; NA, not applicable, signal function not expected at BEmOC facilities; NE, not estimable: in one district the initiation of availability of this signal function coincided with Ebola, in all others there is no variation over months; NV, no variation over time, within each facility, regarding availability of signal function.

not applicable, signal function not expected at BEmOC facilities; NE, not estimable: in one district the initiation of availability of this signal function coincided with Ebola, in all others there is no variation over months; NV, no variation over time, within each facility, regarding availability of signal function. When the analysis treated month–district combinations with no cases of EVD reported as EVD-free months, there was no difference in any of the signal functions except for ability to perform neonatal resuscitation (signal function 7) which increased at both BEmOC and CEmOC level. At both levels, this was attributed to increased availability of equipment required, unrelated to the EVD epidemic, that is, bag and masks for resuscitation. Assisted vaginal delivery (signal function 6) was always available at facilities designated to provide CEmOC during the EVD epidemic. When only BEmOC-level facilities were considered, the estimated OR was 0.45 (95% CI 0.20 to 1.03).

When the analysis treated month–district combinations with no cases of EVD reported as EVD-free months, there was no difference in any of the signal functions except for ability to perform neonatal resuscitation (signal function 7) which increased at both BEmOC and CEmOC level. At both levels, this was attributed to increased availability of equipment required, unrelated to the EVD epidemic, that is, bag and masks for resuscitation. Assisted vaginal delivery (signal function 6) was always available at facilities designated to provide CEmOC during the EVD epidemic. When only BEmOC-level facilities were considered, the estimated OR was 0.45 (95% CI 0.20 to 1.03). Factors affecting availability of signal functions of EmOC Overall, the numbers and cadres of healthcare provider were noted to be in post for each of the signal functions of EmOC across all districts before and after the EVD epidemic, and this was not reported to be a limiting factor for EmOC availability. However, equipment and/or drugs were not always available. Where intravenous/intramuscular antibiotics were not available (72/1716 occasions among 10 facilities), this was equally likely to be due to lack of drugs (42/72 occasions) or lack of equipment (syringes or needles; 41/72 occasions). Non-availability of intravenous/intramuscular oxytocics (110/1716 occasions among 17 facilities) was more commonly due to lack of oxytocics (any type; 87/110 occasions) than equipment (syringes or needles; 40/110 occasions). Non-availability of anticonvulsants (45/1716 occasions among 10 facilities) was because of lack of an anticonvulsant (magnesium sulfate or diazepam; 4/45 occasions) and equipment (syringes or needles; 41/45 occasions).

ack of oxytocics (any type; 87/110 occasions) than equipment (syringes or needles; 40/110 occasions). Non-availability of anticonvulsants (45/1716 occasions among 10 facilities) was because of lack of an anticonvulsant (magnesium sulfate or diazepam; 4/45 occasions) and equipment (syringes or needles; 41/45 occasions). Blood transfusion was available in all but one CEmOC (lack of equipment; cross matching reagents, blood storage refrigerator). Caesarean section was available at 12 out of 13 CEmOC facilities. There were 24 month–facility combinations when it was not available—all 22 months in one facility due to lack of an operating theatre and the first 2 months due to lack of qualified staff. Uptake of services Data for ANC and PNC visits were available for eight districts of which six recorded a statistically significant decrease in the number of ANC and PNC visits during the Ebola epidemic (table 4 and figure 2). Table 4 Association between EVD, number of women attending for ANC, PNC and delivery at a healthcare facility; overall and by district disaggregated by level of healthcare (CEmOC or BEmOC)

Uptake of services Data for ANC and PNC visits were available for eight districts of which six recorded a statistically significant decrease in the number of ANC and PNC visits during the Ebola epidemic (table 4 and figure 2). Table 4 Association between EVD, number of women attending for ANC, PNC and delivery at a healthcare facility; overall and by district disaggregated by level of healthcare (CEmOC or BEmOC) ANC visits PNC visits Facility births CEmOC vs BEmOC Ebola CEmOC vs BEmOC Ebola CEmOC vs BEmOC Ebola District Ratio (95% CI) Ratio (95% CI) p Value Ratio (95% CI) Ratio (95% CI) p Value Ratio (95% CI) Ratio (95% CI) p Value Bo (n=130) 3.72 (0.90 to 15.4) 0.76 (0.71 to 0.81) <0.001 1.52 (0.63 to 3.67) 0.78 (0.72 to 0.85) <0.001 3.09 (1.25 to 7.62) 1.15 (1.07 to 1.24) <0.001 Bombali (n=88) NA 1.04 (0.89 to 1.20) 0.62 NA 1.04 (0.92 to 1.19) 0.53 9.00 (4.70 to 17.1) 0.78 (0.72 to 0.83) <0.001 Bonthe (n=0) ND ND ND ND 1.66 (1.04 to 2.67) 0.76 (0.64 to 0.91) 0.003 Kailahun (n=110) 1.89 (1.64 to 2.18) 0.83 (0.78 to 0.90) <0.001 1.49 (0.97 to 2.31) 0.87 (0.80 to 0.94) <0.001 1.50 (0.99 to 2.27) 0.75 (0.70 to 0.81) <0.001 Kambia (n=126) 1.09 (0.82 to 1.44) 0.69 (0.62 to 0.77) <0.001 1.01 (0.51 to 1.97) 0.78 (0.67 to 0.90) <0.001 3.11 (2.45 to 3.95) 0.60 (0.53 to 0.67) <0.001 Kenema (n=44) 5.94 (5.46 to 6.46) 0.82 (0.76 to 0.88) <0.001 4.80 (4.41 to 5.22) 0.71 (0.65 to 0.78) <0.001 6.47 (2.63 to 15.90) 0.92 (0.86 to 0.98) 0.013 Koinadugu (n=0) ND ND ND ND 5.34 (3.40 to 8.37) 0.81 (0.74 to 0.90) 0.93 (0.82 to 1.06) <0.001 0.27 Kono (n=131) 9.8 (2.6 to 367) 0.74 (0.69 to 0.80) <0.001 3.01 (1.06 to 8.58) 0.53 (0.48 to 0.59) <0.001 3.47 (1.77 to 6.81) 0.87 (0.79 to 0.96) 0.006 Moyamba (n=0) ND ND ND ND 0.98 (0.66 to 1.44) 0.83 (0.76 to 0.90) <0.001 Port Loko (n=101) NA 0.87 (0.80 to 0.93) <0.001 NA 0.65 (0.58 to 0.72) <0.001 1.19 (0.72 to 1.94) 0.58 (0.52 to 0.64) <0.001 Pujehun (n=0) ND ND ND ND 1.03 (0.60 to 1.76) 0.88 (0.80 to 0.98) 0.014 Tonkolili (n=100) NA 0.92 (0.84 to 1.01) 0.09 NA 1.02 (0.90 to 1.15) 0.78 2.59 (1.31 to 5.12) 0.98 (0.89 to 1.08) 0.73 Western area (n=0) ND ND ND ND 11.03 (2.95 to 41.2) 0.99 (0.94 to 1.03) 0.51 All 3.10 (1.88 to 5.11) 0.82 (0.79 to 0.84) <0.001 1.77 (1.16 to 2.68) 0.78 (0.75 to 0.80) <0.001 2.72 (2.10 to 3.52) 0.89(0.87 to 0.91) <0.001 ANC, antenatal care; BEmOC, basic emergency obstetric care; CEmOC, comprehensive emergency obstetric care; EVD, Ebola virus disease; n, number of facility–month combinations; NA, not applicable (no data for the CEmOC in this district); ND, no data available; PNC, postnatal care.

(2.10 to 3.52) 0.89(0.87 to 0.91) <0.001 ANC, antenatal care; BEmOC, basic emergency obstetric care; CEmOC, comprehensive emergency obstetric care; EVD, Ebola virus disease; n, number of facility–month combinations; NA, not applicable (no data for the CEmOC in this district); ND, no data available; PNC, postnatal care. Figure 2 Number of Confirmed Ebola Virus (EBV) cases by district from time of onset of epidemic across Sierra Leone. Overall, for all districts combined, there was a statistically significant reduction in the numbers of ANC and PNC visits after the onset of the EVD epidemic. The estimated reduction for ANC visits was 18% (IRR 0.82, 95% CI 0.79 to 0.84; p<0.001) and for PNC visits was 22% (IRR 0.78, 95% CI 0.75 to 0.80; p<0.001). For both ANC and PNC visits, there were statistically significant differences (G2=29.6, df=1, p<0.001 and G2=6.51, df=1, p=0.01, respectively) between facility type in the IRRs for the impact of the EVD epidemic with greater reductions at CEmOC level than at BEmOC level. For ANC visits at BEmOCs, there was a 14% decrease (IRR 0.86, 95% CI 0.83 to 0.89); whereas for CEmOCs, there was a 25% decrease (IRR 0.75, 95% CI 0.72 to 0.78). For PNC visits, there were 20% and 27% decreases at BEmOCs and CEmOCs, respectively (IRR 0.80, 95% CI 0.77 to 0.83 and IRR 0.73, 95% CI 0.69 to 0.77). ANC and PNC data were not available for the CEmOC-level facility in two of the districts; ANC and PNC data were not available in four districts.

For both ANC and PNC visits, there were statistically significant differences (G2=29.6, df=1, p<0.001 and G2=6.51, df=1, p=0.01, respectively) between facility type in the IRRs for the impact of the EVD epidemic with greater reductions at CEmOC level than at BEmOC level. For ANC visits at BEmOCs, there was a 14% decrease (IRR 0.86, 95% CI 0.83 to 0.89); whereas for CEmOCs, there was a 25% decrease (IRR 0.75, 95% CI 0.72 to 0.78). For PNC visits, there were 20% and 27% decreases at BEmOCs and CEmOCs, respectively (IRR 0.80, 95% CI 0.77 to 0.83 and IRR 0.73, 95% CI 0.69 to 0.77). ANC and PNC data were not available for the CEmOC-level facility in two of the districts; ANC and PNC data were not available in four districts. The number of deliveries that occurred at health facility level (data available for all 13 districts) also showed a statistically significant decrease of 11% (IRR 0.89, 95% CI 0.87 to 0.91). There was a statistically significant difference (G2=11.4, df=1, p=0.0007) between facility type in the IRRs for the impact of the EVD epidemic. For BEmOCs, there was a 14% decrease (IRR 0.86, 95% CI 0.84 to 0.89); whereas for CEmOCs, there was an 8% decrease (IRR 0.92, 95% CI 0.89 to 0.95). Bo was the only district to report an increase in number of deliveries following the start of the EVD epidemic, with a statistically significant increase in numbers occurring mainly at CEmOC level (IRR 1.15, 95% CI 1.07 to 1.24; p<0.001).

9); whereas for CEmOCs, there was an 8% decrease (IRR 0.92, 95% CI 0.89 to 0.95). Bo was the only district to report an increase in number of deliveries following the start of the EVD epidemic, with a statistically significant increase in numbers occurring mainly at CEmOC level (IRR 1.15, 95% CI 1.07 to 1.24; p<0.001). The decrease in number of deliveries at CEmOC level was associated with an overall increase in the caesarean section rate (number of caesarean sections per number of facility births) of 14% (IRR 1.14, 95% CI 1.06 to 1.22) confirming this procedure continued to be available. Maternal mortality ratio and stillbirth rate A total of 464 maternal deaths and 55 095 live births were recorded at healthcare facility level between 1 May 2013 and 31 January 2015; 152 maternal deaths were recorded during the EVD epidemic months and 312 in the 12 months when no EVD was reported in the previous month. For all districts combined, the facility-based maternal mortality ratio increased by 34% after onset of the EVD epidemic (IRR 1.34, 95% CI 1.07 to 1.69; table 5). When type of facility was considered separately (CEmOC or BEmOC), the increase in maternal deaths was significant at CEmOCs level (p<0.001) but not at BEmOC level (p=0.35). However, the interaction between facility type and onset of the EVD epidemic was not statistically significant (G2=2.13, df=2, p=0.35). Table 5 Association between EVD epidemic and facility MMR overall and by district disaggregated by level of healthcare (CEmOC or BEmOC)

For all districts combined, the facility-based maternal mortality ratio increased by 34% after onset of the EVD epidemic (IRR 1.34, 95% CI 1.07 to 1.69; table 5). When type of facility was considered separately (CEmOC or BEmOC), the increase in maternal deaths was significant at CEmOCs level (p<0.001) but not at BEmOC level (p=0.35). However, the interaction between facility type and onset of the EVD epidemic was not statistically significant (G2=2.13, df=2, p=0.35). Table 5 Association between EVD epidemic and facility MMR overall and by district disaggregated by level of healthcare (CEmOC or BEmOC) MMR (maternal deaths/100 000 live births) Facility providing CEmOC Facility providing BEmOC Comparison of CEmOC vs BEmOC Comparison of Ebola vs no Ebola Type of facility or district Number of months with Ebola cases Number of maternal deaths Number of live births No Ebola (n=201) Ebola (n=78) No Ebola (n=1034) Ebola (n=390) IRR (95% CI) p Value IRR (95% CI) p Value All facilities 1962 3097 59 55 42 (26 to 69) <0.001 1.34 (1.07 to 1.69) 0.01 BEmOCs 0.64 (0.18 to 2.24) 0.53 CEmOCs 1.48 (1.21 to 1.81) <0.001 Bo (n=132) 7 54 4389 2638 5519 53 0 93 (13 to 673) <0.001 2.2 (1.3 to 3.8) 0.004 Bombali (n=22) 7 51 4812 1634 1850 0 0 ND 1.2 (0.6 to 2.2) 0.58 Bonthe (n=0) 4 1 1429 0 0 0 455 Only 1 event in this district Kailahun (n=131) 8 13 4233 1687 1266 0 122 43 (6 to 331) <0.001 1.1 (0.3 to 3.6) 0.86 Kambia (n=132) 4 38 3384 2800 11 935 58 0 68 (9 to 495) <0.0001 2.1 (0.8 to 5.3) 0.17 Kenema (n=22) 7 48 5431 1839 1324 0 0 ND 0.7 (0.4 to 1.5) 0.37 Koinadugu (n=132) 4 35 3258 2120 2323 141 0 17 (4 to 70) <0.001 1.1 (0.4 to 2.6) 0.87 Kono (n=131) 7 8 2620 708 332 142 0 5.5 (1.1 to 27) 0.05 0.3 (0.04 to 2.6) 0.23 Moyamba (n=132) 7 16 3227 2987 1242 191 220 24 (7 to 85) <0.001 0.7 (0.2 to 2.5) 0.57 Port Loko (n=22) 7 13 2774 3709 2857 0 0 ND 1.1 (0.2 to 4.9) 0.91 Pujehun (n=21) 5 14 3462 2480 4.697 0 0 ND 1.9 (0.6 to 5.5) 0.29 Tonkolili (n=126) 5 29 3221 1122 8796 102 0 38 (8.9 to 160) <0.001 4.5 (2.0 to 10.0) <0.001 Western area (n=130) 7 143 12 855 1550 2500 78 0 22 (8 to 60) <0.001 1.5 (1.0 to 2.0) 0.03 BEmOC, basic emergency obstetric care; CEmOC, comprehensive emergency obstetric care; EVD, Ebola virus disease; IRR, incidence rate ratio; MMR, maternal mortality ratio; n, number of facility–month combinations; ND, no deaths reported for one level.

7 143 12 855 1550 2500 78 0 22 (8 to 60) <0.001 1.5 (1.0 to 2.0) 0.03 BEmOC, basic emergency obstetric care; CEmOC, comprehensive emergency obstetric care; EVD, Ebola virus disease; IRR, incidence rate ratio; MMR, maternal mortality ratio; n, number of facility–month combinations; ND, no deaths reported for one level. The total number of reported stillbirths was 3589 giving an overall facility-based stillbirth rate of 60.5 per 1000 births. Overall, there was a 24% increase in the incidence of stillbirth (IRR 1.24, 95% CI 1.14 to 1.35; table 6). However, there was an interaction between type of facility and onset of the EVD epidemic (G2=15.6, df=2, p<0.001). When type of facility was considered separately, the increase was significant at CEmOC level (IRR 1.27, 95% CI 1.16 to 1.39; p<0.001) but not at BEmOC level (IRR 1.07, 95% CI 0.85 to 1.39; p=0.57). Table 6 Association between onset of EVD epidemic and facility SBR overall and by district disaggregated by level of healthcare (CEmOC or BEmOC)

The total number of reported stillbirths was 3589 giving an overall facility-based stillbirth rate of 60.5 per 1000 births. Overall, there was a 24% increase in the incidence of stillbirth (IRR 1.24, 95% CI 1.14 to 1.35; table 6). However, there was an interaction between type of facility and onset of the EVD epidemic (G2=15.6, df=2, p<0.001). When type of facility was considered separately, the increase was significant at CEmOC level (IRR 1.27, 95% CI 1.16 to 1.39; p<0.001) but not at BEmOC level (IRR 1.07, 95% CI 0.85 to 1.39; p=0.57). Table 6 Association between onset of EVD epidemic and facility SBR overall and by district disaggregated by level of healthcare (CEmOC or BEmOC) Mean incidence of stillbirths (per 1000 births) Facility providing CEmOC Facility providing BEmOC Comparison of CEmOC vs BEmOC Comparison of Ebola vs no Ebola Type of facility or district Number of months with Ebola Total number of stillbirths Total number of births in district No Ebola (n=201) Ebola (n=78) No Ebola (n=1034) Ebola (n=390) IRR (95% CI) p Value IRR (95% CI) p Value All facilities 138 194 18 17 13.9 (8.7 to 22.9) <0.001 1.24 (1.14 to 1.35) <0.001 Facility providing BEmOC 1.07 (0.85 to 1.35) 0.57 Facility providing CEmOC 1.27 (1.16 to 1.39) <0.001 Bo (n=132) 7 161 4550 68 141 15 13 8.13 (1.8 to 37) 0.001 1.67 (1.22 to 2.28) 0.001 Bombali (n=131) 7 341 5153 105 134 13 8 12.1 (2.9 to 50) 0.001 1.21 (0.97 to 1.54) 0.10 Bonthe (n=132) 4 28 1459 41 17 15 24 3.6 (1.7 to 7.6) 0.001 1.05 (0.4 to 3.0) 0.93 Kailahun (n=131) 8 179 4409 124 94 21 26 5.5 (2.0 to 15.6) 0.001 1.0 (0.7 to 1.4) 0.97 Kambia (n=132) 4 173 3558 120 108 19 16 6.5 (4.4 to 9.5) <0.001 0.9 (0.5 to 1.6) 0.72 Kenema (n=131) 7 369 5802 112 148 15 20 9.8 (3.4 to 28.7) <0.001 1.3 (1.1 to 1.6) 0.01 Koinadugu (n=132) 4 190 3448 88 94 39 34 3.0 (2.1 to 4.2) <0.001 1.0 (0.7 to 1.5) 0.88 Kono (n=131) 7 206 2826 197 242 11 8 16.3 (9.9 to 26.7) <0.001 1.2 (0.9 to 1.5) 0.33 Moyamba (n=132) 7 91 3318 128 245 11 8 30.3 (2.3 to 394) 0.009 1.4 (0.9 to 2.2) 0.13 Port Loko (n=132) 7 177 2951 382 453 13 21 25.9 (6 to 105) <0.001 1.2 (0.8 to 1.7) 0.38 Pujehun (n=131) 5 80 3542 168 162 1 0 398 (55 to 2858) <0.001 1.1 (0.7 to 1.8) 0.73 Tonkolili (n=126) 5 211 3426 137 410 30 27 11.9 (0.9 to 152) 0.06 1.9 (1.4 to 2.4) <0.001 Western area (n=130) 7 1386 14 240 155 183 27 28 7.9 (1.1 to 59) 0.04 1.1 (1.0 to 1.2) 0.06 BEmOC, basic emergency obstetric care; CEmOC, comprehensive emergency obstetric care; EVD, Ebola virus disease; IRR, incidence rate ratio; n, number of facility–month combinations; SBR, stillbirth rate.

6 1.9 (1.4 to 2.4) <0.001 Western area (n=130) 7 1386 14 240 155 183 27 28 7.9 (1.1 to 59) 0.04 1.1 (1.0 to 1.2) 0.06 BEmOC, basic emergency obstetric care; CEmOC, comprehensive emergency obstetric care; EVD, Ebola virus disease; IRR, incidence rate ratio; n, number of facility–month combinations; SBR, stillbirth rate. Discussion Main findings Across Sierra Leone, following the onset of the EVD epidemic, there was a decrease of 18% in the number of women attending for ANC, 22% decrease in attendance for PNC and an 11% decrease in the number of women attending for birth at a healthcare facility able to provide emergency obstetric and newborn care. For women who did access care, there was a corresponding statistically significant 34% increase in the facility maternal mortality ratio and 24% increase in the stillbirth rate. Strengths and weaknesses To the best of our knowledge, this is the only study to have collected data on maternity services uptake and outcomes across all the 13 districts of Sierra Leone and including time both before and during the EVD epidemic. Routinely collected data were obtained from registers at each healthcare facility retrospectively. The EVD epidemic was unexpected with regard to severity and length, and therefore it was not possible to design a prospective study. This was an example of operations research and use of routine data. We were unable to a priori strengthen the data or data collection and recording systems before the study took place.

y. The EVD epidemic was unexpected with regard to severity and length, and therefore it was not possible to design a prospective study. This was an example of operations research and use of routine data. We were unable to a priori strengthen the data or data collection and recording systems before the study took place. Staffing levels required for each type of facility are provided by the Government of Sierra Leone in the Basic Package of Essential Health Services;12 however, there are ongoing shortages of staff. This study did not report if healthcare facilities met the prescribed cadre and numbers of healthcare providers but rather on change, or not, in the number in post during the EVD epidemic. The study did not assess the quality of care provided; for example, there was no assessment of the timeliness with which interventions were provided and whether or not these were provided too late in some cases to be live-saving. Similarly, we did not assess the quality of resuscitation of the newborn to determine if preventable stillbirths occurred as a result of poor or non-resuscitation efforts at the time of birth.

eliness with which interventions were provided and whether or not these were provided too late in some cases to be live-saving. Similarly, we did not assess the quality of resuscitation of the newborn to determine if preventable stillbirths occurred as a result of poor or non-resuscitation efforts at the time of birth. The analysis is defined by reference to the onset (or not) of EVD in any particular district. We did not have access to village-level data which may have provided a more detailed pattern and analysis regarding availability and uptake of care. All healthcare facilities designated to provide EmOC were included in the study but it is conceivable that ANC, PNC or skilled attendance at birth can also be provided at lower level healthcare facilities which were not included in this study. Interpretation in light of other studies Anecdotal reports at the start of the epidemic mentioned healthcare providers ‘leaving their posts’ and ‘refusing to provide patient care’. This study shows that staff did remain in post. The exception was student preservice attachments, reflecting the countrywide closure of schools during the EVD epidemic.13

r studies Anecdotal reports at the start of the epidemic mentioned healthcare providers ‘leaving their posts’ and ‘refusing to provide patient care’. This study shows that staff did remain in post. The exception was student preservice attachments, reflecting the countrywide closure of schools during the EVD epidemic.13 SBA, ANC and PNC continued to be available, and the EVD epidemic did not lead to a decrease in availability of EmOC. However, assisted vaginal delivery (ventouse) was not available across all healthcare facilities designated to provide BEmOC. There were also differences with regard to uptake of care; at hospital or CEmOC levels, the decrease in number of women accessing for ANC or PNC suggested that perhaps women only attended larger healthcare facilities in more urban and populated areas if they had to do so for an emergency and might have attended as late as possible. It is likely that fear of contracting Ebola during visits to healthcare facilities and (unsubstantiated) rumours such as that healthcare providers were injecting patients with the virus, led women to stay away and/or access care late. Perceptions about the quality of patient care by the public during the epidemic may also have reduced numbers attending at facilities. Similar findings have been reported in a study from Kenema district, which showed a reduction in the numbers of pregnant and lactating women accessing services and in Guinea, where there was a 31% reduction in outpatient visits; in both studies, this was attributed to fear of contracting Ebola.14 15

ding at facilities. Similar findings have been reported in a study from Kenema district, which showed a reduction in the numbers of pregnant and lactating women accessing services and in Guinea, where there was a 31% reduction in outpatient visits; in both studies, this was attributed to fear of contracting Ebola.14 15 Women who accessed care at a health facility were significantly more likely to die and more likely to have a stillbirth during the EVD epidemic. However, our results show that healthcare facilities were in principle ‘ready’ and EmOC was available. Healthcare facilities were able to continue to provide all the components (signal functions) of EmOC with only the availability of assisted vaginal delivery showing a statistically significant reduction. This may be due to government guidance to healthcare providers to avoid practising ‘invasive’ procedures which would increase their exposure to the Ebola virus.16 Local by-laws restricting movement between villages may also have limited women's access to healthcare facilities. However, this would have to be considered alongside the Government's requirement for all women to deliver in healthcare facilities rather than in the community.16

rease their exposure to the Ebola virus.16 Local by-laws restricting movement between villages may also have limited women's access to healthcare facilities. However, this would have to be considered alongside the Government's requirement for all women to deliver in healthcare facilities rather than in the community.16 The increase in mortality for mothers and babies could be explained by two factors: (1) women accessed care late, and/or (2) the quality of the care provided was poor. Timely diagnosis and early intervention is needed to save lives in case of an obstetric emergency. Assisted vaginal delivery, manual removal of retained placenta in case of obstetric haemorrhage, caesarean section, manual vacuum aspiration or curettage in case of haemorrhage associated with miscarriage, and even neonatal resuscitation can all be considered life-saving but are ‘invasive’ and require personal protection for the healthcare provider having to deal with such emergencies. There are anecdotal reports and observations of women being ‘left to deliver on their own’, and it is plausible that healthcare providers were reluctant to intervene early and quickly when this was needed thus not providing safe, timely and effective treatment with a resulting poorer outcome.

eal with such emergencies. There are anecdotal reports and observations of women being ‘left to deliver on their own’, and it is plausible that healthcare providers were reluctant to intervene early and quickly when this was needed thus not providing safe, timely and effective treatment with a resulting poorer outcome. Conclusions There is no doubt that the Ebola virus epidemic had a devastating effect on Sierra Leone and lessons are being learnt to improve healthcare delivery in the future. This study shows that maternity care was in principle available and continued to be provided. However, the care may in some cases have been provided late by healthcare providers who were afraid of being infected or because women accessed services late. EmOC was available and is potentially life-saving provided complications are recognised and managed early and quickly. During any epidemic—EVD or other—the public need to be confident that healthcare providers can continue to provide both routine and emergency maternity care while at the same time dealing with the effects of the epidemic. Similarly, healthcare providers need to be supported to be able to provide the highest quality of care, being able to distinguish between women infected with EVD who need isolation and/or referral to special treatment centres and women who need EmOC in a timely and safe manner. Healthcare providers need to be able to safely practise with no risk to their or others' personal health. For countries where the EVD epidemic occurred, this requires urgent attention—both for the immediate restoration of routine health services and to be able to ensure continued access to and uptake of high-quality care during future epidemics.

o be able to safely practise with no risk to their or others' personal health. For countries where the EVD epidemic occurred, this requires urgent attention—both for the immediate restoration of routine health services and to be able to ensure continued access to and uptake of high-quality care during future epidemics. The authors would like to thank Betty Sam, Florence Bull, Steven Bagie Pieh, Mohammed Fornah for collecting the data and the Ministry of Health and Sanitation of Sierra Leone for granting us permission to undertake the study and allowing us access to data. Handling editor: Stephanie Topp Contributors: SAJ, NRvdB and CAA designed the study; SAJ and SG oversaw the data collection; SG, NRvdB, SAJ and SW analysed and interpreted the data; all authors wrote the paper. Funding: This research was funded by WaterAid through a grant from Voluntary Service Overseas (VSO). Competing interests: None declared. Provenance and peer review: Not commissioned; externally peer reviewed. Data sharing statement: No additional data are available.

Summary Box The WHO internship programme is one of the most high-profile junior professional training programme in global health, and previous findings have suggested the programme is inaccessible to young professionals from developing countries. However, the extent of this is unknown. In May 2016, WHO published, for the first time, full statistics concerning Member State participation on the internship programme – they show that only 15% of interns at WHO headquarters were from developing countries; Africa and South-East Asia regional offices offer less than 4% of all WHO internships; and almost 60% of WHO Member States had no nationals participating in the entire programme throughout 2015. The internship programme suffers from inequitable Member State participation, and therefore fails to build future global health capacity in young professionals from developing countries. Reform of the internship programme should focus on overhauling the entire recruitment procedures; providing financial support to interns, particularly from low-income countries; and introducing a semi-structured curriculum to maximise the benefits of the internship.

The internship programme suffers from inequitable Member State participation, and therefore fails to build future global health capacity in young professionals from developing countries. Reform of the internship programme should focus on overhauling the entire recruitment procedures; providing financial support to interns, particularly from low-income countries; and introducing a semi-structured curriculum to maximise the benefits of the internship. The global health workforce is under immense strain.1 For example, Africa—a continent with one-third of the world's disease burden has only ∼3% of global health personnel.2 This year WHO launched its Global Strategy on Human Resources for Health3 (GSHRH)—calling for a redoubling of efforts to better train and equip the global health workforce in order to strengthen public health capacity; echoing the 2006 World Health Assembly (WHA) resolution ‘Rapid scaling up of health workforce production’ passed in 2006.4

d its Global Strategy on Human Resources for Health3 (GSHRH)—calling for a redoubling of efforts to better train and equip the global health workforce in order to strengthen public health capacity; echoing the 2006 World Health Assembly (WHA) resolution ‘Rapid scaling up of health workforce production’ passed in 2006.4 In pursuit of these aims, WHO runs a range of external programmes to train public health professionals, including the WHO Fellowship Programme; and through partnership with over 700 collaborating centres—often at universities—in more than 80 countries. In addition to these external programmes, WHO also runs what should be an internal training programme for future public health leaders: the WHO internship programme. Now in its 50th year, the programme has recently come under increased scrutiny—catalysed by research published in 2014 that suggested it was inaccessible to young professionals from developing countries.5 With both civil society and Member States recognising the potentially corrosive impact of a high-profile internal training programme that contradicts core global health policy;6 the debate surrounding the programme has intersected the wider issue of developing country engagement in international health training opportunities and governance. This commentary presents an overview of the challenges and progress towards making WHO's internship programme equitable.

adicts core global health policy;6 the debate surrounding the programme has intersected the wider issue of developing country engagement in international health training opportunities and governance. This commentary presents an overview of the challenges and progress towards making WHO's internship programme equitable. The internship programme WHO is pre-eminent in global health; central to its mandate is a responsibility to support all 194 of its Member States in addressing their health challenges. By offering ‘concrete professional experience in an international environment’7 to young public health professionals through internships, WHO realises its commitment to states’ future, as well as current, health capabilities. These internships last 3–6 months; they are offered at WHO's headquarters, regional and certain country offices and provide unique exposure and training in global public health. Moreover, through networking opportunities and such else, these internships serve as a key point of career entry for many junior public health professionals.

ps last 3–6 months; they are offered at WHO's headquarters, regional and certain country offices and provide unique exposure and training in global public health. Moreover, through networking opportunities and such else, these internships serve as a key point of career entry for many junior public health professionals. First signs of trouble However, the findings published in 2014 cast a shadow over these ambitions—only 5% of headquarters interns were living in a developing country;5 corroborating the 2009 United Nations Joint Inspection Unit report which found 60% of interns on the WHO internship programme were from just five developed countries.9 What good is offering early career stage public health training, if young professionals from developing countries—where disease burdens and skills shortages are highest—cannot access them? Instead, these opportunities are taken up by nationals from a small number of high-income countries, which have health or academic institutions with long-standing relationships with WHO staff or departments, for reasons explained shortly. For example, in summer 2013 50% of interns at the WHO headquarters’ programme hailed from just two high-income countries.5 Despite the evidence, WHO was slow to act. It did not publish statistics to corroborate or refute the findings, and no statement or comment was offered on how it hoped to address the issues raised. Indeed, in 2013 and 2014 the internship programme was not mentioned in its human resources annual report.

come countries.5 Despite the evidence, WHO was slow to act. It did not publish statistics to corroborate or refute the findings, and no statement or comment was offered on how it hoped to address the issues raised. Indeed, in 2013 and 2014 the internship programme was not mentioned in its human resources annual report. WHO's governing bodies In January 2015, a written statement was submitted by a group of civil society and non-governmental organisation members to the human resources session of the WHO Executive Board.10 This raised concerns about the imbalance in Member State representation on the internship programme, drawing attention to the now mounting evidence. However, it was not until the involvement of WHO's governing bodies in January 2016 that WHO published—for the first time—some of its own centrally collected internship programme statistics. They were stark: among its ∼1000 annual internships offered worldwide in 2014, only 20% were undertaken by candidates from developing countries; the African regional office offered <2% of all WHO internships.11 These results confirmed what many feared—Member State representation on the internship programme was not equitable, nor was it driven by a clear policy objective. At the January 2016 Executive Board, a number of Member States expressed views on the implications of these findings; however, key information concerning the national origin of interns was not released at the time,6 frustrating efforts to achieve consensus on the action to be taken.

driven by a clear policy objective. At the January 2016 Executive Board, a number of Member States expressed views on the implications of these findings; however, key information concerning the national origin of interns was not released at the time,6 frustrating efforts to achieve consensus on the action to be taken. In April 2016, these were finally published:7 only 15% of interns at WHO headquarters were from developing countries; the two regional offices serving regions with the highest disease burdens—Africa and South-East Asia—offered in total <4% of all WHO internships (figure 1); and almost 60% of WHO Member States had no nationals participating in the entire programme throughout 2015 (table 1). These include large emerging economies such as Argentina and South Africa; as well as countries with beleaguered health systems recovering from recent or chronic health emergencies such as Liberia and Haiti. By the time of the 2016 WHA, WHO had provided Member States with a clear picture of the WHO internship programme's shortcomings. Accordingly, during the human resources Committee B discussion several Member States raised their concerns on the internship programme's findings—namely the under-representation of developing countries.8 Table 1 Table details countries with ‘0’ or less than 5 interns on WHO Internship Programme in 2015. Data compiled using WHO 2016 Human Resources Annual Report statistics and list of WHO Member States as of 2016.

In April 2016, these were finally published:7 only 15% of interns at WHO headquarters were from developing countries; the two regional offices serving regions with the highest disease burdens—Africa and South-East Asia—offered in total <4% of all WHO internships (figure 1); and almost 60% of WHO Member States had no nationals participating in the entire programme throughout 2015 (table 1). These include large emerging economies such as Argentina and South Africa; as well as countries with beleaguered health systems recovering from recent or chronic health emergencies such as Liberia and Haiti. By the time of the 2016 WHA, WHO had provided Member States with a clear picture of the WHO internship programme's shortcomings. Accordingly, during the human resources Committee B discussion several Member States raised their concerns on the internship programme's findings—namely the under-representation of developing countries.8 Table 1 Table details countries with ‘0’ or less than 5 interns on WHO Internship Programme in 2015. Data compiled using WHO 2016 Human Resources Annual Report statistics and list of WHO Member States as of 2016. Countries not represented on WHO internship programme in 2015 Afghanistan Bahrain Cameroon Albania Barbados Central African Republic Algeria Belize Chad Andorra Bhutan Comoros Angola Bolivia Cook Islands Antigua and Barbuda Botswana Costa Rica Argentina Brunei Darussalam Croatia Armenia Burundi Cuba Azerbaijan Cabo Verde Cyprus Bahamas Cambodia Czech Republic Georgia Israel Libya Grenada Jamaica Luxembourg Guatemala Kazakhstan Madagascar Guinea Kiribati Malawi Guinea-Bissau Kuwait Maldives Guyana Kyrgyzstan Malta Haiti Lao People's DR Marshall Islands Honduras Latvia Mauritania Iceland Lesotho Mauritius Iraq Liberia Micronesia (FS of) Papua New Guinea Seychelles Tajikistan Paraguay Solomon Islands TFYR of Macedonia Peru Somalia Timor-Leste Qatar South Africa Togo Republic of Moldova South Sudan Tonga Saint Kitts and Nevis Sri Lanka Trinidad and Tobago Saint Lucia St. Vincent Tuvalu Samoa Suriname United Arab Emirates San Marino Swaziland Uzbekistan Sao Tome and Principe Syrian Arab Republic Vanuatu Countries with <5 interns on WHO internship programme in 2015 Bangladesh Cote d'Ivoire Korea, Dem. People's Rep. Of Belarus Ecuador Lebanon Benin Estonia Malaysia Bosnia and Herzegovina Ghana Mali Bulgaria Hungary Moldova, Republic of Burkina Faso Indonesia Mongolia Chile Iran Morocco Colombia Ireland Myanmar Congo, DR Jordan Nepal Congo, Republic of the Kenya New Zealand Slovenia Yemen Figure 1 WHO-IP and regional office opportunities in 2014 and 2015. (A) Distribution (%) of WHO internships across headquarters and regional offices in 2014 and 2015. (B) WHO-IP regional office internship opportunities against regional disease burden disability-adjusted life years (DALYs); circle size denotes proportion (%) of all regional opportunities (same as X-axis). HQ, headquarters; AFRO, African regional office; AMRO*, regional office of the Americas; EURO, European regional office; EMRO, Eastern Mediterranean regional office; IP, internship programme; SEARO, South-east Asian regional office; WPRO, Western pacific regional office. * AMRO data was not included in the 2015 data set published by the WHO.

African regional office; AMRO*, regional office of the Americas; EURO, European regional office; EMRO, Eastern Mediterranean regional office; IP, internship programme; SEARO, South-east Asian regional office; WPRO, Western pacific regional office. * AMRO data was not included in the 2015 data set published by the WHO. The implications While the statistics concerning the internship programme are striking, the reasons for the inequitable representation of young professionals from developing countries are simple. Recruitment to the programme is ad hoc and not policy-driven; departments select candidates informally without reviewing all applications and therefore those with pre-existing connections are far more likely to be selected. Moreover, interns are not financially supported by WHO during their stay, even in expensive locations such as Geneva. This precludes candidates from Member States with significantly lower incomes participating in the programme. In so doing, the internship programme corrodes a vital relationship WHO seeks to cultivate with future health leaders in the developing world.

WHO during their stay, even in expensive locations such as Geneva. This precludes candidates from Member States with significantly lower incomes participating in the programme. In so doing, the internship programme corrodes a vital relationship WHO seeks to cultivate with future health leaders in the developing world. Reform of the internship programme? While staff appointments are subject to a formula that was last revised in 2003,12 the internship programme is subject to no such protocol. Regrettably, it is a lack of objective oversight that has let the programme's structural barriers go unaddressed. However, the attention and transparency now focused on the programme presents a unique opportunity for decision makers. In responding to the concerns of Member States at the 2016 WHA, both WHO's Head of Human Resources and the Assistant Director-General for General Management stated that with Member States’ support and funding, reform of the internship programme would be possible8—but what might this actually entail?

or decision makers. In responding to the concerns of Member States at the 2016 WHA, both WHO's Head of Human Resources and the Assistant Director-General for General Management stated that with Member States’ support and funding, reform of the internship programme would be possible8—but what might this actually entail? First, unguided, recruitment to the internship programme lacks transparency and favours candidates with pre-existing connections.5 While WHO reports that it is working to ‘promote internship opportunities’,7 the appointment of a designated staff officer with responsibility for programme administration will be necessary to ensure recruitment that is equitable. Second, internships are unpaid and there is no financial support offered to candidates from developing countries, precluding many from lower income states participating. WHO states that it is now referring prospective interns to ‘lists of scholarships’;7 however, these are notoriously difficult to locate, often limited in number and only available to nationals from certain countries. Provision of a nominal stipend to compensate interns—considered by Member States in January13 but absent from the WHO report in May7—would comprehensively eliminate this barrier at limited cost. The International Labour Organization has successfully implemented such a scheme. Finally, balanced distribution alone will not guarantee participants gain the experience and knowledge they need. Introduction of a semistructured curriculum would consolidate the programme's capacity building role and improve its credibility in the eyes of Member States and donors.

cessfully implemented such a scheme. Finally, balanced distribution alone will not guarantee participants gain the experience and knowledge they need. Introduction of a semistructured curriculum would consolidate the programme's capacity building role and improve its credibility in the eyes of Member States and donors. Is it worth the effort? In an increasingly complex and fragile global health environment, there are justifiably more immediate concerns for decision makers to prioritise. Yet, if global ambitions for building developing country capacity—as called for in the sustainable development goals—are to be credible, organisations such as WHO can ill afford to run a training programme that all but excludes young professionals from the developing world. With WHO's traditional leadership role under challenge, positive engagement with the next generation of public health professionals is a necessity. By the time WHO reaches its centenary, many of today's young public health professionals should be at the centre of national and regional health security. Reform of the existing internship programme represents a time and resource-efficient way for WHO to use its own organisational resources and international reputation, to directly build equitable future global health capacity among junior public health professionals from the developing world. Recent progress suggests this argument is beginning to be heard; has the time come for WHO and its governing bodies to definitively act?

wn organisational resources and international reputation, to directly build equitable future global health capacity among junior public health professionals from the developing world. Recent progress suggests this argument is beginning to be heard; has the time come for WHO and its governing bodies to definitively act? The authors thank Ms Tara Kedia for her helpful input; and Anais Le Moing, Isabel Degreef Moreno, Mohamed Elzayat and Volha Kryvets for the translation support. Handling editor: Seye Abimbola Contributors: AB-V contributed to the idea and writing. CF contributed to the idea and critical review. MJ was involved in writing and critical review. JJ contributed to the idea, writing and critical review. Competing interests: None declared. Provenance and peer review: Not commissioned; externally peer reviewed. Data sharing statement: No additional data are available.

Key questions What is already known about this topic? There is currently little information available about the experiences, behaviours and attitudes of Ebola survivors. The latest West-African Ebola outbreak led to an unprecedented number of Ebola survivors, who have been shown to suffer from physical and mental sequelae. Studies have now started to investigate the extent and characteristics of those Ebola complications and a few studies have also identified that Ebola survivors may face discrimination and stigmatisation from their communities and their loved ones. What are the new findings? Owing to the recent nature of the outbreak, there is still a lot unknown about the experiences, behaviours and attitudes of Ebola survivors. This study addresses those issues in Sierra Leone and aims to improve our understanding of survivors' experiences during and after Ebola infection. It extends our knowledge of issues that have not yet been discussed extensively such as community acceptance of survivors, attitudes towards sexual transmission or the role survivors can play in Ebola outbreak control activities. Recommendations for policy The results from this study will inform responses to future Ebola outbreaks by increasing our understanding of the patient experience and challenges faced by survivors. By understanding the experience of persons infected with Ebola virus, epidemic control measures can be adapted to reduce mistrust and increase cooperation with affected communities.

Recommendations for policy The results from this study will inform responses to future Ebola outbreaks by increasing our understanding of the patient experience and challenges faced by survivors. By understanding the experience of persons infected with Ebola virus, epidemic control measures can be adapted to reduce mistrust and increase cooperation with affected communities. Introduction The Ebola outbreak that began in 2013 in West Africa was the largest in recorded history. The region faced many challenges in addressing the emerging infectious disease. Contributors to the rapid spread of Ebola included poverty, poor healthcare systems and services, inadequate surveillance systems, community distrust of outbreak response teams and contact with infected people at home, in healthcare facilities and during traditional burials.1 2 Identifying and addressing the knowledge, attitudes and practices of the public played a key role in addressing some of these contributors.

inadequate surveillance systems, community distrust of outbreak response teams and contact with infected people at home, in healthcare facilities and during traditional burials.1 2 Identifying and addressing the knowledge, attitudes and practices of the public played a key role in addressing some of these contributors. In Sierra Leone, the outbreak was declared in May3 and peaked in November 2014,4 but the outbreak response varied from one district to the next, as each district faced the disease at different times and intensities.5 Though mortality associated with this outbreak was high, there have never been so many Ebola survivors before. In Sierra Leone alone, there were over 14 000 cases with an estimated 4051 survivors.6 Many survivors are suffering from physical and mental sequelae.7 ‘Post-Ebola virus disease syndrome’ characterises these complex symptoms, which may affect survivors' daily functions and ability to work and may include chronic joint and muscle pain, fatigue, hearing and vision loss, depression and memory problems.8 9

6 Many survivors are suffering from physical and mental sequelae.7 ‘Post-Ebola virus disease syndrome’ characterises these complex symptoms, which may affect survivors' daily functions and ability to work and may include chronic joint and muscle pain, fatigue, hearing and vision loss, depression and memory problems.8 9 Surveys have also identified that Ebola survivors may face discrimination and stigmatisation from their communities and their loved ones. In August 2014, a national household survey conducted in Sierra Leone showed that 96% of respondents held at least one discriminatory attitude towards Ebola survivors.10 Although declining, 46% of respondents still held at least one discriminatory viewpoint in July 2015.11–14 In Sierra Leone, The Comprehensive Program for Ebola Survivors (CPES) has been developed to protect Ebola survivors. The WHO, together with partner organisations involved in the Ebola response, offered technical support for the development and implementation of this program. It aims to ensure Ebola survivors access the health and social welfare services they require. It includes programs that target stigmatisation and the health-related complications of the disease, and also aims to prevent further transmission, especially through sexual intercourse. It includes counselling and access to health services such as eye clinics, sexual health counselling for individuals and couples, and semen testing.15 16

that target stigmatisation and the health-related complications of the disease, and also aims to prevent further transmission, especially through sexual intercourse. It includes counselling and access to health services such as eye clinics, sexual health counselling for individuals and couples, and semen testing.15 16 In April–May 2015, a qualitative evaluation was conducted to assess survivors' knowledge, attitudes, practices and experiences related to Ebola in Sierra Leone in order to understand survivor needs and stigma directed towards them. These findings may help inform national and local efforts to strengthen services for survivors, address identified barriers and reinforce trust in the healthcare system in a post-Ebola environment in Sierra Leone. Methods This evaluation followed a mixed method design consisting of qualitative interviews followed by short post-interview surveys. The post-interview survey was conducted in order to quantify key thematic areas from the in-depth interviews (IDIs). Triangulating the short survey data with the in-depth qualitative findings aimed to provide a more accurate narrative of survivors' experiences, perceptions, attitudes and practices.

t-interview surveys. The post-interview survey was conducted in order to quantify key thematic areas from the in-depth interviews (IDIs). Triangulating the short survey data with the in-depth qualitative findings aimed to provide a more accurate narrative of survivors' experiences, perceptions, attitudes and practices. Sampling and selection of participants The evaluation consisted of 28 IDIs with survivors at least 18 years of age, followed by a short survey conducted immediately after each IDI. The researchers discussed data saturation and estimated it to be reached at or after 20 interviews. Five districts of Sierra Leone were purposively selected to cover all four geographic regions of the country and to include survivors with varied Ebola Treatment Units (ETUs) discharge dates: Western Urban and Western Rural (both western area), Kambia and Port Loko (both northern province), Kono (eastern province) and Moyamba (southern province). Four interviews were conducted in each district, except in Kambia, where eight individuals participated to include survivors from the Guinea border area. Survivors from Guinea were excluded and only those who were infected in Kambia and were treated in an ETU in Kambia or Port Loko were included in the study. ETUs' psychosocial counsellors and Survivor Psychosocial Networks present in each district were contacted to purposively identify survivors which were interviewed in their homes alone. Snowball sampling was also used to ask participants to refer other potential Ebola survivors. Diversity in participants' sex, discharge date, rural versus urban setting and age was ensured in the selection process. No relationship with participants was established prior to study initiation, and all participants provided informed consent for their participation in this study. Participants had to be ≥18 years old and had to present an Ebola survivor certificate issued by the government of Sierra Leone. No participant refused to participate or dropped out of the study.

established prior to study initiation, and all participants provided informed consent for their participation in this study. Participants had to be ≥18 years old and had to present an Ebola survivor certificate issued by the government of Sierra Leone. No participant refused to participate or dropped out of the study. Data collection Data collection occurred between April and May 2015. Facilitators and note-takers, both males and females, were trained during a 3-day workshop on how to conduct the IDIs. They were led by a senior program manager of FOCUS1000 (MFJ). The interview guide and protocols were first developed by two behavioural scientists with field experience of working in the Ebola response in Sierra Leone. The draft instruments were then reviewed by a senior scientist with extensive experience working with HIV populations in sub-Saharan Africa. Local contextualisation and cultural appropriateness were ensured by a second review from a partner from a local non-governmental organisation (NGO) with experience in conducting knowledge, attitude and practice (KAP) assessments in Sierra Leone. The instruments were then pre-tested by the behavioural scientists with three survivors from the Western Area (which were not included in the final study results) and feedback was used to revise and finalise the instruments. Each IDI (see online supplementary appendix 1) was directly followed by a post-interview survey (see online supplementary appendix 2), which was administered face-to-face by the IDI facilitator using a paper-based instrument. Interviews were conducted in Krio, the most widely spoken language in Sierra Leone, to ensure consistency of the interview questions and to help standardise questions and probes. However, some keywords were translated to the dominant local languages and participants were given the opportunity to express their views on particular subjects in their local languages if necessary. All survivors interviewed were fluent and agreed to be interviewed in Krio. Interviews were tape-recorded and supplemented with hand-written notes. Facilitators and note-takers transcribed and translated the recordings into English. All transcripts were reviewed by a separate supervisory team to ensure accuracy.

ssary. All survivors interviewed were fluent and agreed to be interviewed in Krio. Interviews were tape-recorded and supplemented with hand-written notes. Facilitators and note-takers transcribed and translated the recordings into English. All transcripts were reviewed by a separate supervisory team to ensure accuracy. 10.1136/bmjgh-2016-000108.supp1Supplementary data Coding and data analysis Through a deductive coding process, two of the authors developed and reached consensus on parent codes drawn from the IDI guide. Subcodes were developed using an inductive process. Evaluators coded transcripts using Dedoose, a web-based, qualitative analysis software package and achieved consensus on a final codebook which was used to code all transcripts. Key themes were then identified and summarised. The methodological orientation underpinning the study was a phenomenological inquiry. No repeat interviews could be carried out, and transcripts and findings could not be returned to participants for feedback due to the ongoing Ebola outbreak at the time of the interviews. However, after each interview, major points discussed were repeated to participants for concurrence. Post-interview survey data were entered into Open Data Kit (ODK) and subsequently uploaded to Kobo Collect's web-server for aggregation. The data repository was then imported into SPSS V.22 for analysis. The analysis included the generation of descriptive statistics using frequency tables as well as cross-tabulations using contingency tables.

ta were entered into Open Data Kit (ODK) and subsequently uploaded to Kobo Collect's web-server for aggregation. The data repository was then imported into SPSS V.22 for analysis. The analysis included the generation of descriptive statistics using frequency tables as well as cross-tabulations using contingency tables. Ethical approval The Sierra Leone Ethics and Scientific Review Committee approved the protocol and the US Centre for Disease Control (CDC) Institutional Review Board (IRB) determined the project to be public health non-research. This indicates data collection with human participants was done as a routine public health practice in the context of an unprecedented epidemic and, therefore, does not require ethical approval.

US Centre for Disease Control (CDC) Institutional Review Board (IRB) determined the project to be public health non-research. This indicates data collection with human participants was done as a routine public health practice in the context of an unprecedented epidemic and, therefore, does not require ethical approval. Results Participant characteristics Participants (table 1) included 14 males and 14 females, with a mean age of 31 years (ranging from 18 to 67 years). Survivors were discharged from ETU an average of 4 months (range: 2–7) prior to the interviews, indicating that many were infected and treated around the outbreak peak (November 20144). Unemployment among participants rose from 7.1% prior to Ebola infection to 39.3% at the time of the interviews with a decrease in the number of private business owners (from 10.7% to 3.6%), petty traders (from 32.1% to 10.7%), students (from 10.7% to 7.1%) and teachers (from 7.1% to 3.6%), and an increase in the number of medical or health professionals (from 10.7% to 14.3%) and government employees (from 0% to 7.1%). The increase in the number of health professionals was due to the recruitment and involvement of some Ebola survivors in Ebola control activities. Table 1 Demographic characteristics of interviewed Ebola survivors, Sierra Leone, April–May 2015 (N=28)

Results Participant characteristics Participants (table 1) included 14 males and 14 females, with a mean age of 31 years (ranging from 18 to 67 years). Survivors were discharged from ETU an average of 4 months (range: 2–7) prior to the interviews, indicating that many were infected and treated around the outbreak peak (November 20144). Unemployment among participants rose from 7.1% prior to Ebola infection to 39.3% at the time of the interviews with a decrease in the number of private business owners (from 10.7% to 3.6%), petty traders (from 32.1% to 10.7%), students (from 10.7% to 7.1%) and teachers (from 7.1% to 3.6%), and an increase in the number of medical or health professionals (from 10.7% to 14.3%) and government employees (from 0% to 7.1%). The increase in the number of health professionals was due to the recruitment and involvement of some Ebola survivors in Ebola control activities. Table 1 Demographic characteristics of interviewed Ebola survivors, Sierra Leone, April–May 2015 (N=28) Total (#) Location of interview Western Urban 4 Western Rural 4 Kambia 8 Moyamba 4 Port Loko 4 Kono 4 Age, years 18–25 10 26–64 16 ≥65 2 Marital status (at time of interview) Single 19 Married or with a partner 9 Number of months since release from ETUs (months) 2 6 3 5 4 5 5 6 6 5 7 1 Location at time of illness Same district as where interview was conducted 25 Different district as where interview was conducted 3 Education level No formal education 7 Some primary school 3 Completed primary school 2 Completed Junior secondary school 8 Completed Senior secondary school 4 Completed post-secondary education or training 4 Employment at time of interview Private business 1 Plumber/carpenter/electrician 0 Petty Trader 3 Farmer 3 Teacher/lecturer/instructor 1 Motorcycle taxi driver 1 Medical or health professional 4 Student 2 Other government employee 2 Unemployed 11 Survey results The survey found that 25% (7 of 28) of the survivors interviewed (42.9% males vs 57.1% females) moved home after their return from ETUs and 42.8% (3/7) of those moved to a new district. 89.3% (25 of 28) lived with their relatives at the time of the interviews, of which 88% (22 of 25) had lived with the same people before getting sick.

that 25% (7 of 28) of the survivors interviewed (42.9% males vs 57.1% females) moved home after their return from ETUs and 42.8% (3/7) of those moved to a new district. 89.3% (25 of 28) lived with their relatives at the time of the interviews, of which 88% (22 of 25) had lived with the same people before getting sick. When questioned about potential sources of transmission, 60.7% (17 of 28) of participants stated that they had taken care of sick individuals (58.8%—10 of 17—of which were females), 46.4% (13 of 28) were living in the same household as someone suspected or confirmed to have Ebola, 14.3% (4 of 28) had participated in a burial or funeral (75%—3 of 4—of which were males, all released 6–7 months prior to the interviews) and 7.1% (2 of 28) were unsure of their source of infection. 61.5% (8 of 12) of those released 5–7 months prior to the interviews had shared a household with someone suffering from Ebola compared to 38.5% (5 of 16) of those released 2–4 months prior to the interviews. 75% (21 of 28) of survivors interviewed first sought treatment from a medical professional at a healthcare facility (61.8%—13 of 21—of which were released from an ETU 2–4 months prior to the interview), with only 21.4% (6 of 28) of participants having sought treatment from a medical professional outside a healthcare facility (66.6%—4 of 6—of which were released 5–6 months prior to the interview) and 3.6% (1/28) from a traditional leader (released 4 months prior to the interview). The mean number of days it took for participants to seek treatment after feeling sick was 3.36 (SD 1.7).

from a medical professional outside a healthcare facility (66.6%—4 of 6—of which were released 5–6 months prior to the interview) and 3.6% (1/28) from a traditional leader (released 4 months prior to the interview). The mean number of days it took for participants to seek treatment after feeling sick was 3.36 (SD 1.7). Participants described their day-to-day interaction with community members since their release from an ETU as very good (13 of 28, 46.4%), good (9 of 28, 32.1%), not good (4 of 28, 14.3%) or very bad (2 of 28, 7.1%). More women than men had negative experiences, with 28.5% (4 of 14) of women describing a negative community interaction compared to 14.2% (2 of 14) of men. Furthermore, 66.6% (4 of 6) of those who described a negative experience and 54.5% (12 of 22) of those who described a positive experience were released from an ETU 2–4 months prior to the interview. A minority of survivors interviewed (8 of 28, 28.6%) stated that they had engaged in sexual intercourse since their ETU discharge (50% of which were females), with 50% (4 of 28) of them also reporting not having used a condom during their last intercourse (75% of which were females). The results showed that 87.5% (7 of 8) of those sexually active and 75% (3 of 4) of those who did not use a condom were released from an ETU 5–7 months prior to the interview.

ich were females), with 50% (4 of 28) of them also reporting not having used a condom during their last intercourse (75% of which were females). The results showed that 87.5% (7 of 8) of those sexually active and 75% (3 of 4) of those who did not use a condom were released from an ETU 5–7 months prior to the interview. Survivors' emotional and physical state regarding illness On diagnosis, nearly all survivors recalled feeling terrified about the uncertainty of what would happen to them, and being sick with Ebola made them feel depressed. A few survivors also acknowledged feeling shocked, as they had initially believed Ebola did not exist in Sierra Leone: “I never believed Ebola was real, so I was emotionally confused. (…) I thought nothing was wrong with me” (Moyamba, female). After learning of their diagnosis, most survivors thought death was inevitable, while some felt they were already dead. Having seen others die from Ebola, the prospect of a similar fate was reported to be emotionally challenging. They also explained they were afraid of changes in their physical appearance. Many reported being extremely weak and unable to stand, speak, sit, walk or work; they felt helpless and tormented.

re already dead. Having seen others die from Ebola, the prospect of a similar fate was reported to be emotionally challenging. They also explained they were afraid of changes in their physical appearance. Many reported being extremely weak and unable to stand, speak, sit, walk or work; they felt helpless and tormented. Beliefs on how infection occurred Most survivors believed they contracted the disease by helping or caring for a sick family member: “I believe I got infected when I was caring for my sick wife. When she was sick, she couldn't do anything on her own (…). I used to clean her and helped her when she wanted to do anything” (Western Area Rural, male). Some also discussed attending burials or funerals before contracting the disease. One survivor stated: “After the burial, mourners and other people within the compound got sick and started showing signs [of Ebola]. I, as one of them, also fell sick” (Kambia, female).

e wanted to do anything” (Western Area Rural, male). Some also discussed attending burials or funerals before contracting the disease. One survivor stated: “After the burial, mourners and other people within the compound got sick and started showing signs [of Ebola]. I, as one of them, also fell sick” (Kambia, female). Beliefs and experience with care seeking As national response coordination consolidated, individuals were instructed to report suspected Ebola cases to the national 117 telephone hotline or local call numbers. Before alerting health officials, many survivors tried self-medication such as oral-rehydration therapy, a common practice used for cholera and other endemic diarrheal diseases. A few survivors also described calling the 117 helpline or informing community leaders to report illness as a mean of survival. One survivor described first seeking care from a traditional healer. Survivors reported their family, community members or sometimes staff working in a quarantined area calling 117 for them, without giving them a choice: “Since I was under quarantine, there was no way to seek help elsewhere” (Western Area Rural, male).

One survivor described first seeking care from a traditional healer. Survivors reported their family, community members or sometimes staff working in a quarantined area calling 117 for them, without giving them a choice: “Since I was under quarantine, there was no way to seek help elsewhere” (Western Area Rural, male). Survivors described varied experiences with the ambulances that took them to the Ebola Holding Units (EHU). Many reported being afraid, especially of the smell and use of chlorine to disinfect the vehicles. A woman from Western Area also explained: “[I] went straight to the Ambulance but when I touched the door, one of the ambulance guys shouted at me and sprayed where I touched the door. I was not happy.” A minority of survivors also expressed dissatisfaction with the ambulance services, because the ambulances were reportedly too hot and had locked windows. Some were scared of the health workers' personal protective outfits, or of the negative image associated with the ambulances: “My mother and sisters were all crying bitterly, falling on the ground and giving remarks that I will die if I enter the Ambulance” (Kono, female). However, many survivors talked positively about the ambulance services, and praised the health workers who “started counselling me, saying that I should not be afraid and I will be ok and nothing is going to happen to me” (Western Urban, female).

giving remarks that I will die if I enter the Ambulance” (Kono, female). However, many survivors talked positively about the ambulance services, and praised the health workers who “started counselling me, saying that I should not be afraid and I will be ok and nothing is going to happen to me” (Western Urban, female). Experience with the EHU and ETU Following an initial assessment, patients were generally transported to EHUs, where blood samples were tested. If positive, patients would be transferred to an ETU. Ideally, results were given within 24 hours, but early in the response this process was sometimes delayed by several days. While about half of survivors reported positive experiences at the EHUs and felt that they were well taken care of; the other half described EHU staff as unkind, inattentive and refusing to be in close physical contact with patients. In retrospect, a few survivors believed staff at EHUs lacked knowledge of Ebola because they told them there was no treatment available to cure the disease and yet they survived. Some survivors reported that they did not receive counselling, did not have enough food or drugs, and were left alone in an unsafe and untidy environment. One survivor mentioned: “Because of the poor treatment by the doctor I was afraid to eat the food they served me in the holding centre. I thought they wanted to kill me” (Kambia, female).

d that they did not receive counselling, did not have enough food or drugs, and were left alone in an unsafe and untidy environment. One survivor mentioned: “Because of the poor treatment by the doctor I was afraid to eat the food they served me in the holding centre. I thought they wanted to kill me” (Kambia, female). Individuals who tested positive for Ebola at the EHUs were transferred to an ETU. Nearly all interviewed survivors commented that ETU staff was kind and supportive and they were ‘treated like kings’. Survivors described being pleased with the free food and medication, and appreciated counselling from health workers, including international staff. They frequently expressed enjoying the company of polite and friendly facility staff. They also mentioned being treated fairly and equally, independently of their ‘race, colour or tribe’.

bed being pleased with the free food and medication, and appreciated counselling from health workers, including international staff. They frequently expressed enjoying the company of polite and friendly facility staff. They also mentioned being treated fairly and equally, independently of their ‘race, colour or tribe’. Emotional state after discharge from the ETUs Most survivors explained they were relieved after being discharged from the ETUs, felt like ‘heroes’, and thanked God for their survival. Some saw their recovery as a new start to life: “Once there is life, there is hope: a dead man cannot work, a dead man cannot move” (Western Area Rural, male). However, feelings of sadness, depression and anger were also expressed by a lot of survivors and they reported feeling unsupported. “Since I was discharged I have been crying for help but nobody helps me” (Western Areal Rural, female). Some survivors also expressed frustration after being discharged because their everyday life had deteriorated: “There are times I have the feeling that life is worthless as I now don't engage in anything and now some community people are neglecting me” (Port Loko, male). A few also reported feelings of loneliness and alienation due to loosing family members to the disease.

ischarged because their everyday life had deteriorated: “There are times I have the feeling that life is worthless as I now don't engage in anything and now some community people are neglecting me” (Port Loko, male). A few also reported feelings of loneliness and alienation due to loosing family members to the disease. Family relationships Most survivors described living ‘happily’, and having a good relationship with their families before becoming infected with Ebola. They further shared that such relationships did not change after their return from the ETUs, even though some households were grieving loss of other family members who had died. All interviewed survivors reported that their families welcomed them back with joy, and treated them well. Additionally, families took the roles of carers for interviewed survivors, providing either health, financial or housing support. Their lives were described, once again, as ‘normal’, as if nothing had changed: “My brothers are very much happy to see me back and do things together as we used to do” (Western Urban, male).

dditionally, families took the roles of carers for interviewed survivors, providing either health, financial or housing support. Their lives were described, once again, as ‘normal’, as if nothing had changed: “My brothers are very much happy to see me back and do things together as we used to do” (Western Urban, male). Partner relationships Several survivors who had a partner (boyfriends/girlfriends or husbands/wives) at the time of the interviews (9 of 28) did not report any major changes in their relationships. Only a few survivors mentioned that their partners, mostly boyfriends or girlfriends, did not treat them well when they returned from the ETUs or were afraid of or embarrassed by them. “Since the day [my boyfriend] knew I was Ebola positive, he stopped picking up my calls. (…) I know he was afraid or ashamed of me because I was infected with Ebola” (Kambia, female). While some partners ended their relationship with interviewed survivors, others reunited after ‘counselling sessions’ with social work staff.

e the day [my boyfriend] knew I was Ebola positive, he stopped picking up my calls. (…) I know he was afraid or ashamed of me because I was infected with Ebola” (Kambia, female). While some partners ended their relationship with interviewed survivors, others reunited after ‘counselling sessions’ with social work staff. Sexual behaviour and knowledge of potential risks of Ebola sexual transmission At the time of the interviews, messaging on sexual transmission of Ebola recommended abstinence or condom use out of caution as it was unknown whether Ebola could be sexually transmitted. All survivors reported being counselled to not have unprotected sexual intercourse for the first 90 days on discharge from the ETU, which was consistent with WHO interim advice given to survivors prior to May 2015. Many survivors conveyed that it would be risky for their partners to have sexual relationships with them. They overwhelmingly emphasised that they do not want others to go through what they had suffered. Some of the survivors reported that they decided to wait longer than the recommended 90 days and used condoms to make sure that their partner(s) would not be exposed to Ebola through sexual transmission.

onships with them. They overwhelmingly emphasised that they do not want others to go through what they had suffered. Some of the survivors reported that they decided to wait longer than the recommended 90 days and used condoms to make sure that their partner(s) would not be exposed to Ebola through sexual transmission. In one instance, a male survivor from Port Loko expressed that all Ebola survivors should be quarantined for 3 months ‘so they will not be infecting people with the disease through their sexual activities’. Other survivors mentioned that they advised their community members not to have sexual intercourse with survivors. A handful of survivors expressed that this was mostly important for males or for people who have sex with multiple partners. Finally, a few survivors experienced a loss of libido and some women felt guilty about not having sex with their husbands. One male survivor also mentioned: “My wife denied me sometimes because of the fear of the disease” (Moyamba, male).

was mostly important for males or for people who have sex with multiple partners. Finally, a few survivors experienced a loss of libido and some women felt guilty about not having sex with their husbands. One male survivor also mentioned: “My wife denied me sometimes because of the fear of the disease” (Moyamba, male). Community relationships The potential negative reaction of community members—sometimes including loved ones—emerged as a major concern for interviewed survivors when they initially suspected Ebola, which led some to deny feeling ill. However, the relationships between survivors and their communities were described positively both before and after Ebola by the majority of survivors. Survivors described themselves as social individuals, who were interactive and good ‘religious’ members of their communities, attending either the church or the mosque: “I am a Muslim observing my prayers. I have been a very good social person mingling with people in my community” (Kambia, female). After they were discharged from ETUs, many survivors mentioned they were welcomed back into their communities and continued good relationships and interactions with community members, including religious and local leaders: “People living in my community have no problem with me as we interact together, attend community meetings, play games together and do things together” (Port Loko, male). Survivors shared that community members expressed their sympathy, especially if they had also lost members of their families and prayed for them.

ers: “People living in my community have no problem with me as we interact together, attend community meetings, play games together and do things together” (Port Loko, male). Survivors shared that community members expressed their sympathy, especially if they had also lost members of their families and prayed for them. However, a few survivors also experienced discrimination from members of their communities. They felt people were afraid of them, provoked them and stigmatised them as survivors: “Most of my neighbours abandoned and refused to accept me” (Kono, male). They were deeply affected by the discriminatory attitudes of their community, and felt abandoned, stressed and lonely due to provocations and stigmatisation. Some survivors also described being evicted from their houses by landlords who were scared of possible contagion. Some had reduced their interactions at the church or mosque and others stated that people actively avoided them and their children.

elt abandoned, stressed and lonely due to provocations and stigmatisation. Some survivors also described being evicted from their houses by landlords who were scared of possible contagion. Some had reduced their interactions at the church or mosque and others stated that people actively avoided them and their children. Survivors' financial and employment situation Survivors described their lives before becoming infected with Ebola as normal, comfortable, doing good business. However, most survivors reported losing their jobs, facing financial difficulties and being unable to take care of their families: “Before, I was employed but now, I am not employed. (…) The more you earn, the more you live well with your family” (Western Area Rural, male). Reasons for inability to work were lack of strength and lack of finance to continue or start their businesses. Most of those shifting to unemployment were previously engaged in petty trade and business. After discharge from the ETU, one survivor became a healthcare worker and another one a government official.

, male). Reasons for inability to work were lack of strength and lack of finance to continue or start their businesses. Most of those shifting to unemployment were previously engaged in petty trade and business. After discharge from the ETU, one survivor became a healthcare worker and another one a government official. Other problems mentioned were difficulties paying for their children's school fees and feeling dependent on others: “'I find it difficult now to take care of my children because I have no money, no business (…). I have to beg for our survival, but how long would that continue for?” (Port Loko, male). Moreover, although most survivors reported having received financial benefits from various organisations after being discharged from the ETUs, some reported having received insufficient assistance to help them become financially independent.

eg for our survival, but how long would that continue for?” (Port Loko, male). Moreover, although most survivors reported having received financial benefits from various organisations after being discharged from the ETUs, some reported having received insufficient assistance to help them become financially independent. Improving the situation of survivors Almost all survivors suggested that the government of Sierra Leone should help them by providing jobs, microcredit or training so they could develop necessary skills for employment. They also discussed the need for financial help and their desire to receive money, scholarships and other incentives. They mentioned their need for livelihood support, including the provision of food and supplies as well as housing. Finally, some survivors stated that the government should help in engaging communities and households on the discrimination of survivors. A woman from Moyamba explained that it is important “to sensitise community members (…) on the issue of discriminating Ebola survivors.” Another survivor said that the “government should pass laws in the parliament that no one should discriminate [against] any survivor” (Moyamba, male).

the discrimination of survivors. A woman from Moyamba explained that it is important “to sensitise community members (…) on the issue of discriminating Ebola survivors.” Another survivor said that the “government should pass laws in the parliament that no one should discriminate [against] any survivor” (Moyamba, male). Improving the healthcare system In addition to improving their own situation, a few survivors suggested the need to improve health centres to help the country end the outbreak. In their opinion, health facilities are currently not well equipped, with insufficient drugs and other resources and are difficult to access: “I believe if the health workers have the right skills, facilities and the community have access to the needed resources, it will contribute to moving the country to Ebola free” (Western Area Rural, female). Survivors also recommended improving sanitation, including drinking water, latrines and hand washing.

to access: “I believe if the health workers have the right skills, facilities and the community have access to the needed resources, it will contribute to moving the country to Ebola free” (Western Area Rural, female). Survivors also recommended improving sanitation, including drinking water, latrines and hand washing. Survivors' contribution to the Ebola response All the survivors considered it their role to help their country end the Ebola outbreak and they should be “used as partners in the Ebola fight” (Kambia, male). A few of them mentioned that their first-hand knowledge of Ebola and their lack of fear due to perceived immunity would support their involvement. They strongly believed that they should become involved in social mobilisation. One survivor explained multiple ways in which Ebola survivors could become active in the fight against Ebola: “We can (…) talk to people in our communities (…). On contact tracing, we survivors can go and meet with people that might have come into contact with Ebola infected persons and [have] ran away and [we can] explain to them how we (…) survived because we got early treatment. We will encourage them to comply with the quarantine” (Western Area Rural, female). Many were already involved in sharing their personal experiences with the community in churches and mosques, restaurants, schools or markets. They reported that they had an important role in Ebola prevention and control, and also that this role could help financially and emotionally sustain them as they adjust to being a survivor. Some also thought that they should be involved in contact tracing and others saw themselves as informal enforcers of Ebola response efforts.

ed that they had an important role in Ebola prevention and control, and also that this role could help financially and emotionally sustain them as they adjust to being a survivor. Some also thought that they should be involved in contact tracing and others saw themselves as informal enforcers of Ebola response efforts. Discussion Our findings suggest that survivor experiences, emotions and attitudes changed over time as they moved from disease onset to treatment, discharge and to life post-discharge. Major themes identified across the interviews included acute fear and depression linked to Ebola diagnosis, negative experiences with EHUs, positive experiences with ETUs, feelings of joy and thankfulness on discharge, altruistic motivations for the prevention of Ebola through sexual transmission and concerns about their financial situation and discrimination from their communities.

and depression linked to Ebola diagnosis, negative experiences with EHUs, positive experiences with ETUs, feelings of joy and thankfulness on discharge, altruistic motivations for the prevention of Ebola through sexual transmission and concerns about their financial situation and discrimination from their communities. The findings demonstrate that survivors' knowledge and attitudes about sexual transmission risk largely reflected counselling messages to abstain or use condoms. It should be noted that messaging guidelines on sexual transmission evolved during the outbreak as more information became available which could have led to misperceptions and distrust of information in survivors. It could also explain why interviewed survivors released earlier in the outbreak from ETUs engaged more frequently in sexual activities as survivors were first told to use condoms or abstain for 3 months after their release, which later changed to 6 months and the 12 months. Accurate, evaluated and consistent messages on sexual transmission and appropriate condom use need to be developed and delivered to survivors and their partners. Targeted approaches to HIV and sexually transmitted infection prevention have proved to be highly effective in Sierra Leone and could be applied to the prevention of sexual transmission of Ebola.17 One recent study has shown that Ebola virus RNA can persist in semen for 284 days after the disease onset and evidence on viral persistence in other body fluids is still emerging.18 19 In light of this new evidence, semen testing and counselling services may help reduce the possibility of resurgence. Having survivors registered and part of a longer term care and support system sustaining community engagement to reduce Ebola exposure risks while minimising stigma of survivors may also reduce future risk.20

ght of this new evidence, semen testing and counselling services may help reduce the possibility of resurgence. Having survivors registered and part of a longer term care and support system sustaining community engagement to reduce Ebola exposure risks while minimising stigma of survivors may also reduce future risk.20 While all interviewees reported supportive attitudes from family members, about a third faced discrimination and stigma from their communities after their discharge. Continued promotion of the integration and acceptance of Ebola survivors in communities could reduce stigma and improve health seeking behaviours. An example of a successful reintegration strategy is the Firestone Project in Liberia where teams travelled to survivor homes before their discharge to meet with communities and engage them in plans to welcome survivors home.21 These types of strategies were replicated in some regions of Sierra Leone and some NGOs, such as Partners in Health, also supported employment opportunities for survivors in the response.22 The Sierra Leone Association of Ebola Survivors was created in January 2015 to seek and ensure protection and welfare of its registered members, including Ebola survivors, orphans, widows and widowers.23 Such organisations can strengthen the voice of survivors, giving them more power to demand financial and emotional security and support from governments. They can improve the visibility of survivors, help raise awareness about their experiences and struggles, lobby government and international partners for improvements to services offered to survivors, and provide a forum in which survivors can support each other.

ial and emotional security and support from governments. They can improve the visibility of survivors, help raise awareness about their experiences and struggles, lobby government and international partners for improvements to services offered to survivors, and provide a forum in which survivors can support each other. The Government of Sierra Leone collaborated with the United Nations Development Programme (UNDP) for the implementation of the ‘Social Rehabilitation and Payment to Ebola Survivors Project’ which aims to prevent conflict and build resilience by addressing social marginalisation and discrimination of Survivors.24 Local NGOs, such as Focus1000, and international NGOs, such as Partners in Health (PIH) and Wold Hope International (WHI), have implemented interventions to support survivors. These interventions have already shown that by addressing issues of stigma and discrimination at both the community level and the national level, individual attitudes and actions towards survivors may shift from the need to exclude to the need to include. It is important to maintain and improve these existing activities, as well as to ensure they are implemented to cover all survivors across West Africa. Communication strategies that support survivor reintegration into the community need to be evaluated and further developed.

the need to exclude to the need to include. It is important to maintain and improve these existing activities, as well as to ensure they are implemented to cover all survivors across West Africa. Communication strategies that support survivor reintegration into the community need to be evaluated and further developed. Interviewees emphasised that the improvement of health services is important for ending Ebola transmission. The West-African Ebola outbreak has shed light on issues related to existing medical and epidemiological capacity to respond to emerging disease outbreaks, including problems with the organisation and performance of health systems.25 Furthermore, challenges around building community trust and confidence in the healthcare system persist. Efforts to understand and address trust and confidence in the healthcare systems within particularly heavily affected communities are critical for future planning. This could start with the empowerment and engagement of survivors and entire communities in their health, and ensuring government and international actions are transparent and well communicated to the public.

ce in the healthcare systems within particularly heavily affected communities are critical for future planning. This could start with the empowerment and engagement of survivors and entire communities in their health, and ensuring government and international actions are transparent and well communicated to the public. This study revealed that half of interviewees reported negative experiences in EHUs. These may have been influenced by the fact that they were ill earlier in the outbreak, when services were not yet fully functional and misconceptions about the disease and its transmission where common. These misconceptions, especially when strengthened by fear, might have impacted beliefs, attitudes and behaviours of staff working at EHUs towards Ebola patients. On the other hand, nearly all survivors were positive about their ETU experiences. The difference in experiences at EHUs and ETUs could be explained by the lower staff to patient ratios, poorer pay and lower levels of qualifications of staff in EHUs compared with ETUs. ETUs were restricted to Ebola patients, and had therefore higher levels of expertise and preparation to take care of Ebola patients.26 27

difference in experiences at EHUs and ETUs could be explained by the lower staff to patient ratios, poorer pay and lower levels of qualifications of staff in EHUs compared with ETUs. ETUs were restricted to Ebola patients, and had therefore higher levels of expertise and preparation to take care of Ebola patients.26 27 Ending an Ebola outbreak evidently requires coordinated efforts between government, public health systems and community structures, including strong involvement from trusted community leaders. Survivors were strongly motivated to help end Ebola and to improve the healthcare system. They could serve as valuable resources in connecting the national Ebola response teams to local communities. Some studies have also shown that survivors have a role to play in engaging communities by teaching others how Ebola is transmitted and by helping families understand the need for isolation of individuals with symptoms.28 29 A previous Ebola outbreak saw survivors working alongside safe burial teams, contact tracers and community educators, which, to some extent, also happened in Sierra Leone.11 They can potentially contribute to Sierra Leone's readiness to prevent, detect and respond to future outbreaks as well as in strengthening public confidence in the healthcare system. Ways of incorporating survivors into response leadership roles also need to be identified.

some extent, also happened in Sierra Leone.11 They can potentially contribute to Sierra Leone's readiness to prevent, detect and respond to future outbreaks as well as in strengthening public confidence in the healthcare system. Ways of incorporating survivors into response leadership roles also need to be identified. Acute fear and depression emerged as a major theme among interviewees when discussing their initial experiences with the disease. Many survivors were still facing substantial emotional challenges, financial difficulties, dealing with having lost family members and decreased normal interactions with community members at the time the interviews were conducted. Coupled with some persistent discrimination, survivors faced myriad economic, social and health challenges. Addressing these concerns could facilitate survivors' reintegration into communities. The medical and social service interventions offered within CPES target these issues.16 Emotional challenges faced by survivors could first be addressed by listening to survivors at an individual level (counselling sessions) and a community level (giving them a voice to share their experiences). It is also important to address stigma in communities, by seeking support from community leaders, organising talks and discussions, or publicly supporting survivors in the media. Finally, there could be a benefit from empowering survivors by including them in future outbreak response activities and giving them an active role in preventing Ebola in their communities.21 30

, by seeking support from community leaders, organising talks and discussions, or publicly supporting survivors in the media. Finally, there could be a benefit from empowering survivors by including them in future outbreak response activities and giving them an active role in preventing Ebola in their communities.21 30 More research is needed to identify gaps in the health systems, as well as the education, religion, business and government systems, which may contribute to survivor isolation and stigma. Evaluation of current activities referred to in this discussion will also be important to assess their actual impact on survivors and identify strengths, weaknesses, opportunities and threats. Many of the issues revealed in this evaluation are similar to those identified among survivors from previous Ebola outbreaks;11 31 ensuring that lessons learnt from this and previous evaluations are translated into Ebola response, recovery and health protection activities is essential and could help not only Sierra Leone but also Liberia and Guinea.

this evaluation are similar to those identified among survivors from previous Ebola outbreaks;11 31 ensuring that lessons learnt from this and previous evaluations are translated into Ebola response, recovery and health protection activities is essential and could help not only Sierra Leone but also Liberia and Guinea. Limitations While this evaluation aimed to provide a detailed understanding of the experiences of Ebola survivors to inform Ebola recovery strategies and preparedness for future outbreaks in Sierra Leone; the findings may not be representative of all survivors in Sierra Leone. The outbreak has already lasted over 17 months and impacted all 14 districts, and some of the attitudes and experiences identified in this evaluation may be specific to their geographic location, the ETUs and EHUs they attended, and time of infection. Additionally, survivors were invited to participate in the interviews through local leaders who may have selected the most aware and informed members of the community and also those who may have had the most positive experiences. Finally, EHUs and ETU experiences of those who did not survive are not captured by these interviews.

ly, survivors were invited to participate in the interviews through local leaders who may have selected the most aware and informed members of the community and also those who may have had the most positive experiences. Finally, EHUs and ETU experiences of those who did not survive are not captured by these interviews. Conclusion This evaluation provides a description of the diverse experiences of survivors, following many months of intensified Ebola response efforts, and gives an insight into beliefs of survivors about sources of transmission, healthcare seeking behaviours, life at EHUs and ETUs, acceptance by communities and loved ones and understandings of sexual transmission. This study also stresses the importance of empowering survivors and having them contribute to Sierra Leone's preparedness to face future outbreaks. Addressing the myriad and diverse challenges survivors face such as discrimination, stigma, loss of employment and health problems—should form a centre pillar of an Ebola outbreak recovery strategy. The authors thank Charles Alpren and John T Redd for their invaluable advice and feedback on the manuscript. Handling editor: Seye Abimbola Contributors: EK and HL co-led and conducted the analysis and interpretation of the qualitative data. JW and SL supported the analysis of the qualitative data. MFJ led the statistical analysis of the quantitative data. All authors contributed to the preparation of the manuscript and supported the interpretation of the qualitative and quantitative data.

conducted the analysis and interpretation of the qualitative data. JW and SL supported the analysis of the qualitative data. MFJ led the statistical analysis of the quantitative data. All authors contributed to the preparation of the manuscript and supported the interpretation of the qualitative and quantitative data. Funding: Funding was received from the ERAES programme (funded by the Wellcome Trust, grant number 1504) and FOCUS1000 received funding from the CDC Foundation, administered through eHealth Africa. Competing interests: None declared. Ethics approval: The Sierra Leone Ethics and Scientific Review Committee approved the protocol and the CDC IRB determined the project to be public health non-research. Provenance and peer review: Not commissioned; externally peer reviewed. Data sharing statement: No additional data are available.

Key questions What is already known about this topic? There are little prospective data describing the outcomes of paediatric surgery in low-resource settings. Emergency surgery is associated with more deaths and complications than elective surgery, but most studies carried out until now are in adults. What are the new findings? After accounting for differences in case mix, the odds of death after emergency abdominal surgery could be as high as seven times greater in low-income countries compared with high-income countries. Recommendations for policy The provision of effective essential surgery should be a key priority for global child health agendas and has significant potential to impact on the global burden of disease. Introduction Little data are available addressing the safety profile and risk factors affecting morbidity and mortality in children undergoing surgery globally. Most studies have been in adults and almost invariably were performed in high-resource countries.1–3 Although it is estimated that about 234 million surgical procedures are performed annually worldwide, the percentage of these involving children remains unknown.4 Studies from low- and middle-income countries (LMICs) have shown that in the neonatal period, mortality is associated with sepsis, multiple exposures to anaesthesia (reoperation), postoperative bleeding and complex congenital anomalies.5–8 Other risk factors include non-availability of trained personnel, delayed presentation, childbirth outside a hospital and financial constraints of the caregivers.9–11

period, mortality is associated with sepsis, multiple exposures to anaesthesia (reoperation), postoperative bleeding and complex congenital anomalies.5–8 Other risk factors include non-availability of trained personnel, delayed presentation, childbirth outside a hospital and financial constraints of the caregivers.9–11 Emergency surgery generally carries a higher morbidity and mortality compared with elective procedures.12 13 An estimated 33 000 emergency laparotomies in all ages are performed annually in the UK with a 15–20% mortality, which is 10-fold higher than that of elective cardiac surgery.14 Reasons for this high mortality are multifactorial; as well as patient-related factors, these may include staffing issues, access to operating theatres or access to diagnostic investigations.14 Unfortunately, most of these evidences have been derived from adult populations. To date, no prospective, multicentre, international investigation has evaluated the determinants of morbidity or mortality after emergency abdominal surgery in children on a global scale. The aim of the current study was to evaluate the mortality and morbidity of emergency abdominal surgery in children across countries of different human development indices (HDIs).

onal investigation has evaluated the determinants of morbidity or mortality after emergency abdominal surgery in children on a global scale. The aim of the current study was to evaluate the mortality and morbidity of emergency abdominal surgery in children across countries of different human development indices (HDIs). Methods Study design This was a cohort study of children under the age of 16 years recruited from multiple hospitals performing emergency abdominal surgery. Predefined data items were collected according to a previously published protocol (ClinicalTrials.gov identifier: NCT02179112)15 using the Research Electronic Data Capture (REDCap) which is an online data capture system.16 While the UK National Health Service Research Ethics review considered this study exempt from formal research registration (South East Scotland Research Ethics Service, reference: NR/1404AB12), individual centres obtained their own audit, ethical or institutional approval as appropriate.

n online data capture system.16 While the UK National Health Service Research Ethics review considered this study exempt from formal research registration (South East Scotland Research Ethics Service, reference: NR/1404AB12), individual centres obtained their own audit, ethical or institutional approval as appropriate. The collaborative model used has previously been described elsewhere.17 Investigators from self-selected surgical units identified consecutive patients within 2-week time intervals between 1 July 2014 and 31 December 2014. An open invitation for participation was disseminated through social media, personal contacts, email to authors of published emergency surgery studies and national/international surgical organisations. Short intensive data collection allowed surgical teams within these units to contribute meaningful numbers of patients without requiring additional resources. Multiple 2-week data collection periods within institutions was allowed.

f published emergency surgery studies and national/international surgical organisations. Short intensive data collection allowed surgical teams within these units to contribute meaningful numbers of patients without requiring additional resources. Multiple 2-week data collection periods within institutions was allowed. Patients and procedures Any hospital performing emergency abdominal surgery, which included paediatric patients, could choose to be included (self-selecting). Consecutive patients under age of 16 years undergoing emergency abdominal surgery during a chosen 2-week period between 1 July 2014 and 31 December 2014 were included. Emergency abdominal surgery was defined as any unplanned, non-elective operation, including reoperation after a previous procedure. Abdominal surgery was defined as any open, laparoscopic or laparoscopic-converted procedure that entered the peritoneal cavity. Elective (planned) or semielective procedures (where a patient initially admitted as an emergency was then discharged from hospital and readmitted at a later time for surgery) were excluded.

. Abdominal surgery was defined as any open, laparoscopic or laparoscopic-converted procedure that entered the peritoneal cavity. Elective (planned) or semielective procedures (where a patient initially admitted as an emergency was then discharged from hospital and readmitted at a later time for surgery) were excluded. Data Data were selected to be objective, standardised, easily transcribed and internationally relevant, in order to maximise record completion and accuracy. Recruited patients were followed up to day 30 after surgery or for the length of their inpatient stay where follow-up was not feasible. Records were uploaded by local investigators to the secure online REDCap website. The lead investigator at each site validated all cases prior to data submission. The submitted data were then checked centrally and where missing data were identified, the local lead investigator was contacted and requested to complete the record. Once vetted, the record was accepted into the data set for analysis.

te. The lead investigator at each site validated all cases prior to data submission. The submitted data were then checked centrally and where missing data were identified, the local lead investigator was contacted and requested to complete the record. Once vetted, the record was accepted into the data set for analysis. Outcome measures The primary outcome measure was 30-day postoperative mortality, defined as the number of patients in the cohort who died within 30 days of surgery.18 In the event where 30-day follow-up was unavailable, outcome status at the point of discharge from hospital was recorded. A ‘30-day postoperative mortality/death during hospital stay’, is shortened to ‘30-day mortality’ to aid readability. The secondary outcome measures were 24-hour mortality, major and minor complication, and surgical site infection (SSI). Complications were defined on the Clavien-Dindo scale:19 minor complications as grade I/II (any deviation from the normal postoperative course with or without a need for pharmacological treatment but without requirement for surgical, endoscopic and radiological interventions or critical care admission); reintervention as grade III (surgical, endoscopic or radiological reintervention, without requirement for critical care admission); and major complication as grade IV (complication requiring critical care admission).

ithout requirement for surgical, endoscopic and radiological interventions or critical care admission); reintervention as grade III (surgical, endoscopic or radiological reintervention, without requirement for critical care admission); and major complication as grade IV (complication requiring critical care admission). Statistical analysis The lack of pre-existing literature data in this subject meant that an a priori sample size determination was rendered difficult by unknown factors such as the effect of clustering and variation in mortality by diagnosis. Variation across different international health settings was assessed by stratifying participating centres by country into three tertiles according to the Human Development Index (HDI) rank. This is a composite statistic of life expectancy, education and income indices published by the United Nations (http://hdr.undp.org/en/statistics). Differences between HDI tertiles were tested with the Pearson χ2 test and Kruskal-Wallis test for categorical and continuous variables, respectively.

ment Index (HDI) rank. This is a composite statistic of life expectancy, education and income indices published by the United Nations (http://hdr.undp.org/en/statistics). Differences between HDI tertiles were tested with the Pearson χ2 test and Kruskal-Wallis test for categorical and continuous variables, respectively. Fixed effect binary logistic regression models were explored, and the variables determined to be statistically and clinically important were entered into full multivariable models. Final full model choice was guided by the Akaike information criterion (AIC). Hierarchical multivariable logistic regression models (random intercept) were constructed with country as the first level and patients as the second level. HDI tertile and other explanatory variables were included as fixed effects. Other than HDI tertile, all fixed effects were considered at the level of the patient. Coefficients are expressed as ORs with CIs and p values derived from percentiles of 10 000 bootstrap replications. Level 1 and 2 model residuals were checked and first-order interactions were tested. Goodness of model fit is reported with the Hosmer and Lemeshow test, and predictive ability described by area under the receiver operating characteristic (ROC) curve (c-statistic). All analyses were undertaken using the R Foundation Statistical Programme (R 3.1.1).

were checked and first-order interactions were tested. Goodness of model fit is reported with the Hosmer and Lemeshow test, and predictive ability described by area under the receiver operating characteristic (ROC) curve (c-statistic). All analyses were undertaken using the R Foundation Statistical Programme (R 3.1.1). Results Patients A total of 1409 patients aged under 16 years, from 253 centres in 43 countries, were included in this study (figure 1). At the time of operation, 282 (20.0%) were under the age of 2 years. Of all children, 694 (49.3%) were from high-HDI, 450 (31.9%) from middle-HDI and 265 (18.8%) from low-HDI groups. There were slightly more males than females in all HDI groups (table 1) (55.9% in high-HDI, 61.1% in middle-HDI and 58.1% in low-HDI groups). Missing data rates were low, with one missing outcome for 24-hour mortality and one missing outcome for 30-day mortality. In 1140/1409 patients, 30-day outcomes, which otherwise represent status at discharge, were confirmed by direct patient contact (80.9%; high 572/694, 82.4%; middle 358/450, 79.6%; low 210/265, 79.2%; χ2 test, p=0.361). Table 1 Patient characteristics

Results Patients A total of 1409 patients aged under 16 years, from 253 centres in 43 countries, were included in this study (figure 1). At the time of operation, 282 (20.0%) were under the age of 2 years. Of all children, 694 (49.3%) were from high-HDI, 450 (31.9%) from middle-HDI and 265 (18.8%) from low-HDI groups. There were slightly more males than females in all HDI groups (table 1) (55.9% in high-HDI, 61.1% in middle-HDI and 58.1% in low-HDI groups). Missing data rates were low, with one missing outcome for 24-hour mortality and one missing outcome for 30-day mortality. In 1140/1409 patients, 30-day outcomes, which otherwise represent status at discharge, were confirmed by direct patient contact (80.9%; high 572/694, 82.4%; middle 358/450, 79.6%; low 210/265, 79.2%; χ2 test, p=0.361). Table 1 Patient characteristics HDI tertile p Value High Middle Low Age in completed years Mean (SD) 8.9 (5.1) 9.1 (5.0) 7.0 (5.6) <0.001 Gender Male 388 (55.9) 275 (61.1) 152 (57.4) 0.216 Female 306 (44.1) 175 (38.9) 113 (42.6) Missing 0 (0.0) 0 (0.0) 0 (0.0) ASA grade 1 507 (73.1) 354 (78.7) 154 (58.1) <0.001 2 105 (15.1) 65 (14.4) 58 (21.9) 3 51 (7.3) 12 (2.7) 37 (14.0) 4 23 (3.3) 6 (1.3) 12 (4.5) 5 8 (1.2) 13 (2.9) 4 (1.5) Missing 0 (0.0) 0 (0.0) 0 (0.0) Surgical safety checklist used No, not available in this hospital 35 (5.0) 192 (42.7) 95 (35.8) <0.001 No, but available in this hospital 6 (0.9) 39 (8.7) 74 (27.9) Yes 653 (94.1) 217 (48.2) 96 (36.2) Missing 0 (0.0) 2 (0.4) 0 (0.0) Perforated viscus No 596 (85.9) 399 (88.7) 190 (71.7) <0.001 Yes 97 (14.0) 49 (10.9) 68 (25.7) Missing 1 (0.1) 2 (0.4) 7 (2.6) Prophylactic antibiotics No, not available 6 (0.9) 16 (3.6) 0 (0.0) 0.404* No, but available 90 (13.0) 55 (12.2) 36 (13.6) Yes 598 (86.2) 377 (83.8) 228 (86.0) Missing 0 (0.0) 2 (0.4) 1 (0.4) Whole blood/products No, but available in this hospital 661 (95.2) 385 (85.6) 201 (75.8) <0.001* No, not available in this hospital 8 (1.2) 7 (1.6) 1 (0.4) Yes, whole blood 2 (0.3) 30 (6.7) 54 (20.4) Yes, blood products 23 (3.3) 26 (5.8) 9 (3.4) Missing 0 (0.0) 2 (0.4) 0 (0.0) *χ2 test is for yes versus no.

.4) 1 (0.4) Whole blood/products No, but available in this hospital 661 (95.2) 385 (85.6) 201 (75.8) <0.001* No, not available in this hospital 8 (1.2) 7 (1.6) 1 (0.4) Yes, whole blood 2 (0.3) 30 (6.7) 54 (20.4) Yes, blood products 23 (3.3) 26 (5.8) 9 (3.4) Missing 0 (0.0) 2 (0.4) 0 (0.0) *χ2 test is for yes versus no. ASA, American Society of Anesthesiologists; HDI, Human Development Index. Figure 1 World map showing participating countries and number of enrolled patients. Demographics Children undergoing emergency abdominal surgery in low-HDI countries had higher American Society of Anaesthesiologists (ASA) grades than children in middle-HDI or high-HDI groups (table 1). Furthermore, the WHO surgical safety checklist was employed prior to surgery in less than half of children undergoing emergency abdominal surgery from the low-HDI and middle-HDI groups compared with over 90% in the high-HDI group. At operation, 214/1406 (15.2%) of the children were found to have a perforated viscus, and this varied with HDI group (high 97/694, 14.0%; middle 49/450, 10.9%; low 68/265, 25.7%). Use of laparoscopy was widespread in high-HDI nations (341/694, 49.1%), whereas in middle-HDI (30/450, 6.7%) and low-HDI (8/257, 3.0%) settings, rates were much lower (p<0.001).

children were found to have a perforated viscus, and this varied with HDI group (high 97/694, 14.0%; middle 49/450, 10.9%; low 68/265, 25.7%). Use of laparoscopy was widespread in high-HDI nations (341/694, 49.1%), whereas in middle-HDI (30/450, 6.7%) and low-HDI (8/257, 3.0%) settings, rates were much lower (p<0.001). Appendicitis was the most common indication for undergoing surgery across all groups, followed by congenital abnormalities, intussusception and hernia (figure 2A and online supplementary table S1). Emergency abdominal surgery for congenital abnormalities was significantly higher in low-HDI groups compared with middle-HDI and high-HDI groups (14.3% cf. 1.8% and 3.2%, respectively). 10.1136/bmjgh-2016-000091.supp1supplementary tables Figure 2 (A) Indications for emergency abdominal surgery in children across Human Developmental Index groups; (B) Surgical outcomes by Human Development Index group; (C) Adjusted 30-day mortality according to age groups. HDI, Human Developmental Index; SSI, surgical site infection.

10.1136/bmjgh-2016-000091.supp1supplementary tables Figure 2 (A) Indications for emergency abdominal surgery in children across Human Developmental Index groups; (B) Surgical outcomes by Human Development Index group; (C) Adjusted 30-day mortality according to age groups. HDI, Human Developmental Index; SSI, surgical site infection. Mortality Overall, 30-day mortality following surgery was 2.9% (n=41/1409) (figure 3). Of these deaths, 29.3% (n=12/41) occurred within 24 hours and 70.7% (n=29/41) between 24 hours and 30 days. Mortality varied significantly with HDI, with significantly higher proportions in low-HDI countries at 24 hours (0.3% in high-HDI, 0.7% in middle-HDI and 2.6% in low-HDI groups, p=0.005) and 30 days (0.9% in high-HDI, 2.9% in middle-HDI and 8.3% in low-HDI groups, p<0.001). Other associations with 24-hour and 30-day mortality in univariable analyses included neonatal age, >1 ASA grade and non-appendicitis procedures. Perforated viscus was significantly associated with 30-day mortality. An inversely proportional relationship is seen between 30-day mortality and age in all HDI groups even after adjustment in models (figure 2C). Figure 3 Patient complications and mortality profile according to Human Development Index. HDI, Human Developmental Index; SSI, surgical site infection.

Mortality Overall, 30-day mortality following surgery was 2.9% (n=41/1409) (figure 3). Of these deaths, 29.3% (n=12/41) occurred within 24 hours and 70.7% (n=29/41) between 24 hours and 30 days. Mortality varied significantly with HDI, with significantly higher proportions in low-HDI countries at 24 hours (0.3% in high-HDI, 0.7% in middle-HDI and 2.6% in low-HDI groups, p=0.005) and 30 days (0.9% in high-HDI, 2.9% in middle-HDI and 8.3% in low-HDI groups, p<0.001). Other associations with 24-hour and 30-day mortality in univariable analyses included neonatal age, >1 ASA grade and non-appendicitis procedures. Perforated viscus was significantly associated with 30-day mortality. An inversely proportional relationship is seen between 30-day mortality and age in all HDI groups even after adjustment in models (figure 2C). Figure 3 Patient complications and mortality profile according to Human Development Index. HDI, Human Developmental Index; SSI, surgical site infection. In multilevel models, the association between low-HDI country, and 24-hour (OR 7.08, 95% CI 1.39 to 36.10, p=0.018) (table 2) and 30-day mortality (OR 7.79, 95% CI 2.96 to 20.48, p<0.001) (table 3) persisted. Middle-HDI country was associated with a 30-day mortality (OR 5.57, 95% CI 1.90 to 16.39, p=0.002) but not 24-hour mortality. A perforated viscus was significantly associated with increased 30-day mortality, whereas appendicitis was associated with lower 24-hour and 30-day mortality compared with other indications. Table 2 Factors associated with 24-hour mortality

In multilevel models, the association between low-HDI country, and 24-hour (OR 7.08, 95% CI 1.39 to 36.10, p=0.018) (table 2) and 30-day mortality (OR 7.79, 95% CI 2.96 to 20.48, p<0.001) (table 3) persisted. Middle-HDI country was associated with a 30-day mortality (OR 5.57, 95% CI 1.90 to 16.39, p=0.002) but not 24-hour mortality. A perforated viscus was significantly associated with increased 30-day mortality, whereas appendicitis was associated with lower 24-hour and 30-day mortality compared with other indications. Table 2 Factors associated with 24-hour mortality Alive Died Univariate logistic regression OR (95% CI, p value) Multilevel logistic regression OR (95% CI, p value) HDI tertile High 692 (99.7) 2 (0.3) – – Middle 446 (99.3) 3 (0.7) 2.33 (0.38 to 17.72, p=0.356) 3.71 (0.56 to 24.56, p=0.174) Low 258 (97.4) 7 (2.6) 9.39 (2.25 to 63.28, p=0.005) 7.08 (1.39 to 36.10, p=0.018) Age Child (>2 years <16 years) 1104 (99.4) 7 (0.6) – – Infant (>1 month <2 years) 148 (99.3) 1 (0.7) 1.07 (0.06 to 6.05, p=0.953) 0.16 (0.02 to 1.45, p=0.102) Neonate (≤1 month) 143 (97.3) 4 (2.7) 4.41 (1.14 to 14.79, p=0.019) 0.74 (0.16 to 3.33, p=0.694) Gender Male 811 (99.5) 4 (0.5) – – Female 585 (98.7) 8 (1.3) 2.77 (0.87 to 10.43, p=0.097) 3.47 (0.99 to 12.22, p=0.053) ASA 1 975 (99.8) 2 (0.2) – – >1 421 (97.7) 10 (2.3) 11.58 (3.04 to 75.55, p=0.002) 5.22 (0.96 to 28.23, p=0.055) Perforated viscus No 1177 (99.3) 8 (0.7) – – Yes 209 (98.1) 4 (1.9) 2.82 (0.75 to 9.02, p=0.093) 1.57 (0.40 to 6.21, p=0.520) Primary operation Non-appendicectomy 475 (97.7) 11 (2.3) – – Appendicectomy 921 (99.9) 1 (0.1) 0.05 (0.00 to 0.24, p=0.003) 0.07 (0.01 to 0.59, p=0.015) n=1398, AIC=120.2, c-statistic=0.922, H and L GOF=χ2=3.438, df=8, p value=0.904.

) – – Yes 209 (98.1) 4 (1.9) 2.82 (0.75 to 9.02, p=0.093) 1.57 (0.40 to 6.21, p=0.520) Primary operation Non-appendicectomy 475 (97.7) 11 (2.3) – – Appendicectomy 921 (99.9) 1 (0.1) 0.05 (0.00 to 0.24, p=0.003) 0.07 (0.01 to 0.59, p=0.015) n=1398, AIC=120.2, c-statistic=0.922, H and L GOF=χ2=3.438, df=8, p value=0.904. AIC, Akaike information criterion; ASA, American Society of Anesthesiologists; df, degree of freedom; H and L GOF, Hosmer-Lemeshow Goodness of fit; HDI, Human Development Index. Table 3 Factors associated with 30-day mortality

) – – Yes 209 (98.1) 4 (1.9) 2.82 (0.75 to 9.02, p=0.093) 1.57 (0.40 to 6.21, p=0.520) Primary operation Non-appendicectomy 475 (97.7) 11 (2.3) – – Appendicectomy 921 (99.9) 1 (0.1) 0.05 (0.00 to 0.24, p=0.003) 0.07 (0.01 to 0.59, p=0.015) n=1398, AIC=120.2, c-statistic=0.922, H and L GOF=χ2=3.438, df=8, p value=0.904. AIC, Akaike information criterion; ASA, American Society of Anesthesiologists; df, degree of freedom; H and L GOF, Hosmer-Lemeshow Goodness of fit; HDI, Human Development Index. Table 3 Factors associated with 30-day mortality Alive Died Univariate logistic regression OR (95% CI, p value) Multilevel logistic regression OR (95% CI, p value) HDI tertile High 688 (99.1) 6 (0.9) – – Middle 436 (97.1) 13 (2.9) 3.42 (1.34 to 9.79, p=0.013) 5.57 (1.90 to 16.39, p=0.002) Low 243 (91.7) 22 (8.3) 10.38 (4.42 to 28.46, p<0.001) 7.79 (2.96 to 20.48, p<0.001) Age Child (>2 years <16 years) 1095 (98.6) 16 (1.4) – – Infant (>1 month<2 years) 140 (94.0) 9 (6.0) 4.40 (1.83 to 9.95, p=0.001) 0.91 (0.35 to 2.38, p=0.849) Neonate (≤1 month) 131 (89.1) 16 (10.9) 8.36 (4.06 to 17.22, p<0.001) 2.27 (0.92 to 5.62, p=0.075) Gender Male 794 (97.4) 21 (2.6) – – Female 573 (96.6) 20 (3.4) 1.32 (0.70 to 2.47, p=0.382) 1.98 (1.00 to 3.94, p=0.051) ASA 1 964 (98.7) 13 (1.3) – – >1 403 (93.5) 28 (6.5) 5.15 (2.69 to 10.37, p<0.001) 1.47 (0.67 to 3.25, p=0.337) Perforated viscus No 1157 (97.6) 28 (2.4) – – Yes 200 (93.9) 13 (6.1) 2.69 (1.33 to 5.17, p=0.004) 2.63 (1.21 to 5.73, p=0.015) Primary operation Non-appendicectomy 447 (92.0) 39 (8.0) – – Appendicectomy 920 (99.8) 2 (0.2) 0.02 (0.00 to 0.08, p<0.001) 0.04 (0.01 to 0.18, p<0.001) n=1398, AIC=282.7, c-statistic=0.902, H&L GOF=χ2=6.418, df=8, p value=0.601.

2.4) – – Yes 200 (93.9) 13 (6.1) 2.69 (1.33 to 5.17, p=0.004) 2.63 (1.21 to 5.73, p=0.015) Primary operation Non-appendicectomy 447 (92.0) 39 (8.0) – – Appendicectomy 920 (99.8) 2 (0.2) 0.02 (0.00 to 0.08, p<0.001) 0.04 (0.01 to 0.18, p<0.001) n=1398, AIC=282.7, c-statistic=0.902, H&L GOF=χ2=6.418, df=8, p value=0.601. AIC, Akaike information criterion; ASA, American Society of Anesthesiologists; df, degree of freedom; H and L, Hosmer-Lemeshow Goodness of fit; HDI, Human Development Index. An analysis of predicted excess deaths was performed using the final multilevel 30-day mortality model. Based on this model, if all children in low-HDI and middle-HDI countries were considered to have been in high-HDI countries but otherwise had the same characteristics, 29 lesser deaths are predicted (40 per 1000 procedures). Major complications and reintervention The overall rate of major complications following emergency abdominal surgery was 7.2% (n=102/1409) (figure 2B and online supplementary table S2). Major complications were significantly more common in low-HDI countries (11.3%, 30/265) compared with middle-HDI and high-HDI countries (6.4%, 29/450 and 6.2% 43/694, respectively, p=0.017). The rate of reintervention across the HDI groups mirrors these complications rates (low 6.8%, middle 4.4%, high 4.2%, p=0.222; online supplementary table S3).

ificantly more common in low-HDI countries (11.3%, 30/265) compared with middle-HDI and high-HDI countries (6.4%, 29/450 and 6.2% 43/694, respectively, p=0.017). The rate of reintervention across the HDI groups mirrors these complications rates (low 6.8%, middle 4.4%, high 4.2%, p=0.222; online supplementary table S3). Minor complications Across all HDI groups, the minor complication rate (Clavien-Dindo I-II) was 14.8% (n=208). This varied across HDI groups, with higher rates in low-HDI countries (20.9%) compared with middle-HDI and high-HDI countries (13.1% and 13.8% respectively, p=0.010), but these differences did not persist in multivariable analysis (see online supplementary table S4). Surgical site infection The overall SSI rate was 9.3% (n=131). This varied significantly across HDI groups (low 21.1%, middle 9.6%, high 4.6%, p<0.001, online supplementary table S5). Discussion The main findings of this study are sevenfold and fourfold higher 30-day mortalities in low-HDI and middle-HDI countries, respectively, compared with high-HDI countries. These rates are considerably greater than the threefold higher mortality previously reported among adult patients in low-HDI countries and account for an excess 40 deaths per thousand procedures in low-HDI and middle-HDI compared with high-HDI countries in this study alone.20 The risk factors for this excess mortality are necessarily multifactorial, including a higher intestinal perforation rate, which may reflect delayed access to surgery and different patterns of disease.

ss 40 deaths per thousand procedures in low-HDI and middle-HDI compared with high-HDI countries in this study alone.20 The risk factors for this excess mortality are necessarily multifactorial, including a higher intestinal perforation rate, which may reflect delayed access to surgery and different patterns of disease. The twofold higher rate of major and minor postoperative complications and the fivefold difference in SSIs are also noteworthy. Our study does not allow us to identify the main factors responsible for these differences, but other studies in the literature point out a variety of aetiological factors including sepsis, multiple exposure to anaesthesia in the neonatal period, postoperative bleeding, as well as complexity of congenital anomaly, delayed presentation, non-availability of trained personnel and financial constraints on the part of the caregivers.5–11 21 While the overall commonest surgical procedure in children remains appendicectomy, other complex procedures for congenital anomalies and intestinal obstruction are commonly performed in children in resource-limited settings. The similarity in procedures performed across resource settings was not expected, but it does demonstrate the depth of training required by surgical personnel to be able to handle such complex cases. Minimal access surgery was infrequently used in low-HDI and middle-HDI countries, showing inequality in access to contemporary technology through lack of resources including training in use of such technology.22

es demonstrate the depth of training required by surgical personnel to be able to handle such complex cases. Minimal access surgery was infrequently used in low-HDI and middle-HDI countries, showing inequality in access to contemporary technology through lack of resources including training in use of such technology.22 The study was able to draw from a large and diverse patient population, spanning wide geographical and resource areas globally. Despite the convenience sampling employed, it offers a snapshot of essential paediatric surgery across the globe. The main body of data from the study highlights the differences in pathology, patient premorbid status, operative findings and outcomes based on HDI grouping. The higher ASA status of children requiring emergency abdominal surgery in low-HDI and middle-HDI countries settings is concerning, and it potentially reflects delayed access to care with the consequent negative impact on postoperative outcomes. Similarly, the percentage of perforated viscus encountered at surgery was also significantly higher in low-HDI and middle-HDI countries. The delay in access to care has been previously reported by studies from LMICs.9–11 This may account for the poor survival of neonates with severe congenital anomalies in these settings, such as intestinal atresia, abdominal wall defects and oesophageal atresia.11 23 A study from Nigeria indicated that delayed intervention time >72 hours, neonatal age and severe postoperative complications are associated with higher mortality in paediatric surgical emergencies.21

congenital anomalies in these settings, such as intestinal atresia, abdominal wall defects and oesophageal atresia.11 23 A study from Nigeria indicated that delayed intervention time >72 hours, neonatal age and severe postoperative complications are associated with higher mortality in paediatric surgical emergencies.21 This study has some limitations. Being based on convenience sampling of hospitals, the data collected may not be truly representative of other sites which may be more poorly resourced. Collection bias, however, may result in the true outcomes being even worse in LMICs, as the lowest resource sites would be less likely to participate. In addition, other factors such as availability of personnel, availability of complex anaesthetic and intensive care support, and delay time before surgery were not analysed in this study but may significantly impact on postoperative mortality. The current study has documented differences in surgical outcomes in children based on HDI groups, but has not explored in depth the reasons for these differences. This will form the agenda for future studies, together with outcome studies, focusing on elective essential surgical procedures in children.

tive mortality. The current study has documented differences in surgical outcomes in children based on HDI groups, but has not explored in depth the reasons for these differences. This will form the agenda for future studies, together with outcome studies, focusing on elective essential surgical procedures in children. The main conclusion of this study is that emergency abdominal surgery in children is associated with significantly worse outcomes in LMICs. The documentation provided by this study is essential to the process of scaling up surgical services for children in low-resource settings. Good surgical outcomes require a multitude of factors, including trained personnel, good facilities and surgical supplies, as well as prompt access to surgical care. Thus, any single intervention in this multifaceted system has a high likelihood of failing to fully address these complex issues. This relates to many well-meaning efforts from high-income countries (HICs) to assist surgically in resource-limited settings. For instance, temporary platforms in the form of ‘surgical safaris’, the provision of surgical equipment alone, or short-term training courses outside one's normal work setting will likely have little long-term impact.24 25 The likeliest context in which broad systematic change can occur is likely that of a long-lasting institutional partnership. In such a context of relationship with mutual understanding and trust, appropriate change can be implemented in whichever areas are most needed, and progress can be monitored and evaluated.26

25 The likeliest context in which broad systematic change can occur is likely that of a long-lasting institutional partnership. In such a context of relationship with mutual understanding and trust, appropriate change can be implemented in whichever areas are most needed, and progress can be monitored and evaluated.26 The recent global recognition of surgery as an essential healthcare component has provided a unique impetus for provision of essential surgical services, especially in LMICs.27 28 The task ahead is a huge one, in terms of access to and quality of care. The current study has documented relatively poor outcomes of emergency abdominal surgery in children in low-HDI and middle-HDI countries. Such data are essential in guiding efforts to improve the surgical care of children globally and prioritise it in the global health agenda.

in terms of access to and quality of care. The current study has documented relatively poor outcomes of emergency abdominal surgery in children in low-HDI and middle-HDI countries. Such data are essential in guiding efforts to improve the surgical care of children globally and prioritise it in the global health agenda. The authors would like to acknowledge Jacky Hong Chieh Chen, Lawani Ismail, Dylan Roi, Eugenio Grasset Escobar for protocol translation. Organisations assisting in dissemination (alphabetical) are as follows: Asian Medical Students' Association (AMSA), Association of Surgeons in Training (ASiT), College of Surgeons of East, Central and Southern Africa (COSECSA), Cutting Edge Manipal, Egyptian Medical Student Research Association (EMRA), International Collaboration For Essential Surgery (ICES), International Federation of Medical Student Associations (IFMSA), Lifebox Foundation, School of Surgery, Student Audit and Research in Surgery (STARSurg), The Electives Network, UK National Research Collaborative, World Korean Medical Students Association (WKMSA), World Society of Emergency Surgery (WSES), World Surgical Association (WSA). Individuals assisting in dissemination (alphabetical) are as follows: Douglas Bowley, Vimal Gokani, Jaymie Ang Henry, Chia Kong, Chris Lavy, Jane Lim, Laura Luque, Mahiben Maruthappu, Praveen Mogan, Dmitri Nepogodiev, Raza Sayyed, Joseph Shalhoub, Ravi Vohra. Handling editor: Seye Abimbola Twitter: Follow GlobalSurg at @GlobalSurg

The authors would like to acknowledge Jacky Hong Chieh Chen, Lawani Ismail, Dylan Roi, Eugenio Grasset Escobar for protocol translation. Organisations assisting in dissemination (alphabetical) are as follows: Asian Medical Students' Association (AMSA), Association of Surgeons in Training (ASiT), College of Surgeons of East, Central and Southern Africa (COSECSA), Cutting Edge Manipal, Egyptian Medical Student Research Association (EMRA), International Collaboration For Essential Surgery (ICES), International Federation of Medical Student Associations (IFMSA), Lifebox Foundation, School of Surgery, Student Audit and Research in Surgery (STARSurg), The Electives Network, UK National Research Collaborative, World Korean Medical Students Association (WKMSA), World Society of Emergency Surgery (WSES), World Surgical Association (WSA). Individuals assisting in dissemination (alphabetical) are as follows: Douglas Bowley, Vimal Gokani, Jaymie Ang Henry, Chia Kong, Chris Lavy, Jane Lim, Laura Luque, Mahiben Maruthappu, Praveen Mogan, Dmitri Nepogodiev, Raza Sayyed, Joseph Shalhoub, Ravi Vohra. Handling editor: Seye Abimbola Twitter: Follow GlobalSurg at @GlobalSurg Collaborators: Writing group consisted of Adesoji O Ademuyiwa, Alexis P Arnaud, Thomas M Drake, J Edward F Fitzgerald, Dan Poenaru, Aneel Bhangu, Ewen M Harrison (Guarantor), on behalf of the GlobalSurg Collaborative. Steering group (alphabetical) consisted of Adesoji O Ademuyiwa, Aneel Bhangu, Thomas M Drake, J Edward F Fitzgerald, Stuart Fergusson, James C Glasbey, Ewen M Harrison, Chetan Khatri, Midhun Mohan, Dmitri Nepogodiev, Kjetil Søreide.

, Aneel Bhangu, Ewen M Harrison (Guarantor), on behalf of the GlobalSurg Collaborative. Steering group (alphabetical) consisted of Adesoji O Ademuyiwa, Aneel Bhangu, Thomas M Drake, J Edward F Fitzgerald, Stuart Fergusson, James C Glasbey, Ewen M Harrison, Chetan Khatri, Midhun Mohan, Dmitri Nepogodiev, Kjetil Søreide. Statistical analysis was carried out by Thomas M Drake, Ewen M Harrison. National leads were involved in recruitment of multiple centres (in some cases all centres) from the countries listed. Chetan Khatri (Lead Coordinator for GlobalSurg), Neel Gobin (Australia), Ana Vega Freitas (Brazil), Nigel Hall (Canada), Sung-Hee Kim (Hong Kong, China), Ahmed Negeida, Hosni Khairy (Egypt), Zahra Jaffry, Stephen J Chapman (England), Alexis P Arnaud (France), Stephen Tabiri (Ghana), Gustavo Recinos (Guatemala), Midhun Mohan (India), Radhian Amandito (Indonesia), Marwan Shawki (Iraq), Michael Hanrahan (Ireland), Francesco Pata (Italy), Justas Zilinskas (Lithuania), April Camilla Roslani, Cheng Chun Goh (Malaysia), Adesoji O Ademuyiwa (Nigeria), Gareth Irwin (Northern Ireland), Sebastian Shu, Laura Luque (Peru), Hunain Shiwani, Afnan Altamimi, Mohammed Ubaid Alsaggaf (Saudi Arabia), Stuart Fergusson (Scotland), Richard Spence, Sarah Rayne (South Africa), Jenifa Jeyakumar (Sri Lanka), Yucel Cengiz (Sweden), Dmitri A Raptis (Switzerland), James C Glasbey (Wales). Patient enrolment and data collection Argentina: Claudio Fermani, Ruben Balmaceda, Maria Marta Modolo (Hospital Luis Lagomaggiore);

National leads were involved in recruitment of multiple centres (in some cases all centres) from the countries listed. Chetan Khatri (Lead Coordinator for GlobalSurg), Neel Gobin (Australia), Ana Vega Freitas (Brazil), Nigel Hall (Canada), Sung-Hee Kim (Hong Kong, China), Ahmed Negeida, Hosni Khairy (Egypt), Zahra Jaffry, Stephen J Chapman (England), Alexis P Arnaud (France), Stephen Tabiri (Ghana), Gustavo Recinos (Guatemala), Midhun Mohan (India), Radhian Amandito (Indonesia), Marwan Shawki (Iraq), Michael Hanrahan (Ireland), Francesco Pata (Italy), Justas Zilinskas (Lithuania), April Camilla Roslani, Cheng Chun Goh (Malaysia), Adesoji O Ademuyiwa (Nigeria), Gareth Irwin (Northern Ireland), Sebastian Shu, Laura Luque (Peru), Hunain Shiwani, Afnan Altamimi, Mohammed Ubaid Alsaggaf (Saudi Arabia), Stuart Fergusson (Scotland), Richard Spence, Sarah Rayne (South Africa), Jenifa Jeyakumar (Sri Lanka), Yucel Cengiz (Sweden), Dmitri A Raptis (Switzerland), James C Glasbey (Wales). Patient enrolment and data collection Argentina: Claudio Fermani, Ruben Balmaceda, Maria Marta Modolo (Hospital Luis Lagomaggiore); Australia: Ewan Macdermid, Neel Gobin, Roxanne Chenn, Cheryl Ou Yong, Michael Edye (Blacktown Hospital), Martin Jarmin, Scott K D’amours, Dushyant Iyer (Liverpool Hospital, The University Of New South Wales), Daniel Youssef, Nicholas Phillips, Jason Brown (Royal Brisbane & Women’s Hospital), Isaac Hanley (The Tweed Hospital), Marilla Dickfos (Toowoomba Hospital); Bangladesh: Ashrarur Rahman Mitul, Khalid Mahmud (Dhaka Shishu (Children) Hospital), Antje Oosterkamp (Lamb Hospital);

Australia: Ewan Macdermid, Neel Gobin, Roxanne Chenn, Cheryl Ou Yong, Michael Edye (Blacktown Hospital), Martin Jarmin, Scott K D’amours, Dushyant Iyer (Liverpool Hospital, The University Of New South Wales), Daniel Youssef, Nicholas Phillips, Jason Brown (Royal Brisbane & Women’s Hospital), Isaac Hanley (The Tweed Hospital), Marilla Dickfos (Toowoomba Hospital); Bangladesh: Ashrarur Rahman Mitul, Khalid Mahmud (Dhaka Shishu (Children) Hospital), Antje Oosterkamp (Lamb Hospital); Benin: Pamphile A Assouto, Ismail Lawani, Yacoubou Imorou Souaibou (Centre National Hospitalier Et Universitaire Hubert Koutoukou Maga); Brunei: Giridhar H Devadasar, Chean Leung Chong, Muhammad Rashid Minhas Qadir, (Ssb Hospital), Kyaw Phyo Aung, Lee Shi Yeo, Chean Leung Chong (RIPAS Hospital); Brazil: Vanessa Dina Palomino Castillo, Monique Moron Munhoz, Gisele Moreira (Conjunto Hospitalar De Sorocaba), Luiz Carlos Barros De Castro Segundo, Salim Anderson Khouri Ferreira, Maíra Cassa Careta (Hospital Da Santa Casa De Misericórdia De Vitória), Rafael Araujo, Juliana Menegussi, Marisa Leal, Caio Vinícius Barroso de Lima, Luiza Sarmento Tatagiba, Antônio Leal (Hospital Infantil Nossa Senhora Da Gloria); Cameroon: Samuel Nigo, Juana Kabba, Tagang Ebogo Ngwa, James Brown (Mbingo Baptist Hospital);

Brazil: Vanessa Dina Palomino Castillo, Monique Moron Munhoz, Gisele Moreira (Conjunto Hospitalar De Sorocaba), Luiz Carlos Barros De Castro Segundo, Salim Anderson Khouri Ferreira, Maíra Cassa Careta (Hospital Da Santa Casa De Misericórdia De Vitória), Rafael Araujo, Juliana Menegussi, Marisa Leal, Caio Vinícius Barroso de Lima, Luiza Sarmento Tatagiba, Antônio Leal (Hospital Infantil Nossa Senhora Da Gloria); Cameroon: Samuel Nigo, Juana Kabba, Tagang Ebogo Ngwa, James Brown (Mbingo Baptist Hospital); Canada: Sebastian King, Augusto Zani, Georges Azzie, Mohammed Firdouse, Sameer Kushwaha, Arnav Agarwal (The Hospital For Sick Children, Toronto), Karen Bailey, Brian Cameron, Michael Livingston (McMaster Children’s Hospital), Alexandre Horobjowsky, Dan L Deckelbaum, Tarek Razek (Centre for Global Surgery, McGill University Health Centre); Colombia: Irene Montes, Sebastian Sierra, Manuela Mendez (Clinica CES), Maria Isabel Villegas, Maria Clara Mendoza Arango, Ivan Mendoza, (Clínica Las Vegas), Fred Alexander Naranjo Aristizã ¡bal, Jaime Andres Montoya Botero, Victor Manuel Quintero Riaza (El Hospital Pablo Tobón Uribe), Jakeline Restrepo, Carlos Morales, Maria Clara Mendoza Arango (Hospital Universitario San Vicente Fundación), Herman Cruz, Alejandro Munera, Maria Clara Mendoza Arango (Ips Universitaria Clínica León Xiii); Croatia: Robert Karlo, Edgar Domini, Jakov Mihanovic (Zadar General Hospital), Mihael Radic, Kresimir Zamarin, Nikica Pezelj (General Hospital Sibenik);

Colombia: Irene Montes, Sebastian Sierra, Manuela Mendez (Clinica CES), Maria Isabel Villegas, Maria Clara Mendoza Arango, Ivan Mendoza, (Clínica Las Vegas), Fred Alexander Naranjo Aristizã ¡bal, Jaime Andres Montoya Botero, Victor Manuel Quintero Riaza (El Hospital Pablo Tobón Uribe), Jakeline Restrepo, Carlos Morales, Maria Clara Mendoza Arango (Hospital Universitario San Vicente Fundación), Herman Cruz, Alejandro Munera, Maria Clara Mendoza Arango (Ips Universitaria Clínica León Xiii); Croatia: Robert Karlo, Edgar Domini, Jakov Mihanovic (Zadar General Hospital), Mihael Radic, Kresimir Zamarin, Nikica Pezelj (General Hospital Sibenik); Egypt: Ahmed Khyrallh, Ahamed Hassan, Gamal Shimy, Mohamed A Baky Fahmy (Al-Azher University Hospital); Ayman Nabawi, Muhammad Saad Ali Muhammad Gohar, Mohamed Elfil, Mohamed Ghoneem, Muhammad El-Saied Ahmad Muhammad Gohar, Mohamed Asal, Mostafa Abdelkader, Mahmoud Gomah, Hayssam Rashwan, Mohamed Karkeet, Ahmed Gomaa (Alexandria Main University Hospital); Amr Hasan, Ahmed Elgebaly, Omar Saleh, Ahmad Abdel Fattah, Abdullah Gouda, Abd Elrahman Elshafay, Abdalla Gharib, Mohammed Hanafy, Abdullah Al-Mallah, Mahmoud Abdulgawad, Mohamad Baheeg, Mohammed Alhendy, Ibrahim Abdel Fattah, Abdalla Kenibar, Omar Osman, Mostafa Gemeah, Ahmed Mohammed, Abdalrahman Adel, Ahmed Maher Menshawy Mesreb, Abdelrahman Mohammed, Abdelrahman Sayed, Mohamed Abozaid (Al-Hussein Hospital); Ahmed Hafez El-Badri Kotb, Ali Amin Ahmed Ata, Mohammed Nasr, Abdelrahman Alkammash, Mohammed Saeed, Nader Abd El Hamid, Attia Mohamed Attia, Ahmed Abd El Galeel, Eslam Elbanby, Khalid Salah El-Dien, Usama Hantour, Omar Alahmady, Billal Mansour, Amr Muhammad Elkorashy (Bab El-Sharia Hospital); Emad Mohamed Saeed Taha, Kholod Tarek Lasheen, Salma Said Elkolaly, Nehal Yosri Elsayed Abdel-Wahab, Mahmoud Ahmed Fathi Abozyed, Ahmed Adel, Ahmed Moustafa Saeed, Gehad Samir El Sayed, Jehad Hassan Youssif (Banha University Hospital); Soliman Magdy Ahmed, Nermeen Soubhy El-Shahat, Abd El-Rahman Hegazy Khedr (Belbeis Central Hospital); Abdelrhman Osama Elsebaaye, Mohamed Elzayat, Mohamed Abdelraheim, Ibrahim Elzayat, Mahmoud Warda, Khaled Naser El Deen, Abdelrhman Essam, Omar Salah, Mohamed Abbas, Mona Rashad, Ibrahim Elzayyat, Dalia Hemeda, Gehad Tawfik, Mai Salama, Hazem Khaled, Mohamed Seisa, (El Dawly Hospital—Mansoura); Kareem Elshaer, Abdelfatah Hussein, Mahmoud Elkhadrawi (El Mahalla General Hospital); Ahmed Mohamed Afifi, Osama Saadeldeen Ebrahim, Mahmoud Mohamed Metwally (El Mataria Educational Hospital); Rowida Elmelegy, Diaa Moustafa Elbendary Elsawahly, Hisham Safa, Eman Nofal, Mohamed Elbermawy, Metwally Abo Raya, Ahmed Abdelmotaleb Ghazy, Hisham Samih, Asmaa Abdelgelil, Sarah Abdelghany, Ahmed El Kholy, Fatma Elkady, Mahmoud Salma, Sarah Samy, Reem Fakher, Aya Aboarab, Ahmed Samir, Ahmed Sakr, Abdelra

al Hospital); Rowida Elmelegy, Diaa Moustafa Elbendary Elsawahly, Hisham Safa, Eman Nofal, Mohamed Elbermawy, Metwally Abo Raya, Ahmed Abdelmotaleb Ghazy, Hisham Samih, Asmaa Abdelgelil, Sarah Abdelghany, Ahmed El Kholy, Fatma Elkady, Mahmoud Salma, Sarah Samy, Reem Fakher, Aya Aboarab, Ahmed Samir, Ahmed Sakr, Abdelra hman Haroun, Asmaa Abdel-Rahman Al-Aarag, Ahmed Elkholy, Sally Elshanwany (El-Menshawy General Hospital); Esraa Ghanem (El-Shohadaa Central Hospital); Ahmed Tammam, Ali Mohamed Hammad, Yousra El Shoura, Gehad El Ashal, Hosni Khairy (Kasr Al-Ainy School Of Medicine); Sarah Antar, Sara Mehrez, Mahmoud Abdelshafy, Maha Gamal Mohamad Hamad, Mona Hosh, Emad Abdallah, Basma Magdy, Thuraya Alzayat, Elsayed Gamaly, Hossam Elfeki, Amany Abouzahra, Shereen Elsheikh, Fatimah I Elgendy (Mansoura University Hospitals); Fathia Abd El-Salam, Osama Seifelnasr, Mohamed Ammar, Athar Eysa, Aliaa Sadek, Aliaa Gamal Toeema, Aly Nasr, Mohamed Abuseif, Hagar Zidan, Sara Abd Elmageed Barakat, Nadin Elsayed, Yasmin Abd Elrasoul, Ahmed El-Kelany, Mohamed Sabry Ammar, Mennat-Allah Mustafa, Yasmin Makhlouf, Mohamed Etman, Samar Saad, Mahmoud Alrahawy, Ahmed Raslan, Mahmoud Morsi, Ahmed Sabry, Hager Elwakil, Heba Shaker, Hagar Zidan, Ahmed Elkelany (Menoufia University Hospitals); Hussein El-Kashef, Mohamed Shaalan, Areej Tarek (Minia University Hospital); Ayman Elwan, Ahmed Ragab Nayel, Mostafa Seif, Doaa Emadeldin Shafik, Mohamed Ali Ghoname, Ahmad Almallah, Ahmed Fouad, Ayman Elwan, Eman Adel Sayma (New Damietta University Hospital); Ahmad Elbatahgy, Angham Solaiman El-Ma’doul, Ahmed Mosad, Hager Tolba, Diaa Eldin Abdelazeem Amin Elsorogy, Hassan Ali Mostafa, Amira Atef Omar, Ola Sherief Abd El Hameed, Ahmed Lasheen (Quweisna Central Hospital, Quweisna); Yasser Abd El Salam, Ashraf Morsi, Mohammed Ismail (Ras El Tin General Hospital); Hager Ahmed, Mohamed A Amer, Ahmed Elkelany, Ahmed Sabry El-Hamouly, Noura Attallah, Omnia Mosalum, Ahmed Afandy, Ahmed Mokhtar, Alaa Abouelnasr, Sara Ayad, Ramdan Shaker, Rokia Sakr, Mahmoud Amreia, Soaad Elsobky, Mohamed Mustafa, Ahmed Abo El Magd, Abeer Marey, Amr Tarek, Mohamed Fadel (Shebin El Kom Teaching Hospital, Menoufia); Mohamed Moamen Mohamed, Amr Fadel, Emad Ali Ahmed (Sohag University Hospital); Ahmad Ali, Mohammad Ghassan Alwafai, Ehab Abdulkader Hemida Ghazy Alnawam, Abdullah Dwydar, Sara Kharsa, Ehab Mamdouh, Hatem El-Sheemy, Ibrahim Alyoussef, Abouelatta Khairy Aly, Ahmad Aldalaq, Ehab Alnawam, Dalia Alkhabbaz (Souad Kafafi University Hosp

men Mohamed, Amr Fadel, Emad Ali Ahmed (Sohag University Hospital); Ahmad Ali, Mohammad Ghassan Alwafai, Ehab Abdulkader Hemida Ghazy Alnawam, Abdullah Dwydar, Sara Kharsa, Ehab Mamdouh, Hatem El-Sheemy, Ibrahim Alyoussef, Abouelatta Khairy Aly, Ahmad Aldalaq, Ehab Alnawam, Dalia Alkhabbaz (Souad Kafafi University Hosp ital); Mahmoud Saad, Shady Hussein, Ahmed Abo Elazayem, Ahmed Meshref, Marwa Elashmawy, Mohammed Mousa, Ahmad Nashaat, Sara Ghanem, Zaynab M Elsayed, Aya Elwaey, Iman Elkadsh (Suez Canal University Hospitals); Mariam Darweesh, Ahmed Mohameden, Mennaallah Hafez (Suez General Hospital); Ahmed Badr, Assmaa Badwy, Mohamed Abd El Slam (Talla Q7 Central Hospital); Mohamed Elazoul, Safwat Al-Nahrawi, Lotfy Eldamaty, Fathee Nada, Mohamed Ameen, Aya Hagar, Mohamed Elsehimy, Mohammad Abo-ryia, Hossam Dawoud, Shorouk El Mesery, Abeer El Gendy, Ahmed Abdelkareem, Ahmed Safwan Marey, Mostafa Allam, Sherif Shehata, Khaled Abozeid, Marwa Elshobary, Ahmed Fahiem, Sameh Sarsik, Amel Hashish, Mohamed Zidan, Mohamed Hashish, Shaimaa Aql, Abdelaziz Osman Abdelaziz Elhendawy (Tanta University Hospital); Mohamed Husseini, Omar Khater, Esraa Abdalmageed Kasem, Ahmed Gheith, Yasmin Elfouly, Ahmed Ragab Soliman, Yasmein Hani, Nesma Elfouly, Ahmed Fawzy, Ahmed Hassan, Mohammad Rashid, Abdallah Salah Elsherbiny, Basem Sieda, Nermin Mohamed Badwi, Mohammed Mustafa Hassan Mohammed, Osama Mohamed, Mohammad Abdulkhalek Habeeb (Zagazig University Hospitals);

Esraa Abdalmageed Kasem, Ahmed Gheith, Yasmin Elfouly, Ahmed Ragab Soliman, Yasmein Hani, Nesma Elfouly, Ahmed Fawzy, Ahmed Hassan, Mohammad Rashid, Abdallah Salah Elsherbiny, Basem Sieda, Nermin Mohamed Badwi, Mohammed Mustafa Hassan Mohammed, Osama Mohamed, Mohammad Abdulkhalek Habeeb (Zagazig University Hospitals); Ethiopia: Mengistu Worku, Nichole Starr (Dessie Referral Hospital), Semay Desta, Sahlu Wondimu, Nebyou Seyoum Abebe (Menelik IiHospital), Efeson Thomas, Frehun Ayele Asele, Daniel Dabessa (Myungsung Christian Medical Center), Nebiyou Seyoum Abebe, Abebe Bekele Zerihun (Tikur Anbessa Hospital); France: Aurelien Scalabre, Fernanda Frade, Sabine Irtan (Trousseau Hospital, Sorbonnes Universités, UPMC Univ Paris), Valentine Parent, Amandine Martin, Alexis P Arnaud, Vivien Graffeille, Elodie Gaignard, Quentin Alimi (Rennes University Hospital), Olivier Abbo, Sofia Mouttalib, Ourdia Bouali (Hôpital des Enfants, Toulouse), Erik Hervieux, Yves Aigrain, Nathalie Botto (Hôpital Necker-Enfants Malades, Paris), Alice Faure, Lucile Fievet, Nicoleta Panait (Hôpital Nord, Marseille), Emilie Eyssartier, Francoise Schmitt, Guillaume Podevin (Pediatric Surgery Department, University Hospital, Angers), Cecile Muller, Arnaud Bonnard, Matthieu Peycelon (Robert Debré Children University Hospital);

ôpital Necker-Enfants Malades, Paris), Alice Faure, Lucile Fievet, Nicoleta Panait (Hôpital Nord, Marseille), Emilie Eyssartier, Francoise Schmitt, Guillaume Podevin (Pediatric Surgery Department, University Hospital, Angers), Cecile Muller, Arnaud Bonnard, Matthieu Peycelon (Robert Debré Children University Hospital); Ghana: Francis Abantanga, Kwaku Boakye-Yiadom, Mohammed Bukari (Komfo Anokye Teaching Hospital), Frank Owusu (Offinso District Hospital), Joseph Awuku-Asabre, Stephen Tabiri, Lemuel Davies Bray (University For Development Studies, School Of Medicine And Health Sciences, General Surgery Department, Tamale Teaching Hospital); Greece: Dimitrios Lytras, Kyriakos Psarianos, Anastasia Bamicha (Achillopoyleio General Hospital Of Volos), Christos Anthoulakis, Nikolaos Nikoloudis, Nikolaos Mitroudis (Serres General Hospital); Guatemala: Gustavo Recinos, Sergio Estupinian, Walter Forno (Hospital De Accidentes Ceibal), Romeo Guevara, Maria Aguilera, Napoleon Mendez, Cesar Augusto Azmitia Mendizabal, Pablo Ramazzini, Mario Contreras Urquizu (Hospital General San Juan De Dios), Daniel Estuardo Marroquín Rodríguez, Carlos Iván Pérez Velásquez, Sara María Contreras Mérida (Hospital Regional de Retalhuleu), Francisco Regalado, Mario Lopez, Miguel Siguantay (Hospital Roosevelt, Guatemala);

ar Augusto Azmitia Mendizabal, Pablo Ramazzini, Mario Contreras Urquizu (Hospital General San Juan De Dios), Daniel Estuardo Marroquín Rodríguez, Carlos Iván Pérez Velásquez, Sara María Contreras Mérida (Hospital Regional de Retalhuleu), Francisco Regalado, Mario Lopez, Miguel Siguantay (Hospital Roosevelt, Guatemala); India: SS Prasad, Anand Kirishnan, Nidhi Gyanchandani (KMC Hospital), Sriram Bhat, Anjana Sreedharan, S.V. Kinnera (Kasturba Medical College), Shravan Nadkarni, Harish Neelamraju Lakshmi, Puneet Malik (Sawai Man Singh Medical College & Hospitals, Jaipur, Rajasthan), Abid Bin Mahamood (Travancore Medical College Hospital), Monty Khajanchi, Savni Satoskar, Rajeev Satoskar (Seth Gordhandas Sunderdas Medical College And King Edward Memorial Hospital), Yella Reddy, Caranj Venugopal, Sunil Kumar (PES Institute Of Medical Sciences and Research); Indonesia: Eldaa Prisca Refianti Sutanto, Daniel Ardian Soeselo, Chintya Tedjaatmadja (Atmajaya Hospital), Fitriana Nur Rahmawati, Radhian Amandito, Maria Mayasari (Dr Cipto Mangunkusumo General Hospital, Jakarta); Iraq: Ruqaya Kadhim Mohammed Jawad Al-Hasani, Hasan Ismael Ibraheem Al-Hameedi, Israa Abdullah Aziz Al-Azraqi (Al Sader Medical City), Lubna Sabeeh, Rahma Kamil, Marwan Shawki (Baghdad Medical City);

India: SS Prasad, Anand Kirishnan, Nidhi Gyanchandani (KMC Hospital), Sriram Bhat, Anjana Sreedharan, S.V. Kinnera (Kasturba Medical College), Shravan Nadkarni, Harish Neelamraju Lakshmi, Puneet Malik (Sawai Man Singh Medical College & Hospitals, Jaipur, Rajasthan), Abid Bin Mahamood (Travancore Medical College Hospital), Monty Khajanchi, Savni Satoskar, Rajeev Satoskar (Seth Gordhandas Sunderdas Medical College And King Edward Memorial Hospital), Yella Reddy, Caranj Venugopal, Sunil Kumar (PES Institute Of Medical Sciences and Research); Indonesia: Eldaa Prisca Refianti Sutanto, Daniel Ardian Soeselo, Chintya Tedjaatmadja (Atmajaya Hospital), Fitriana Nur Rahmawati, Radhian Amandito, Maria Mayasari (Dr Cipto Mangunkusumo General Hospital, Jakarta); Iraq: Ruqaya Kadhim Mohammed Jawad Al-Hasani, Hasan Ismael Ibraheem Al-Hameedi, Israa Abdullah Aziz Al-Azraqi (Al Sader Medical City), Lubna Sabeeh, Rahma Kamil, Marwan Shawki (Baghdad Medical City); Ireland: Amoudtha Rasendran, Jacqueline Sheehan, Robert Kerley, Caoimhe Normile, Richard William Gilbert, Jiheon Song, Linnea Mauro, Mohammed Osman Dablouk, Michael Hanrahan, Paul Kielty, Eleanor Marks (Cork University Hospital), Simon Gosling, Michelle Mccarthy, Amoudtha Rasendran (Cork University Hospital and University College Cork), Diya Mirghani, Syed Altaf Naqvi, Chee Siong Wong (Limerick University Hospital), Simon George Gosling, Michelle Mccarthy, Amoudtha Rasendran, Ciara Fahy, Jiheon Song, Michael Hanrahan, Diana Duarte Cadogan, Anna Powell, Richard Gilbert, Caroline Clifford, Caoimhe Normile, Aoife Driscoll (Mercy University Hospital), Stassen Paul, Chris Lee, Ross Bowe (Midlands Regional Hospital Mullingar), William Hutch, Michael Hanrahan (University College Cork), Helen Mohan, Maeve O’Neill, Kenneth Mealy (Wexford General Hospital);

dogan, Anna Powell, Richard Gilbert, Caroline Clifford, Caoimhe Normile, Aoife Driscoll (Mercy University Hospital), Stassen Paul, Chris Lee, Ross Bowe (Midlands Regional Hospital Mullingar), William Hutch, Michael Hanrahan (University College Cork), Helen Mohan, Maeve O’Neill, Kenneth Mealy (Wexford General Hospital); Italy: Piergiorgio Danelli, Andrea Bondurri, Anna Maffioli (Azienda Ospedaliera Luigi Sacco—Polo Universitario), Luigi Bonavina, Yuri Macchitella, Chiara Ceriani (University of Milan, IRCCS Policlinico San Donato), Ezio Veronese, Luca Bortolasi, Alireza Hasheminia (San Bonifacio Hospital), Francesco Pata, Angelo Benevento, Gaetano Tessera (Sant’Antonio Abate Hospital, Gallarate), Luca Turati, Giovanni Sgroi, Emanuele Rausa (Treviglio Hospital); Lithuania: Donatas Venskutonis, Saulius Bradulskis, Linas Urbanavicius, Aiste Austraite, Romualdas Riauka, Justas Zilinskas, Zilvinas Dambrauskas (Lithuanian University Of Health Sciences); Malawi: Ross Coomber, Kenneth Johnson, Jennifer Nowers (Queen Elizabeth Hospital); Malaysia: Dineshwary Periasammy, Afizah Salleh, Andre Das (Hospital Kajang), Reuben Goh Ern Tze, Milaksh Nirumal Kumar, Nik Azim Nik Abdullah (Sarawak General Hospital), Hoong Yin Chong, April Camilla Roslani, Cheng Chun Goh (University Malaya Medical Centre); Malta: Marija Agius, Elaine Borg, Maureen Bezzina, Roberta Bugeja, Martinique Vella-Baldacchino, Andrew Spina, Josephine Psaila (Mater Dei Hospital, Malta);

Malaysia: Dineshwary Periasammy, Afizah Salleh, Andre Das (Hospital Kajang), Reuben Goh Ern Tze, Milaksh Nirumal Kumar, Nik Azim Nik Abdullah (Sarawak General Hospital), Hoong Yin Chong, April Camilla Roslani, Cheng Chun Goh (University Malaya Medical Centre); Malta: Marija Agius, Elaine Borg, Maureen Bezzina, Roberta Bugeja, Martinique Vella-Baldacchino, Andrew Spina, Josephine Psaila (Mater Dei Hospital, Malta); Martinique: Helene Francois-Coridon, Cecilia Tolg, Jean-Francois Colombani (Department of Pediatric Surgery, Mother and Children’s Hospital, University Hospital Of Martinique); Mozambique: Mário Jacobe, Domingos Mapasse, Elizabeth Snyder (Hospital Central Maputo); New Zealand: Ramadan Oumer, Mohammed Osman (Whangarei Hospital, Northland District Health Board);

Martinique: Helene Francois-Coridon, Cecilia Tolg, Jean-Francois Colombani (Department of Pediatric Surgery, Mother and Children’s Hospital, University Hospital Of Martinique); Mozambique: Mário Jacobe, Domingos Mapasse, Elizabeth Snyder (Hospital Central Maputo); New Zealand: Ramadan Oumer, Mohammed Osman (Whangarei Hospital, Northland District Health Board); Nigeria: Aminu Mohammad, Lofty-John Anyanwu, Abdulrahman Sheshe (Aminu Kano Teaching Hospital), Alaba Adesina, Olubukola Faturoti, Ogechukwu Taiwo (Babcock University Teaching Hospital), Muhammad Habib Ibrahim, Abdulrasheed A Nasir, Siyaka Itopa Suleiman (Federal Medical Centre, Birnin Kebbi), Adewale Adeniyi, Opeoluwa Adesanya, Ademola Adebanjo (Federal Medical Centre), Roland Osuoji, Kazeem Atobatele, Ayokunle Ogunyemi, Omolara Wiiliams, Mobolaji Oludara, Olabode Oshodi (Lagos State University Teaching Hospital), Adesoji O Ademuyiwa, Abdul Razzaq, Oluwagbemiga Lawal, Felix Alakaloko, Olumide Elebute, Adedapo Osinowo, Christopher Bode (Lagos University Teaching Hospital), Abidemi Adesuyi (National Hospital, Abuja), Adesoji Tade, Adeleke Adekoya, Collins Nwokoro (Olabisi Onabanjo University Teaching Hospital), Omobolaji O Ayandipo, Taiwo Akeem Lawal, Akinlabi E Ajao (University College Hospital), Samuel Sani Ali, Babatunde Odeyemi, Samson Olori (University of Abuja Teaching Hospital), Ademola Popoola, Ademola Adeyeye, James Adeniran (University of Ilorin Teaching Hospital);

ro (Olabisi Onabanjo University Teaching Hospital), Omobolaji O Ayandipo, Taiwo Akeem Lawal, Akinlabi E Ajao (University College Hospital), Samuel Sani Ali, Babatunde Odeyemi, Samson Olori (University of Abuja Teaching Hospital), Ademola Popoola, Ademola Adeyeye, James Adeniran (University of Ilorin Teaching Hospital); Norway: William J. Lossius (Department Of Gastrointestinal Surgery, St. Olavs Hospital, Trondheim University Hospital), Ingemar Havemann (Søerlandet Hospital Kristiansand), Kenneth Thorsen, Jon Kristian Narvestad, Kjetil Soreide (Stavanger University Hospital), Trude Beate Wold, Linn Nymo (University Hospital Of North Norway, Troms); Oman: Mohammed Elsiddig, Manzoor Dar (Sohar Hospital); Pakistan: Kamran Faisal Bhopal, Zainab Iftikhar, Muhammad Mohsin Furqan (Bahawal Victoria Hospital), Bakhtiar Nighat, Masood Jawaid, Abdul Khalique (Dow University Hospital), Ahsan Zil-E-Ali, Anam Rashid (Fatima Memorial Hospital);

Norway: William J. Lossius (Department Of Gastrointestinal Surgery, St. Olavs Hospital, Trondheim University Hospital), Ingemar Havemann (Søerlandet Hospital Kristiansand), Kenneth Thorsen, Jon Kristian Narvestad, Kjetil Soreide (Stavanger University Hospital), Trude Beate Wold, Linn Nymo (University Hospital Of North Norway, Troms); Oman: Mohammed Elsiddig, Manzoor Dar (Sohar Hospital); Pakistan: Kamran Faisal Bhopal, Zainab Iftikhar, Muhammad Mohsin Furqan (Bahawal Victoria Hospital), Bakhtiar Nighat, Masood Jawaid, Abdul Khalique (Dow University Hospital), Ahsan Zil-E-Ali, Anam Rashid (Fatima Memorial Hospital); Peru: Wendy Leslie Messa Aguilar, Jose Antonio Cabala Chiong, Ana Cecilia, Manchego Bautista (Carlos Alberto Seguin Escobedo National Hospital, EsSalud), Eduardo Huaman, Sergio Zegarra, Rony Camacho (Hospital Nacional Guillermo Almenara), Jose María Vergara Celis, Diego Alonso Romani Pozo (Hospital De Emergencias Pediátricas), José Hamasaki, Edilberto Temoche, Jaime Herrera-Matta (Hospital De Policia), Carla Pierina García Torres, Luis Miguel Alvarez Barreda, Ronald Renato Barrionuevo Ojeda (Hospital Goyeneche), Octavio Garaycochea (Hospital Regional Ii-Ii Minsa Moyobamba), Melanie Castro Mollo, Mitchelle Solange De Fã Tima Linares Delgado, Francisco Fujii (Hospital Maria Auxiliadora), Ana Cecilia Manchego Bautista, Wendy Leslie Messa Aguilar, Jose Antonio Cabala Chiong (Hospital Nacional Carlos Alberto Seguin), Susana Yrma Aranzabal Durand, Carlos Alejandro Arroyo Basto, Nelson Manuel Urbina Rojas (Hospital Nacional Edgardo Rebagliati Martins-EsSalud), Sebastian Bernardo Shu Yip, Ana Lucia Contreras Vergara, Andrea Echevarria Rosas Moran, Giuliano Borda Luque, Manuel Rodriguez Castro, Ramon Alvarado Jaramillo (Hospital Nacional Cayetano Heredia), George Manrique Sila, Crislee Elizabeth Lopez, Mardelangel Zapata Ponze De Leon, Massiell Machaca, Ronald Coasaca Huaraya, Andy Arenas, Clara Milagros Herrera Puma, Wilfredo Pino, Christian Hinojosa, Melanie Zapata Ponze De Leon, Susan Limache, George Manrrique Sila, Layza-Alejandra Mercado Rodriguez (Hospital Regional Honorio Delgado Espinoza);

ee Elizabeth Lopez, Mardelangel Zapata Ponze De Leon, Massiell Machaca, Ronald Coasaca Huaraya, Andy Arenas, Clara Milagros Herrera Puma, Wilfredo Pino, Christian Hinojosa, Melanie Zapata Ponze De Leon, Susan Limache, George Manrrique Sila, Layza-Alejandra Mercado Rodriguez (Hospital Regional Honorio Delgado Espinoza); Réunion: Frederique Sauvat (Chu Réunion); Romania: Lucian Corneliu Vida, Liviu Iuliu Muntean, Aurel Sandu Mironescu (Spitalul Clinic De Copii Brasov);

ee Elizabeth Lopez, Mardelangel Zapata Ponze De Leon, Massiell Machaca, Ronald Coasaca Huaraya, Andy Arenas, Clara Milagros Herrera Puma, Wilfredo Pino, Christian Hinojosa, Melanie Zapata Ponze De Leon, Susan Limache, George Manrrique Sila, Layza-Alejandra Mercado Rodriguez (Hospital Regional Honorio Delgado Espinoza); Réunion: Frederique Sauvat (Chu Réunion); Romania: Lucian Corneliu Vida, Liviu Iuliu Muntean, Aurel Sandu Mironescu (Spitalul Clinic De Copii Brasov); Saudi Arabia: Ibrahim N. Alomar, Saleh A. Alnuqaydan, Abdulrahman M. Altwigry (Buraydah Central Hospital), Moayad Othman, Nohad Osman (Imam Abdulrahman Al Faisal Hospital), Enas Alqahtani (King Abdulaziz Hospital Al Ahsa National Guard), Mohammed Alzahrani, Rifan Alyami, Emad Aljohani (King Abdulaziz Medical City), Ibrahim Alhabli, Zaher Mikwar, Sultan Almuallem (King Abdulaziz Medical City (King Khalid National Guard Hospital), Jeddah), Emad Aljohani, Rifan Alyami, Mohammed Alzahrani (King Abdulaziz Medical City, Riyadh), Abrar Nawawi, Mohamad Bakhaidar, Ashraf A. Maghrabi, Mohammed Alsaggaf, Murad Aljiffry, Abdulmalik Altaf, Ahmad Khoja, Alaa Habeebullah, Nouf Akeel (Department of Surgery, Faculty of Medicine, King Abdulaziz University, Jeddah, Saudi Arabia), Nashat Ghandora, Abdullah Almoflihi, Abdulmalik Huwait (King Fahad General Hospital), Abeer Al-shammari, Mashael Al-Mousa (King Fahad Hospital), Masood Alghamdi, Walid Adham, Bandar Albeladi, Muayad Ahmed Alfarsi, Atif Mahdi, Saad Al Awwad (King Fahd Hospital), Afnan Altamimi, Thamer Nouh, Mazen Hassanain (King Khalid University Hospital, King Saud University), Salman Aldhafeeri, Nawal Sadig, Osama Algohary (King Khalid General Hospital), Mohannad Aledrisy, Ahmad Gudal, Ahmad Alrifaie (King Khalid National Guard Hospital), Mohammed AlRowais, Amani Althwainy (Department of Surgery, King Saud University), Alaa Shabkah, Uthman Alamoudi, Mawaddah Alrajraji (National Guard Hospital), Basim Alghamdi, Saud Aljohani, Abdullah Daqeeq (RCYMC), Jubran J Al-Faifi (Security Forces Hospital);

udal, Ahmad Alrifaie (King Khalid National Guard Hospital), Mohammed AlRowais, Amani Althwainy (Department of Surgery, King Saud University), Alaa Shabkah, Uthman Alamoudi, Mawaddah Alrajraji (National Guard Hospital), Basim Alghamdi, Saud Aljohani, Abdullah Daqeeq (RCYMC), Jubran J Al-Faifi (Security Forces Hospital); South Africa: Vicky Jennings, Nyawira Ngayu, Rachel Moore (Chris Hani Baragwanath Academic Hospital), Victor Kong (Edendale Hospital), Colleen Sampson, Richard Spence, Eugenio Panieri (Groote Schuur), Myint Tun, Albert Mohale Mphatsoe, Jo-Anne Carreira (Leratong Hospital), Ella Teasdale, Mark Wagener (Ngwelezana Hospital), Stefan Botes, Danelo Du Plessis (Rob Ferreira Hospital); Spain: Janet Pagnozzi, Jimy Harold Jara Quezada, Jose Luis Rodicio, German Minguez, Raquel Rodríguez-Uría, Paul Ugalde, Camilo Lopez-Arevalo, Luis Barneo, Jessica Patricia Gonzales Stuva (Hospital Universitario Central de Asturias), Jose Aguilar-Jimenez, Jose Andres Garcia-Marin (Hospital Morales Meseguer. SMS), Irene Ortega-Vazquez, Lorena Rodriguez, Norberto Herrera (Severo Ochoa University Hospital); Sri Lanka: Prasad Pitigala Arachchi, Wanigasekara Senanayake Mudiyanselage Kithsiri Janakantha Senanayake, Lalith Asanka Jayasooriya Jayasooriya Arachchige (Department Of General Surgery, Teaching Hospital Kandy), Sivasuriya Sivaganesh, Dulan Irusha Samaraweera, Vimalakanthan Thanusan (The National Hospital Of Sri Lanka);

Sri Lanka: Prasad Pitigala Arachchi, Wanigasekara Senanayake Mudiyanselage Kithsiri Janakantha Senanayake, Lalith Asanka Jayasooriya Jayasooriya Arachchige (Department Of General Surgery, Teaching Hospital Kandy), Sivasuriya Sivaganesh, Dulan Irusha Samaraweera, Vimalakanthan Thanusan (The National Hospital Of Sri Lanka); Sudan: Ahmed Elgaili Khalid Musa, Reem Mohammed Hassan Balila, Mohamed Awad Elkarim Hamad Mohamed (Ibrahim Malik Teaching Hospital), Hussein Ali, Hagir Zain Elabdin, Alaa Hassan ( Jarash International Specialized Hospital), Sefeldin Mahdi, Hala Ahmed, Sahar Abdoun Ishag Idris (Khartoum Teaching Hospital), Makki Elsayed, Mohammed Elsayed, Mohamed Mahmoud (Omdurman Teaching Hospital); Sweden: Hildur Thorarinsdottir, Maria Utter (Helsingborgs Lasarett), Sami Martin Sundstrom (Hudiksvall Sjukhus), Cecilia Wredberg, Ann Kjellin (Karolinska Universitetssjukhuset), Johanna Nyberg, Bjorn Frisk (Skaraborg Hospital Skovde), Yücel Cengiz, Sandra Ahlqvist, Ida Björklund (Sundsvall Hospital), Maria Hjertberg (Vrinnevi Hospital), Malin Sund, Linda Andersson, Ulf Gunnarsson (Department Of Surgical And Perioperative Sciences, Umeå University and Umea University Hospital), Hanna Royson, Per Weber (Vaxjo Central Hospital);

aborg Hospital Skovde), Yücel Cengiz, Sandra Ahlqvist, Ida Björklund (Sundsvall Hospital), Maria Hjertberg (Vrinnevi Hospital), Malin Sund, Linda Andersson, Ulf Gunnarsson (Department Of Surgical And Perioperative Sciences, Umeå University and Umea University Hospital), Hanna Royson, Per Weber (Vaxjo Central Hospital); Switzerland: Roger Schmid, Debora Schivo, Vasileios Despotidis (Bürgerspital Solothurn), Stefan Breitenstein, Ralph F Staerkle, Erik Schadde (Kantonsspital Winterthur), Fabian Deichsel, Alexandra Gerosa, Antonio Nocito (Kantonsspital Baden), Dimitri Aristotle Raptis, Barbara Mijuskovic, Markus Zuber, Lukas Eisner (Kantonsspital Olten), Swantje Kruspi, Katharina Beate Reinisch, Christin Schoewe (Kreisspital für das Freiamt Muri AG), Allan Novak, Adrian F. Palma, Gerfried Teufelberger (Kreisspital Muri, Department Of Surgery); Turkey: Ali Zeynel Abidin Balkan, Mehmet Gumar, Mehmet Ali Yavuz (Harran University Research and Treatment Hospital), Ufuk Karabacak, Gokhan Lap, Bahar Busra Ozkan (Ondokuz Mayis University, Medical Faculty);

Switzerland: Roger Schmid, Debora Schivo, Vasileios Despotidis (Bürgerspital Solothurn), Stefan Breitenstein, Ralph F Staerkle, Erik Schadde (Kantonsspital Winterthur), Fabian Deichsel, Alexandra Gerosa, Antonio Nocito (Kantonsspital Baden), Dimitri Aristotle Raptis, Barbara Mijuskovic, Markus Zuber, Lukas Eisner (Kantonsspital Olten), Swantje Kruspi, Katharina Beate Reinisch, Christin Schoewe (Kreisspital für das Freiamt Muri AG), Allan Novak, Adrian F. Palma, Gerfried Teufelberger (Kreisspital Muri, Department Of Surgery); Turkey: Ali Zeynel Abidin Balkan, Mehmet Gumar, Mehmet Ali Yavuz (Harran University Research and Treatment Hospital), Ufuk Karabacak, Gokhan Lap, Bahar Busra Ozkan (Ondokuz Mayis University, Medical Faculty); UK: Ryan Adams, Robert Morton, Liam Henderson, Ruth Gratton, Keiran David Clement, Kate Yu-Ching Chang, David McNish, Ryan McIntosh, William Milligan (Aberdeen Royal Infirmary), Brendan Skelly, Hannah Anderson-Knight, Roger Lawther (Altnagelvin Area Hospital), Jemina Onimowo, Veereanna Shatkar, Shivanee Tharmalingam (Barking, Havering And Redbridge University Hospitals National Health Services (NHS) Trust, Romford), Evelina Woin, Tessa Fautz, Oliver Ziff (Barnet General Hospital), Shiva Dindyal, Sam Arman, Shagorika Talukder, Sam Arman, Vijay Gadhvi, Shagorika Talukder (Basildon and Thurrock University Foundation Trust), Luen Shaun Chew, Jonathan Heath (Blackpool Victoria Teaching Hospital), Gurdeep Singh Mannu, Dimitris-Christos Zachariades, Ailsa Claire Snaith (Buckinghamshire Healthcare NHS Trust), Thusitha Sampath Hettiarachchi, Arjun Nesaratnam, James Wheeler (Cambridge University Hospitals NHS Foundation Trust), Mark Sykes, Nebil Behar, Harriet Jordan (Chelsea And Westminster Hospital), Tan Arulampalam, Apar Shah, Damien Brown (Colchester Hospital University NHS Foundation Trust), Emma Blower, Paul Sutton, Konstantinos Gasteratos, Dale Vimalachandran (Countess Of Chester Hospital), Cathy Magee, Gareth Irwin, Andrew Mcguigan (Craigavon Area Hospital), Stephen Mcaleer, Clare Morgan (Daisy Hill Hospital), Sarah Braungart (Department of Paediatric Surgery, Leeds General Infirmary), Kirsten Lafferty, Peter Labib, Andrei Tanase, Clodagh Mangan, Lillian Reza (Derriford Hospital), Helen Woodward, Craig Gouldthorpe, Megan Turner (Diana, Princess Of Wales Hospital), Jonathan R L Wild, Tom AM Malik, Victoria K Proctor (Doncaster Royal Infirmary NHS Foundation Trust), Kalon Hewage, James Davies (Dorset County Hospital), Andre Dubois, Sayed Sarwary, Ali Zardab, Alan Grant, Robert Mcintyre (Dr Gray’s Hospital), Shirish Tewari, Gemma Humm, Eriberto Farinella, Sunil Parthiban (East And North Hertfordshire NHS Trust Lister Hospital) Nigel J Hall, Naomi J Wright, Christina P Major (Evelina London Children’s Hospital), Thelma Xerri, Phoebe De Bono, Jasim Amin, Mustafa Farhad, John F.

an Grant, Robert Mcintyre (Dr Gray’s Hospital), Shirish Tewari, Gemma Humm, Eriberto Farinella, Sunil Parthiban (East And North Hertfordshire NHS Trust Lister Hospital) Nigel J Hall, Naomi J Wright, Christina P Major (Evelina London Children’s Hospital), Thelma Xerri, Phoebe De Bono, Jasim Amin, Mustafa Farhad, John F. Camilleri-Brennan, Andrew G N Robertson, Joanna Swann, James Richards, Aijaz Jabbar, Myranda Attard, Hannah Burns, Euan Macdonald, Matthew Baldacchino, Jennifer Skehan, Julian Camilleri-Brennan (Forth Valley Royal Hospital), Tom Falconer Hall, Madelaine Gimzewska, Greta Mclachlan (Frimley Park Hospital), Jamie Shah, James Giles (George Eliot Hospital), Maleeha Hassan, William Beasley, Apostolos Vlachogiorgos, Stephen Dias, Geta Maharaj, Rosie McDonald (Glangwili General Hospital), Kate Cross, Clare M Rees, Bernard Van Duren (Great Ormond Street Hospital for Children NHS Foundation Trust), Emma Upchurch (Great Western Hospital), Sharad Karandikar, Doug Bowley, Ahmed Karim (Heart of England Foundation Trust), Witold Chachulski, Liam Richardson, Giles Dawnay, Ben Thompson, Ajayesh Mistry, Aneel Bhangu, Millika Ghetia, Sudipta Roy, Ossama Al-Obaedi, Millika Ghetia, Kaustuv Das (Hereford County Hospital), Ash Prabhudesai, DM Cocker, Jessica Juliana Tan (Hillingdon Hospital), Sayinthen Vivekanantham, Michael Gillespie, Katrin Gudlaugsdottir (Inverclyde Royal Hospital), Theodore Pezas, Chelise Currow, Matthew Young-Han Kim (Ipswich Hospital NHS Trust), Yahya Salama, Rohi Shah, Ahmad Aboelkassem Ibrahem, Hamdi Ebdewi, Gianpiero Gravante, Saleem El-Rabaa (Kettering General Hospital), Zoe Chan, Zaffar Hassan (King’s College Hospital), Misty Makinde, David Hemingway, Ramzana Dean, Alexander Boddy, Ahmed Aber, Vijay Patel, Deevia Kotecha (Leicester Royal Infirmary), Harmony Kaur Ubhi, Simon-Peter Hosein (Luton and Dunstable Hospital), Simon Ward, Kamran Malik (Macclesfield District General Hospital), Leifa Jennings, Tom Newton, Mirna Alkhouri, Min Kyu Kang, Christopher Houlden, Jonathan Barry (Morriston Hospital), Michael S J Wilson, Yan Ning Neo, Ibrahim Ibrahim, Emily Chan, Fraser S Peck, Pei J Lim, Alexander S North, Rebecca Blundell, Adam Williamson (Ninewells Hospital, NHS Tayside), Dina Fouad, Ashish Minocha (Norfolk And Norwich University Hospital), Kathryn Mccarthy, Emma Court, Alice Chambers (North Bristol NHS Trust), Jenna Yee, Ji Chung Tham, Ceri Beaton (North Devon District Hospital), Una Walsh, Joseph Lockey, Salman Bokhari, Lara Howells, Megan Griffiths, Laura

ells Hospital, NHS Tayside), Dina Fouad, Ashish Minocha (Norfolk And Norwich University Hospital), Kathryn Mccarthy, Emma Court, Alice Chambers (North Bristol NHS Trust), Jenna Yee, Ji Chung Tham, Ceri Beaton (North Devon District Hospital), Una Walsh, Joseph Lockey, Salman Bokhari, Lara Howells, Megan Griffiths, Laura Yallop (Northwick Park Hospital), Shailinder Singh, Omar Nasher, Paul Jackson (Nottingham Children’s Hospital, Queen’s Medical Centre Campus), Saed Ramzi, Shady Zeidan, Jennifer Doughty (Plymouth Hospitals NHS Trust), Sidhartha Sinha, Ross Davenport, Jason Lewis (Princess Alexandra Hospital), Leo Duffy, Elizabeth Mcaleer, Eleanor Williams (Princess Of Wales Hospital), Rhalumi Daniel Obute, Thomas E Glover, David J Clark (Queen Elizabeth Hospital King’s Lynn), Mohamed Boshnaq, Mansoor Akhtar, Pascale Capleton, Samer Doughan, Mohamed Rabie, Ismail Mohamed (Queen Elizabeth The Queen Mother Hospital), Duncan Samuel, Lauren Dickson, Matthew Kennedy, Eleanor Dempster, Emma Brown, Natalie Maple, Eimear Monaghan, Bernhard Wolf, Alicia Garland (Raigmore Hospital), Jonathan Lund, Catherine Boereboom, Jennifer Murphy, Gillian Tierney, Samson Tou (Royal Derby Hospital), Eleanor Franziska Zimmermann, Neil James Smart, Andrea Marie Warwick, Theodora Stasinou, Ian Daniels, Kim Findlay-Cooper (Royal Devon and Exeter NHS Foundation Trust), Stefan Mitrasinovic, Swayamjyoti Ray, Massimo Varcada, Rovan D’souza, Sharif Omara (Royal Free Hospital), Tamsin Boyce, Harriet Whewell, Elin Jones, Jennifer Ma, Emily Abington, Meera Ramcharn, Gethin Williams (Royal Gwent Hospital), Joseph Winstanley, Ewan D.

ndlay-Cooper (Royal Devon and Exeter NHS Foundation Trust), Stefan Mitrasinovic, Swayamjyoti Ray, Massimo Varcada, Rovan D’souza, Sharif Omara (Royal Free Hospital), Tamsin Boyce, Harriet Whewell, Elin Jones, Jennifer Ma, Emily Abington, Meera Ramcharn, Gethin Williams (Royal Gwent Hospital), Joseph Winstanley, Ewan D. Kennedy, Emily NW Yeung (Royal Hospital For Sick Children), Stuart J Fergusson, Catrin Jones, Stephen O’neill, Shujing Jane Lim, Ignatius Liew, Hari Nair, Cameron Fairfield, Julia Oh, Samantha Koh, Andrew Wilson, Catherine Fairfield, Francesca Th’ng, Nichola Robertson (Royal Infirmary of Edinburgh), Delran Anandkumar, Ashok Kirupagaran, Timothy F Jones, Hew D Torrance, Alexander J Fowler, Charmilie Chandrakumar, Priyank Patel, Syed Faaz Ashraf, Sonam M. Lakhani, Aaron Lawson Mclean, Sonia Basson (Royal London Hospital), Jeremy Batt, Catriona Bowman, Michael Stoddart, Natasha Benons (Royal United Hospital Bath), Tom Barker, Virginia Summerour, Edward Harper (Sandwell and West Birmingham Hospitals NHS Trust), Caroline Smith, Matthew Hampton (Sheffield Children’s Hospital), Doug Mckechnie, Ayaan Farah, Anita Chun (Southend University Hospital), Bernadette Pereira, Kristof Nemeth, Emily Decker, Stefano Giuliani, Aly Shalaby (St. George’s Healthcare NHS Trust and University), Aleksandra Szczap, Swathikan Chidambaram, Chee Yang Chen, Kavian Kulasabanathan, Srishti Chhabra, Elisabeth Kostov, Philippe Harbord, James Barnacle (St.

un (Southend University Hospital), Bernadette Pereira, Kristof Nemeth, Emily Decker, Stefano Giuliani, Aly Shalaby (St. George’s Healthcare NHS Trust and University), Aleksandra Szczap, Swathikan Chidambaram, Chee Yang Chen, Kavian Kulasabanathan, Srishti Chhabra, Elisabeth Kostov, Philippe Harbord, James Barnacle (St. Mary’s Hospital), Madan Mohan Palliyil, Mina Zikry, Johnathan Porter, Charef Raslan, Mohammed Saeed, Shazia Hafiz, Niksa Soltani, Katie Baillie (Stockport NHS Foundation Trust), Ahmad Mirza, Haroon Saeed, Simon Galloway (The University Hospital of South Manchester), Gia Elena, Mohammad Afzal, Mohamed Zakir (United Lincolnshire Hospitals—Pilgrim Hospital), Peter Sodde, Charles Hand, Aiesha Sriram, Tamsyn Clark, Patrick Holton, Amy Livesey (University Hospital Coventry And Warwickshire), Yashashwi Sinha, Fahad Mujtaba Iqbal, Indervir Singh Bharj, Adriana Rotundo, Cara Jenvey, Robert Slade (University Hospital Of North Staffordshire NHS Trust), David Golding, Samuel Haines, Ali Adel Ne’ma Abdullah, Thomas W Tilston, Dafydd Loughran, Danielle Donoghue, Lorenzo Giacci, Mohamed Ashur Sherif, Peter Harrison, Alethea Tang (University Hospital Of Wales), Mohamed Elshaer, Tomas Urbonas, Amjid Riaz, Annie Chapman, Parisha Acharya, Joseph Shalhoub (Watford General Hospital), Cathleen Grossart, David McMorran (Western General Hospital), Makhosini Mlotshwa, William Hawkins, Sofronis Loizides (Western Sussex Hospitals NHS Trust), Peter Thomson, Shahab Khan, Fiona Taylor, Jalak Shukla, Emma Elizabeth Howie (Whipps Cross University Hospital), Linda Macdonald, Olusegun Komolafe, Neil Mcintyre (Wishaw General Hospital), James Cragg, Jody Parker, Duncan Stewart (Wrexham Maelor Hospital), Luke Lintin, Julia Tracy, Tahir Farooq (Yeovil District Hospital);

r Thomson, Shahab Khan, Fiona Taylor, Jalak Shukla, Emma Elizabeth Howie (Whipps Cross University Hospital), Linda Macdonald, Olusegun Komolafe, Neil Mcintyre (Wishaw General Hospital), James Cragg, Jody Parker, Duncan Stewart (Wrexham Maelor Hospital), Luke Lintin, Julia Tracy, Tahir Farooq (Yeovil District Hospital); The USA: Melanie Sion, Michael S. Weinstein, Viren Punja (Thomas Jefferson University Hospital), Nikolay Bugaev, Monica Goodstein, Shadi Razmdjou (Tufts Medical Center). Competing interests: None declared. Provenance and peer review: Not commissioned; externally peer reviewed. Data sharing statement: This is the paediatric data from a larger study—GlobalSurg 1. Part of the data has been published under the group name GlobalSurg Collaborative.

Key questions What is already known about this topic? A gross lack of adequate infection prevention and control (IPC) practice in health facilities was a main driver of the Ebola virus disease (EVD) epidemic in Sierra Leone. Given the rarity of these epidemics, it is likely that IPC strategies are not frequently documented in the scientific literature and have not undergone formal evaluation in situ. What are the new findings? We comprehensively evaluate attitudes and self-efficacy towards IPC, and adherence to practice using the appropriate combination of qualitative, quantitative, observational and participatory approaches. The study was carried out during the height of the national epidemic, thereby presenting a unique opportunity to examine actual healthcare worker behaviours and attitudes under duress, and also to inform policy and practice. Recommendations for policy Sierra Leone's National Recovery Plan for 2015–2017 has put US$33 million towards scaling up and maintaining IPC across all healthcare facilities in order to prevent a recurrence of EVD. The practice gaps identified provide the rationale to improve current training packages by providing insight into contextual, emotional, psychological and behavioural factors that influence adherence to IPC practice and the motivations of healthcare workers.

Recommendations for policy Sierra Leone's National Recovery Plan for 2015–2017 has put US$33 million towards scaling up and maintaining IPC across all healthcare facilities in order to prevent a recurrence of EVD. The practice gaps identified provide the rationale to improve current training packages by providing insight into contextual, emotional, psychological and behavioural factors that influence adherence to IPC practice and the motivations of healthcare workers. Introduction Sierra Leone was profoundly impacted by the Ebola virus disease (EVD) epidemic in West Africa, documenting 14 122 cases and 3955 deaths.1 Its first confirmed case in May 2014 led to the initial outbreak in the eastern districts of Kailahun and Kenema. From June to December, transmission spread to all districts and peaked at 600 confirmed cases weekly.2 The incidence among healthcare workers (HCWs) became 100 times that of the general population, leading to the deaths of nearly 10% of the workforce.3 4

initial outbreak in the eastern districts of Kailahun and Kenema. From June to December, transmission spread to all districts and peaked at 600 confirmed cases weekly.2 The incidence among healthcare workers (HCWs) became 100 times that of the general population, leading to the deaths of nearly 10% of the workforce.3 4 Poor infection prevention and control (IPC) serves as an efficient amplifier of transmission of viral haemorrhagic fevers (VHF).5–7 In primary healthcare facilities, also called peripheral health units (PHUs), HCWs lacked the supplies and training to apply rigorous symptom screening and IPC practices recommended for Ebola treatment units (ETU).8 Such deficits increased the risk of occupational and nosocomial infection for HCWs and non-EVD patients, respectively. The majority (66%) of HCW infections occurred in PHUs and hospitals.4 As HCWs became infected, colleagues became frightened and demoralised, and the community's trust of the health system was further eroded.9

ficits increased the risk of occupational and nosocomial infection for HCWs and non-EVD patients, respectively. The majority (66%) of HCW infections occurred in PHUs and hospitals.4 As HCWs became infected, colleagues became frightened and demoralised, and the community's trust of the health system was further eroded.9 By August, grossly insufficient IPC led to the infection of 43 HCWs in Kenema district, mainly in Kenema Government Hospital, which had become a de facto ETU.3 10 To prevent EVD transmission in PHUs, the International Rescue Committee (IRC), WHO and Kenema's District Health Management Team provided IPC supplies including light personal protective equipment (PPE), and training to Kenema's PHUs near the peak of the district's outbreak in August 2014. The training covered screening, isolation, referral, hand hygiene, use of light PPE, sharps management, environmental cleaning and waste disposal.11 12 The epidemic continued to spread rapidly and geographically. Nearly all PHUs remained open, albeit with substantially reduced staffing and services.13 A rapid assessment of PHUs in six districts found deficiencies in the identification and isolation of suspected cases, scarcity of supplies (PPE, chlorine, water and incinerators) and delays in referral of suspected cases to ETUs.14 This led the Ministry of Health and Sanitation, the IRC-led Ebola Response Consortium, UNICEF and the US Centers for Disease Control and Prevention (CDC) to train HCWs in IPC in all 1180 PHUs across 14 districts nationally, between October and December 2014.12 15 The effort was paired with a quality assurance programme to monitor inventory, structures and practices on an ongoing basis. To learn from this experience and evaluate attitudes, experiences and the effects of an improvement workshop on behaviours, we conducted a mixed-methods study with multiple objectives. The primary objective was to generate insights into how IPC behaviours can be improved in a short time frame during an EVD outbreak. A secondary objective was to assess HCW attitudes, self-efficacy and experiences with IPC practice. Another secondary objective was to evaluate the effectiveness of participatory workshops to develop improvement plans, through the measurement of changes in adherence to IPC protocols. The primary outcome measures of effectiveness were the proportion of correct IPC behaviours within the domains of prescreening, donning, screening, doffing and consultation.

luate the effectiveness of participatory workshops to develop improvement plans, through the measurement of changes in adherence to IPC protocols. The primary outcome measures of effectiveness were the proportion of correct IPC behaviours within the domains of prescreening, donning, screening, doffing and consultation. Methods Study design, setting and participants Using a participatory action framework and a mixed-methods approach, we conducted a single group, pretest post-test study (also called an uncontrolled before and after intervention study) in Bo and Kenema districts in December 2014 and January 2015.16 17 The districts were at different phases of the epidemic. In Kenema, the epidemic had peaked, and by December, there were fewer than two cases per week. Bo's first cases were reported in July 2014, and by December, transmission dropped from 20 to 40 cases to 10 cases per week. The national IPC trainings led by the Ministry of Health and Sanitation and the Ebola Response Consortium were completed ∼1 week before the data collection for this study began in December 2014.

Bo's first cases were reported in July 2014, and by December, transmission dropped from 20 to 40 cases to 10 cases per week. The national IPC trainings led by the Ministry of Health and Sanitation and the Ebola Response Consortium were completed ∼1 week before the data collection for this study began in December 2014. There were two phases of the study where data were collected: a baseline period (10–20 December 2014) and a follow-up period 3 weeks later (7–16 January 2015). The study's intervention consisted of a participatory workshop in each district immediately following the baseline period and attended by HCWs, district health officials, community health officers (CHOs, who are main healthcare provider at the PHU level) and community representatives. At this workshop, participants reviewed baseline data on IPC practices, attitudes and risk perception, and they developed improvement plans for each PHU. At baseline and follow-up, we conducted self-administered surveys with HCWs exposed to the intervention and who were present at the PHUs to assess demographics, attitudes and self-efficacy towards IPC. Also, at baseline and follow-up, we measured HCW's adherence to IPC protocols using structured observations of patient encounters. During both periods, in-depth interviews (IDIs) were conducted to explore attitudes and self-efficacy towards IPC, and experiences with IPC (without attempts to compare periods). This included vignettes where HCWs were asked how they would act in three situations related to IPC in their professional and personal lives.

During both periods, in-depth interviews (IDIs) were conducted to explore attitudes and self-efficacy towards IPC, and experiences with IPC (without attempts to compare periods). This included vignettes where HCWs were asked how they would act in three situations related to IPC in their professional and personal lives. We used stratified random sampling to select PHUs from a sampling frame of 121 PHUs in Kenema district and of 110 PHUs in Bo district. We stratified by urban/rural setting and any/no suspected cases at the PHU level, to maximise variation. One facility was randomly chosen from each stratum in each district resulting in a total of eight participating PHUs. At least four HCWs across a range of roles were included in the IDIs at each facility as most facilities had no more than four staff. This formed the purposive sample for the survey. Sample sizes for the observations were not calculated a priori due to the fact that observers could be present in PHUs for a limited time period and therefore could capture a limited number of observations. A timeline of the methods is presented in figure 1. Figure 1 Timeline of the methods.

to community HIV services. Box 1 Key study definitions in relation to community HIV services ▸ Community health worker (CHW): Any individual delivering healthcare, trained in the skills needed for the intervention but with no certificate or degree in tertiary education. In Kenya this term includes both CHVs and CHEWs. ▸ Community health volunteer (CHV): A volunteer CHW trained in a government-approved curriculum, who is responsible for 20 households, offering advice on disease prevention and control, providing family and maternal health services, promoting environmental health and sanitation, and performing basic curative tasks ▸ Community health extension worker (CHEW): A trained health worker employed by the Kenyan government in a link health facility, providing support and supervision to CHVs ▸ Lay counsellor: An individual who has completed secondary education and been trained specifically to conduct HIV testing and counselling; usually employed by an NGO. ▸ Community HIV services: Services provided in the community including home-based HIV counselling and testing, linkage for care and treatment, and home-based care. ▸ Integration: This refers to the incorporation of community HIV services traditionally carried out by vertical programmes into the existing Ministry of Health community health structures.

We used stratified random sampling to select PHUs from a sampling frame of 121 PHUs in Kenema district and of 110 PHUs in Bo district. We stratified by urban/rural setting and any/no suspected cases at the PHU level, to maximise variation. One facility was randomly chosen from each stratum in each district resulting in a total of eight participating PHUs. At least four HCWs across a range of roles were included in the IDIs at each facility as most facilities had no more than four staff. This formed the purposive sample for the survey. Sample sizes for the observations were not calculated a priori due to the fact that observers could be present in PHUs for a limited time period and therefore could capture a limited number of observations. A timeline of the methods is presented in figure 1. Figure 1 Timeline of the methods. Data collection and measurement Two observers and eight qualitative interviewers per district were trained for 2 and 3 days, respectively. Three co-investigators trained the interviewers and supervised data collection (LSH, RA and HB). Research tools were piloted in PHUs that were not selected for study. The survey was self-administered to the HCWs available on that day. For the structured observations, teams of two observers watched HCW–patient encounters for 5 hours on a single day at each PHU. Behaviours were recorded for each domain in the national protocol (patient screening, donning and doffing of PPE, patient consultation, isolation of patients screened positive, donning and doffing of PPE for isolation, and dead body management).11 Data were collected with smartphones using Magpi software (Datadyne, Washington, DC, USA). If a behaviour was clearly a hazard (ie, HCW attempts to touch the patient without gloves), observers were instructed to intervene. IDIs were conducted in Krio and Mende by one supervisor and three interviewers per district, digitally recorded and typed verbatim in Krio or Mende. They lasted for 30–60 min. The transcripts were translated from Krio and Mende to English.

ts to touch the patient without gloves), observers were instructed to intervene. IDIs were conducted in Krio and Mende by one supervisor and three interviewers per district, digitally recorded and typed verbatim in Krio or Mende. They lasted for 30–60 min. The transcripts were translated from Krio and Mende to English. Data analysis Data were analysed and interpreted concurrently using a convergent-parallel design to integrate findings across methods.18 Quantitative analysis of the survey and structured observations was conducted using Stata V.14 (StataCorp LP, College Station, Texas, USA). For the survey, responses on a four-point Likert item scale were summarised using the median and the IQR. Since HCWs were selected based on their availability, some HCWs may have changed between rounds. Since pairing was not possible, distributions of responses at baseline and at follow-up were compared using the Wilcoxon rank-sum test. For the structured observations, the proportion of correct behaviours for each task and the changes between rounds were computed. The main exposure and outcome were the time period (baseline vs follow-up) and the proportion of correct behaviours, respectively. A log-binomial model was used to estimate risk ratios (RR) for each correct behaviour at baseline and follow-up. Generalised estimating equations (GEE) with robust SEs accounted for repeated measures among HCWs and clustering within PHUs.19 An exchangeable working correlation structure was assumed. For all statistical tests, a significance level of p<0.05 was chosen. For the qualitative components, an initial phase of inductive coding on a selection of rich, diverse and representative transcripts was performed based in part on Grounded theory.20 Coding and analysis were conducted using Dedoose 5.011 (SocioCultural Research Consultants, LLC, Los Angeles, California, USA).

as chosen. For the qualitative components, an initial phase of inductive coding on a selection of rich, diverse and representative transcripts was performed based in part on Grounded theory.20 Coding and analysis were conducted using Dedoose 5.011 (SocioCultural Research Consultants, LLC, Los Angeles, California, USA). Ethics The study received ethics approval from Durham University's Institutional Review Board and the Sierra Leone Ethics and Scientific Research Committee. HCWs provided written informed consent. If any potentially hazardous behaviours were observed, observers were required to intervene immediately through a verbal notification to the HCW. Results The survey was administered to 35 HCWs at baseline and 33 HCWs at follow-up in 8 PHUs (table 1). Twenty-two (63%) of the 35 HCWs were the same between rounds, based on profession, age and sex. There were no confirmed cases among HCWs in the sampled PHUs during the study period. Participants included CHOs, community health nurses (CHNs), maternal child health aides (MCHAs) and community health assistants (CHA). Half were below 40 years of age, and half were women. The majority (77%) were trained through the national IPC training, and 43% had already screened patients. In total, 54 IDIs were analysed. Three recordings were lost, but saturation had been reached before completion of the available transcripts. All field notes were reviewed to ensure no new themes emerged. Table 1 Characteristics of survey participants, baseline (N=35)

Results The survey was administered to 35 HCWs at baseline and 33 HCWs at follow-up in 8 PHUs (table 1). Twenty-two (63%) of the 35 HCWs were the same between rounds, based on profession, age and sex. There were no confirmed cases among HCWs in the sampled PHUs during the study period. Participants included CHOs, community health nurses (CHNs), maternal child health aides (MCHAs) and community health assistants (CHA). Half were below 40 years of age, and half were women. The majority (77%) were trained through the national IPC training, and 43% had already screened patients. In total, 54 IDIs were analysed. Three recordings were lost, but saturation had been reached before completion of the available transcripts. All field notes were reviewed to ensure no new themes emerged. Table 1 Characteristics of survey participants, baseline (N=35) Characteristic N (%) Sex, male 14 (40) Age* <30 8 (23) 30–39 11 (31) 40–49 11 (31) 50+ 3 (8) Profession* CHN 11 (31) MCHA 9 (26) CHA 4 (11) CHO 3 (9) Community health worker 1 (3) Endemic disease control unit assistant 1 (3) Laboratory technician 1 (3) Other 4 (11) Workplace Community health post 17 (49) Community health centre 16 (46) Maternal and child health post 2 (6) District Bo 16 (46) Kenema 19 (54) Trained in national IPC programme* 27 (77) Screened patients in past 6 months 15 (43) *Missing data for n=2 (age), n=1 (profession) and n=4 (training). CHA, community health assistant; CHN, community health nurse; CHO, community health officer; IPC, infection prevention and control; MCHA, maternal child health aide.

Characteristic N (%) Sex, male 14 (40) Age* <30 8 (23) 30–39 11 (31) 40–49 11 (31) 50+ 3 (8) Profession* CHN 11 (31) MCHA 9 (26) CHA 4 (11) CHO 3 (9) Community health worker 1 (3) Endemic disease control unit assistant 1 (3) Laboratory technician 1 (3) Other 4 (11) Workplace Community health post 17 (49) Community health centre 16 (46) Maternal and child health post 2 (6) District Bo 16 (46) Kenema 19 (54) Trained in national IPC programme* 27 (77) Screened patients in past 6 months 15 (43) *Missing data for n=2 (age), n=1 (profession) and n=4 (training). CHA, community health assistant; CHN, community health nurse; CHO, community health officer; IPC, infection prevention and control; MCHA, maternal child health aide. Implementation of the workshop intervention Each district conducted a daylong workshop. HCWs, health authorities and community members identified key themes in the data. They developed causal diagrams and matrices, to link IPC challenges to potential solutions, and improvement plans for each PHU that aimed to improve IPC within 3 weeks (table 2). Solutions ranged from specific and attainable (eg, obtaining PPE for safe deliveries) to broad and more distal (eg, improving the water supply). Owing to the competing priorities of the emergency response, improvement plans were not always completed within 3 weeks. Table 2 Key IPC challenges and solutions outlined by workshop participants in action plans

Implementation of the workshop intervention Each district conducted a daylong workshop. HCWs, health authorities and community members identified key themes in the data. They developed causal diagrams and matrices, to link IPC challenges to potential solutions, and improvement plans for each PHU that aimed to improve IPC within 3 weeks (table 2). Solutions ranged from specific and attainable (eg, obtaining PPE for safe deliveries) to broad and more distal (eg, improving the water supply). Owing to the competing priorities of the emergency response, improvement plans were not always completed within 3 weeks. Table 2 Key IPC challenges and solutions outlined by workshop participants in action plans Problem Potential solution Frequency, n=8 (%) Lack plan and physical materials for screening booth Build screening materials or booth 7 (88) Lack plan/materials for deliveries Procure elbow gloves, delivery aprons, etc 4 (50) No latrines for suspect cases Build a dedicated latrine 4 (50) Routine care requires contact Obtain an electronic blood pressure machine 4 (50) Community members do not understand rationale for IPC Increase community sensitisation on IPC and handwashing 3 (38) Handwashing among staff and patients is poor Reinforce handwashing through signage; increase soap supply 3 (38) Lack a working incinerator Build an incinerator or burning pit 3 (38) Lack an isolation area Build an isolation area 3 (38) Lack fencing for facility Put in fencing 3 (38) Water supply is inconsistent Increase the supply of water 3 (38) Need to reinforce supervision, training or mentorship for IPC Implement IPC supervision or peer mentoring 2 (25) Lack space for women postdelivery Obtain mattresses for postnatal care 2 (25) Concerned PPE will run out Ensure additional PPE is available 1 (13) Electricity is inconsistent Address generator problems 1 (13) Lack safe area for PPE removal Make space for a PPE removal area 1 (13) HCW, healthcare worker; IPC, infection prevention and control; PPE, personal protective equipment.

tnatal care 2 (25) Concerned PPE will run out Ensure additional PPE is available 1 (13) Electricity is inconsistent Address generator problems 1 (13) Lack safe area for PPE removal Make space for a PPE removal area 1 (13) HCW, healthcare worker; IPC, infection prevention and control; PPE, personal protective equipment. Risk perception, attitudes and self-efficacy Survey results did not change significantly between rounds; we report the baseline results in the text and the full results in table 3. Respondents believed that they had an increased risk of infection compared to the public (median=4 (strongly agree), IQR, 3–4). There was slight disagreement with the false statement that children posed a lesser risk of transmission as adults (median=2 (disagree), IQR, 2–3). HCWs described difficulty in recognising how the risks of infection for EVD and other diseases differed. As EVD was described as an epidemic, ‘it would not last for long and that maybe after one or 2 months it will all be over and gone’ (Female state enrolled nurse, Bo). When asked if they would avoid the use of gloves to treat ‘non-Ebola’ patients and PPE to treat family members for any condition, HCWs indicated strong disagreement with these statements (median=1 (strongly disagree), IQR, 1–2). Table 3 Self-efficacy, risk perception and attitudes among HCWs

Risk perception, attitudes and self-efficacy Survey results did not change significantly between rounds; we report the baseline results in the text and the full results in table 3. Respondents believed that they had an increased risk of infection compared to the public (median=4 (strongly agree), IQR, 3–4). There was slight disagreement with the false statement that children posed a lesser risk of transmission as adults (median=2 (disagree), IQR, 2–3). HCWs described difficulty in recognising how the risks of infection for EVD and other diseases differed. As EVD was described as an epidemic, ‘it would not last for long and that maybe after one or 2 months it will all be over and gone’ (Female state enrolled nurse, Bo). When asked if they would avoid the use of gloves to treat ‘non-Ebola’ patients and PPE to treat family members for any condition, HCWs indicated strong disagreement with these statements (median=1 (strongly disagree), IQR, 1–2). Table 3 Self-efficacy, risk perception and attitudes among HCWs Overall Bo Kenema Baseline 35 Follow-up 33 Baseline 16 Follow-up 16 Baseline 19 Follow-up 17 No. of respondents Median* (IQR) Median (IQR) p Value† Median (IQR) Median (IQR) Median (IQR) Median (IQR) Self-efficacy I can correctly identify suspected Ebola cases using the screening flow chart. 4 (3–4) 3 (3–4) 0.35 4 (3–4) 4 (3–4) 4 (3–4) 4 (3–4) I can remove PPE after isolating a suspected Ebola case without infecting myself. 4 (3–4) 3 (3–4) 0.52 4 (3–4) 3 (3–4) 4 (3–4) 3 (3–4) I can safely disinfect a room where a suspected Ebola case has been isolated to remove any risk of infection to myself or other. 4 (3–4) 4 (3–4) 0.25 4 (3–4) 4 (3–4) 4 (3–4) 3 (3–4) There is enough PPE at my facility to protect us from being infected with Ebola. 4 (3–4) 3 (2–4) 0.21 3 (3–4) 3 (2–4) 4 (3–4) 4 (3–4) Attitudes and risk perception I am at higher risk of becoming infected with Ebola because I work in a health facility. 4 (3–4) 4 (3–4) 0.51 4 (3–4) 4 (3–4) 4 (3–4) 4 (3–4) I am less likely to become infected with Ebola when taking care of children than adults. 2 (2–3) 2 (1–3) 0.87 2 (2–3) 2 (2–4) 2 (1–2) 2 (1–3) If my colleague is sick it would be cruel to use PPE when treating him/her. 2 (1–4) 1 (1–3) 0.4 2 (1–4) 1 (1–2) 2 (1–4) 2 (1–4) I do not need to use PPE when taking care of a family member with a fever, headache, diarrhoea and nausea. 1 (1–2) 1 (1–2) 0.87 1 (1–2) 1 (1–2) 1 (1–4) 1 (1–2) I do not need to wear gloves when I take care of non-Ebola patients. 1 (1–2) 2 (1–2) 0.29 1 (1–2) 1 (1–2) 2 (1–2) 2 (1–2) *Responses were given on a four-point Likert item scale from strongly disagree 1 to strongly agree 4.

th a fever, headache, diarrhoea and nausea. 1 (1–2) 1 (1–2) 0.87 1 (1–2) 1 (1–2) 1 (1–4) 1 (1–2) I do not need to wear gloves when I take care of non-Ebola patients. 1 (1–2) 2 (1–2) 0.29 1 (1–2) 1 (1–2) 2 (1–2) 2 (1–2) *Responses were given on a four-point Likert item scale from strongly disagree 1 to strongly agree 4. †Evaluated using the Wilcoxon rank-sum test. HCW, healthcare worker; IQR, interquartile range; PPE, personal protective equipment.

th a fever, headache, diarrhoea and nausea. 1 (1–2) 1 (1–2) 0.87 1 (1–2) 1 (1–2) 1 (1–4) 1 (1–2) I do not need to wear gloves when I take care of non-Ebola patients. 1 (1–2) 2 (1–2) 0.29 1 (1–2) 1 (1–2) 2 (1–2) 2 (1–2) *Responses were given on a four-point Likert item scale from strongly disagree 1 to strongly agree 4. †Evaluated using the Wilcoxon rank-sum test. HCW, healthcare worker; IQR, interquartile range; PPE, personal protective equipment. HCWs described PPE as uncomfortable, hot and causing sweating and itching, yet at the same time, ‘precious, lifesaving, necessary for protecting oneself and one's family’. On balance, “it's better that you overheat but are protected than that you get fresh air and become contaminated. I choose to be hot but protected” (Female CHO, Bo). A recurrent theme was that HCWs regretted the physical distance with their patients caused by PPE. There was disagreement among HCWs regarding the statement, ‘it would be cruel to use PPE when treating a sick colleague’ (median=2 (disagree), IQR, 1–4) (table 3). However, a vignette to elicit perspectives on the management of an ill HCW suggested correct behaviours. HCWs most often reported that they would tell an infected colleague to isolate herself (‘put her in observation’, ‘don't touch her’, ‘tell her not to touch anybody’) or they would refer her to an ETU (‘call the emergency line’, ‘get that ambulance to take her away’, ‘encourage her with kind words while she is being referred’). While acknowledging that it would be an upsetting experience (‘she will feel the stigma of the Ebola, she will be shedding tears, as will we’), most insisted on isolating or using PPE to treat her: “She is my colleague and friend and when the Ebola finishes…I will apologize to her, but (for now) I will not touch her, I won't do it, before all of us die, let one die so that others can live” (Female MCHA, Kenema).

Ebola, she will be shedding tears, as will we’), most insisted on isolating or using PPE to treat her: “She is my colleague and friend and when the Ebola finishes…I will apologize to her, but (for now) I will not touch her, I won't do it, before all of us die, let one die so that others can live” (Female MCHA, Kenema). Most HCWs expressed self-efficacy in identifying cases, removing PPE, and disinfecting a room after identification of a suspected case (see table 3). HCWs described five prevailing emotions that influenced the maintenance of care: disbelief, dread, fear, sadness and determination. Fear was described with the most depth and nuance, followed by sadness. Their self-efficacy developed after a gradual acceptance of the threat and after receiving training, supplies and undergoing practice. HCWs described how their own attitude or knowledge has changed after the training saying, for instance, ‘Now I feel like I have to be careful in everything I do’ (Female CHN Bo). Several HCWs, particularly those engaged in childbirth, described discontinuing work at the outset, but resuming services with confidence once they received training and PPE stocks:Let me say the truth, before Ebola, we were working hard but we were careless in terms of IPC. As for me, the only time I used to wear gloves was during delivery…the use of chlorine for hand washing was not common…We had no idea about the use of wearing of goggles, facemasks, PPE and gowns…Now with the epidemic of Ebola, hand washing is widely practiced. (Female MCHA, Kenema)

ut we were careless in terms of IPC. As for me, the only time I used to wear gloves was during delivery…the use of chlorine for hand washing was not common…We had no idea about the use of wearing of goggles, facemasks, PPE and gowns…Now with the epidemic of Ebola, hand washing is widely practiced. (Female MCHA, Kenema) Most HCWs mentioned that for their IPC to be effective, community sensitisation was essential. PPE induced fear among patients, evoking images of burial teams and ‘memories of brothers and sisters taken by Ebola’ and ‘buried by these people’. Sensitisation by HCWs was reportedly impeded by restrictions on their movement, inaccessibility of communities, finances and a resistance from community members:They are really been panicked to come…they will stand at the gate and start to talk to themselves in fear of the booths that we have constructed. But we are still sensitizing them to continue coming. (Female MCHA, Kenema)

n their movement, inaccessibility of communities, finances and a resistance from community members:They are really been panicked to come…they will stand at the gate and start to talk to themselves in fear of the booths that we have constructed. But we are still sensitizing them to continue coming. (Female MCHA, Kenema) HCWs tried to counteract patients’ fears by counselling them individually to understand the rationale behind the use of PPE:When the patients come, they sit down. Before we start our work, we talk to them, “Now, you see me as I am, I am alright. I am going to dress in order to protect myself, and protect you. May be I am sick but you are not aware. I would be talking to you may be the spit from my mouth jumps to your face or whatsoever or your nose or your eye being that they are closer to me, if I had the disease, you will have it. Or in case I am asking you questions then your child throws up or coughs, I will be infected. So for this reason I am going to put on these dressings. Don't see me and be afraid. I am trying to protect myself and protect you so that I won't infect you and you also will not infect me. (Male MCHA, Bo)

, you will have it. Or in case I am asking you questions then your child throws up or coughs, I will be infected. So for this reason I am going to put on these dressings. Don't see me and be afraid. I am trying to protect myself and protect you so that I won't infect you and you also will not infect me. (Male MCHA, Bo) HCWs mentioned three further threats to self-efficacy. First, HCWs doubted the differential diagnosis for suspect cases: “typhoid…malaria…Lassa have signs of Ebola” (Female CHO, Bo). Second, respondents at follow-up remained concerned about PPE shortages (median=3 (agree), IQR, 2–3). Third, HCWs emphasised that while conducting IPC, they continued to deal with a disrupted health system:There is no toilet, no water well, no network coverage, no means of transportation… these are our problems. … And you tell a person to wash their hands at the facility, but this is not easy without water. (HCW, Bo)

. Third, HCWs emphasised that while conducting IPC, they continued to deal with a disrupted health system:There is no toilet, no water well, no network coverage, no means of transportation… these are our problems. … And you tell a person to wash their hands at the facility, but this is not easy without water. (HCW, Bo) Adherence to IPC behaviours The proportions of correct behaviours and RRs comparing the proportion of correct behaviours between baseline (90 screenings and 54 consultations) and follow-up (131 screenings and 32 consultations) are shown in table 4 (see online supplementary material annexe (Final annex_ratnayake.pdf) for results stratified by district). No suspected cases or dead bodies were observed; therefore, all observations relate to the screening of patients and subsequent consultations. During prescreenings, only one instance of HCW handwashing was observed. The proportion of HCWs asking patients to wash their hands (RR 1.45, 95% CI 1.16 to 1.8) and patients doing so on prompting from the HCW (1.49, 1.19 to 1.86) increased. Patient handwashing, with or without HCW prompting, increased though not significantly from 82% to 99% (RR 1.21, 95% CI 0.95 to 1.71). HCWs frequently mentioned patient handwashing as straining on the HCW–patient relationship: When they come and you tell them to wash their hands, they make comments like, “What about [you], do you wash your hands every day?”…the concept that behaviour should be changed, it is not really easy, it is difficult. (Female CHO, Kenema) Table 4 Proportions of correct IPC events before and after the workshop

Adherence to IPC behaviours The proportions of correct behaviours and RRs comparing the proportion of correct behaviours between baseline (90 screenings and 54 consultations) and follow-up (131 screenings and 32 consultations) are shown in table 4 (see online supplementary material annexe (Final annex_ratnayake.pdf) for results stratified by district). No suspected cases or dead bodies were observed; therefore, all observations relate to the screening of patients and subsequent consultations. During prescreenings, only one instance of HCW handwashing was observed. The proportion of HCWs asking patients to wash their hands (RR 1.45, 95% CI 1.16 to 1.8) and patients doing so on prompting from the HCW (1.49, 1.19 to 1.86) increased. Patient handwashing, with or without HCW prompting, increased though not significantly from 82% to 99% (RR 1.21, 95% CI 0.95 to 1.71). HCWs frequently mentioned patient handwashing as straining on the HCW–patient relationship: When they come and you tell them to wash their hands, they make comments like, “What about [you], do you wash your hands every day?”…the concept that behaviour should be changed, it is not really easy, it is difficult. (Female CHO, Kenema) Table 4 Proportions of correct IPC events before and after the workshop Baseline n=90 Follow-up n=131 RR* Correct Per cent Correct Per cent  95% CI Prescreening Patient went directly, or HCW-directed patient, to screening area 51 57 31 24 0.53 0.37 to 0.77 Attendant washed hands 1 1 0 0 – – Screener asked patient to wash hands 56 62 105 80 1.45 1.16 to 1.80 Patient washed hands on direction from HCW 54 60 105 80 1.49 1.19 to 1.86 Patient washed hands directly or washed on direction from HCW 74 82 130 99 1.27 0.95 to 1.71† Donning Wore rubber boots or covers 60 67 111 85 1.51 1.14 to 1.99 Wore face shield or mask 69 77 109 83 1.27 1.03 to 1.58 Completed in correct order 3 3 73 56 8.94 0.84 to 95.61 Took off /did not wear jewellery 89 99 114 87 0.83 0.72 to 0.97 Wore new gloves 17 19 40 31 2.56 1.37 to 4.79 Continued to wear gloves 63 70 87 66 0.75 0.6 to 0.94 Screening No other HCWs were in screening area 86 96 104 79 0.86 0.69 to 1.07† Stood 1.5 m from patient 82 91 130 99 1.11 0.83 to 1.48† Sat sideways to patient 21 23 75 57 2.3 1.34 to 3.95 Held digital thermometer 5–6 cm from patient 82 91 15 12 0.23 0.12 to 0.43† Doffing Removed any light PPE 13 14 42 32 2.54 1.32 to 4.88 Removed gloves 9 10 29 22 4.09 1.34 to 12.49 Washed gloved or ungloved hands 10 11 25 19 2.58 1.0 to 6.66 Removed face shield or goggles 8 9 2 2 0.21 0.05 to 0.94 Completed in correct order (if removed gloves) 3 3 29 22 6.64 2.09 to 21.14 Baseline n=54 Follow-up n=32 Correct Per cent Correct Per cent RR* 95% CI Consultations Washed hands before treating patient 8 15 3 10 0.63 0.18 to 2.21 Washed hands after treating patient 21 39 5 16 0.91 0.5 to 1.65 Put on new gloves before treating patient 50 93 29 91 0.97 0.85 to 1.1 Did not remove gloves after treating patient 6 11 8 25 1.51 0.55 to 4.12 Stood 1.5 m from patient 35 65 29 91 1.18 0.92 to 1.51 *Risk ratio using binomial regression (family: binomial, link: log) accounting for clustering at the health facility level (GEE). Hyphens indicate where parameter was not estimable.

.85 to 1.1 Did not remove gloves after treating patient 6 11 8 25 1.51 0.55 to 4.12 Stood 1.5 m from patient 35 65 29 91 1.18 0.92 to 1.51 *Risk ratio using binomial regression (family: binomial, link: log) accounting for clustering at the health facility level (GEE). Hyphens indicate where parameter was not estimable. †Indicates that a Poisson regression (family: Poisson, link: log) was used due to the failure of the binomial model to converge. HCW, healthcare worker; IPC, infection prevention and control. 10.1136/bmjgh-2016-000103.supp1Supplementary data

.85 to 1.1 Did not remove gloves after treating patient 6 11 8 25 1.51 0.55 to 4.12 Stood 1.5 m from patient 35 65 29 91 1.18 0.92 to 1.51 *Risk ratio using binomial regression (family: binomial, link: log) accounting for clustering at the health facility level (GEE). Hyphens indicate where parameter was not estimable. †Indicates that a Poisson regression (family: Poisson, link: log) was used due to the failure of the binomial model to converge. HCW, healthcare worker; IPC, infection prevention and control. 10.1136/bmjgh-2016-000103.supp1Supplementary data HCWs wore boots and face masks more than 60% of the time at baseline and more than 80% at follow-up (boots, RR 1.51, 95% CI 1.14 to 1.99; face masks, RR 1.27, 95% CI 1.03 to 1.58). Donning in the correct order increased ninefold from baseline (3%) to follow-up (56%) (RR 8.94, 95% CI 0.84 to 95.61). In 20% of screenings at follow-up, additional HCWs were present in the screening area (which is not recommended; RR 0.86, 95% CI 0.69 to 1.07). Virtually all HCWs stood 1.5 m from patients, increasing from 91% to 99% at follow-up (RR 1.11, 95% CI 0.83 to 1.48). Twice as many HCWs sat sideways towards patients to avoid bodily fluids (23% vs 57%, RR 2.3, 95% CI 1.34 to 3.95). There was a marked decrease from 91% to 12% of HCWs holding thermometers at the recommended distance of 5–6 cm from patients (RR 0.23, 95% CI 0.12 to 0.43). Across rounds, the temperature check was applied without questioning for symptoms and risk factors if afebrile. In no case did a screener ask a patient about all symptoms and risk factors. HCWs described questioning as necessary to ‘determine the [epidemiological] link’ for case identification. Still, questioning patients was not viewed as particularly effective because individuals could ‘deny and hide the (link)’.

if afebrile. In no case did a screener ask a patient about all symptoms and risk factors. HCWs described questioning as necessary to ‘determine the [epidemiological] link’ for case identification. Still, questioning patients was not viewed as particularly effective because individuals could ‘deny and hide the (link)’. Some differences between baseline and follow-up regarding the doffing procedure were significant, including removing light PPE and gloves (light PPE, RR 2.54, 95% CI 1.32 to 4.88 and gloves, RR 4.09, 95% CI 1.34 to 12.49) and completion in correct order (RR 6.64, 95% CI 2.09 to 21.14). Doffing was compromised by the fact that a low proportion of HCWs removed PPE between screenings (14% at baseline and 32% at follow-up). Proportions of glove removal postscreening increased, but remained low (10% at baseline, 22% at follow-up). This was accompanied by a lack of handwashing of gloved or ungloved hands between screenings (11% at baseline, 19% at follow-up). HCWs expressed concern about PPE stock-outs, as well as the strain on incinerators that frequent glove and PPE disposal would cause. Among the 29 HCWs that removed gloves, all completed doffing in the correct order at follow-up. For consultations, low proportions of HCWs washed their hands before treating a patient (15% at baseline, 10% at follow-up) or after (39% at baseline, 16% at follow-up). Most HCWs put on a new pair of gloves at baseline (93%) and follow-up (91%), and a few kept the gloves on after treating the patient. Most HCWs stayed 1.5 m from patients (65% at baseline, 91% at follow-up).

ands before treating a patient (15% at baseline, 10% at follow-up) or after (39% at baseline, 16% at follow-up). Most HCWs put on a new pair of gloves at baseline (93%) and follow-up (91%), and a few kept the gloves on after treating the patient. Most HCWs stayed 1.5 m from patients (65% at baseline, 91% at follow-up). Discussion The EVD epidemic could be considered an overwhelming emergency in a series of severe epidemics (shigellosis and cholera) and endemic diseases (Lassa fever) in Sierra Leone that have required rigorous IPC.21–23 In the midst of the emergency response, we studied IPC in PHUs. This provided an exceptional opportunity to directly observe and evaluate adherence to IPC, and to work with HCWs to improve practice and discuss in detail the determinants of practice. The conviction among HCWs that IPC is lifesaving overrides the strong physical discomfort and distance with patients that it causes. During workshops, HCWs focused on improving screening, maintaining physical distance and encouraging patient handwashing; changes in these domains were reflected in the improvements seen in these behaviours at follow-up. Significant improvements were not consistent across behaviours, partly due to several high baseline values (>80%). While HCWs also discussed HCW handwashing, glove changing and the questioning for symptoms and risk factors, these were poorly adhered to across rounds.

improvements seen in these behaviours at follow-up. Significant improvements were not consistent across behaviours, partly due to several high baseline values (>80%). While HCWs also discussed HCW handwashing, glove changing and the questioning for symptoms and risk factors, these were poorly adhered to across rounds. Our study had important limitations. Uncontrolled before and after study designs lack a control group, thus limiting the ability to attribute changes observed to the intervention.17 Since we had a prior belief that the workshop and IPC improvement intervention would be beneficial, we believed that it would be unethical to observe IPC behaviours without intervening in a control group. Owing to the need to rapidly implement the study during a crisis, sample sizes of PHUs were intentionally small. The results are generalisable only to the PHUs included in the sample. The delay between the baseline and follow-up was short, though given the rapid progression of the epidemic, a study of short-term behaviour changes was warranted. The lack of pairing of HCWs between rounds is due to data collection being based on the availability of HCWs on the day of data collection rather than an explicit goal to conduct data collection on days when HCWs could be matched at follow-up. The implication of this limitation is that we cannot be sure that the all of those at follow-up were as exposed as those in the baseline. This likely leads to an underestimation of the intervention's effect. It is notable that staffing in PHUs is limited to a small pool of HCWs, and therefore, 63% of HCWs were the same at baseline and follow-up. As well, IPC improvement plans targeted changes at the PHU level, affecting all HCWs, not just those included in the baseline. There were gaps in fully implementing and prospectively monitoring the IPC improvement plans. Instead, we investigated changes in IPC retrospectively. At least one part of the observation protocol was apparently not adequately pretested; we think that the observations of thermometer placement at follow-up are likely specious. Transmission declined by December, limiting opportunities to assess IPC for isolation and body management; the number of HCWs observed was therefore small. Finally, HCWs who were interviewed may have been more motivated to practise IPC than those who fled during the peak of the epidemic.

follow-up are likely specious. Transmission declined by December, limiting opportunities to assess IPC for isolation and body management; the number of HCWs observed was therefore small. Finally, HCWs who were interviewed may have been more motivated to practise IPC than those who fled during the peak of the epidemic. Nonetheless, the quantitative and qualitative results were consistent. Attitudes towards IPC were favourable, but adherence with guidelines was markedly better for some behaviours than for others. HCWs consistently wore light PPE despite reporting persistent community fears. They described their own fear in detail, relating it to the unprecedented geographic expansion of the epidemic and the common experience of losing colleagues.9 We interpret this fear as being a driver for some IPC protocols. It is notable that during VHF outbreaks in Uganda and Democratic Republic of Congo, HCWs cited community resistance as a major reason for not wearing PPE in health facilities.5 24 In contrast, PPE use in this study was high, while glove changing and handwashing among HCWs, whether gloved or ungloved, were poor. This may also reflect a gap in knowledge among HCWs about how putting on or changing gloves before making contact with patients is necessary to improve patient safety.25 26 As gloves are fomites, changing and washing should be universal. HCW practices may be governed by the rules of rationality in disrupted health systems under normal circumstances, where chronic supply chain issues lead to widespread stock-out of PPE. Another area of uncertainty was the reported hesitation to use PPE for the management of ill colleagues. When faced with a real-life situation of an ill colleague, providers’ emotions may override their knowledge of safe practices, as seen during previous VHF epidemics.5 27 This presents an occupational risk for HCWs who are socially and emotionally challenged by their social group's tendency to not use PPE for one of their own. Overall, as transmission had abated, underlying emotions and competing priorities may foster a waning adherence to IPC.

as seen during previous VHF epidemics.5 27 This presents an occupational risk for HCWs who are socially and emotionally challenged by their social group's tendency to not use PPE for one of their own. Overall, as transmission had abated, underlying emotions and competing priorities may foster a waning adherence to IPC. Our findings reveal difficulties with screening protocols in PHUs. Identifying suspect cases before they enter the PHU is the foundation for IPC in the context of EVD.8 Across rounds, the protocol was followed incorrectly by applying the temperature check without questioning for symptoms and risk factors if afebrile. As HCWs cited the importance of establishing epidemiologic links, one explanation for their insufficient history taking may be low confidence in the protocol's effectiveness in detecting symptoms and epidemiological links due to patients’ assumed tendency to hide them. In PHUs, the majority of patients presenting for vaccination, antenatal care, and endemic diseases would not have been infected. Making the differential diagnosis of a suspect case relies heavily on the WHO case definition that specifies symptoms similar to malaria and typhoid.28 The lack of questioning may indicate that HCWs exercise prescreening to judge whether a patient appears ‘well’ or ‘ill’. Patients presenting for routine services in this study may have appeared well and HCWs may have given them a cursory temperature check without appropriately questioning for risk factors (in the absence of fever). This reliance on fever may be misguided; a cohort study of confirmed cases in a holding unit at Connaught Hospital in Freetown found a reduced sensitivity of the WHO case definition with 16% of confirmed cases presenting without fever.29

heck without appropriately questioning for risk factors (in the absence of fever). This reliance on fever may be misguided; a cohort study of confirmed cases in a holding unit at Connaught Hospital in Freetown found a reduced sensitivity of the WHO case definition with 16% of confirmed cases presenting without fever.29 The development of IPC systems in developing countries must address several core challenges to health systems: cost, procurement, a lack of knowledge and experience with IPC, and other cultural issues.26 In addition, IPC protocols may vary as the evidence base for some practices is lacking.30 31 It follows that the rapid scale-up of the Ebola IPC protocol in Sierra Leone has been a singular challenge. In the wake of the epidemic, the importance of IPC in primary care settings elsewhere in West Africa is gaining recognition through efforts to systematically address IPC in health facilities such as the Efficiency and Edification project in Burkina Faso, Senegal and Côte d'Ivoire.32 Notwithstanding the structural support and costs covered by Sierra Leone's national IPC programme, there are several opportunities to improve adherence via structural, social and behavioural interventions (table 5).33 First, the Ebola Response Consortium's longitudinal postintervention monitoring of structures, practices and supplies is necessary for identifying improvements needed and maintaining highly specialised supervision for staff and reiterating the importance of IPC.12 15 Second, training needs to address more complex determinants of adherence, for example, the dual aims of hand hygiene and glove changing in addressing different circumstances for contact with bodily fluids of an Ebola patient for occupational and nosocomial transmission. Explaining that gloves must be clean to protect HCWs, and their patients, is most imperative. Generating positive peer pressure through participation by colleagues and senior managers can also be a driver for adherence to hand hygiene.26 34 Using this logic, a group of HCWs’ belief in IPC and their ability to perform it may be key to achieving consistency. Third, during the foundational training, HCWs should be engaged early in discussing the care of ill colleagues and the need to implement IPC without compromise.

river for adherence to hand hygiene.26 34 Using this logic, a group of HCWs’ belief in IPC and their ability to perform it may be key to achieving consistency. Third, during the foundational training, HCWs should be engaged early in discussing the care of ill colleagues and the need to implement IPC without compromise. After an initial training, supportive supervision could probe and quell any doubts and assure the exhaustive screening of apparently healthy patients.5 Fourth, as community fears affect self-efficacy, sensitisation on PPE use in PHUs should be integrated into community engagement.6 Finally, other areas that we did not address in our study relate to the improvement of the tools of IPC which may increase HCW confidence in protocols. For instance, more research is needed to assess the effectiveness of different types of light PPE for healthcare settings31 35 36 and on the use of rapid diagnostic tests for clinical screening to improve the overall predictive value of screening for EVD.37–39 Table 5 Challenges to adherence to IPC in a primary health system Major challenge How addressed in December 2014–January 2015 Potential additional solutions Communities are unprepared for the systematic use of IPC and PPE in PHUs. HCWs sensitise community members as they come to PHU. Targeted communication campaign in community to set expectations Counselling approaches for HCWs to use in screening and consultation HCWs may not initially believe in the high risk of infection. Training to raise awareness of risks for HCW infection. Integrated IPC training in preservice education curricula

Major challenge How addressed in December 2014–January 2015 Potential additional solutions Communities are unprepared for the systematic use of IPC and PPE in PHUs. HCWs sensitise community members as they come to PHU. Targeted communication campaign in community to set expectations Counselling approaches for HCWs to use in screening and consultation HCWs may not initially believe in the high risk of infection. Training to raise awareness of risks for HCW infection. Integrated IPC training in preservice education curricula Reinforcement of in-service IPC training in particular for new staff Ongoing supportive supervision Low confidence in the identification of suspect cases. Training in screening. Research on new diagnostic techniques (eg, rapid diagnostic tests to increase sensitivity of the case definition and the overall effectiveness of screening) PPE causes separation in bond between HCWs and patients. HCWs found ways to motivate patients to recognise them. Guidance for HCW to increase communication and bonding with patients Regular meetings between HCW and health committee to discuss issues Discomfort while using light PPE on a routine basis. Training in PPE use. Technical improvements to light PPE Poor glove changing practices. Poor handwashing. Training in PPE use. Spot checking. Training that emphasises reasoning for appropriate use of PPE (including risks of not changing gloves) Peer systems that emphasise changing of gloves Monitoring for feelings of high self-efficacy in core behaviours among groups of HCWs Fear of PPE stock-out hinders use. Routine stocking of PPE. Improved supply chain

Poor glove changing practices. Poor handwashing. Training in PPE use. Spot checking. Training that emphasises reasoning for appropriate use of PPE (including risks of not changing gloves) Peer systems that emphasise changing of gloves Monitoring for feelings of high self-efficacy in core behaviours among groups of HCWs Fear of PPE stock-out hinders use. Routine stocking of PPE. Improved supply chain Training that emphasises reasoning for appropriate use of PPE Mixed attitudes towards using PPE with fellow HCWs. No specific actions known by the authors. Training that specifies HCW treatment scenarios and addresses doubts Implementation within a weak and fractured health system. IPC treated as emergency response. Improved supply chain systems Improved payment systems for human resources Improved coverage of functional water and sanitation infrastructure HCWs, healthcare workers; IPC, infection prevention and control; PHUs, peripheral health units; PPE, personal protective equipment. As Sierra Leone's recovery plan intends to make all PHUs compliant with national IPC protocol, understanding how behaviours can be optimised will be paramount in achieving this goal.40 EVD's re-emergence in Sierra Leone in January 2016 may have led to nosocomial transmission due to the patient's treatment seeking at a hospital.41 42 This underlines that the international community must continue to develop and support IPC in West Africa, in addition to surveillance and outbreak response mechanisms, to address future epidemics.

ra Leone in January 2016 may have led to nosocomial transmission due to the patient's treatment seeking at a hospital.41 42 This underlines that the international community must continue to develop and support IPC in West Africa, in addition to surveillance and outbreak response mechanisms, to address future epidemics. The authors thank the Ministry of Health and Sanitation of Sierra Leone for their support. They also thank Tamba Sam, Erin Stone and Paul Amendola from the IRC for their valuable help in facilitating this research and William E. Oswald for helpful discussions on the analysis. Handling editor: Valery Ridde. Twitter: Follow Lara Ho at @LaraSYHo Contributors: LSH, LM and RR developed the research idea. LSH, RR, HB, MB, RA and TK designed the study. HB, RA, LSH and LM undertook the implementation and data collection. RR, SM and LSH analysed the data. All authors interpreted the data, drafted or revised the paper and gave final approval for the paper to be published.

ibutors: LSH, LM and RR developed the research idea. LSH, RR, HB, MB, RA and TK designed the study. HB, RA, LSH and LM undertook the implementation and data collection. RR, SM and LSH analysed the data. All authors interpreted the data, drafted or revised the paper and gave final approval for the paper to be published. Funding: This work was supported by the Research for Health in Humanitarian Crises (R2HC) Programme, managed by ELRHA (SCUK—accountable grant number 13488). The Research for Health in Humanitarian Crises (R2HC) programme aims to improve health outcomes by strengthening the evidence base for public health interventions in humanitarian crises. Visit http://www.elrha.org/work/r2hc for more information. The £8 million R2HC programme is funded equally by the Wellcome Trust and DFID, with Enhancing Learning and Research for. Humanitarian Assistance (ELRHA) overseeing the programme's execution and management. The funder had no role in study design, data collection, analysis, interpretation or writing. Competing interests: MB reports grants from the International Rescue Committee (IRC), during the conduct of the study. SM reports personal fees for conducting analysis from the IRC, during the conduct of the study. Patient consent: Written consent was obtained from healthcare workers. Ethics approval: The study received ethics approval from Durham University's Institutional Review Board and the Sierra Leone Ethics and Scientific Research Committee. Provenance and peer review: Not commissioned; externally peer reviewed.

Competing interests: MB reports grants from the International Rescue Committee (IRC), during the conduct of the study. SM reports personal fees for conducting analysis from the IRC, during the conduct of the study. Patient consent: Written consent was obtained from healthcare workers. Ethics approval: The study received ethics approval from Durham University's Institutional Review Board and the Sierra Leone Ethics and Scientific Research Committee. Provenance and peer review: Not commissioned; externally peer reviewed. Data sharing statement: Owing to ethical restrictions related to confidentiality, data are available on request by contacting Ruwan Ratnayake (ruwan.ratnayake@rescue.org).

Key questions What is already known about this topic? Rapid diagnostic tests for malaria (mRDT) now form a central element of malaria care. The scale-up of mRDTs through retail outlets is being pursued by several governments in Eastern and Southern Africa. Existing evaluations focus on the intended consequences of mRDT use. What are the new findings? This study examined the unintended consequences of the introduction of mRDTs into registered drug shops in Mukono, Uganda. The popularity of the tests appeared to be linked to a rise in the status of drug shop vendors. Drug shops with mRDTs appeared more attractive places to seek care, sales of medicines increased and treatment seeking appeared to shift out of the formal sector providers and into these shops. Recommendations for policy Policy makers need to be aware of the unintended consequences of interventions before they decide on whether they are appropriate or not. This is especially the case in settings in which there is little regulation and oversight of practice. If mRDTs are introduced into drug shops in Uganda, a significant increase in surveillance and supervision is likely to be necessary in order to guard against poor quality, inequitable practices that may increase out of pocket payments among those who can least afford them.

Recommendations for policy Policy makers need to be aware of the unintended consequences of interventions before they decide on whether they are appropriate or not. This is especially the case in settings in which there is little regulation and oversight of practice. If mRDTs are introduced into drug shops in Uganda, a significant increase in surveillance and supervision is likely to be necessary in order to guard against poor quality, inequitable practices that may increase out of pocket payments among those who can least afford them. Introduction Across East Africa, urban and rural citizens’ use of retail outlets to access pharmaceutical medicines to treat episodes of febrile illness is well established.1–6 Current framings of the global drive to control and eliminate malaria often place this observation at the core research agendas, programme financing and national decision-making processes.7

al citizens’ use of retail outlets to access pharmaceutical medicines to treat episodes of febrile illness is well established.1–6 Current framings of the global drive to control and eliminate malaria often place this observation at the core research agendas, programme financing and national decision-making processes.7 This interest in providing malaria medication through the retail sector has coincided with a shift in global policy from the presumptive treatment of all fevers as malaria to the targeted treatment of parasitologically confirmed malaria, a policy to be realised in large part through the introduction of a rapid diagnostic test for malaria (mRDT).8 For many, this has made the successful introduction of mRDTs into the retail sector imperative.9 10 For others, the importance of introducing mRDTs has been underscored by the results of the Affordable Medicines Facility for Malaria (AMFm) project, which sought to flood private sector markets with affordable, quality assured artemisinin combination therapy (ACT), to crowd out less efficacious antimalarials.11 12 AMFm achieved its goal of improving access through changes in the private for profit sector, but in so doing, it precipitated a substantial increase in the use of ACTs among those who did not have malaria.13 This raised concerns about the cost-effectiveness of the subsidy, as well as whether it would favour the development and spread of parasite resistance to artemisinin and risk the future efficacy of the drug.

t in so doing, it precipitated a substantial increase in the use of ACTs among those who did not have malaria.13 This raised concerns about the cost-effectiveness of the subsidy, as well as whether it would favour the development and spread of parasite resistance to artemisinin and risk the future efficacy of the drug. Despite the fact that introducing mRDTs into private sector markets appears a pragmatic solution to the overuse of ACTs in retail outlets9 10 and has been enthusiastically taken up by many national-level and global-level policymakers, studies on the impact of the mRDT have shown varied results and been limited in scope.11 13 14 In the retail sector in Kenya, while mRDTs were reported to have been popular among clients, they did not substantially reduce the numbers of ACTs sold to people who tested negative.11 In Uganda, two studies reported that it is feasible and safe to introduce mRDTs into registered drug shops. In the first study, the response to the introduction of mRDTs was heterogeneous in terms of uptake and adherence to test results.13 In the second study, however, mRDTs were taken up enthusiastically by registered drug shop vendors (DSVs) and ACTs were appropriately targeted at 79% of patients in the mRDT arm as opposed to 33.7% in the presumptive treatment arm.15

duction of mRDTs was heterogeneous in terms of uptake and adherence to test results.13 In the second study, however, mRDTs were taken up enthusiastically by registered drug shop vendors (DSVs) and ACTs were appropriately targeted at 79% of patients in the mRDT arm as opposed to 33.7% in the presumptive treatment arm.15 While these studies have been concerned with ascertaining whether and how mRDTs can be introduced, holistic accounts that describe the mechanisms that their successful introduction is based on and the unintended consequences of their introduction are lacking. This constitutes a considerable gap in the literature as all interventions have impacts on society that cannot be foreseen and which must be attended to in the overall evaluation of an intervention.16 17 This is especially the case in places such as drug shops that are poorly regulated and in which poor practice is unlikely to be influenced or curtailed by government policy and surveillance.

have impacts on society that cannot be foreseen and which must be attended to in the overall evaluation of an intervention.16 17 This is especially the case in places such as drug shops that are poorly regulated and in which poor practice is unlikely to be influenced or curtailed by government policy and surveillance. This paper adds to the literature on drug shops and mRDTs by providing an anthropologically informed analysis of the unintended consequences of the introduction of mRDTs into the retail sector in Uganda. It explores the introduction of mRDTs into drug shops as a process of assemblage, focusing on how mRDTs were arranged, organised and interacted with other objects, people, medicines, desires, forms of regulation and everyday practice found in drug shops in Mukono District. In so doing, the paper shows that the popularity of mRDTs among both DSVs and their clients was not a straightforward process of the adoption of a technology that rationalised everyday practice. Instead, we show how the use and popularity of the test was intimately connected to local concerns about the trustworthiness of DSVs, the quality of services on offer in these shops, and the efficacy of medicines and relationship between drug shops and formal health providers.

ology that rationalised everyday practice. Instead, we show how the use and popularity of the test was intimately connected to local concerns about the trustworthiness of DSVs, the quality of services on offer in these shops, and the efficacy of medicines and relationship between drug shops and formal health providers. Theoretical background Anthropological approaches to evaluation, with a commitment to inductive, holistic research, are well suited to analysing the consequences of programmes beyond those specified in proposals and protocols. It has been suggested by Kleinman17 that this type of analysis is one of the key contributions that medical anthropologists can make to global health.18 19 His interest stems from Merton's original formulation of the unanticipated consequences of purposive social action as the unforeseen effects of actions with a motive, a choice among alternatives, a goal and a process (such large-scale government programmes, but also complex intervention trials).16 17 Merton offers several reasons for their emergence including unforeseen shifts in context (such as ongoing economic or social changes); the immediacy of the interests of those formulating the programme; and the institutional values that shape the thoughts of programme managers or policymakers which make it hard to see beyond their particular paradigm.16 While this approach is useful in pointing to the complexity of introducing programmes to promote social change, it offers relatively little in terms of understanding how local, everyday practices shape interventions. Moreover, it offers no means of understanding how interconnections between the intended and unintended consequences of an intervention may come about and, importantly for this paper, how one may influence the other.

it offers relatively little in terms of understanding how local, everyday practices shape interventions. Moreover, it offers no means of understanding how interconnections between the intended and unintended consequences of an intervention may come about and, importantly for this paper, how one may influence the other. This paper draws on post-structuralist theories of the microprocesses of sociomaterial relations, sometimes referred to in the literature as theories of assemblage, to better understand and develop an analysis of the unintended consequences of introducing mRDTs into the retail sector. Drawing on the French philosopher Deleuze, these theories have been influential in anthropology and sociology for their ability to account for the interaction between society, science and technology and local processes of social change.20 21 This gives ideas of assemblage an obvious relevance to the analysis of the introduction of mRDTs and also provides a pertinent frame for understanding Ugandan drug shops more generally. Like other ethnographic and social analyses of drug sellers and informal private providers,22 23 the literature on Ugandan drug shops has shown these to be spaces that provide an awkward mix of services, combining disparate elements from different social arenas and in which those selling drugs do not necessarily adhere to a single logic (biomedical or economic, scientific or religious) but are more likely to draw on and display forms of practice that come from multiple institutional settings.3 24 Analyses that attend to the structural position of drug sellers tend to present them as existing in an odd and often insecure position, as liminal actors occupying in an in-between social space.3 Post-structuralist accounts attend to the ways in which social fields are assembled rather than structured and suggest that the analyst shifts the focus. Instead of mapping the contradictions that shoot through everyday practice, findings seek to account for the ways in which heterogeneous elements of the social and material world (from medicines, to social norms, desires, people, buildings and illness) come together and cohere, influencing and impacting on one another.25

stead of mapping the contradictions that shoot through everyday practice, findings seek to account for the ways in which heterogeneous elements of the social and material world (from medicines, to social norms, desires, people, buildings and illness) come together and cohere, influencing and impacting on one another.25 In terms of accounting for the intended consequences of the introduction of mRDTs into drug shops, theories of assemblage draw our attention to the ways in which programmes or projects work by establishing relationships between the mRDT and the use of malaria medicines and that they fail when they do not. Yet, as this intended relationship is established, relations between the other social and material elements within the shop emerge and it is these processes (sometimes quite self-consciously arranged and organised by DSVs and their clients) that create the unintended consequences of the intervention.

when they do not. Yet, as this intended relationship is established, relations between the other social and material elements within the shop emerge and it is these processes (sometimes quite self-consciously arranged and organised by DSVs and their clients) that create the unintended consequences of the intervention. A focus on the unintended consequences through the lens of assemblage theories therefore makes the connections (or sociomaterial relations) between already existing elements of the drug shop and those that are put in place through the actions of those implementing a project or programme the key sites for analysis. Through a focus on everyday practice, this paper analyses the processes through which mRDTs became part of the assemblages that constitute the drug shops in Mukono. Combining qualitative and quantitative data, it explores three sets of connections that were key to shaping the uses of the mRDT in the drug shops: local markets, regulation, diagnosis and trust; blood tests, disease and the generation of trust between DSV and their clients; mRDTs, pharmaceutical markets and the desire for effective medicine.

d quantitative data, it explores three sets of connections that were key to shaping the uses of the mRDT in the drug shops: local markets, regulation, diagnosis and trust; blood tests, disease and the generation of trust between DSV and their clients; mRDTs, pharmaceutical markets and the desire for effective medicine. Study setting Between October 2010 and December 2011, a cluster-randomised trial was carried out in 59 registered drug shops grouped into 20 geographical clusters in a periurban area in Mukono, Uganda. Staff from 65 drug shops had been invited to join the trial, though 6 shops failed to attend the training and so were excluded. Following invitation and consent to join the project, clusters of drug shops were randomised to receive training in one of two methods to diagnose malaria: 10 clusters were trained to use clinical signs and symptoms (the presumptive arm), and 10 clusters were trained to use mRDTs (the mRDT arm). All DSVs attended a 3-day training in taking patient history, recognising malaria, prescribing ACTs and rectal artesunate, when and how to refer patients into the health service. In addition, the DSVs in the intervention arm received an additional day’s training on when and how to use an mRDT. DSVs were also trained in record keeping and preparing blood slides for the reference microscopy conducted on each client, which were used by the trial to evaluate whether the treatment decisions were appropriate or not. All DSVs were given ACTs for free (artemether-lumefantrine) to sell to patients at an agreed low retail price (child: 1000 Uganda shilling, adult: 3000 Uganda shilling, equivalent to US$0.40 and US$1.20, respectively), disposable gloves, cotton wool, lancets, blood slides, safety boxes for sharps disposal, triplicate carbon copy patient registers for record keeping for the project, and to provide a referral form should the client be considered too ill to be treated in the drug shop. All DSVs in the mRDT arm were additionally provided with a continuous supply of mRDTs for free, and were asked to sell these to patients at a fixed price (500 Uganda shilling, equivalent to US$0.20). The DSVs in the mRDT intervention arm were provided with signs to place outside their shop advertising the availability of mRDTs.

in the mRDT arm were additionally provided with a continuous supply of mRDTs for free, and were asked to sell these to patients at a fixed price (500 Uganda shilling, equivalent to US$0.20). The DSVs in the mRDT intervention arm were provided with signs to place outside their shop advertising the availability of mRDTs. In both arms, community sensitisation through Village Health Teams informed local residents about mRDTs underlining the fact that not all fevers are malaria; that it was beneficial to test for malaria before providing treatment and that mRDTs were available locally from trained DSVs and public health facilities. A detailed description of the various elements of the intervention is provided in the main trial paper.15

RDTs underlining the fact that not all fevers are malaria; that it was beneficial to test for malaria before providing treatment and that mRDTs were available locally from trained DSVs and public health facilities. A detailed description of the various elements of the intervention is provided in the main trial paper.15 Methods Towards the end of the trial, between November 2011 and February 2012, research was undertaken to provide an anthropologically informed, holistic account of the intervention by exploring the processes and everyday practices involved with the introduction and incorporation of mRDTs into drug shops and their affect on the referral of clients from drug shops to other health facilities. The research was particularly concerned with the relationships between the different elements of the trial, and the ways in which they impacted on the local context of health seeking, the provision of healthcare and the perception and experiences of conducting an mRDT among providers and care seekers. The methods used reflect the COREQ recommendations for qualitative research26 and the relationships between the researchers and the participants have been described in detail elsewhere.27 Focus group discussions Focus group discussions (FGDs) were conducted with three subgroups: drug shop clients, DSVs and health workers at local public facilities and a private not for profit hospital. In all, 21 FGDs were conducted (see table 1) and each FGD had at least 5 and no more than 13 participants.

Methods Towards the end of the trial, between November 2011 and February 2012, research was undertaken to provide an anthropologically informed, holistic account of the intervention by exploring the processes and everyday practices involved with the introduction and incorporation of mRDTs into drug shops and their affect on the referral of clients from drug shops to other health facilities. The research was particularly concerned with the relationships between the different elements of the trial, and the ways in which they impacted on the local context of health seeking, the provision of healthcare and the perception and experiences of conducting an mRDT among providers and care seekers. The methods used reflect the COREQ recommendations for qualitative research26 and the relationships between the researchers and the participants have been described in detail elsewhere.27 Focus group discussions Focus group discussions (FGDs) were conducted with three subgroups: drug shop clients, DSVs and health workers at local public facilities and a private not for profit hospital. In all, 21 FGDs were conducted (see table 1) and each FGD had at least 5 and no more than 13 participants. Table 1 Focus group discussions: numbers, participants, recruitment, stratification and exclusion criteria

Focus group discussions Focus group discussions (FGDs) were conducted with three subgroups: drug shop clients, DSVs and health workers at local public facilities and a private not for profit hospital. In all, 21 FGDs were conducted (see table 1) and each FGD had at least 5 and no more than 13 participants. Table 1 Focus group discussions: numbers, participants, recruitment, stratification and exclusion criteria DSV Health worker Drug shop client Number of FGDs in the RDT arm 4 5 4 Number of FGDs in the presumptive arm 3 3 2 Total FGDs 7 8 6 Total FGD participants 54 71 54 Recruitment method From project records From visits to the health facilities From project records Stratification By arm and frequency of referral By arm and level of health facility By referral for carers (for children) or adult seeking care for febrile illness; or those who had visited a shop for care for febrile illness Exclusion criteria Those who had worked in the area for <6 months Those who had worked in the area for <6 months Those who had not visited a drug shop in the past 6 months, children under the age of 18 DSV, drug shop vendor; FGD, focus group discussion; RDT, rapid diagnostic test.

ed a shop for care for febrile illness Exclusion criteria Those who had worked in the area for <6 months Those who had worked in the area for <6 months Those who had not visited a drug shop in the past 6 months, children under the age of 18 DSV, drug shop vendor; FGD, focus group discussion; RDT, rapid diagnostic test. The participants were given a description of the project when they were invited to participate in the FGD. On the day of the FGD, the project was described again and the participants were asked to sign a sheet to consent in the research. They were advised that at any point they could leave the FGD and that there was no requirement for them to stay. Focus groups were conducted in Luganda and/or English, and in all FGDs notes were taken and the discussion recorded digitally.

gain and the participants were asked to sign a sheet to consent in the research. They were advised that at any point they could leave the FGD and that there was no requirement for them to stay. Focus groups were conducted in Luganda and/or English, and in all FGDs notes were taken and the discussion recorded digitally. The FGDs were conducted in a Sunday school building except for two that were conducted at health facilities. Each focus group was attended by three social scientists acting as facilitator, note taker or organiser of the session. The topics discussed were local treatment-seeking behaviour, experiences at drug shops, experiences with rapid diagnostic tests and experiences with referral. The latter was a particular interest in the study and this focus on referral from drug shops into public sector health facilities was the reason behind the inclusion of health workers from local health centres and hospitals in the research. An emphasis was placed on describing and understanding actual experience rather than eliciting generalised or ideal statements. The facilitators also allowed conversations to develop between participants when important but unanticipated themes arose. Participants received money to cover the cost of their travel and refreshments. Recordings were transcribed, translated in English and imported into Nvivo (QSR International, Cambridge, Massachusetts, USA), a qualitative data analysis software package for the aggregation of data by codes.

The FGDs were conducted in a Sunday school building except for two that were conducted at health facilities. Each focus group was attended by three social scientists acting as facilitator, note taker or organiser of the session. The topics discussed were local treatment-seeking behaviour, experiences at drug shops, experiences with rapid diagnostic tests and experiences with referral. The latter was a particular interest in the study and this focus on referral from drug shops into public sector health facilities was the reason behind the inclusion of health workers from local health centres and hospitals in the research. An emphasis was placed on describing and understanding actual experience rather than eliciting generalised or ideal statements. The facilitators also allowed conversations to develop between participants when important but unanticipated themes arose. Participants received money to cover the cost of their travel and refreshments. Recordings were transcribed, translated in English and imported into Nvivo (QSR International, Cambridge, Massachusetts, USA), a qualitative data analysis software package for the aggregation of data by codes. Coding and analysing qualitative data is an interpretive endeavour and each FGD was coded by one social scientist who had collected the data with close support and supervision from the anthropologist (EH). A coding scheme for each group was drawn up through a combination of themes emerging from the transcript coupled with predefined areas of interest identified from the main intervention (mostly the interaction between different components of the trial) and from anthropological theory on informal health providers. The coding scheme was adapted as necessary to reflect new themes emerging and the anthropologist provided input at the conceptual level of analysis. The main findings were discussed with the principal investigator of the main trial, the anthropologist and the project's social scientists during a series of face-to-face meetings that took place everyday over the course of a week.

emes emerging and the anthropologist provided input at the conceptual level of analysis. The main findings were discussed with the principal investigator of the main trial, the anthropologist and the project's social scientists during a series of face-to-face meetings that took place everyday over the course of a week. Follow-up questionnaires DSVs in all clusters recorded the names and contact details of clients who visited their shops for a fever or history of fever. Interviewers then visited DSVs according to a randomised schedule to identify recent clients and carried out questionnaire interviews (n=646) at the client's home after 4 days, and again between 10 and 14 days after visiting a drug shop. The questionnaires were conducted to elicit reports on the economic costs associated with treatment of the recent febrile episode at a drug shop, whether they had been offered and had accepted to be tested with an mRDT, their reported results; the type and quantities of drugs purchased and any subsequent costs over the next 10–14 days; and exploring perceptions of mRDTs, their knowledge and understanding of the tests, treatment-seeking behaviour and referrals. For clients who were under 18, questionnaires were conducted with the adult who attended the drug shop with the child; clients under 18 who attended the DSV alone were excluded from the study. Clients who did participate or consent to both questionnaires or did not meet the overall study inclusion criteria were excluded from the follow-up analysis. The analysis of open questions was coded by CH using Nvivo (QSR International, Cambridge, Massachusetts, USA) and, where appropriate, imported into Stata V.13 (StataCorp LP) and converted into ordinal or categorical variables for descriptive quantitative analysis in conjunction with quantitative responses.

up analysis. The analysis of open questions was coded by CH using Nvivo (QSR International, Cambridge, Massachusetts, USA) and, where appropriate, imported into Stata V.13 (StataCorp LP) and converted into ordinal or categorical variables for descriptive quantitative analysis in conjunction with quantitative responses. Ethical clearance was granted by the London School of Hygiene and Tropical Medicine and the Uganda National Council for Science and Technology. Results Context: diagnostic practice and the generation of trust According to government policy, registered drug shops are places in which over-the-counter medication may be bought and sold. Legally, drug shops are distinct from private clinics and expected to provide a limited range of products for sale, which excludes prescription medication such as antibiotics. In Mukono, however, categories of drug sellers were fluid and clients were unable to categorise private sector providers effectively during FGDs. The fluidity of this categorisation was also reflected in the heterogeneous nature of the drug shops in the trial. Most DSVs reported that they provided diagnostic advice and sales of medicine only but at least 2 out of the 59 study shops also offered in-patient services and provided intravenous medication. At some, the registered owner (who by law had to have some formal biomedical training) would be in the shop for most of its opening hours, providing relatively constant access to a health provider.

dicine only but at least 2 out of the 59 study shops also offered in-patient services and provided intravenous medication. At some, the registered owner (who by law had to have some formal biomedical training) would be in the shop for most of its opening hours, providing relatively constant access to a health provider. As has been reported elsewhere among clients in Uganda, the care seekers attending shops in Mukono appreciated the good social relations often found in these spaces3 but maintained a generalised suspicion of DSVs’ motives and skills.27 DSVs could be respectful and kind; attentive to the psychological, social and spiritual needs of their clients; understanding of and responsive to their economic constraints (providing of credit and selling cheaper or incomplete doses) and keen to use their skill and expertise. As one woman who had visited a drug shop in the mRDT arm of the trial with her child explained, these social relations often revolved around a desire for powerful pharmaceutical drugs to manage illness in them and in their children.We had five thousand shillings (2 US$) in our pocket. But because the sickness was severe, she wrote us a bill of eighteen thousand shillings (7 US$). Yet, she had given us a plenty of drugs in a polythene bag. Then I told her, “I have come with five thousand [shillings],” and she said, “Aha, there is no problem. You can even come back tomorrow”, and she didn't even know us…[later in the discussion]…This health worker gives you drugs and you also look at it and say, “as I have never seen such a quantity of drugs!” and you simply say that. Oh, but the drugs are worth the money you pay!

,” and she said, “Aha, there is no problem. You can even come back tomorrow”, and she didn't even know us…[later in the discussion]…This health worker gives you drugs and you also look at it and say, “as I have never seen such a quantity of drugs!” and you simply say that. Oh, but the drugs are worth the money you pay! FGD with mothers in the mRDT arm of the trial Attending a drug shop was also a risky act. Businesses were infrequently visited by government regulators. In all three groups of participants in the FGDs, the following narratives echoed: drug shops are often run by those who claimed expertise but had none, who knowingly sold stolen, dangerous, expired or counterfeit drugs and were only interested in making as large a profit as possible. Accounts of particular instances of receiving poor quality care at drug shops and its consequences for children's health were commonplace. As previously described by Birungi,28 injections were a popular form of medicine, a sign of good care but raised particular concerns about the infection and illness that could be caused when unskilled DSVs administered injections to small children. Moreover, in the context of ongoing concerns about how to pay for treatment, clients expressed concern that they could be overcharged for medication by exploitative DSVs.

re but raised particular concerns about the infection and illness that could be caused when unskilled DSVs administered injections to small children. Moreover, in the context of ongoing concerns about how to pay for treatment, clients expressed concern that they could be overcharged for medication by exploitative DSVs. Where some FGD participants considered it to be almost impossible to distinguish between skilled and unskilled, trustworthy and untrustworthy DSVs, others described how they acted to protect themselves against the risks associated with attending a drug shop. Previous treatment success, recommendations from neighbours and the DSV's diagnostic assessments played a central role in identifying an effective, good quality and caring provider. Below, a group of women described how they evaluated a DSV's skill through the questions that s/he asked about the child's symptoms.P6: When you look there, when you are just passing by, or maybe if you have had your child fall sick, some people base [their assessment of a drug shop] on the way the drug shop looks and then see whether they [the DSVs] are skilled, drugs are in plenty. But, when you pass there, [you may see that] the drugs that are there are few. Now you can despise the place yet in the real, actual sense she has the idea. SS1: What do you mean by that thing of ‘idea’? P2: What she wants to say is that she knows what she does. SS2: How can you come to know that someone is knowledgeable?

Where some FGD participants considered it to be almost impossible to distinguish between skilled and unskilled, trustworthy and untrustworthy DSVs, others described how they acted to protect themselves against the risks associated with attending a drug shop. Previous treatment success, recommendations from neighbours and the DSV's diagnostic assessments played a central role in identifying an effective, good quality and caring provider. Below, a group of women described how they evaluated a DSV's skill through the questions that s/he asked about the child's symptoms.P6: When you look there, when you are just passing by, or maybe if you have had your child fall sick, some people base [their assessment of a drug shop] on the way the drug shop looks and then see whether they [the DSVs] are skilled, drugs are in plenty. But, when you pass there, [you may see that] the drugs that are there are few. Now you can despise the place yet in the real, actual sense she has the idea. SS1: What do you mean by that thing of ‘idea’? P2: What she wants to say is that she knows what she does. SS2: How can you come to know that someone is knowledgeable? P6: For me what I can explain, you can take a child and you tell her, “Health worker, I feel the child has high temperature.” You have not yet told her that it seems the child is sick with malaria or what but for her she starts asking you that…that is to say she knows the conditions of children. Most of the time she asks you, “Has he done this, do you feel his head is a bit hot (some high temperature), how was he at night, did he at any time have diarrhoea?” Okay things like that. That is to say, you have not told her but she says correct things and you tell her, “yes, health worker that is how it was”, before she measures his temperature to see the temperature…and she asks you about the condition of the child, or how he was, whether it is malaria or it is another thing.

” Okay things like that. That is to say, you have not told her but she says correct things and you tell her, “yes, health worker that is how it was”, before she measures his temperature to see the temperature…and she asks you about the condition of the child, or how he was, whether it is malaria or it is another thing. FGD with mothers in the non-RDT arm The diagnostic process underway in these drug shops that existed prior to or without the introduction of mRDTs had a productive role that went beyond the imperative of identifying disease in order to provide treatment. In this heterogeneous and poorly regulated market, it was an important demonstration of medical knowledge, and acted on the relationship between the DSV and the client, engendering trust in a space in which questions around motivation of the vendor and the type of service that they provided were under constant scrutiny.

n this heterogeneous and poorly regulated market, it was an important demonstration of medical knowledge, and acted on the relationship between the DSV and the client, engendering trust in a space in which questions around motivation of the vendor and the type of service that they provided were under constant scrutiny. mRDTs’ impact on decision-making about where to seek care According to the FGDs, mRDTs had not been available in the drug shops in Mukono prior to the intervention but when they were introduced, they were popular among DSVs and clients. Characteristics of the clients interviewed on day 4 after drug shop consultation are presented in table 2. Following refusal of consent (12) and loss to follow-up (130), clients or carers comprised 251 who had visited a drug shop in the presumptive diagnosis arm and 253 in the mRDT arm of the trial. In both arms of the trial, half of those seeking care and medication from a drug shop were children under the age of 5, at 49.4% in the presumptive arm and 51.8% in the mRDT arm. Of the 253 clients presenting with symptoms of a febrile illness in the mRDT arm, 97.6% reported that they had purchased an mRDT. There were occasional reports of mRDTs also being used in the presumptive arm (7/251 cases), findings that correspond to those from the FGDs. This suggests that certain DSVs were making the choice to purchase and offer these tests for sale (see table 2). Table 2 Characteristics of drug shop clients followed up and interviewed at home 4 days after the drug shop visit

mRDTs’ impact on decision-making about where to seek care According to the FGDs, mRDTs had not been available in the drug shops in Mukono prior to the intervention but when they were introduced, they were popular among DSVs and clients. Characteristics of the clients interviewed on day 4 after drug shop consultation are presented in table 2. Following refusal of consent (12) and loss to follow-up (130), clients or carers comprised 251 who had visited a drug shop in the presumptive diagnosis arm and 253 in the mRDT arm of the trial. In both arms of the trial, half of those seeking care and medication from a drug shop were children under the age of 5, at 49.4% in the presumptive arm and 51.8% in the mRDT arm. Of the 253 clients presenting with symptoms of a febrile illness in the mRDT arm, 97.6% reported that they had purchased an mRDT. There were occasional reports of mRDTs also being used in the presumptive arm (7/251 cases), findings that correspond to those from the FGDs. This suggests that certain DSVs were making the choice to purchase and offer these tests for sale (see table 2). Table 2 Characteristics of drug shop clients followed up and interviewed at home 4 days after the drug shop visit Clients who visited drug shops in the presumptive arm N=251 Clients who visited drug shops in the RDT arm N=253 n Per cent n Per cent Age of client (years) 0–1 28 11.2 42 16.6 1–5 96 38.2 89 35.2 5–18 75 29.9 54 21.3 18–45 40 15.9 49 19.4 45+ 12 4.8 19 7.5 Sex of participant Female 137 54.6 122 48.2 Male 114 45.4 131 51.8 Role of respondent Client (18+) 45 17.9 59 23.3 Mother 157 62.5 141 55.6 Head of household 19 7.6 22 8.7 Other guardian/carer* 30 12.0 31 12.3 Reported symptom(s) Diarrhoea 23 9.2 23 9.2 Cough or influenza 93 37.0 90 35.7 Fever 130 51.8 115 45.6 Headache 57 22.7 50 19.8 Vomiting 19 7.6 34 23.5 Malaria 26 10.4 21 8.4 Went elsewhere before this DSV Yes 107 42.6 114 45.1 No 140 55.8 128 50.6 Did not report anything 4 1.6 11 4.3 Purchased an mRDT Yes 7 2.8 247 97.6 No 244 97.2 6 2.42 *Where the adult patient or carer of the sick child was not available for interview, and the household head identified themselves as sufficiently knowledgeable about the illness episode to be able to answer questions about it.

Did not report anything 4 1.6 11 4.3 Purchased an mRDT Yes 7 2.8 247 97.6 No 244 97.2 6 2.42 *Where the adult patient or carer of the sick child was not available for interview, and the household head identified themselves as sufficiently knowledgeable about the illness episode to be able to answer questions about it. DSV, drug shop vendor; mRDT, rapid diagnostic test for malaria; RDT, rapid diagnostic test. Almost three-quarters of these care seekers in the mRDT arm of the trial (72% of the 253 respondents) reported in the questionnaires that they knew about the availability of mRDTs prior to the recent visit to the shop. Of those, the majority (73% of the 172 respondents) considered that this had an impact on their decision-making and why they had chosen to seek care in a particular drug shop. During the FGDs, health workers also noted changes in treatment-seeking patterns.P4: It [the introduction of mRDTs] has not affected us badly because it has helped to reduce the number of patients [that we see] and they [the DSVs] also get some money…[laughter]… P2: In my opinion, if a person has a sick child and they go to the drug shop and they take off some blood in order to really confirm that she has malaria they get to know the truth. They [the client] are happy because instead of going to the hospital they are being treated nearer their home. FGD with health workers from HC II in the intervention arm site

P2: In my opinion, if a person has a sick child and they go to the drug shop and they take off some blood in order to really confirm that she has malaria they get to know the truth. They [the client] are happy because instead of going to the hospital they are being treated nearer their home. FGD with health workers from HC II in the intervention arm site P5: We now get fewer cases of malaria at our hospital. You know what that means for our income, which is a bit negative for us but for the community it's okay because they have the services nearer. FGD with health workers in HC IV and the not-for-profit hospital As mRDTs were introduced into drug shops, the process of finding out about disease maintained its ability to engender trust between the vendor and the client and the benefits of mRDTs were interpreted through this lens. The first quotation below links this to a process of being able to see and then act on a disease, while the second connected the care available at drug shops to the larger health facilities.Okay for me the different thing that I have seen, we have been suffering and what made us like the hospital are the blood tests. Yet, in these drug shops, they would simply inject you without testing your blood to see what type of fever it is. But for me the different thing that I have seen now, [is that] they [DSVs] test blood. Now we have come to completely dislike the public sector hospital because that [blood testing] is what has been mostly taking us there. FGD with mothers in the mRDT arm of the trial

As mRDTs were introduced into drug shops, the process of finding out about disease maintained its ability to engender trust between the vendor and the client and the benefits of mRDTs were interpreted through this lens. The first quotation below links this to a process of being able to see and then act on a disease, while the second connected the care available at drug shops to the larger health facilities.Okay for me the different thing that I have seen, we have been suffering and what made us like the hospital are the blood tests. Yet, in these drug shops, they would simply inject you without testing your blood to see what type of fever it is. But for me the different thing that I have seen now, [is that] they [DSVs] test blood. Now we have come to completely dislike the public sector hospital because that [blood testing] is what has been mostly taking us there. FGD with mothers in the mRDT arm of the trial P1: The majority of us used to go to those big health facilities but now we go there to drug shops and clinics and they give you treatment. P5: I think it [testing] has been good for them [the drug shops vendors] because the majority of people have trust in them. Previously, they would despise those clinics. Facilitator: Previously you never used to put trust in them. Why were you going to other health facilities in the past? P4: Blood, they never used to test blood. Many Ps: Hmmm, blood FGD with mothers in the mRDT arm of the trial

P5: I think it [testing] has been good for them [the drug shops vendors] because the majority of people have trust in them. Previously, they would despise those clinics. Facilitator: Previously you never used to put trust in them. Why were you going to other health facilities in the past? P4: Blood, they never used to test blood. Many Ps: Hmmm, blood FGD with mothers in the mRDT arm of the trial Refracted through this new logic of testing before treating illness, previous diagnostic practices in the drug shop were conceptually reconfigured. The practice of describing and suggesting symptoms so that they could be matched to available medicines was often recast as guess work and was contrasted with the definitive knowledge that the mRDT provided.I think work has been simplified because those days, before this project started, we would just try to guess what's going on, how to treat the patient. But after testing, I think we know exactly what we are doing and even the patients themselves realize it and that's why they come. But of course, those days they [patients] would have that negative…what? [attitude]…but now they know if you go there, they test you and they give you…what?…[They] will cure you. Of course that confidence has increased in them. Ya, they are confident that they going to get what? [That they are going to get] okay. FGD with DSVs in the mRDT arm of the trial

Refracted through this new logic of testing before treating illness, previous diagnostic practices in the drug shop were conceptually reconfigured. The practice of describing and suggesting symptoms so that they could be matched to available medicines was often recast as guess work and was contrasted with the definitive knowledge that the mRDT provided.I think work has been simplified because those days, before this project started, we would just try to guess what's going on, how to treat the patient. But after testing, I think we know exactly what we are doing and even the patients themselves realize it and that's why they come. But of course, those days they [patients] would have that negative…what? [attitude]…but now they know if you go there, they test you and they give you…what?…[They] will cure you. Of course that confidence has increased in them. Ya, they are confident that they going to get what? [That they are going to get] okay. FGD with DSVs in the mRDT arm of the trial The claims by DSVs and clients of the benefits of using mRDTs were significantly broader than those stated in the project protocol. A few clients reported that the mRDTs in the shops they visited diagnosed malaria, typhoid and syphilis (eg, “what came out of the test was that my child had malaria and syphilis,” FGD with mothers in the mRDT arm of the trial). Mostly, however, the capacity of the tests was not specified but was described by DSVs and their clients as creating a general change in practice that enabled DSVs to know the causes of and necessary treatment for ‘illness’ rather than just identifying those patients with/without malaria.Ever since we started this project, it has made us treat the patients well because before we have been treating a person simply. A person comes, “I am suffering from malaria.” You would touch on him, measure the temperature and then give him drugs. But now you treat something you know and he also leaves understanding that really, they have treated me because you tested his blood.

atients well because before we have been treating a person simply. A person comes, “I am suffering from malaria.” You would touch on him, measure the temperature and then give him drugs. But now you treat something you know and he also leaves understanding that really, they have treated me because you tested his blood. FGD with DSVs in the mRDT arm of the trial From the questionnaires in the mRDT arm of the trial, just under half who had received an mRDT reported that they had not seen the results of the test (see table 3). Moreover, when they reported on their interpretation of the role in the care-seeking process, 47% of the 253 care seekers reported that the use of a test either enabled the DSV ‘to know’ what was causing their illness or fever, or ‘to treat what they know’. Table 3 Client reports of perceptions and experiences with mRDT Presumptive arm N=7 RDT arm N=247 n Per cent n Per cent Report being shown the mRDT result Yes 0 0 129 52.2 No 7 100 116 47.0 Did not report anything 0 0 2 0.8 Reported the mRDT result Positive 4 57.1 142 57.5 Negative 0 0 80 32.4 Did not know 3 42.9 22 8.9 Did not report anything 0 0 3 1.2 Reported the role of the mRDT as providing Definitive diagnosis of malaria 2 40 88 35.6 Knowledge on illness of patient 5 60 115 46.6 Did not report anything 0 0 44 17.8 mRDT, rapid diagnostic test for malaria; RDT, rapid diagnostic test.

Negative 0 0 80 32.4 Did not know 3 42.9 22 8.9 Did not report anything 0 0 3 1.2 Reported the role of the mRDT as providing Definitive diagnosis of malaria 2 40 88 35.6 Knowledge on illness of patient 5 60 115 46.6 Did not report anything 0 0 44 17.8 mRDT, rapid diagnostic test for malaria; RDT, rapid diagnostic test. Reports of the use of the mRDT to enable a definitive diagnosis (ie, of any illness) emerged in focus groups with DSVs and clients. In the quotation below, a mother who attended a drug shop in the mRDT arm underscores how the mRDT was imagined as enabling decisive knowledge, the effective targeting of treatment and that this would provide better value for money than in a system where malaria was treated presumptively.That five hundred shillings [for the mRDT], and she treats what she has known. Because some time back someone could come and she (DSV) would say, “it seems that he has malaria,” and she would prescribe different drugs for you and she would tell you that you had to pay this and this amount of money. But now, she takes blood from you and she gets to know that she has tested your blood and you have malaria and the drugs she gives you are for malaria. FGD with mothers in the mRDT arm of the trial The DSV and the client quoted above also provide details of what makes up a good visit to a drug shop, that the client has been tested and purchased effective medication. The final section of the results considers the ways in which mRDTs acted on clients’ desires for effective medicines and reshaped local pharmaceutical markets in Mukono.

nt quoted above also provide details of what makes up a good visit to a drug shop, that the client has been tested and purchased effective medication. The final section of the results considers the ways in which mRDTs acted on clients’ desires for effective medicines and reshaped local pharmaceutical markets in Mukono. mRDTs increase medicine sales in drug shops Like many of those working on the introduction of mRDTs in other sectors of health systems in Africa, the researchers involved in the mRDT trial in Mukono were concerned with the ways in which DSVs would manage negative test results. A negative mRDT could mean that drug sellers would not make a sale, reducing their income, and disincentivise testing among DSVs. We have previously shown how RDT use can also affect a DSV's reputation.27 Negative results carry a risk of revealing diagnostic uncertainty which could be interpreted as a lack of knowledge and skill in managing a patient's illness. During the FGDs, some clients did express concern about the ability of the test to always provide an accurate result. During the FGDs with the DSVs, however, concerns about negative test results were relatively rare. A close look at the interpretation of the capacity of the mRDT, its perceived ability to create definitive knowledge around disease and the way in which its introduction set a new standard for care based on testing blood helps to make sense of this.

s, however, concerns about negative test results were relatively rare. A close look at the interpretation of the capacity of the mRDT, its perceived ability to create definitive knowledge around disease and the way in which its introduction set a new standard for care based on testing blood helps to make sense of this. When DSVs described anxiety around the management of a negative mRDT result, this was most often linked to encounters with clients who were convinced that they had malaria and wanted to purchase an ACT, but what also appears to have been critical in managing a negative result was the action taken by the DSV following the generation of the result. Referring a client who was not considered to be dangerously sick, especially if the DSV did not provide any medication first, was considered to be economically perilous and also hazardous to their reputation and so was infrequent.27 Instead, DSVs could act on a negative test result by providing what was understood by the client to be effective treatment. Below, two different mothers from different focus groups explained a positive outcome of their trip to the drug shop using mRDTs with their children. In the first description, the child is diagnosed with malaria, and in the second the child tests negative to malaria but both received an injection and other medication.After testing the child's blood, she gave him an injection and ended there. The malaria was very severe and she gave him an injection. She even gave me tablets that were yellow and we went back home. She even told me that I have to give to the child plenty to drink.

a but both received an injection and other medication.After testing the child's blood, she gave him an injection and ended there. The malaria was very severe and she gave him an injection. She even gave me tablets that were yellow and we went back home. She even told me that I have to give to the child plenty to drink. FGD with Mothers in the mRDT arm of the trial For me, I took a child there [to the drug shop] who had diarrhoea and had fever. They tested him but there was no malaria. They gave him drugs and syrup and they also gave him four injections. They gave him four injections and he got well. FGD with Mothers in the mRDT arm of the trial

FGD with Mothers in the mRDT arm of the trial For me, I took a child there [to the drug shop] who had diarrhoea and had fever. They tested him but there was no malaria. They gave him drugs and syrup and they also gave him four injections. They gave him four injections and he got well. FGD with Mothers in the mRDT arm of the trial The imperative to receive medicines as an outcome of drug shop encounters in both arms of the trial was underscored by data from the questionnaires with care seekers, where 100% of those interviewed during follow-up reported having purchased some form of medication (table 4). Unlike in the FGDs, where medicines were often identified by their colour, whether they were in tablet form, syrup, injection or crushed into a drink, for the questionnaires clients were asked to specify wherever possible the type of medicine that they had purchased. In concordance with the overall findings from the trial,15 reported sales of ACTs were substantially lower in the mRDT arm than in the presumptive arm, though not entirely ruling out the sale of malaria medicines to patients with a negative test. In the mRDT arm of the trial, for the clients who reported having a positive mRDT result, the majority said that they had purchased an ACT (94%) either with (17.9%) or without (76.4%) an antibiotic. A similar percentage to those being treated for febrile illness in the presumptive arm purchased an antibiotic (18%; see table 4). As one of the FGD respondents explained, past experiences of giving malaria medication to her son during an illness episode continued to influence the clients’ preference and demands for particular medicines. Table 4 Treatments purchased in registered drug shops by method of diagnosis and mRDT test result

see table 4). As one of the FGD respondents explained, past experiences of giving malaria medication to her son during an illness episode continued to influence the clients’ preference and demands for particular medicines. Table 4 Treatments purchased in registered drug shops by method of diagnosis and mRDT test result Type of drug treatment mRDT arm mRDT negative N=101 (%) mRDT positive N=123 (%) RDT total (%) Presumptive arm N=246 (%) No drugs purchased 0 (0.0) 0 (0.0) 0 (0.0) 0 (0.0) Antipyretic only 22 (21.8) 4 (3.3) 26 (11.6) 0 (0.0) Other drugs only (eg, cough syrups, deworming) 13 (12.9) 1 (0.8) 14 (6.3) 0 (0.0) Antibiotic without any antimalarial 29 (28.7) 0 (0.0) 29 (12.9) 1 (0.4) Other non-ACT antimalarial treatment without antibiotic 4 (4.0) 2 (1.6) 6 (2.7) 0 (0.0) Other non-ACT antimalarial treatment with antibiotic 3 (3.0) 0 (0.0) 3 (1.3) 1 (0.4) ACT without antibiotic 24 (23.8) 94 (76.4) 118 (52.7) 210 (85.4) ACT with antibiotic 6 (5.9) 22 (17.9) 28 (12.5) 34 (13.8) Median cost (UGX) (minimum–maximum) 5000 (200–33 000) 4050 (300–32 000) 5000 (200–33 000) 3300 (800–45 000) Median cost US$ (minimum–maximum) 2 (0.08–13.20) 1.62 (0.12–12.80) 2 (0.08–13.20) 1.32 (0.32–18.00) ACT, artemisinin combination therapy; mRDT, rapid diagnostic test for malaria; RDT, rapid diagnostic test.

13.8) Median cost (UGX) (minimum–maximum) 5000 (200–33 000) 4050 (300–32 000) 5000 (200–33 000) 3300 (800–45 000) Median cost US$ (minimum–maximum) 2 (0.08–13.20) 1.62 (0.12–12.80) 2 (0.08–13.20) 1.32 (0.32–18.00) ACT, artemisinin combination therapy; mRDT, rapid diagnostic test for malaria; RDT, rapid diagnostic test. Now, when you get to the health worker he tests and tells you there is no malaria and as a result you tell him, “Health worker, the other time you gave me these tablets and these tablets. So, still just give me maybe Camoquine or Chloroquine and I go and give [them to] him and I see if he will get better.” FGD with mothers in the mRDT arm of the trial While the results of the FGDs suggest that there was a good deal of polypharmacy among those testing both positive and negative for malaria, the questionnaires suggested that the range of medicines sold to clients with a negative test result was greater than either those with a positive result or those who had not been tested at all. It appears that more mRDT-negative clients purchased an antibiotic compared with mRDT-positive or untested patients—37.6–18% and 15%, respectively. Overall, the median amount of money spent by clients in drug shops using mRDTs was higher than those in the presumptive arm and the difference was marginally greater among clients in the mRDT arm who received a negative test result.

biotic compared with mRDT-positive or untested patients—37.6–18% and 15%, respectively. Overall, the median amount of money spent by clients in drug shops using mRDTs was higher than those in the presumptive arm and the difference was marginally greater among clients in the mRDT arm who received a negative test result. Discussion This paper has provided an analysis of the unintended consequences of the introduction of mRDTs into drug shops in Mukono district, Uganda. The study took place alongside a cluster-randomised control trial with the intention of providing an analysis of the broader impacts of introducing rapid diagnostic tests in drug shops. For the FGDs, our sampling strategy ensured that we included a range of participants providing and seeking care within Mukono and the semistructured interviews were based on a random sample. In both cases, however, the study was limited by its use of data that relied on participants’ own accounts of their practices and experiences and it is possible that both the FGD participants and those responding to the interview questions shaped their responses in a particular way that did not always reflect actual practice. The approach could have been enhanced by incorporating participant observation, thus enabling researchers to observe the dynamic interactions between test results, medicines, formulations of expertise and good quality care.

uestions shaped their responses in a particular way that did not always reflect actual practice. The approach could have been enhanced by incorporating participant observation, thus enabling researchers to observe the dynamic interactions between test results, medicines, formulations of expertise and good quality care. To conduct the analysis, we analysed qualitative and quantitative data, using the lens of assemblage theories to prompt a focus on the ways in which mRDTs were arranged and organised in relation to other objects, people, medicines, desires, forms of regulation and everyday practice found in these spaces. In so doing, we suggest that the success of the intervention and its intended effect of facilitating high rates of uptake of the test and adherence to mRDT results was most likely connected to the development of trust between the DSVs and their clients, increased sales of unauthorised medicines and the attraction of clients to drug shops during illness episodes which would have previously resulted in a trip to the public and private not for profit formal sector.

e to mRDT results was most likely connected to the development of trust between the DSVs and their clients, increased sales of unauthorised medicines and the attraction of clients to drug shops during illness episodes which would have previously resulted in a trip to the public and private not for profit formal sector. Our findings that the introduction of mRDTs appeared to make drug shops more attractive places to seek care resonate with other observations from the main trial. In comparison with both the exit interviews conducted before the trial began29 and results from the presumptive arm in the main trial, a significantly higher proportion of clients seen by DSVs in the mRDT arm were parasite positive by microscopy, suggesting that there may have been a differential change in treatment-seeking behaviour following the introduction of mRDTs (Mbonye, personal communication). These are different from research findings in Kenya in which the introduction of subsidised ACTs in drug shops, rather than the introduction of subsidised mRDTs, was considered to impact on decision-making about where to receive care.11 These results, however, came from a study in which the results of mRDTs overall were reported to have had little impact on decision-making about treatment and the authors suggested that this lack of effect on treatment seeking was likely to reflect their limited impact on treatment decisions more generally.11

e care.11 These results, however, came from a study in which the results of mRDTs overall were reported to have had little impact on decision-making about treatment and the authors suggested that this lack of effect on treatment seeking was likely to reflect their limited impact on treatment decisions more generally.11 Our findings challenge widespread concerns that adherence to negative mRDT results would be undermined in the retail sector because of the importance of making a sale in these places.9 First, recorded adherence to positive and negative mRDT results in relation to the prescription of ACTs was high in the main trial findings.15 Second, questionnaires with clients found that none had left the shop empty-handed and that costs to clients were higher in the mRDT arm of the trial. Reported sales of antibiotics were substantially more common in patients who tested negative, with a slight increase also in non-ACT antimalarials, suggesting that the ability to sell other medications (including the illegal sale of antibiotics and injections) was a factor in facilitating adherence to negative test results.

ted sales of antibiotics were substantially more common in patients who tested negative, with a slight increase also in non-ACT antimalarials, suggesting that the ability to sell other medications (including the illegal sale of antibiotics and injections) was a factor in facilitating adherence to negative test results. Finally, the perception that mRDTs can identify other causes of illness and not just malaria has also been noted in other settings. In a qualitative study of patients receiving care in public sector facilities in Ghana, the authors were concerned that once patients learnt of its limitations, this would lead to disappointment with the test.30 Work on the acceptability of introducing mRDTs into drug retail outlets in Kenya also suggests that the capacity of mRDTs may be misinterpreted in these settings—drug sellers there reported that their introduction would be able to remove the ‘guesswork’ from diagnostic practice (with no reference to malaria per se), enabling drug sellers to target drugs more effectively.10 Whether the mRDT will appear as attractive to DSVs or their clients if its limitations were better understood remains an important area for further investigation. Further research will also be needed to ascertain whether the trust in the test as a decisive tool through which disease can be known will change and diminish over time and the effects that it will have on the attraction of the test to DSVs and their clients.

erstood remains an important area for further investigation. Further research will also be needed to ascertain whether the trust in the test as a decisive tool through which disease can be known will change and diminish over time and the effects that it will have on the attraction of the test to DSVs and their clients. Conclusion Interest in the potential of markets and the retail sector in particular to deliver healthcare in low-income settings is evident across many debates within global health.31 For those concerned with improving access to quality assured, targeted treatment for malaria, investing in public and donor funds in the retail sector through subsidies for first-line antimalarial medication and mRDTs appears a logical step.9 10 32 33 While the retail sector is constituted differently in different countries, in many low-income countries the lack of regulatory oversight of drug shops makes the need for holistic accounts of interventions in this sector imperative before policy change is put into effect. In this paper, we constructed such an analysis by drawing on Merton's idea of the unintended consequences of purposive action and recent theories of assemblage. We have shown how the mRDT interacted with the desires of care seekers for trustworthy providers and targeted pharmaceutical cure for their illness; and the lack of regulation of medicine sales makes the mRDT appear more powerful than it is and drug shops more attractive places to seek care. Once these dynamics are taken into account, national policymakers may consider that introducing mRDTs will have to go alongside significant concomitant improvements in the regulation of drug shops. These could possibly be achieved through accreditation schemes34 35 and/or through the workings of national health insurance programmes that may come into place in some low-income settings over the next decade.

oducing mRDTs will have to go alongside significant concomitant improvements in the regulation of drug shops. These could possibly be achieved through accreditation schemes34 35 and/or through the workings of national health insurance programmes that may come into place in some low-income settings over the next decade. The authors are grateful to the study participants for their time and the insights that they gave, and to the social scientists who conducted the interviews. The study was funded by a grant from the ACT Consortium at the London School of Hygiene and Tropical Medicine funded by the Bill and Melinda Gates Foundation. SEC was supported by the Wellcome Trust through a Research Career Development Fellowship (084933). CIRC is supported by a Wellcome Trust Grant. Handling editor: Valery Ridde

The authors are grateful to the study participants for their time and the insights that they gave, and to the social scientists who conducted the interviews. The study was funded by a grant from the ACT Consortium at the London School of Hygiene and Tropical Medicine funded by the Bill and Melinda Gates Foundation. SEC was supported by the Wellcome Trust through a Research Career Development Fellowship (084933). CIRC is supported by a Wellcome Trust Grant. Handling editor: Valery Ridde Contributors: EH supervised the field research, drafted the paper, wrote the theoretical background, supervised the coding of the FGD data and provided the main analysis; CH wrote the theoretical background and analysed the quantitative results; SL gave substantial comments on the paper and the interpretation of the quantitative results; KH designed the questionnaire and gave substantial comments on the interpretation of the quantitative results; MK collected and coded the FGD data and provided comments on the interpretation of the FGD data; PM, SEC and AM provided substantial comments on the paper and the overall interpretation of findings; CIRC supervised the field research and provided substantial comments on the paper. AM, SEC, PM and KH were involved in the design of the intervention. AM, SEC, PM and SL were involved in the implementation of the intervention. Funding: Wellcome Trust (084933 (career development fellowship (SEC)); Institutional Strategic Support Fund (CIRC). Competing interests: None declared. Ethics approval: LSHTM Ethics Committee, Uganda National Council for Science and Technology.

Contributors: EH supervised the field research, drafted the paper, wrote the theoretical background, supervised the coding of the FGD data and provided the main analysis; CH wrote the theoretical background and analysed the quantitative results; SL gave substantial comments on the paper and the interpretation of the quantitative results; KH designed the questionnaire and gave substantial comments on the interpretation of the quantitative results; MK collected and coded the FGD data and provided comments on the interpretation of the FGD data; PM, SEC and AM provided substantial comments on the paper and the overall interpretation of findings; CIRC supervised the field research and provided substantial comments on the paper. AM, SEC, PM and KH were involved in the design of the intervention. AM, SEC, PM and SL were involved in the implementation of the intervention. Funding: Wellcome Trust (084933 (career development fellowship (SEC)); Institutional Strategic Support Fund (CIRC). Competing interests: None declared. Ethics approval: LSHTM Ethics Committee, Uganda National Council for Science and Technology. Provenance and peer review: Not commissioned; externally peer reviewed. Data sharing statement: No additional data are available.

Key questions What is already known about this topic? In many settings HIV services at the community level are provided through vertical programmes whose sustainability is under threat. Studies on decentralising HIV services in Sub-Saharan Africa have demonstrated that while using an integrated service delivery approach is feasible and effective, questions remain about whether and how this can be implemented within a holistic community health platform. What are the new findings? This is the first study in Kenya to comprehensively explore the perspectives of a range of stakeholders on the acceptability and feasibility of integrating HIV services into existing community health structures. We found widespread support for integrating HIV service at the community level. Recommendations for policy Integrating HIV services such as testing and counselling, linkage and adherence support with community health programmes could prove to be a sustainable approach to scaling-up and normalising community HIV services. This task-shifting approach, however, needs to build on existing frameworks in order to strengthen—and not distort—community health systems.

ting and counselling, linkage and adherence support with community health programmes could prove to be a sustainable approach to scaling-up and normalising community HIV services. This task-shifting approach, however, needs to build on existing frameworks in order to strengthen—and not distort—community health systems. Introduction Many African countries are adopting the UNAIDS 90–90–90 target of ensuring that 90% of people living with HIV know their status, 90% of the people with diagnosed HIV infection receive antiretroviral therapy and 90% of those on treatment have viral suppression.1 Reaching these targets requires a substantial increase in service coverage but is simply impossible to achieve through traditional facility-based HIV services. Community-based approaches have been shown to increase HIV testing uptake, increase the proportion of first-time testers, identify persons earlier in the course of HIV-infection,2–4 and improve linkage to care.5–7 Currently, data on the effectiveness of these community-based approaches are derived largely from donor-funded projects delivered through parallel vertical programmes. For example in Kenya, where our study was conducted, about 38% of HIV testing and counselling has been carried out in the community by lay counsellors salaried on specific projects.8 These lay counsellors do not have a clinical background, they are trained only to conduct testing and counselling, are employed by non-governmental organisations and are not a recognised cadre in Kenya's Ministry of Health scheme of service.9

rried out in the community by lay counsellors salaried on specific projects.8 These lay counsellors do not have a clinical background, they are trained only to conduct testing and counselling, are employed by non-governmental organisations and are not a recognised cadre in Kenya's Ministry of Health scheme of service.9 Despite evidence of being effective, donor funding for community HIV services in Sub-Saharan Africa is flat-lining or decreasing,10 with calls for integrating HIV services into community health as a strategy to reduce stigma and sustainably increase HIV services.11 12 However, integration may not be as straightforward as it seems; a recent review of integrating sexual health services into primary care revealed a number of challenges in coordination, logistics, human resources (HR), training, supervision and financing.13

gy to reduce stigma and sustainably increase HIV services.11 12 However, integration may not be as straightforward as it seems; a recent review of integrating sexual health services into primary care revealed a number of challenges in coordination, logistics, human resources (HR), training, supervision and financing.13 At the same time as donor funding declines in Kenya, two changes in health provision offer a window of opportunity for integration. First, Kenyan health services have devolved from a single central government to 47 county governments with decisions around service delivery priorities now being determined and funded at the county level,14 15 and second, the Kenyan community health strategy is being revised.16 Figure 1 depicts government community health structures according to the revised community health strategy (2014), including how lay counsellors work in parallel but with referral links to government health structures and services. The strategy defines community health volunteer roles as raising awareness, promoting early service seeking behaviour, defaulter tracing and caring for the chronically ill.16 Community health extension workers are employed by the government to provide support and supervision to community health volunteers and to provide diagnosis and treatment such as for malaria and other childhood illnesses, but they have no specific HIV-related tasks.16

ulter tracing and caring for the chronically ill.16 Community health extension workers are employed by the government to provide support and supervision to community health volunteers and to provide diagnosis and treatment such as for malaria and other childhood illnesses, but they have no specific HIV-related tasks.16 Figure 1 Revised community health structures for government workers and lay counsellors. CHEW, community health extension worker; HBTC, home-based testing and counselling; NGO, non-governmental organisation. Our aim of this study was to provide timely information on possibilities for integrating HIV services at this critical juncture in Kenya. We set out to describe perceptions of current policy and practice for HIV service delivery at the community level and to explore opportunities to integrate HIV services among the key community health actors at the various levels of the Kenyan community health system. Box 1 presents key definitions in relation to community HIV services. Box 1 Key study definitions in relation to community HIV services ▸ Community health worker (CHW): Any individual delivering healthcare, trained in the skills needed for the intervention but with no certificate or degree in tertiary education. In Kenya this term includes both CHVs and CHEWs.

▸ Community HIV services: Services provided in the community including home-based HIV counselling and testing, linkage for care and treatment, and home-based care. ▸ Integration: This refers to the incorporation of community HIV services traditionally carried out by vertical programmes into the existing Ministry of Health community health structures. Methods We used qualitative methods to explore the perceptions of actors at the macro level (referred to as policymaker level), the meso level (county managers implementing services) and the micro level (community health extension workers, community health volunteers, lay counsellors and community members). This macro-meso-micro framework has been used in studies of policy implementation and draws on existing theory that participants at each level ‘frame’ their understanding differently.17–22 We used a range of interview types shown in table 1. With policymakers and health managers we used in-depth interviews to gain insights into their perspectives. We also interviewed home-based testing and counselling clients individually to explore sensitive issues in private. These interviews were supplemented by anonymous online semistructured questionnaires with lay counsellors—a method selected to avoid potential biases resulting from individuals being interviewed by researchers employed in their own organisation. Finally, we used focus-group discussions with community health volunteers and community members as group interaction can generate ideas and conversations that enable understanding of similar and diverging views within a group.23 24

ng from individuals being interviewed by researchers employed in their own organisation. Finally, we used focus-group discussions with community health volunteers and community members as group interaction can generate ideas and conversations that enable understanding of similar and diverging views within a group.23 24 Table 1 Characteristics of study respondents National Nairobi Kitui In-depth interview respondents Policymakers 3 male 1 female 0 0 Subcounty managers 0 0 male 3 female 3 male 0 female Facility in-charges 0 0 male 2 female 1 male 1 female Community health extension workers 0 4 male 4 female 4 male 4 female Home-based testing and counselling clients 0 0 male 5 female 1 male 4 female Total (n=40) 4 18 18 Characteristics of semistructured questionnaire respondents Lay counsellors 6 male 7 female 3 male 9 female Total (n=25) 13 12 Characteristics of focus-group discussion respondents Community members (2 FGDs per county) 5 male 15 female 10 male 12 female Community health volunteers (3 FGDs per county) 12 male 24 female 11 male 25 female Total (n=114) 56 58 FGDs, focus group discussions.

sellors 6 male 7 female 3 male 9 female Total (n=25) 13 12 Characteristics of focus-group discussion respondents Community members (2 FGDs per county) 5 male 15 female 10 male 12 female Community health volunteers (3 FGDs per county) 12 male 24 female 11 male 25 female Total (n=114) 56 58 FGDs, focus group discussions. Study sites:The study was conducted in 2013 in two counties in Kenya selected for their geographical and social variation and recent exposure to community HIV services in the form of home-based HIV testing and counselling. Kitui, a rural county located in southeast Kenya, has an estimated population of 1 065 329 with poor health services and outcome indicators, high rates of childhood malnutrition and an HIV prevalence rate of 4.3%.25 26 Nairobi is a densely populated urban county with 3.7 million people, numerous slum areas, inequitable health service usage and an HIV prevalence of 8%.25–27 We selected community units in three of six subcounties in Kitui and four of nine subcounties of Nairobi on the basis of their having received home-based testing and counselling services from lay counsellors within the previous 2 years.

rous slum areas, inequitable health service usage and an HIV prevalence of 8%.25–27 We selected community units in three of six subcounties in Kitui and four of nine subcounties of Nairobi on the basis of their having received home-based testing and counselling services from lay counsellors within the previous 2 years. We selected participants purposively and continued interviewing until saturation was reached (see table 1). We selected all policymakers and subcounty managers directly involved in decision-making on community health services. We interviewed facility in-charges and community health extension workers (CHEWs) in the associated link facilities and asked all lay counsellors from the non-governmental (NGO) employees who had offered services in the study areas to complete the questionnaire. Community health volunteers (CHVs) and general community members were convenience sampled with the help of local administrators and focus group discussions (FGDs) were conducted on an ‘all welcome basis’. Male and female home-based HIV testing clients were approached in the communities by CHVs who had been involved in community mobilisation. Our survey with lay counsellors had 100% uptake.

nvenience sampled with the help of local administrators and focus group discussions (FGDs) were conducted on an ‘all welcome basis’. Male and female home-based HIV testing clients were approached in the communities by CHVs who had been involved in community mobilisation. Our survey with lay counsellors had 100% uptake. Data collection: Topic guides for FGDs explored a broad range of priorities and concerns at the community level and included questions on the types of services available in the community, perceptions of the quality of services, potential roles of lay counsellors and CHVs and what they thought about integrating HIV services. We included some similar questions across methods to ensure triangulation and comparison. The survey with lay counsellors specifically sought to determine how they link with the community, their knowledge of other community duties and willingness to take on more community roles. Topic guides were developed, translated into Kiswahili, back translated to English and piloted prior to use. Trained interviewers facilitated all discussions. Data from in-depth interviews and focus-group discussions were digitally recorded and transcriptions counterchecked with audio files. Regular meetings were held between interviewers and senior researchers to review emerging themes, and a coding framework was developed jointly through an iterative process starting with the major themes covered in the topic guides.23 Findings were presented to county stakeholders as well as members of the national community health operations research technical working group, and feedback from stakeholders was integrated into the analysis through alterations to the coding framework and final analysis. The study protocol was approved by the Kenya Medical Research Institute Ethics and Review Committee and the Royal Tropical Institute, Amsterdam (KIT) Research Ethics Committee.

working group, and feedback from stakeholders was integrated into the analysis through alterations to the coding framework and final analysis. The study protocol was approved by the Kenya Medical Research Institute Ethics and Review Committee and the Royal Tropical Institute, Amsterdam (KIT) Research Ethics Committee. Results Themes emerged in three key areas: (1) current and potential HIV testing services at the community level; (2) the issues and challenges around vertical programmes; (3) the perceptions of HIV service integration at community level. For each area we present and examine commonly held views across the range of participants at the policymaker, implementation and community levels.

ial HIV testing services at the community level; (2) the issues and challenges around vertical programmes; (3) the perceptions of HIV service integration at community level. For each area we present and examine commonly held views across the range of participants at the policymaker, implementation and community levels. HIV services delivery at the community level: policy, practice and potential for integration Our findings revealed that community-level actors (primarily community health volunteers) do interpret their general roles as outlined in the official policies as applying to providing community HIV services. These included health promotion “(CHVs) usually visit us at home, and they tell us how we can protect ourselves from HIV”. (Kitui Community 1); referral of pregnant women for prevention of mother-to-child transmission (PMTCT) “If a woman is HIV positive and they are pregnant you keep reminding them to go and give birth in the hospital so that the baby does not get the virus”. (Kitui CHV 1); defaulter tracing for antiretroviral therapy, PMTCT and tuberculosis (TB) treatment (more commonly described in Nairobi than Kitui county) “If the doctor finds there is a defaulter, they will ask the CHV to follow up and know what is wrong”. (Kitui CHV 1).

spital so that the baby does not get the virus”. (Kitui CHV 1); defaulter tracing for antiretroviral therapy, PMTCT and tuberculosis (TB) treatment (more commonly described in Nairobi than Kitui county) “If the doctor finds there is a defaulter, they will ask the CHV to follow up and know what is wrong”. (Kitui CHV 1). In addition to the commonly described roles, a minority of respondents also described other roles including condom distribution; referral and linkage of HIV-positive clients to existing government HIV services; assisting with couples' disclosure; referral of patients with signs and symptoms of TB; and home-based care for HIV-positive clients. “Clients who test HIV positive are linked to community health workers with a mandate to ensure that they access care” (HBTC Counsellor). All of the roles identified (including those commonly and less commonly described) align with the Kenya Community Health Strategy (2006) and are summarised in table 2 below. Table 2 HIV roles for community health volunteers: policy, practice and opportunities for integration identified by respondents

In addition to the commonly described roles, a minority of respondents also described other roles including condom distribution; referral and linkage of HIV-positive clients to existing government HIV services; assisting with couples' disclosure; referral of patients with signs and symptoms of TB; and home-based care for HIV-positive clients. “Clients who test HIV positive are linked to community health workers with a mandate to ensure that they access care” (HBTC Counsellor). All of the roles identified (including those commonly and less commonly described) align with the Kenya Community Health Strategy (2006) and are summarised in table 2 below. Table 2 HIV roles for community health volunteers: policy, practice and opportunities for integration identified by respondents Area of focus in the community strategy* HIV-related tasks described in community health policy† HIV-related tasks described by meso-level and micro-level respondents Additional HIV-related tasks described by CHVs (micro-level) Suggestions made by all levels of respondents for potential roles for lay counsellors, CHVs and CHEWs in an integrated approach to HIV services Disease prevention and control to reduce morbidity, disability and mortality Raise awareness on disease causation, control and prevention, in particular STI/HIV/AIDS HIV prevention education Condom distribution Community level respondents suggest that CHVs and CHEWs continue providing HIV prevention education and explore opportunities to expand CHV distribution of condoms. They feel lay counsellors were acceptable to their clients who described seeking them out in the event of any problems. Lay counsellors identified roles which they could take on in addition to their current HBTC roles. Family health services to expand family planning, maternal, child and youth services Promote early service-seeking behaviour Referral of pregnant women for prevention of mother-to-child transmission (PMTCT) and hospital deliveries Referral and linkage of HIV positive to care currently is carried out by NGO-supported lay counsellors Policymakers felt home-based HIV testing could be conducted by CHEWs if appropriately trained, with referral and linkage to care (as currently carried out by NGO-supported lay counsellors). CHVs expressed a desire to be trained to conduct HIV testing and felt it would help extend coverage of testing services, particularly among youth.

t home-based HIV testing could be conducted by CHEWs if appropriately trained, with referral and linkage to care (as currently carried out by NGO-supported lay counsellors). CHVs expressed a desire to be trained to conduct HIV testing and felt it would help extend coverage of testing services, particularly among youth. Information education communication (IEC) for community health promotion and disease prevention Sensitise, mobilise and organise community to ensure leadership and awareness of rights and responsibilities in health Mobilisation and referral for HBTC Aid in couples disclosure Respondents at macro and meso levels expressed the need for greater community engagement around HIV issues, opportunities for assisting with couples’ disclosure and for normalisation of HIV through HBTC Disease control Community-based referral system Conduct community directly observed treatment (C-DOTS) and defaulter tracing Defaulter tracing for antiretroviral therapy (ART), PMTCT and TB medication Referral of patients with signs and symptoms of TB Policymaker respondents (macro level) see the potential for community-based HIV management through a decentralised approach Care for chronically ill None Home-based care for HIV-positive community members CHEWs and CHVs see an opportunity for holistic care and expanding home-based care *Strategic Plan of Kenya Taking the Kenya Essential Package for Health to the Community. †A Strategy for the Delivery of Level One Services, 2006 pages 10–13.

Information education communication (IEC) for community health promotion and disease prevention Sensitise, mobilise and organise community to ensure leadership and awareness of rights and responsibilities in health Mobilisation and referral for HBTC Aid in couples disclosure Respondents at macro and meso levels expressed the need for greater community engagement around HIV issues, opportunities for assisting with couples’ disclosure and for normalisation of HIV through HBTC Disease control Community-based referral system Conduct community directly observed treatment (C-DOTS) and defaulter tracing Defaulter tracing for antiretroviral therapy (ART), PMTCT and TB medication Referral of patients with signs and symptoms of TB Policymaker respondents (macro level) see the potential for community-based HIV management through a decentralised approach Care for chronically ill None Home-based care for HIV-positive community members CHEWs and CHVs see an opportunity for holistic care and expanding home-based care *Strategic Plan of Kenya Taking the Kenya Essential Package for Health to the Community. †A Strategy for the Delivery of Level One Services, 2006 pages 10–13. CHEWs, community health extension workers; CHVs, community health volunteers; HBTC, home-based testing and counselling; NGOs, non-governmental organisations; STI, sexually transmitted infections; TB, tuberculosis.

Information education communication (IEC) for community health promotion and disease prevention Sensitise, mobilise and organise community to ensure leadership and awareness of rights and responsibilities in health Mobilisation and referral for HBTC Aid in couples disclosure Respondents at macro and meso levels expressed the need for greater community engagement around HIV issues, opportunities for assisting with couples’ disclosure and for normalisation of HIV through HBTC Disease control Community-based referral system Conduct community directly observed treatment (C-DOTS) and defaulter tracing Defaulter tracing for antiretroviral therapy (ART), PMTCT and TB medication Referral of patients with signs and symptoms of TB Policymaker respondents (macro level) see the potential for community-based HIV management through a decentralised approach Care for chronically ill None Home-based care for HIV-positive community members CHEWs and CHVs see an opportunity for holistic care and expanding home-based care *Strategic Plan of Kenya Taking the Kenya Essential Package for Health to the Community. †A Strategy for the Delivery of Level One Services, 2006 pages 10–13. CHEWs, community health extension workers; CHVs, community health volunteers; HBTC, home-based testing and counselling; NGOs, non-governmental organisations; STI, sexually transmitted infections; TB, tuberculosis. Community-health extension workers, community health volunteers and lay counsellors all expressed willingness to take on additional professional roles. Lay counsellors were thought by policymakers to be able to take on community health extension workers tasks in relation to community health, and counsellors endorsed this in their own responses. It was also commonly accepted that community health extension workers could take additional roles for home-based testing and counselling if provided with the necessary training. This was echoed by some community health volunteers who wanted to receive home-based testing and counselling training in response to community demand for those services.So that when we are attending to this client, we attend to all issues of nutrition, home-based care issues, issues of TB, like that, so that when I come I come fully, not I come, then another person comes for TB, then another person comes, I just want to go and do everything… because these people in the community need care, they need people, who can follow them up. (NBO CHEW 08)

l issues of nutrition, home-based care issues, issues of TB, like that, so that when I come I come fully, not I come, then another person comes for TB, then another person comes, I just want to go and do everything… because these people in the community need care, they need people, who can follow them up. (NBO CHEW 08) Challenges of vertical programming at the policymaker level play out at the county and community level Participants from all levels were aware of the multiple HIV-related tasks at the community level, noting that they were often driven by vertical programmes in response to local need or external funding.HBTC has been run vertically in this country … (National Policymaker 3)

vel play out at the county and community level Participants from all levels were aware of the multiple HIV-related tasks at the community level, noting that they were often driven by vertical programmes in response to local need or external funding.HBTC has been run vertically in this country … (National Policymaker 3) Challenges in coordination, coverage, duplication and lack of clarity on roles were highlighted as well as concerns that HIV services were a major driver of such issues. Policymakers and subcounty managers highlighted that the absence of government-funded community health services in some areas resulted in a vacuum leading to NGO-driven scale-up and resultant inequities in coverage and focus—with NGOs providing disease-specific (often HIV-specific) services and selecting geographical areas convenient to their organisation. One national policymaker reflected a common opinion that the high level of vertical programming for both community health services and HIV at the national level had a harmful impact on overall service delivery at the community level:At the top there, the structures are parallel, when it is at the top there, it is one problem, but when it gets back to the community becomes a big problem. (National Policymaker 2)

g for both community health services and HIV at the national level had a harmful impact on overall service delivery at the community level:At the top there, the structures are parallel, when it is at the top there, it is one problem, but when it gets back to the community becomes a big problem. (National Policymaker 2) At the implementation level respondents highlighted that system integration required a variety of county structures and processes to be put in place, with a focus on communication between the levels, coordination among stakeholders and robust systems for supply chains and referral. There was an identified need to improve interorganisational relationships between stakeholders at the community level. In Nairobi, a member of the subcounty health management team described poor communication between the subcounty health management team and NGOs. There was also variation between HIV services provided by NGOs, with some NGOs providing a focus on services for children who are HIV positive, others concentrating on HIV-positive pregnant women and others on the provision of home-based testing and counselling, all with patchy coverage. In addition to this there were differences described in the remuneration of community health volunteers at the community level within and beyond HIV services:We sent out a circular that they (all community health volunteers) should be remunerated…but most partners are inclined towards HIV, hence those CHVs inclined to HIV are the ones remunerated…. (National Policymaker 4)

the remuneration of community health volunteers at the community level within and beyond HIV services:We sent out a circular that they (all community health volunteers) should be remunerated…but most partners are inclined towards HIV, hence those CHVs inclined to HIV are the ones remunerated…. (National Policymaker 4) Perceptions of HIV service integration at the community level There was support across all respondents for an integrated approach incorporating home-based testing and counselling within existing community health activities carried out by government co-ordinated community health actors (community health extension workers and community health volunteers). This was summarised by a national policymaker who said:HBTC has been run vertically in this country … the only way to handle that issue is to make it integrated so that the community health extension worker becomes the person who is responsible in the HBTC. (National Policymaker 3)

ers and community health volunteers). This was summarised by a national policymaker who said:HBTC has been run vertically in this country … the only way to handle that issue is to make it integrated so that the community health extension worker becomes the person who is responsible in the HBTC. (National Policymaker 3) Participants saw opportunities and benefits as well as challenges in integration at each level of the health system and for a range of cadres (see table 3). On the one hand, all groups of respondents identified the need for more holistic care at the community level with potential benefits perceived for the micro level. The benefits of normalising and sustaining an approach to HIV testing at the community level were identified by lay counsellors and were also noted by policymakers who raised concerns over funding flat-lining and the need for a unified cadre able to provide HIV testing as part of the package of care. On the other hand, concerns were raised about the practical feasibility of integration including issues of political backing, the need for consistent budget allocation for community health activities, strengthening supply chain structures to ensure community providers have adequate supplies and the need for recognising and investing in training and supervising community providers of HIV testing and counselling (HTC) services. Table 3 Perceptions of the potential impacts of an integrated model at macro, meso and micro level

Participants saw opportunities and benefits as well as challenges in integration at each level of the health system and for a range of cadres (see table 3). On the one hand, all groups of respondents identified the need for more holistic care at the community level with potential benefits perceived for the micro level. The benefits of normalising and sustaining an approach to HIV testing at the community level were identified by lay counsellors and were also noted by policymakers who raised concerns over funding flat-lining and the need for a unified cadre able to provide HIV testing as part of the package of care. On the other hand, concerns were raised about the practical feasibility of integration including issues of political backing, the need for consistent budget allocation for community health activities, strengthening supply chain structures to ensure community providers have adequate supplies and the need for recognising and investing in training and supervising community providers of HIV testing and counselling (HTC) services. Table 3 Perceptions of the potential impacts of an integrated model at macro, meso and micro level Potential impacts on CHEWs Potential impacts on existing lay counsellors Potential impacts on CHVs Potential impacts at the community level Perceptions at macro level national and county policymakers CHEWs trained and competent in HTC Too many tasks could dilute quality and accountability Skilled group taken up as part of Community Health Services and ‘home testing’ within health system Increased clarity on HBTC support functions of CHVs Concerns about workload Normalises HIV testing Enables holistic services to be delivered at home Perceptions at meso-level and county-level implementers Improved county coordination of vertical programmes and of interorganisational relationships Integrated approach to training, supervision, data management, commodities and supplies. Potential for stock outs Improved supervision and support for HIV services offered, able to conduct current HIV tasks within an official framework Opportunity to extend HIV services within the community Perceptions at micro level—community level actors CHEWs able to offer HBTC at home for pregnant women, TB patient contacts, families of HIV-positive individuals Offer services additional to current HBTC roles Holistic picture of the household; able to mobilise for HIV testing, support linkage from community to health facility; able to respond to community demand for HTC or to provide condoms Availability of condoms at the community level Easier access to HTC, increased uptake especially among men Holistic care available at the community level Concerns about confidentiality and stigma CHEWs, community health extension workers; CHVs, community health volunteers; HBTC, home-based testing and counselling; HTC, HIV testing and counselling; TB, tuberculosis.

Easier access to HTC, increased uptake especially among men Holistic care available at the community level Concerns about confidentiality and stigma CHEWs, community health extension workers; CHVs, community health volunteers; HBTC, home-based testing and counselling; HTC, HIV testing and counselling; TB, tuberculosis. Implementers, county staff and national policymakers raised the importance of interorganisational relationships (eg, strategic alliances and common governance mechanisms) as well as partnerships between professionals both within and between organisations. Community members and lay counsellors raised confidentiality as a concern in a model integrating home-based testing and counselling in the community strategy, although contrasting opinions were expressed and solutions also offered as illustrated here:They should be trained to ensure confidentiality. They can visit us, give us counselling and test us…but there must be some precautions on how they will be trained. (Kitui Community 1). Respondents had additional suggestions such as using an alternative provider for home-based testing and counselling at the community level; using a CHV from a different community; and including confidentiality as a selection criterion for community health volunteers.

Community members and lay counsellors raised confidentiality as a concern in a model integrating home-based testing and counselling in the community strategy, although contrasting opinions were expressed and solutions also offered as illustrated here:They should be trained to ensure confidentiality. They can visit us, give us counselling and test us…but there must be some precautions on how they will be trained. (Kitui Community 1). Respondents had additional suggestions such as using an alternative provider for home-based testing and counselling at the community level; using a CHV from a different community; and including confidentiality as a selection criterion for community health volunteers. Discussion Our study shows that current HIV service provision at the community level already goes beyond policy with community-level HIV services that are supported by NGOs including adherence counselling, referral, defaulter tracing, home-based care and HIV testing in a responsive manner. Study respondents described a range of challenges presented by the current vertical programming and a expressed a desire for better integration. Our findings revealed support for integrating HIV services at the community level, and opportunities and some benefits of integration were outlined across the health system from policymaker to implementation and community levels. Participants also raised concerns over the feasibility of practical, workload and financial aspects of integration within existing health systems and HIV-specific issues to do with unintended harm from community-based HIV services that would need to be addressed proactively in any integrated platform.

community levels. Participants also raised concerns over the feasibility of practical, workload and financial aspects of integration within existing health systems and HIV-specific issues to do with unintended harm from community-based HIV services that would need to be addressed proactively in any integrated platform. Achieving the ambitious UNAIDS 90–90–90 targets requires a strong community platform Community platforms are likely to prove an essential component of HIV service expansion in Kenya in order to address the low uptake of services (particularly HIV testing services) among men, youth and couples.27 Providing services at home eliminates associated transport costs for clients, increases access and creates an environment that normalises HIV and HIV testing.24 A strong community platform would need to work outwards from a focus on coordination, competency, training and supervisory support systems at the health worker level to clarity on roles, tasks and remuneration at the policymaker level, as well as ensuring uptake and quality of service delivery at the community level. Should county governments choose to expand access to HIV services through community platforms there will also be a need for indicators to ensure equitable coverage and quality of services.28 Combining funding envelopes for decentralised HIV and existing community services is likely to increase the sustainability of these services. Furthermore, a strong community platform that includes HIV services will provide a governance and accountability framework for innovations like HIV self-testing that are currently being expanded in Kenyan communities to achieve UNAIDS targets.29

unity services is likely to increase the sustainability of these services. Furthermore, a strong community platform that includes HIV services will provide a governance and accountability framework for innovations like HIV self-testing that are currently being expanded in Kenyan communities to achieve UNAIDS targets.29 Integrated approaches are desired at the community level and Kenya has a window of opportunity The desire for a holistic approach to community-based care that includes HIV services and community-based testing is supported by integration at the community level that seeks to improve overall well-being rather than a particular condition.18 Integration also provides an opportunity for building on the pre-existing relationships with their communities that community health volunteers and community health extension workers enjoy by virtue of their unique interface between the community and the health system.30 Community engagement in selecting who is trained to provide HIV services, in promoting the programme and in sharing the content of training on confidentiality will play a vital part in building trusting relationships as highlighted in studies from Zambia and Kenya on community perceptions of home-based HIV testing approaches.24 31 Community health workers visiting the homes of clients in other contexts have been shown to be effective in performing HIV testing and in delivering a range of other HIV services along the cascade such as adherence counselling and defaulter tracing.6 32–34 These were paid and trained staff more akin to the community health extension workers than the community health volunteers in Kenya. A recent review of the literature on the evidence for integrating HIV and other health services, such as TB care, into national health systems, showed that the effect on health outcomes and quality of services was mixed.35 As Kenya moves forward with expanding HIV testing as a priority health service there is a need to seek synergies between vertical and horizontal programmes,36 ensuring that the integration of HIV services leads to a strengthening of community systems.37 38

effect on health outcomes and quality of services was mixed.35 As Kenya moves forward with expanding HIV testing as a priority health service there is a need to seek synergies between vertical and horizontal programmes,36 ensuring that the integration of HIV services leads to a strengthening of community systems.37 38 In Kenya community outreach has largely been provided by lay counsellors, but as external funding flat-lines, decisions need to be made on whether or not these lay counsellors are taken up into existing community structures and whether their tasks are shifted to existing community cadres or both.39–41 The current revision of the community health strategy and the process of devolution in Kenya provide a unique policy window for adopting a person-focused perspective that builds on current structures and expertise, provides a more coordinated approach between stakeholders and retains a highly skilled cadre of lay counsellors.42

nt revision of the community health strategy and the process of devolution in Kenya provide a unique policy window for adopting a person-focused perspective that builds on current structures and expertise, provides a more coordinated approach between stakeholders and retains a highly skilled cadre of lay counsellors.42 Feasibility and sustainability of integration Participants highlighted multiple current and potential HIV-related roles for community providers, but the feasibility, appropriateness and sustainability of each role needs to be examined at each level of the system. Adding tasks to community health workers as part of scale-up can undermine the quality of services and divert scarce resources from priority interventions.43 44 Findings from Ethiopia, where paid health extension workers have multiple tasks, reveal there is a need for clearly defined roles, standardised support, monitoring and accountability, strong referral links, supervision and training in order to realise benefits from the value of the interface role.30 For the integration of HIV services at community level to be feasible in Kenya it needs to be seen as a priority by communities and to use remunerated healthcare providers who are supported by strong policies and systems. Any proposed methods must take into account the community and the organisational settings.45 Integration will face constraints, such as the strain that financing newly converted lay counsellor-CHEWS may place on county budgets, and caution should be exercised in assuming that the current NGO-facilitated service landscape can necessarily be easily mapped onto government-run structures.

rganisational settings.45 Integration will face constraints, such as the strain that financing newly converted lay counsellor-CHEWS may place on county budgets, and caution should be exercised in assuming that the current NGO-facilitated service landscape can necessarily be easily mapped onto government-run structures. Limitations There are a number of limitations to our study. The study was carried out in only 2 of the 47 counties and both had been exposed to home-based testing and counselling services and a range of other HIV services provided by NGOs. The selection of community participants was carried out purposively by local administrators and home-based testing and counselling clients were selected by community health volunteers, thereby introducing a selection bias. Overall we had more female than male respondents and this was particularly marked among clients who had been tested, where more women were likely to be found at home. This gender bias may have influenced perceptions of the need for a holistic approach since women are the main beneficiaries of community health programmes that focus on maternal and child health. All of the lay counsellors interviewed were from a single NGO (that also employed the researchers) and had previously worked in the same counties. We used an anonymous online survey to avoid social desirability bias or power dynamics from the researchers being in the same institution. The lay counsellors would have all been exposed to the same way of working with CHVs and CHEWs in the community as it was agreed that CHVs would conduct community mobilisation, set up appointments and support the services, roles they did not necessarily perform in other counties or for other NGOs.

the same institution. The lay counsellors would have all been exposed to the same way of working with CHVs and CHEWs in the community as it was agreed that CHVs would conduct community mobilisation, set up appointments and support the services, roles they did not necessarily perform in other counties or for other NGOs. Conclusion HIV policymakers in Kenya are at a crossroads as they respond to a decline in funding for community programmes at the same time as increasing evidence indicates that they are effective and valued. The new community strategy provides a timely opportunity to incorporate and scale up HIV services into existing community health platforms in an equitable manner, an approach that appears to be well supported among study participants at all levels. Policymakers can take advantage of the opportunity of the new community strategy to incorporate and scale up HIV services in an equitable manner that is cognizant of potential constraints and challenges. For this to work, however, there is a need to address key health systems, funding and coordination issues at each level of the system: macro, meso and micro. One option would be to incorporate existing lay counsellors into the CHEW cadre and vice versa in line with national recommendations to formalise lay counsellors in the framework of the community strategy.8 Community health volunteers could then work alongside a capacitated community health extension worker cadre in support roles that increase uptake and outcomes of HIV services.3 6

the CHEW cadre and vice versa in line with national recommendations to formalise lay counsellors in the framework of the community strategy.8 Community health volunteers could then work alongside a capacitated community health extension worker cadre in support roles that increase uptake and outcomes of HIV services.3 6 The authors would like to acknowledge the policymakers, county and subcounty Health Management Team members, facility managers, community-health extension worker, Community Health Workers and community members who gave of their time to be interviewed. Thanks to the qualitative research team who conducted and transcribed interviews and discussions, including Millicent Kiruki, Veronica Mwania, Felista Kilunda, Carolyn Chebet Terer, Joel Ratemo, Henry Wera and Peter Kilonzo. The authors also thank Professor Shabbar Jaffar for reviewing the final draft of the manuscript. Handling editor: Stephanie Topp. Twitter: Follow Lilian Otiso @taiso29 and Miriam Taegtmeyer @MiriamTaegtmeye Contributors: LO and MT and KK conceptualised the study and this article and LO oversaw a team of data collectors. MM was involved in data collection and provided input on the results and Kenyan community context. The analysis was performed by RM, LO, MM, KdK and MT. MT prepared the final draft, with contributions and revisions made by LO, RM, KdK and RK. All authors have read and approved the final version.

eam of data collectors. MM was involved in data collection and provided input on the results and Kenyan community context. The analysis was performed by RM, LO, MM, KdK and MT. MT prepared the final draft, with contributions and revisions made by LO, RM, KdK and RK. All authors have read and approved the final version. Funding: This research forms part of a multicountry context analysis for REACHOUT, a 5-year multicountry research consortium aiming to maximise the equity, effectiveness and efficiency of close-to-community services in rural areas and urban slums in six countries (Bangladesh, Ethiopia, Indonesia, Kenya, Malawi and Mozambique). The REACHOUT Consortium is funded by the European Union FP7 grant (number 306090). This document reflects only the authors' views and the European Union is not liable for any use that may be made of the information contained therein. Competing interests: None declared. Ethics approval: Kenya Medical Research Institute Ethics and Review Committee. Provenance and peer review: Not commissioned; externally peer reviewed. Data sharing statement: No additional data are available.

Key questions What is already known about this topic? We sought to evaluate the relative cost-effectiveness of introducing the RTS,S malaria vaccine in sub-Saharan Africa compared with further scale-up of existing interventions. We did not identify any studies modelling the impact and costs of intervention packages with respect to scaling-up coverage. What are the new findings? We find that implementing the RTS,S malaria vaccine generally only enters the optimal pathway of scale-up of interventions once very high coverage of vector control interventions, along with seasonal malaria chemoprevention in settings where it is recommended, has been achieved. Recommendations for policy While the RTS,S malaria vaccine can be an effective tool for reducing burden, enhancing coverage of vector control should generally remain of higher priority across sub-Saharan Africa. This is particularly the case within settings where universal coverage of long-lasting insecticide-treated nets has not yet been achieved.

le the RTS,S malaria vaccine can be an effective tool for reducing burden, enhancing coverage of vector control should generally remain of higher priority across sub-Saharan Africa. This is particularly the case within settings where universal coverage of long-lasting insecticide-treated nets has not yet been achieved. Introduction Following the adoption of the Millennium Development Goals (MDGs) in 2000, the concurrent targets of increasing the proportion of children under 5 years that sleep under a bed net and that have access to appropriate antimalarial drugs1 have been associated with an estimated 37% decline in malaria case incidence and 60% decline in malaria mortality.2 3 These gains have been achieved with substantial investment in malaria which has more than doubled over the last decade, from less than $1 billion in 2005 to over $2.5 billion in 2015.2 However, since 2011, the acceleration in funding has slowed, plateauing in recent years.2 Funding gaps therefore remain an inevitable issue for future control and elimination efforts and thus optimising the use of available resources is paramount. Within this context, new interventions must be evaluated not only on their direct cost-effectiveness, but comparative to the other intervention options.

t years.2 Funding gaps therefore remain an inevitable issue for future control and elimination efforts and thus optimising the use of available resources is paramount. Within this context, new interventions must be evaluated not only on their direct cost-effectiveness, but comparative to the other intervention options. In early 2015, the final results from the large phase III trial of the RTS,S malaria vaccine across 11 sites in Africa were published.4 Reported efficacy against clinical and severe malaria disease in the presence of high use of bed nets for prevention and alongside a high level of access to care was moderate. The trial reported 36.3% (95% CI 31.8% to 40.5%) efficacy against clinical disease in children aged 5–17 months over 4 years under a four-dose schedule and 32.2% (13.7% to 46.9%) efficacy against severe disease.4 Nevertheless, in the high transmission sites contributing most of the disease episodes, there was a significant public health impact, with between 1000 and 6000 cases estimated per 1000 population over 4 years of follow-up in the six higher transmission sites.4 To estimate the wider public health impact and cost-effectiveness of the vaccine in settings representative of current levels of ongoing malaria transmission, a WHO working group was formed to compare the outputs from four mathematical models parameterised using the trial data. The results from this comparison demonstrated that the vaccine could have a substantial public health impact across settings ranging from 10% to 60% parasite prevalence in 2–10 year olds.5 Furthermore, assuming a midrange cost of $5 per dose under a four-dose schedule, the vaccine was considered to be highly cost-effective (incremental cost-effectiveness ratio (ICER) of $44–$279 per disability-adjusted life year (DALY)) within these same transmission levels.5

o 60% parasite prevalence in 2–10 year olds.5 Furthermore, assuming a midrange cost of $5 per dose under a four-dose schedule, the vaccine was considered to be highly cost-effective (incremental cost-effectiveness ratio (ICER) of $44–$279 per disability-adjusted life year (DALY)) within these same transmission levels.5 Existing malaria interventions are also highly cost-effective.6 Although there are substantial differences between the methodologies used to make these estimates, a 2011 study estimated the median ICER per DALY at $27 (range $8.15–110) for long-lasting insecticide-treated nets (LLINs) and $143 (range $135–150) for indoor residual spraying (IRS), while for seasonal malaria chemoprevention (SMC), the cost per case averted has been estimated at $68 (95% CI $62 to $75), similar to comparable metrics for LLINs and IRS.7 Thus, in areas in which coverage of these interventions is not yet universal, it is important to understand the relative cost-effectiveness of the full suite of interventions and where the RTS,S malaria vaccine could contribute. Importantly, this needs to take into account the diminishing marginal returns associated with the scale-up of interventions that may lead to a higher unit cost at high levels of coverage.8 9

derstand the relative cost-effectiveness of the full suite of interventions and where the RTS,S malaria vaccine could contribute. Importantly, this needs to take into account the diminishing marginal returns associated with the scale-up of interventions that may lead to a higher unit cost at high levels of coverage.8 9 Here, we use a well-established transmission model for Plasmodium falciparum malaria and its associated interventions10 to estimate the cost and impact of different intervention packages at varying levels of scale-up. We evaluate these packages over a wide range of transmission settings and use the estimates to derive the most cost-effective pathways for scaling-up malaria interventions in order to inform decisions about the introduction of the RTS,S malaria vaccine.

of different intervention packages at varying levels of scale-up. We evaluate these packages over a wide range of transmission settings and use the estimates to derive the most cost-effective pathways for scaling-up malaria interventions in order to inform decisions about the introduction of the RTS,S malaria vaccine. Methods Transmission and intervention model We used an established model of P. falciparum malaria transmission that incorporates the full suite of interventions.10 In brief, individuals that are initially susceptible are exposed to infection with P. falciparum malaria from bites of infectious mosquitoes. The rate at which susceptible individuals become infected is influenced by mosquito density and infectivity and is moderated by immunity. Infants are partially protected for the first 6 months of their life by passively acquired maternal immunity and individuals acquire natural immunity with repeated exposures through time. Infected individuals may be treated, after which follows a period of prophylaxis before returning to the susceptible class. Infected individuals can develop clinical disease,11 which may progress to severe disease and possibly death.12 A proportion of infected individuals harbour asymptomatic infections, some of which may also be subpatent. The model explicitly incorporates mosquito-population dynamics,13 allowing the modelling of various protective, repellent and killing aspects of the vector-control interventions. We adopt the same methodology for incorporating the RTS,S vaccine into the model as used for a large model-comparison exercise.5 The RTS,S vaccine model dynamics use a biphasic antibody decay model fitted to the phase III individual-level trial data, allowing the level of protection against clinical disease to be captured with respect to antibody titre postvaccination.14

TS,S vaccine into the model as used for a large model-comparison exercise.5 The RTS,S vaccine model dynamics use a biphasic antibody decay model fitted to the phase III individual-level trial data, allowing the level of protection against clinical disease to be captured with respect to antibody titre postvaccination.14 Transmission settings To capture the range of transmission settings across Africa, we generated a baseline set of ‘strata’. These were characterised by the parasite prevalence in the absence of interventions other than treatment, the annual seasonal pattern of transmission and the mosquito vector species present and their associated bionomics (which affect the predicted impact of LLINs and IRS). The mean parasite prevalence over the full year in 2–10 year-olds (PfPR2_10) was simulated in 17 bands between 0.1% and 80% to capture the range of transmission levels observed prior to intervention scale-up.3 Four seasonality profiles were simulated based on Fourier-transformed average rainfall patterns obtained from satellite data between 2003 and 2006:15 (A) highly seasonal—with a single strong peak in rainfall characteristic of the Sahel region; (B) seasonal—with a less strong peak in rainfall characteristic of West Africa coastal areas; (C) bimodal—with one large and a second smaller peak in rainfall characteristic of East and Southern Africa; and (D) non-seasonal characteristic of perennial levels of rainfall observed in Central Africa. The four mosquito vector profiles capture behavioural difference in the levels of anthropophagy (the human biting index), endophily (indoor biting), endophagy (indoor resting) and the timing of bites relative to sleeping hours. Rather than simulating species and their combinations, these profiles represent a range of vector species bionomics moving from behaviours associated with Anopheles gambiae s.s./Anopheles funestus to those associated with Anopheles arabiensis. The combination of parasite prevalence bands, seasonality of transmission and vector bionomics resulted in 272 baseline strata. Further details are provided in the online supplementary materials.

ing from behaviours associated with Anopheles gambiae s.s./Anopheles funestus to those associated with Anopheles arabiensis. The combination of parasite prevalence bands, seasonality of transmission and vector bionomics resulted in 272 baseline strata. Further details are provided in the online supplementary materials. 10.1136/bmjgh-2016-000090.supp1supplementary materials

ing from behaviours associated with Anopheles gambiae s.s./Anopheles funestus to those associated with Anopheles arabiensis. The combination of parasite prevalence bands, seasonality of transmission and vector bionomics resulted in 272 baseline strata. Further details are provided in the online supplementary materials. 10.1136/bmjgh-2016-000090.supp1supplementary materials Interventions For each strata, we simulated the impact of all possible applicable combinations from a set of four interventions (LLINs, IRS, SMC and RTS,S) at a range of coverage levels (see online supplementary materials for details), resulting in a total of 306 000 simulations. Throughout we assumed that an LLIN would cover 1.8 people (consistent with the approach taken in the World Malaria Report2) and that nets would be distributed on a 3-yearly cycle. The bed net model further captures loss of adherence, decay in insecticides and wear-and-tear over time16 and we define coverage of bed nets as usage as reported in DHS/MIS surveys.17 18 For IRS, we assumed that a DDT-like insecticide was used and applied once a year. For this intervention, coverage was defined as the proportion of the population residing in a house that was sprayed in the previous year. SMC was simulated following WHO recommendations to children between 6 months and 5 years of age, with 3 monthly doses of SP-amodiaquine in seasonal settings, with the second dose timed to occur at the seasonal peak in transmission.19 Coverage was defined as the proportion of eligible children who received all three doses and we did not model the effect of partial doses. Following the recent WHO recommendations, we considered a four-dose vaccine schedule in children aged 5–27 months. We assumed children would be vaccinated at 6, 7.5 and 9 and 27 months, with the timings of the first doses chosen to coincide with other contacts with healthcare at 6 and 9 months. Coverage was defined as the proportion of eligible children receiving the full four doses. A 20% drop off between those receiving the first three does and the fourth dose was included to capture loss of follow-up.5

the timings of the first doses chosen to coincide with other contacts with healthcare at 6 and 9 months. Coverage was defined as the proportion of eligible children receiving the full four doses. A 20% drop off between those receiving the first three does and the fourth dose was included to capture loss of follow-up.5 Receipt of a single intervention was assumed to be correlated across the population. When two or more interventions were included, we explored two options: no correlation in receipt (ie, independent random distribution of both interventions) or full correlation (ie, those that receive the first intervention are will also receive the second). Throughout we assumed that 60% of those with clinical disease received prompt and effective first-line treatment. Costing In the absence of detailed country-level data for all interventions, we adopted a unit costing approach. These were derived from the literature and inflated to 2015 US$ (table 1). Table 1 Unit cost values for interventions

Receipt of a single intervention was assumed to be correlated across the population. When two or more interventions were included, we explored two options: no correlation in receipt (ie, independent random distribution of both interventions) or full correlation (ie, those that receive the first intervention are will also receive the second). Throughout we assumed that 60% of those with clinical disease received prompt and effective first-line treatment. Costing In the absence of detailed country-level data for all interventions, we adopted a unit costing approach. These were derived from the literature and inflated to 2015 US$ (table 1). Table 1 Unit cost values for interventions Intervention Unit cost (2015 US $) Reference/notes LLINs 7.03 per LLIN delivered White et al.20 The 2009 costing was not inflated as more recent estimates are similar21 IRS 5.41 per person protected PMI AIRS project22 SMC 1.65 per child per round CHAI/MSF estimates23 24 RTS,S 39.25 per fully vaccinated child Under the assumption of $5 per dose5 25 Treatment (clinical cases) 2.59 per person (test+treatment) 26 Treatment (severe cases) 33.54 per person 20 We considered two approaches for costing increasing coverage of the four interventions. The first approach assumed increases in coverage were associated with linear increases in cost, while in the second approach, we derived non-linear relationships between coverage and unit costs. For this second approach, the number of nets required to achieve a given coverage level (defined by usage) was obtained from Bhatt et al,27 assuming a net-retention half-life of 3 years and the business-as-usual net allocation process. We estimated similar health production functions for IRS, SMC and the vaccine by fitting a model to data relating the cost and coverage of these interventions in a Bayesian framework (see online supplementary materials for full details). The total cost (P) of delivering an intervention to an individual is assumed to consist of two components: the commodity cost (U) and the delivery cost (D)

cine by fitting a model to data relating the cost and coverage of these interventions in a Bayesian framework (see online supplementary materials for full details). The total cost (P) of delivering an intervention to an individual is assumed to consist of two components: the commodity cost (U) and the delivery cost (D) The commodity cost remains fixed per person (under the assumption that economies of scale have been reached) with respect to coverage (C). The delivery cost per person is fixed at a baseline amount, N, until coverage reaches a given threshold, Cτ, above which the delivery costs increase logarithmically Outcome measures We considered a number of outcome measures to evaluate the incremental cost-effectiveness of each intervention. The primary outcome measure presented throughout is the number of cases averted over a 10-year period. Other outcome measures considered include the DALYs averted and the number of cases averted in children aged 6 months–5 years old and over a 10-year period (see online supplementary materials for more details on calculation).

ary outcome measure presented throughout is the number of cases averted over a 10-year period. Other outcome measures considered include the DALYs averted and the number of cases averted in children aged 6 months–5 years old and over a 10-year period (see online supplementary materials for more details on calculation). Estimating cost-effective scale-up For each transmission strata, the cost-effective scale-up of interventions was estimated using the following steps: (i) start with 0% usage/coverage of all interventions, (ii) for each available intervention, calculate the ICER of scaling up to the next usage/coverage level, (iii) implement the step with the lowest associated ICER (the most cost-effective), (iv) repeat the process (i–iii) until the scale-up of all interventions has been maximised or elimination is achieved. This process is summarised in figure 1. It should be noted that while this approach results in the most cost-effective next level of intervention coverage, the resulting intervention packages are not necessarily maximally efficient for a given budget since this reflects a gradient descent optimisation (stepwise) rather than a multidimensional optimisation approach. Nevertheless, it is used here to illustrate the pathways at each point in the absence of defined budget limits.

resulting intervention packages are not necessarily maximally efficient for a given budget since this reflects a gradient descent optimisation (stepwise) rather than a multidimensional optimisation approach. Nevertheless, it is used here to illustrate the pathways at each point in the absence of defined budget limits. Figure 1 Schematic of the cost-effective scale-up pathway. For each transmission strata, the cost-effective scale-up of interventions was estimated, starting with no intervention coverage then scaling up coverage based on the most favourable incremental cost-effectiveness ratio. Scale-up ceased when all interventions were at full coverage or elimination had occurred. Cost sensitivity analysis We assessed the sensitivity of the order of scale-up to uncertainty in the costs of interventions and their associated production functions. The analysis was repeated with 100 random draws from the posterior predictive interval for the IRS, SMC and RTS,S production functions. LLIN costs were randomly drawn from the interval between the least optimistic (net retention half-life of 2 years and current allocation process) and most optimistic (net retention half-life of 3 years and improved allocation process) net-allocation models.27 For each 100 runs and across all transmission strata, the relative occurrence of each intervention at each scale-up step was measured.

tic (net retention half-life of 2 years and current allocation process) and most optimistic (net retention half-life of 3 years and improved allocation process) net-allocation models.27 For each 100 runs and across all transmission strata, the relative occurrence of each intervention at each scale-up step was measured. The sensitivity of the scale-up order to the assumed $5 per dose cost of the vaccine was also examined by incrementally decreasing its price (increases were not included, given the initial results). At each step, the proportion of settings where the vaccine appeared in the pathway before LLINs, IRS and SMC was then recorded.

nsitivity of the scale-up order to the assumed $5 per dose cost of the vaccine was also examined by incrementally decreasing its price (increases were not included, given the initial results). At each step, the proportion of settings where the vaccine appeared in the pathway before LLINs, IRS and SMC was then recorded. Results Figure 2 shows the scale-up pathways for the combination of the four interventions across a range of transmission levels characterised by their baseline PfPR2–10 (in the absence of interventions other than treatment of clinical cases). For the majority of settings, across the full-range of baseline PfPR2–10, for vector bionomics characteristic of the three main species found in Africa (A. gambiae s.s., A. arabiensis and A. funestus) and for seasonal and non-seasonal settings, LLINs appear first in the most cost-effective pathway. At baseline PfPR2–10 <5%, LLINs and RTS,S are predicted to reduce ongoing transmission to pre-elimination levels. If scale-up of these interventions is unable to achieve pre-elimination transmission levels, our results suggest that LLINs should be scaled up to very high usage levels (75%) prior to introducing additional interventions if cost-effectiveness is the single deciding factor. After this level has been achieved, in non-seasonal settings, IRS is the second-most cost-effective intervention at lower baseline PfPR2–10 (5%< PfPR2–10 <65%), while RTS,S is estimated to be the second most cost-effective intervention in settings with baseline PfPR2–10 ≥65%. In seasonal settings, SMC is generally the second most cost-effective intervention to introduce prior to IRS or RTS,S.

cond-most cost-effective intervention at lower baseline PfPR2–10 (5%< PfPR2–10 <65%), while RTS,S is estimated to be the second most cost-effective intervention in settings with baseline PfPR2–10 ≥65%. In seasonal settings, SMC is generally the second most cost-effective intervention to introduce prior to IRS or RTS,S. Figure 2 Costs-effective scale-up pathways with linear costs. Each row represents a cost-effective scale-up pathway for a specific transmission setting (baseline PfPR2_10, seasonal profile, vector profile, intervention correlation) ordered by PfPR2_10 on the y-axis. Interventions are scaled-up in the order reading along the row from left to right, with the fill colour representing the intervention being scaled-up. Panels split the output into (A) non-seasonal settings and (B) seasonal settings, with the latter including seasonal malaria chemoprevention as an option. The results shown in figure 2 assume a single unit cost that does not change with increasing coverage. However, as higher coverage levels are sought, costs tend to increase as it becomes more difficult to fill the remaining coverage gaps and to access the hardest-to-reach populations. Figure 3 shows our estimated empirical production functions for IRS, SMC and vaccination (based on DTP3 data) alongside the previously published estimated for LLIN usage.27 All four functions have a similar shape, with increasing costs at high coverage. However, for IRS and SMC, the recorded coverage was consistently high (>80%) and the limited number of data points mean that this function is uncertain.

on (based on DTP3 data) alongside the previously published estimated for LLIN usage.27 All four functions have a similar shape, with increasing costs at high coverage. However, for IRS and SMC, the recorded coverage was consistently high (>80%) and the limited number of data points mean that this function is uncertain. Figure 3 Non-linear production functions. Production functions estimate the non-linear relationships between intervention usage or coverage and the cost per person. (A) Cost per person protected by long-lasting insecticide-treated nets, taken from Bhatt et al.27 Three scenarios: assuming current allocation model and a 3-year net-retention half-life, an improved allocation model and 3-year net retention half-life and a current allocation model and 2-year net retention half-life are shown by the solid black, dotted grey and dashed grey lines, respectively. (B) DTP3 coverage as a function of the price per person (standardised by the cost of a fully vaccinated child). (C) IRS coverage as a function of the cost per person protected based on President's Malaria Initiative data.22 (D) Seasonal malaria chemoprevention coverage as a function of the costs per child per dose based on Clinton Health Access Initiative23 and Medicins San Frontiers estimates.24 For (B–D), black lines represent the best-fit model and shaded areas the 95% prediction interval.

ased on President's Malaria Initiative data.22 (D) Seasonal malaria chemoprevention coverage as a function of the costs per child per dose based on Clinton Health Access Initiative23 and Medicins San Frontiers estimates.24 For (B–D), black lines represent the best-fit model and shaded areas the 95% prediction interval. Including the non-linear production functions shown in figure 3 leads to a substantially more complicated pattern of scale-up pathways (figure 4). With these additional non-linearities, alternative interventions are always introduced before LLIN usage is increased to the maximum level due to the high estimated cost of achieving high LLIN usage. In non-seasonal settings, LLINs are estimated to be the most cost-effective initial interventions. Across the majority of settings with baseline 5%<PfPR2–10<60%, IRS appears second with RTS,S also introduced once moderate levels of IRS coverage have been achieved. In seasonal settings, for baseline PfPR2–10<50%, LLINs remain the first most cost-effective intervention. However, in settings with higher baseline transmission, SMC and RTS,S appear earlier. Following this, the second and third most cost-effective additional interventions vary by setting, with IRS being favoured in locations where the vector bionomics are more amenable to insecticidal control, whereas RTS,S or SMC are favoured in settings with less favourable vector bionomics.

SMC and RTS,S appear earlier. Following this, the second and third most cost-effective additional interventions vary by setting, with IRS being favoured in locations where the vector bionomics are more amenable to insecticidal control, whereas RTS,S or SMC are favoured in settings with less favourable vector bionomics. Figure 4 Costs-effective scale-up pathways with non-linear costs. Each row represents a cost-effective scale-up pathway for a specific transmission setting (baseline PfPR2_10, seasonal profile, vector profile, intervention correlation) ordered by PfPR2_10 on the y-axis. Interventions are scaled-up in the order reading along the row from left to right, the fill colour representing the intervention being scaled-up. Panels split the output into (A) non-seasonal settings and (B) seasonal settings, with the latter including seasonal malaria chemoprevention as an option.

2_10 on the y-axis. Interventions are scaled-up in the order reading along the row from left to right, the fill colour representing the intervention being scaled-up. Panels split the output into (A) non-seasonal settings and (B) seasonal settings, with the latter including seasonal malaria chemoprevention as an option. Figure 5 illustrates the translation of these generic results to locations in sub-Saharan Africa using estimates of vector species presence, baseline PfPR2–10 and seasonality. Across the majority of settings, LLINs are the first most cost-effective intervention. Using the results from figure 4, the LLIN usage at which a second intervention is estimated to be more cost-effective than further LLIN scale-up is shown in figure 4A. In the majority of settings, we estimate a switch is cost-effective at 55–65% LLIN usage, although in some seasonal areas in West Africa and in areas with high estimated baseline PfPR2–10, other interventions are estimated to be more cost-effective at lower levels of LLIN usage. This threshold is similar to the levels of usage that have now been achieved in many parts of Africa (figure 5B).

–65% LLIN usage, although in some seasonal areas in West Africa and in areas with high estimated baseline PfPR2–10, other interventions are estimated to be more cost-effective at lower levels of LLIN usage. This threshold is similar to the levels of usage that have now been achieved in many parts of Africa (figure 5B). Figure 5 Long-lasting insecticide-treated net (LLIN) primary scale-up and usage statistics. (A) The LLIN usage at which an alternative intervention is first introduced for sub-Saharan Africa. For much of sub-Saharan Africa, LLIN scale-up to medium or high usage levels before any other intervention is implemented is the most cost-effective. In seasonal areas, indoor residual spraying or seasonal malaria chemoprevention can be the first most cost-effective intervention. (B) The distribution of country level LLIN usage estimates for 2015.3

, LLIN scale-up to medium or high usage levels before any other intervention is implemented is the most cost-effective. In seasonal areas, indoor residual spraying or seasonal malaria chemoprevention can be the first most cost-effective intervention. (B) The distribution of country level LLIN usage estimates for 2015.3 While the results vary for different assumed unit costs for each intervention, in a sensitivity analysis of these costs, we found that the order of scale-up is generally maintained (figure 6A). Since the price of the RTS,S vaccine has not been released, we additionally assessed the sensitivity to the assumed price per dose. Figure 6B shows the proportion of scenarios in which RTS,S, at a given price per dose, is estimated to occur before other interventions in the scale-up pathway. In general, RTS,S remains late in the pathway. However, this pattern changes if the price per dose drops below US$3 where the relative cost-effectiveness becomes comparable to IRS and SMC. However, even at a very low cost per dose (<US$1.00), LLINs are estimated to remain a more cost-effective intervention than RTS,S in approximately half of the settings.

pathway. However, this pattern changes if the price per dose drops below US$3 where the relative cost-effectiveness becomes comparable to IRS and SMC. However, even at a very low cost per dose (<US$1.00), LLINs are estimated to remain a more cost-effective intervention than RTS,S in approximately half of the settings. Figure 6 Sensitivity of the results to variations in costs of the interventions. (A) The sensitivity of the scale-up pathway to uncertainties in the health production functions determining the cost of each intervention. Columns represent 10 equally spaced samples (from left to right) along the scale-up pathway. Numbered cells denote the number of instances, out of 100 realisations, that a given intervention was implemented at that step. (B) Sensitivity of outcome to assumed cost per dose of RTS,S. The assumed cost per dose was decreased in incremental amounts. At each step, the proportion of settings in which the RTS,S was implemented before either long-lasting insecticide-treated nets, indoor residual spraying or seasonal malaria chemoprevention (blue, red and yellow lines, respectively) is shown.

ose of RTS,S. The assumed cost per dose was decreased in incremental amounts. At each step, the proportion of settings in which the RTS,S was implemented before either long-lasting insecticide-treated nets, indoor residual spraying or seasonal malaria chemoprevention (blue, red and yellow lines, respectively) is shown. Discussion Our analysis demonstrates that LLINs remain the most cost-effective first intervention to reduce malaria transmission across the broad range of transmission settings observed in sub-Saharan Africa. The high consumption and competitive market for LLINs has driven costs down over the last 10 years.21 This, coupled with their dual effect of personal-level and community-level protection,28 makes them highly cost-effective. This finding was consistent for all outcome measures considered (see online supplementary materials S1–S4). Furthermore, our results indicate that, based on cost-effectiveness considerations, the RTS,S vaccine should be considered a secondary intervention alongside the two other WHO-recommended malaria interventions for this region—SMC and IRS. The recommended schedule for the RTS,S vaccine that will be tested in pilot implementation is for four doses given in children aged 5–27 months, who constitute a small subset of the exposed population. The vaccine offers partial protection to this group over a duration of ∼4 years.4 14 As a result, this vaccine does not have the benefit of inducing herd-immunity in the population and is considerably more expensive per person than the other interventions considered here (at the assumed cost of $5 per dose), lowering its relative cost-effectiveness. Thus, other than in high transmission settings where there is a high burden of disease in young children, we find that RTS,S enters the cost-effectiveness scale-up pathway later on when other potential options for reducing transmission and/or protecting from disease are already maximised.

s relative cost-effectiveness. Thus, other than in high transmission settings where there is a high burden of disease in young children, we find that RTS,S enters the cost-effectiveness scale-up pathway later on when other potential options for reducing transmission and/or protecting from disease are already maximised. While increasing usage of LLINs is identified as the most cost-effective first intervention, once these levels reach 50–60%, we estimate that the three alternative interventions—IRS, SMC and RTS,S—become increasingly competitive when comparing the relative cost-effectiveness. This level of LLIN usage is similar to the levels reported for many countries in sub-Saharan Africa in 2015,3 suggesting that context-specific cost-effectiveness considerations may become increasingly important as investment in current or new interventions are considered. The inclusion of IRS as a secondary vector control option could further reduce onward transmission through providing additional protection for those that do not consistently use nets or over the periods between net distribution rounds in which the integrity of the net and/or efficacy of the insecticide has decayed. However, the trial data on the combination of these two vector control interventions remain inconclusive29 30 and hence close monitoring would be required to fully understand the operational impact of their combined use. Furthermore, any recommendations in favour of vector control must be made in the light of current and potential future insecticide resistance, the effects of which were not included in this analysis. In a small number of settings, characterised by a high level of seasonal transmission and intense transmission, we identified either SMC or IRS as the most cost-effective first intervention. SMC and IRS are temporally targeted at the peak transmission season, and this therefore increases their cost-effectiveness relative to non-seasonally targeted interventions.31 A comparison of the impact of pairwise combinations of interventions is included in online supplementary material S5.

ective first intervention. SMC and IRS are temporally targeted at the peak transmission season, and this therefore increases their cost-effectiveness relative to non-seasonally targeted interventions.31 A comparison of the impact of pairwise combinations of interventions is included in online supplementary material S5. We explicitly excluded treatment scale-up options from the cost-effectiveness analysis for several reasons. First, treatment has been previously shown to be highly cost-effective.6 Therefore, equitable access to treatment for severe disease is an ethical priority and universal scale-up of treatment coverage is important to preserve. Second, while increasing treatment coverage, and therefore costs, affects the absolute cost-effectiveness of other interventions (with lower treatment coverage making them more cost-effective), the relative cost-effectiveness when comparing interventions (and therefore scale-up pathways) will be influenced far less. Third, the ability of a country to increase coverage of treatment will depend critically on health system capacity and hence vary geographically. Thus, a simple unit cost approach is unlikely to be appropriate. For other interventions, the commodities (particularly LLINs) are purchased through Global Fund pathways for which there is a coordinated tendering process. We have therefore adopted a unit costing approach for the interventions considered in this broad-scale comparison.

nit cost approach is unlikely to be appropriate. For other interventions, the commodities (particularly LLINs) are purchased through Global Fund pathways for which there is a coordinated tendering process. We have therefore adopted a unit costing approach for the interventions considered in this broad-scale comparison. Systems-level inefficiencies,32 overallocation27 and systematic under-representation in hard-to-reach populations33 34 may all contribute to diminishing marginal returns when investing in increasing the coverage of an intervention to very high levels. Our results demonstrate that it is important to capture these non-linearities when considering the relative costs-effectiveness of introducing new interventions such as the RTS,S malaria vaccine. With the simple assumption that costs associated with increasing the usage or coverage of an intervention increases linearly, we observe a very clear picture of cost-effective scale-up, with LLIN usage increased to the maximum level (75%) in nearly all settings before any other intervention is implemented. However, the inclusion of non-linear productions functions that capture the increasing cost associated with achieving high levels of coverage of any given intervention leads to a more complicated picture of the cost-effective scale-up pathway. While there was considerable uncertainty in our estimated production functions for each intervention, general patterns in scale-up remained fairly robust to this. However, further data on these patterns are critical to inform local planning. Subtle differences in the inflection points of the production functions affect when a switch between interventions is made. This is especially of note for the SMC production function where lack of data lead to considerable uncertainties in the resultant production function.

tterns are critical to inform local planning. Subtle differences in the inflection points of the production functions affect when a switch between interventions is made. This is especially of note for the SMC production function where lack of data lead to considerable uncertainties in the resultant production function. While cost-based uncertainties were explored in this analysis, we did not additionally explore the uncertainty in model structure or parameterisation due to the computational complexity in undertaking such an analysis. Clearly, model parameters and structures could affect the relative impact of interventions as well as the combinations of interventions needed to reach pre-elimination levels. Determining the cost-effective scale-up in a stepwise manner informed by the ICER always chooses the next most cost-effective option. This is analogous to a gradual scale-up of interventions over time, where future options are considered, given an established intervention landscape. However, for a given spend, the resulting intervention package estimated from a stepwise approach may differ from a global optima if all scale-up options were considered in unison.

alogous to a gradual scale-up of interventions over time, where future options are considered, given an established intervention landscape. However, for a given spend, the resulting intervention package estimated from a stepwise approach may differ from a global optima if all scale-up options were considered in unison. Context-specific challenges to scaling-up a given intervention will always be present and cannot be represented in this style of analysis. To this end, policy decisions must take into account such challenges when considering recommendations in a specific setting. While a number of countries have achieved the high levels of LLIN usage that our analysis suggests to be cost-effective,3 in other settings, there may be barriers to achieving and sustaining the levels in the long term, nuancing the decision to target LLIN scale-up. Nevertheless, the ambitious targets set by the WHO for universal coverage may be integral in driving trends in net usage upwards, even if the target cannot be reached. Handling editor: Sanni Yaya. Contributors: PW prepared and performed the analysis and drafted the manuscript. PW, PGTW, JTG and ACG contributed to conceiving and designing the analysis and writing the final draft. PGTW, JTG and ACG contributed to analysis and interpretation of the data. PW is the corresponding author. Funding: The Bill & Melinda Gates Foundation (PW, ACG), UK Medical Research Council fellowships (PGTW, JTG), Malaria Vaccine Initiative (JTG and ACG) and MRC Centre Funding + DFID (all). Competing interests: None declared.

Contributors: PW prepared and performed the analysis and drafted the manuscript. PW, PGTW, JTG and ACG contributed to conceiving and designing the analysis and writing the final draft. PGTW, JTG and ACG contributed to analysis and interpretation of the data. PW is the corresponding author. Funding: The Bill & Melinda Gates Foundation (PW, ACG), UK Medical Research Council fellowships (PGTW, JTG), Malaria Vaccine Initiative (JTG and ACG) and MRC Centre Funding + DFID (all). Competing interests: None declared. Provenance and peer review: Not commissioned; externally peer reviewed. Data sharing statement: No additional data are available.

Key questions What is already known about this topic? Tuberculosis (TB) drug resistance is a major public health challenge globally and in India; a key contributing factor is poor knowledge of TB diagnosis and treatment among providers. Evidence from systematic reviews shows long diagnostic delays for TB, with patients frequently switching providers—suggesting they do not receive appropriate care that they are satisfied with. Recent studies in India have also demonstrated gaps between what providers know and what they do in clinical practice—the know-do gap. What are the new findings? This study measures providers' knowledge of diagnosis and treatment of TB, using a sampling method designed to estimate competence of providers who are most commonly visited by households in study areas. In addition to finding low levels of provider competence in diagnosis and treatment of TB, we also find evidence suggesting that, among all providers who report that they treat TB cases, there is no significant association between having formal medical training and provider competence. Providers with medical training were not more likely to diagnose or treat TB correctly compared with those without formal training. Our analysis also demonstrates the severe information asymmetry problems in the healthcare market because of which patients in these settings are unable to rely on common signals of quality such as medical degrees or experience to infer provider competence.

In addition to finding low levels of provider competence in diagnosis and treatment of TB, we also find evidence suggesting that, among all providers who report that they treat TB cases, there is no significant association between having formal medical training and provider competence. Providers with medical training were not more likely to diagnose or treat TB correctly compared with those without formal training. Our analysis also demonstrates the severe information asymmetry problems in the healthcare market because of which patients in these settings are unable to rely on common signals of quality such as medical degrees or experience to infer provider competence. Recommendations for policy Our results highlight the need for policies to improve training, incentives, task shifting and regulation to improve knowledge and performance of existing providers in the healthcare system. Introduction The scale of India's tuberculosis (TB) burden looms large, contributing almost a quarter of the 9.6 million cases worldwide.1 Bihar, one of India's largest and poorest states, bears a substantial share of India's TB burden, as active TB disease is associated with poverty.2–4 With over 100 million inhabitants and a per-capita annual income of $502i in 2011, less than half of India's national average, Bihar's healthcare system registered almost 68 000 new TB patients in 2015 alone.5 6

states, bears a substantial share of India's TB burden, as active TB disease is associated with poverty.2–4 With over 100 million inhabitants and a per-capita annual income of $502i in 2011, less than half of India's national average, Bihar's healthcare system registered almost 68 000 new TB patients in 2015 alone.5 6 A major challenge to achieving improved outcomes in Bihar is the generally poor quality of medical care available to patients in rural areas.7 In the public sector, lack of availability of trained providers and absenteeism among medical care providers is a key limitation—as high as 67% in primary health centres among doctors and 52% among nurses.8 9 Not surprisingly then, over 90% of the healthcare used by households (outpatient care) in rural Bihar is provided by the private sector, of which 70% is from informal sector providers.10 The net result of widespread absenteeism in the formal public sector, untrained informal sector providers and low levels of provider effort is the alarmingly low quality of care provided to patients.

ds (outpatient care) in rural Bihar is provided by the private sector, of which 70% is from informal sector providers.10 The net result of widespread absenteeism in the formal public sector, untrained informal sector providers and low levels of provider effort is the alarmingly low quality of care provided to patients. Active TB disease is treatable, and multidrug resistance potentially avoidable, provided that cases can be correctly diagnosed and appropriate treatment regimens are administered for adequate periods of time. However, systematic reviews from India have shown long diagnostic delays with patients frequently switching providers,11 and have also shown poor quality of TB care in India.12 Previous studies of rural patients in Bihar treated for TB using DOTS (Directly Observed Treatment, Short course) in the public sector show high rates of drop-out and symptom persistence, despite completing treatment.13 Another study that sampled 371 174 individuals in 30 districts across India shows that nearly half of those with TB who sought care did so in private sector and non-DOTS settings.14 Yet, the potential for higher quality TB diagnosis and treatment as represented by the limits of provider knowledge remains ill characterised in settings like Bihar, especially among private sector providers who are the first point of medical contact for most rural patients. There are no statewide efforts to engage the informal private sector in TB control.

B diagnosis and treatment as represented by the limits of provider knowledge remains ill characterised in settings like Bihar, especially among private sector providers who are the first point of medical contact for most rural patients. There are no statewide efforts to engage the informal private sector in TB control. This paper contributes to the literature on quality of care in rural areas in developing countries, focusing on the diagnosis and treatment of TB—a chronic, communicable disease of global significance.1 We estimate the knowledge of healthcare providers in rural Bihar, India, in terms of providing a correct diagnosis of TB and prescribing the appropriate treatment when interviewed using clinical vignettes. We also analyse how healthcare provider characteristics that are observable to patients predict the probability of a correct diagnosis and treatment for the vignettes. Findings from these analyses have important public health and policy implications for improving low levels of provider knowledge and increasing the quality of TB diagnosis and treatment. Methods Setting We analyse data from provider quality assessments undertaken as part of baseline surveys conducted for the Bihar Evaluation of Social Franchising and Telemedicine (BEST) project—an evaluation of a large telemedicine programme funded by the Bill and Melinda Gates Foundation in Bihar.15

This paper contributes to the literature on quality of care in rural areas in developing countries, focusing on the diagnosis and treatment of TB—a chronic, communicable disease of global significance.1 We estimate the knowledge of healthcare providers in rural Bihar, India, in terms of providing a correct diagnosis of TB and prescribing the appropriate treatment when interviewed using clinical vignettes. We also analyse how healthcare provider characteristics that are observable to patients predict the probability of a correct diagnosis and treatment for the vignettes. Findings from these analyses have important public health and policy implications for improving low levels of provider knowledge and increasing the quality of TB diagnosis and treatment. Methods Setting We analyse data from provider quality assessments undertaken as part of baseline surveys conducted for the Bihar Evaluation of Social Franchising and Telemedicine (BEST) project—an evaluation of a large telemedicine programme funded by the Bill and Melinda Gates Foundation in Bihar.15 Sampling method The provider surveys and quality assessments were conducted in 80 randomly selected clusters (out of 360 rural clusters) in the study, representing rural areas from 11 districts. Clusters in the study were defined to represent market catchment areas with a population of ∼20 000 as described in Mohanan et al.16 The objective of the provider surveys was to assess the knowledge of the providers who deliver care to households surveyed at baseline. In each cluster, data from interviews with 64 randomly selected households were used to generate a list of all providers visited in the past 6 months, regardless of medical training of the providers. The number of households was chosen based on power calculations to estimate the impact of the programme on population health outcomes.16 We selected the five most frequently visited providers as reported by the 64 randomly selected households in each cluster for inclusion in the present study, in order to have 90% power to detect a 20% improvement in quality of care provided. The total number of providers in the study areas ranged between 6 and 70;16 the current study focuses on the 5 most commonly visited providers in these clusters. Since some clusters had fewer than 5 providers, our final sample includes 395 providers. We administered surveys as well as a series of clinical vignettes to each sampled provider, including a vignette for a case of suspected pulmonary TB. The BEST study protocol was approved by Duke University (29755) and India's Health Ministry Steering Committee (number 12/2008/30-HMSC/4).

sample includes 395 providers. We administered surveys as well as a series of clinical vignettes to each sampled provider, including a vignette for a case of suspected pulmonary TB. The BEST study protocol was approved by Duke University (29755) and India's Health Ministry Steering Committee (number 12/2008/30-HMSC/4). Data collection instruments We use data from the provider surveys and provider responses to vignettes to assess the quality of care available for TB patients. Trained interviewers first administered a detailed structured survey to each sampled provider. The survey collected information on provider characteristics, such as age, education, medical qualifications, experience, types of clinical activities in practice, types of illnesses treated and infrastructure in the facility in which they practice. To measure provider knowledge, we analyse data from clinical vignettes. The vignette method involves presenting a hypothetical case to the provider in an interview setting with two interviewers, one reading out scripted answers to the provider's questions and the other recording all of the provider's responses to the vignette. The case intends to represent a new pulmonary TB patient, who is visiting a healthcare provider for the first time, with productive cough of more than 2 weeks, accompanied by chest pain, haemoptysis, loss of appetite, weight loss, night sweats and fever.

other recording all of the provider's responses to the vignette. The case intends to represent a new pulmonary TB patient, who is visiting a healthcare provider for the first time, with productive cough of more than 2 weeks, accompanied by chest pain, haemoptysis, loss of appetite, weight loss, night sweats and fever. The TB vignette starts with the interviewer telling the provider to assume that a man aged 40 years visits the provider and that he will comply with all tests and medications that the provider might recommend and will return if required. The patient reports, “Doctor, I have been suffering from fever, cough and weakness, and I have been losing weight.” The provider then proceeds to ask history questions (eg, ‘Do you have fever with chills?’) and the interviewer reads out the scripted responses (‘No’). If the provider says she would examine the patient and check his pulse, respiratory rate or auscultate his chest, the interviewer will read out the appropriate responses (‘80 bpm’, ‘20 breaths/minute’, ‘normal breath sounds’, respectively). Similarly, the vignettes also provide results on tests that the provider might recommend—if specific blood tests were recommended, the enumerators would read out test results. An X-ray was provided on request showing opacity in the right apex. After pilot testing, and based on previous studies conducted in a range of settings,17 as well as inputs from local clinicians, the vignettes were designed to include clinically relevant information as well as information that were commonly asked for by providers for social or cultural reasons (such as marital status or number of children). (See vignette modules included in online supplementary appendix).

s well as inputs from local clinicians, the vignettes were designed to include clinically relevant information as well as information that were commonly asked for by providers for social or cultural reasons (such as marital status or number of children). (See vignette modules included in online supplementary appendix). The vignette responses provide information about whether the doctor is able to ask the most clinically relevant questions, to establish the correct diagnosis and also detailed information about the investigations recommended and treatment prescribed. Analyses of the providers' characteristics in relation to their performance on the vignettes (ie, correctly diagnosing and offering appropriate treatment) form the core of our analysis. Statistical analysis Descriptive statistics for provider characteristics were computed to compare groups of providers with and without medical qualifications. Differences between these two groups were tested using unpaired two-tailed t-tests and χ2 tests of proportions, with SEs adjusted for the study design.

The vignette responses provide information about whether the doctor is able to ask the most clinically relevant questions, to establish the correct diagnosis and also detailed information about the investigations recommended and treatment prescribed. Analyses of the providers' characteristics in relation to their performance on the vignettes (ie, correctly diagnosing and offering appropriate treatment) form the core of our analysis. Statistical analysis Descriptive statistics for provider characteristics were computed to compare groups of providers with and without medical qualifications. Differences between these two groups were tested using unpaired two-tailed t-tests and χ2 tests of proportions, with SEs adjusted for the study design. We assessed providers' performance on vignettes based on: (1) correct diagnosis; (2) correct treatment and (3) recommendation of a referral to another provider or hospital. Of note, a large share of providers reported not seeing or treating TB patients: 24.6% of those with medical qualifications and 67.2% of those without. While it is possible to analyse provider TB vignette performance only for those reporting seeing and treating TB patients, we chose instead to analyse all providers based on the following rationale. Since TB is endemic in rural Bihar, patients may present with symptoms of their illness even to providers who do not claim to treat TB since patients will not know that they are suffering from TB when they visit the provider. In such a situation, if the provider does not recognise the symptoms, and ask the right diagnostic questions or perform the minimum necessary examinations, TB patients would still receive delayed TB diagnosis and poor quality healthcare. We include providers with and without formal qualifications. Practitioners with little to no formal training provide most healthcare in rural areas in India, and previous studies on quality of care in rural India report that the quality of care provided by practitioners with formal qualifications is also poor .ii 7 18–21 Hence, our analysis of provider performance on vignettes includes all providers in our sample. We also include analogous analyses and findings after restricting to providers who claim to treat TB in the online supplementary appendix.

of care provided by practitioners with formal qualifications is also poor .ii 7 18–21 Hence, our analysis of provider performance on vignettes includes all providers in our sample. We also include analogous analyses and findings after restricting to providers who claim to treat TB in the online supplementary appendix. For diagnosis and treatment, we concentrated our analyses on three aspects: (1) diagnostic process; (2) providing a correct diagnosis and (3) whether or not the correct treatment was prescribed. We assessed the diagnostic process each provider stated that he/she would undertake (ie, the questions, examinations and tests used to form a diagnosis) relative to standard diagnostic procedures to measure ‘knowledge’. We summarised provider knowledge using Item Response Theory (IRT) to calculate a knowledge score for each provider using previously developed methods.17 The IRT methodology is a model-based measurement used to describe the relation between how the provider responds to a set of questions and the level of the ‘latent variable’ (knowledge) being measured by the scale. It is widely used in settings to assess items in questionnaires where participants are scored on multiple items to recover an underlying latent trait or ability. In the context of our paper, a correct response to an item is obtained every time a provider asks a key diagnostic question (such as duration of cough), or for results of diagnostic tests (such as sputum smear) or performs a relevant examination (listed in table 2), and our latent variable or trait is the provider's overall knowledge. We use a three-parametric logistic (3PL) model to construct our knowledge index following Das and Hammer.17 Regardless of diagnostic process, we assessed whether each provider correctly stated a diagnosis of the case in the vignette as being TB.

2), and our latent variable or trait is the provider's overall knowledge. We use a three-parametric logistic (3PL) model to construct our knowledge index following Das and Hammer.17 Regardless of diagnostic process, we assessed whether each provider correctly stated a diagnosis of the case in the vignette as being TB. In identifying the appropriate TB treatment, we followed the WHO 2010 guidelines and WHO's 2014 Standards for TB Care in India.22 The WHO treatment guidelines state that the treatment of new TB cases include 6 months of rifampicin as part of a multidrug regimen (2HRZE+4 HR).iii However, since our investigators were unable to collect information on duration of each of the prescribed drugs consistently, with some providers using generic pharmaceutical names of drugs and others using brand names, we employ a broader definition of ‘correct’ treatment. We defined correct treatment to include all prescriptions that included 6 months of rifampicin as part of a multidrug treatment that also included any duration of isoniazid.

th some providers using generic pharmaceutical names of drugs and others using brand names, we employ a broader definition of ‘correct’ treatment. We defined correct treatment to include all prescriptions that included 6 months of rifampicin as part of a multidrug treatment that also included any duration of isoniazid. We examined how a provider's observable characteristics are associated with their knowledge using multivariable linear regression models because these characteristics are the features that are available to patients to choose between providers. We also examined how knowledge along with a provider's observable characteristics related to the probability of making a correct diagnosis, providing correct treatment and making a referral using multivariable probit regression models.iv Further, we conducted analyses where we restricted the assessment of correct treatment provision only to those providers who offered any treatment. All regression models controlled for the age of the provider, years of experience and medical qualification. Additionally, in separate specifications, we controlled for the type of medicine practiced, the number of working hours per week, average patient caseload per day, whether providers engaged in public events like running medical camps, whether the clinic was public or private, whether provider claimed to treat TB, sold medicines at the clinic, infrastructure index and average fee charged by provider. All analyses adjusted SEs for survey design by clustering at the level of a cluster in our study.v

engaged in public events like running medical camps, whether the clinic was public or private, whether provider claimed to treat TB, sold medicines at the clinic, infrastructure index and average fee charged by provider. All analyses adjusted SEs for survey design by clustering at the level of a cluster in our study.v Results Provider characteristics Of the 395 providers most commonly visited by representative households in our study areas and interviewed in our study, 79.5% (314) did not have any formal medical qualifications. Among those with medical qualifications, less than half (35) had MBBS (Bachelor of Medicine, Bachelor of Surgery—the equivalent of MD in the USA) degrees or higher, while the remaining 46 had degrees or diplomas in Ayurveda, Homeopathy or Unani systems of medicine (BAMS/BHMS/BUMS),23 hereafter abbreviated as BA/H/UMS. Among the 314 providers without formal qualifications, 20.7% (65 out of 314) had some training such as pharmacist or registered medical practitioner (RMP); a small fraction reported informal trainingvi where they had worked with other doctors in the past, while the vast majority (228) had no formal or informal medical qualification, as seen in figure 1. Figure 1 Distribution of providers by qualification.

Results Provider characteristics Of the 395 providers most commonly visited by representative households in our study areas and interviewed in our study, 79.5% (314) did not have any formal medical qualifications. Among those with medical qualifications, less than half (35) had MBBS (Bachelor of Medicine, Bachelor of Surgery—the equivalent of MD in the USA) degrees or higher, while the remaining 46 had degrees or diplomas in Ayurveda, Homeopathy or Unani systems of medicine (BAMS/BHMS/BUMS),23 hereafter abbreviated as BA/H/UMS. Among the 314 providers without formal qualifications, 20.7% (65 out of 314) had some training such as pharmacist or registered medical practitioner (RMP); a small fraction reported informal trainingvi where they had worked with other doctors in the past, while the vast majority (228) had no formal or informal medical qualification, as seen in figure 1. Figure 1 Distribution of providers by qualification. Table 1 describes provider characteristics, among those with and without medical qualifications. While providers with formal qualifications have comparable years of experience as those without qualifications as well as comparable levels of ownership of the facilities that they practice in, they vary significantly in the range of services they provide. MBBS providers report longer working hours per week, higher participation in camps and significantly lower rates of providing treatment as part of their consultation and selling drugs than providers with training in other systems and those without training. In rural settings such as the one where this study was conducted, providers frequently carry their own stock of medicines and offer medication (treatment) as part of the consultation and charge a combined fee. We recorded this characteristic of the provider/facility as ‘administering treatment’ in the survey (90.8% among the unqualified and 76.1% among the BA/H/UMS group compared with 57.1% among MBBS providers). Some providers sell drugs separately as well (53.2% among the unqualified and 41.3% among the BA/H/UMS group compared with 8.6% among MBBS providers). Providers with MBBS training work in clinics with higher levels of infrastructure and also command fees that are almost twice as high as BA/H/UMS providers and over three times as high as those without formal medical qualifications.

nqualified and 41.3% among the BA/H/UMS group compared with 8.6% among MBBS providers). Providers with MBBS training work in clinics with higher levels of infrastructure and also command fees that are almost twice as high as BA/H/UMS providers and over three times as high as those without formal medical qualifications. Table 1 Providers characteristics according to medical education

nqualified and 41.3% among the BA/H/UMS group compared with 8.6% among MBBS providers). Providers with MBBS training work in clinics with higher levels of infrastructure and also command fees that are almost twice as high as BA/H/UMS providers and over three times as high as those without formal medical qualifications. Table 1 Providers characteristics according to medical education Variables [1] MBBS [2] BA/H/UMS [3] Other [4] DIFF 1–2 [5] DIFF 1–3 Age (years) 45.7 (42.2 to 49.2) 46.8 (43.6 to 50) 43.5 (42.2 to 44.7) −1.1 2.2 Education >high School (%) 100 (– to –) 97.8 (93.6 to 102.1) 70.7 (65.7 to 75.7) 2.2 29.30*** Has ever used a computer (%) 62.9 (46.6 to 79.1) 30.4 (17 to 43.9) 12.4 (8.8 to 16.1) 32.42*** 50.44*** Experience (years) 18.5 (15 to 22.1) 19 (16 to 21.9) 18.2 (17.1 to 19.4) −0.4 0.3 Average patient caseload (day) 29.3 (23.5 to 35.1) 17.1 (15.6 to 18.6) 17.2 (16.5 to 18) 12.17*** 12.06*** Working hours (per week) 61.8 (56.3 to 67.4) 52.2 (47.2 to 57.2) 48.3 (46.4 to 50.3) 9.61** 13.51*** Run camps (%) 40 (23.5 to 56.5) 6.5 (−0.7 to 13.7) 4.5 (2.2 to 6.7) 33.48*** 35.54*** Public health facility (%) 22.9 (8.7 to 37) 2.2 (−2.1 to 6.4) 0.3 (−0.3 to 0.9) 20.68*** 22.54*** Infrastructure Index 3.1 (2 to 4.3) 0.2 (−0.3 to 0.7) −0.4 (−0.5 to −0.3) 2.92*** 3.53*** Consultation fee (Rs) 64.2 (44.8 to 83.6) 38.6 (26.4 to 50.9) 15.5 (13.3 to 17.6) 25.60** 48.75*** Task (% of providers) Consultation with patients 100 (– to –) 100 (– to –) 99.7 (99.1 to 100.3) 0.0 0.3 Administering treatment 57.1 (40.5 to 73.8) 76.1 (63.6 to 88.5) 90.8 (87.6 to 94) – –33.62*** Selling drugs 8.6 (−0.8 to 18) 41.3 (26.9 to 55.7) 53.2 (47.7 to 58.7) –32.73*** –44.61*** Laboratory-related duties 17.1 (4.5 to 29.8) 6.5 (−0.7 to 13.7) 3.8 (1.7 to 5.9) 10.6 13.3 Administrative duties 48.6 (31.8 to 65.4) 71.7 (58.6 to 84.9) 61.8 (56.4 to 67.2) –23.17** −13.2 Ownership 51.4 (34.6 to 68.2) 76.1 (63.6 to 88.5) 72 (67 to 77) –24.66** –20.55** Type of medicine practiced (%) Allopathic 94.3 (86.5 to 100) 89.1 (80 to 98.2) 91.4 (88.3 to 94.5) 5.2 2.9 Homeopathic/Ayurvedic 20 (6.6 to 33.4) 63 (48.9 to 77.1) 32.8 (27.6 to 38) –43.04*** −12.8 Type of diseases treated (%) Tuberculosis 88.6 (77.9 to 99.3) 65.2 (51.3 to 79.1) 32.8 (27.6 to 38) 23.35*** 55.77*** VL 2.9 (−2.7 to 8.5) 4.3 (−1.6 to 10.3) 1.9 (0.4 to 3.4) −1.5 1.0 Observations 35 46 314 Columns 1–3 report mean (95% CI) and columns 4 and 5 report differences. BA/H/UMS includes BAMS, BUMS and BHMS degrees as well as Diploma in Ayurvedic and some others MD degrees.

2 (51.3 to 79.1) 32.8 (27.6 to 38) 23.35*** 55.77*** VL 2.9 (−2.7 to 8.5) 4.3 (−1.6 to 10.3) 1.9 (0.4 to 3.4) −1.5 1.0 Observations 35 46 314 Columns 1–3 report mean (95% CI) and columns 4 and 5 report differences. BA/H/UMS includes BAMS, BUMS and BHMS degrees as well as Diploma in Ayurvedic and some others MD degrees. Providers classified as ‘Other’ includes all providers with NO medical training or those with coursework related in some way to medicine such as pharmacist or informal training. The infrastructure index was computed according to the following variables: electricity, power backup, number of consulting rooms, number of bed for day observation, provision of tests, provision of X-rays and computer system. Asterisks represent statistically significant differences, with ***p<0.01, **p<0.05, *p<0.1. Source: Providers Interview.

Providers classified as ‘Other’ includes all providers with NO medical training or those with coursework related in some way to medicine such as pharmacist or informal training. The infrastructure index was computed according to the following variables: electricity, power backup, number of consulting rooms, number of bed for day observation, provision of tests, provision of X-rays and computer system. Asterisks represent statistically significant differences, with ***p<0.01, **p<0.05, *p<0.1. Source: Providers Interview. When asked whether they provide treatment for TB, 88.6% of MBBS providers reported treating patients, as did 65.2% of BA/H/UMS providers and 32.8% of those without qualifications. While it is tempting to conclude that most providers in the sample—especially those without medical qualifications—do not manage TB, such a conclusion would miss a critical point that patients often do not know what their underlying illness is, at least in the early stages of the care-seeking pathway. Patients experience symptoms and seek care from providers whom they frequently visit.11 24–27 The provider then has to diagnose the condition, and decide whether to treat the patient or refer them to a hospital. If providers incorrectly diagnose TB as another condition, they may incorrectly also report that they do not treat TB when surveyed.

toms and seek care from providers whom they frequently visit.11 24–27 The provider then has to diagnose the condition, and decide whether to treat the patient or refer them to a hospital. If providers incorrectly diagnose TB as another condition, they may incorrectly also report that they do not treat TB when surveyed. Provider knowledge: diagnostic questions, treatment and referral Table 2 shows the overall fraction of providers and fraction by type of provider, who asked or performed key diagnostic questions and examinations, made the right diagnosis and provided the correct treatment. The most common question asked by providers was duration of fever (48.9%, 95% CI 43.8% to 53.9%). Only 32.2% (95% CI 27.6% to 37.0%) of providers asked a key question related to TB diagnosis about duration of cough (30.9% (95% CI 24.9% to 36.9%) among unqualified, relative to 37.1% (95% CI 22.7% to 51.5%) among MBBS providers). Table 2 Fraction of providers who asked or performed key diagnostic questions and examinations, diagnosis and treatment and average competence score

Provider knowledge: diagnostic questions, treatment and referral Table 2 shows the overall fraction of providers and fraction by type of provider, who asked or performed key diagnostic questions and examinations, made the right diagnosis and provided the correct treatment. The most common question asked by providers was duration of fever (48.9%, 95% CI 43.8% to 53.9%). Only 32.2% (95% CI 27.6% to 37.0%) of providers asked a key question related to TB diagnosis about duration of cough (30.9% (95% CI 24.9% to 36.9%) among unqualified, relative to 37.1% (95% CI 22.7% to 51.5%) among MBBS providers). Table 2 Fraction of providers who asked or performed key diagnostic questions and examinations, diagnosis and treatment and average competence score All providers MBBS [1] BA/H/U/MS [2] Other [3] DIFF 1–2 DIFF 1–3 Questions and examinations Since when has he had the fever 48.9 (43.8 to 53.9) 51.4 (32.7 to 70.1) 47.8 (33.1 to 62.5) 48.7 (42.7 to 54.7) 3.6 2.7 Fever with chills 31.4 (26.8 to 36.2) 31.4 (16 to 46.9) 32.6 (18.6 to 46.6) 31.2 (25.5 to 37.0) −1.2 0.2 Is fever continuous 33.4 (28.8 to 38.3) 34.3 (16.5 to 52.1) 39.1 (23.5 to 54.8) 32.5 (26.8 to 38.2) −4.8 1.8 Are there any night sweats present 6.1 (3.9 to 8.9) 11.4 (2.2 to 20.6) 6.5 (−0.5 to 13.6) 5.4 (2.9 to 7.9) 4.9 6.0 Cough since when 32.2 (27.6 to 37) 37.1 (22.7 to 51.5) 37.0 (20 to 53.9) 30.9 (24.9 to 36.9) 0.2 6.3 Pain in the chest 13.7 (10.4 to 17.5) 20.0 (5.1 to 34.9) 21.7 (9.5 to 34) 11.8 (8.2 to 15.4) −1.7 8.2 Is there sputum 24.6 (20.4 to 29.1) 34.3 (20.2 to 48.4) 23.9 (10.4 to 37.4) 23.6 (19.0 to 28.2) 10.4 10.7 How is the sputum 24.6 (20.4 to 29.1) 37.1 (20 to 54.3) 23.9 (9.7 to 38.1) 23.2 (18.8 to 27.7) 13.2 13.9 Blood in the sputum 11.1 (8.2 to 14.7) 14.3 (0.7 to 27.9) 13.0 (3.3 to 22.8) 10.5 (7.3 to 13.7) 1.2 3.8 How much blood in the sputum 10.9 (8.0 to 14.4) 8.6 (−0.4 to 17.6) 10.9 (1.4 to 20.3) 11.1 (7.4 to 14.8) −2.3 −2.6 Have you been eating less 13.2 (10 to 16.9) 28.6 (12.3 to 44.8) 23.9 (10.8 to 37.1) 9.9 (6.3 to 13.4) 4.7 18.7** Have you visited other doctors before coming here 15.9 (12.5 to 19.9) 20.0 (4.2 to 35.8) 8.7 (0.2 to 17.2) 16.6 (11.7 to 21.4) 11.3 3.4 Weight 10.6 (7.8 to 14.1) 25.7 (8.5 to 43) 13.0 (2.8 to 23.3) 8.6 (5.2 to 12.0) 12.7 17.1* Temperature 16.7 (13.2 to 20.8) 20.0 (7.1 to 32.9) 13.0 (3.0 to 23.1) 16.9 (11.8 to 22.0) 7.0 3.1 Blood for tlc/dlc 32.4 (27.8 to 37.3) 45.7 (27.3 to 64.1) 28.3 (14.8 to 41.7) 31.5 (25.2 to 37.8) 17.5 14.2 Blood test hb 15.9 (12.5 to 19.9) 34.3 (18.5 to 50) 10.9 (1.2 to 20.5) 14.6 (10.0 to 19.3) 23.4** 19.6** Blood for fasting ESR (erythrocytic sedimentation rate) 32.4 (27.8 to 37.3) 37.1 (20.3 to 54) 26.1 (12.6 to 39.6) 32.8 (26.1 to 39.5) 11.1 4.3 Mantaux tuberculin skin test 12.7 (9.5 to 16.3) 17.1 (3.5 to 30.8) 13.0 (2.8 to 23.3) 12.1 (8.4 to 15.8) 4.1 5.0 Sputum for AFB (acid-fast bacilli) 20.0 (16.2 to 24.3) 54.3 (35.6 to 73) 15.2 (4.5 to 26.0) 16.9 (12.4 to 21.3) 39.1***

32.4 (27.8 to 37.3) 37.1 (20.3 to 54) 26.1 (12.6 to 39.6) 32.8 (26.1 to 39.5) 11.1 4.3 Mantaux tuberculin skin test 12.7 (9.5 to 16.3) 17.1 (3.5 to 30.8) 13.0 (2.8 to 23.3) 12.1 (8.4 to 15.8) 4.1 5.0 Sputum for AFB (acid-fast bacilli) 20.0 (16.2 to 24.3) 54.3 (35.6 to 73) 15.2 (4.5 to 26.0) 16.9 (12.4 to 21.3) 39.1*** 37.4*** Chest X-ray 31.9 (27.3 to 36.7) 57.1 (38.2 to 76.1) 28.3 (13.6 to 42.9) 29.6 (23.9 to 35.3) 28.9** 27.5*** TB test 2.3 (1.0 to 4.3) 2.9 (−2.8 to 8.5) 4.3 (−1.7 to 10.4) 1.9 (0.4 to 3.4) −1.5 1.0 Diagnosis Gave any diagnosis 92.4 (89.3 to 94.8) 97.1 (91.3 to 103) 95.7 (89.4 to 101.9) 91.4 (88.3 to 94.5) 1.5 5.7* Correct diagnosis 60.0 (55.0 to 64.9) 91.4 (81.8 to 101) 73.9 (60.8 to 87) 54.5 (48.8 to 60.1) 17.5** 37.0*** Correct diagnosis, if any 64.9 (59.8 to 69.8) 94.1 (86.1 to 102.2) 77.3 (64.4 to 90.2) 59.6 (53.6 to 65.6) 16.8** 34.5*** Treatment Gave any treatment 66.6 (61.7 to 71.2) 82.9 (71.0 to 94.8) 80.4 (68.1 to 92.8) 62.7 (57.2 to 68.3) 2.4 20.1*** Correct treatment 14.4 (11.1 to 18.3) 45.7 (26.0 to 65.4) 19.6 (8.6 to 30.6) 10.2 (6.8 to 13.6) 26.2** 35.5*** Correct treatment, if any 21.7 (16.8 to 27.1) 55.2 (33.4 to 76.9) 24.3 (11.2 to 37.5) 16.2 (11.1 to 21.4) 30.9** 38.9*** Others Recommend referral 48.9 (43.8 to 53.9) 77.1 (63.3 to 91.0) 63.0 (48.8 to 77.3) 43.6 (37.4 to 49.9) 14.1 33.5*** Knowledge score −1.0 (−1.2 to −0.73) 0.0 (−0.6 to 0.6) −1.3 (−2.1 to −0.45) −1.0 (−1.3 to −0.7) 1.3** 1.0*** Values are percentage except for the competence score variable. Source: Vignette survey. Observations from 395 providers. CIs reported in parentheses.

37.4*** Chest X-ray 31.9 (27.3 to 36.7) 57.1 (38.2 to 76.1) 28.3 (13.6 to 42.9) 29.6 (23.9 to 35.3) 28.9** 27.5*** TB test 2.3 (1.0 to 4.3) 2.9 (−2.8 to 8.5) 4.3 (−1.7 to 10.4) 1.9 (0.4 to 3.4) −1.5 1.0 Diagnosis Gave any diagnosis 92.4 (89.3 to 94.8) 97.1 (91.3 to 103) 95.7 (89.4 to 101.9) 91.4 (88.3 to 94.5) 1.5 5.7* Correct diagnosis 60.0 (55.0 to 64.9) 91.4 (81.8 to 101) 73.9 (60.8 to 87) 54.5 (48.8 to 60.1) 17.5** 37.0*** Correct diagnosis, if any 64.9 (59.8 to 69.8) 94.1 (86.1 to 102.2) 77.3 (64.4 to 90.2) 59.6 (53.6 to 65.6) 16.8** 34.5*** Treatment Gave any treatment 66.6 (61.7 to 71.2) 82.9 (71.0 to 94.8) 80.4 (68.1 to 92.8) 62.7 (57.2 to 68.3) 2.4 20.1*** Correct treatment 14.4 (11.1 to 18.3) 45.7 (26.0 to 65.4) 19.6 (8.6 to 30.6) 10.2 (6.8 to 13.6) 26.2** 35.5*** Correct treatment, if any 21.7 (16.8 to 27.1) 55.2 (33.4 to 76.9) 24.3 (11.2 to 37.5) 16.2 (11.1 to 21.4) 30.9** 38.9*** Others Recommend referral 48.9 (43.8 to 53.9) 77.1 (63.3 to 91.0) 63.0 (48.8 to 77.3) 43.6 (37.4 to 49.9) 14.1 33.5*** Knowledge score −1.0 (−1.2 to −0.73) 0.0 (−0.6 to 0.6) −1.3 (−2.1 to −0.45) −1.0 (−1.3 to −0.7) 1.3** 1.0*** Values are percentage except for the competence score variable. Source: Vignette survey. Observations from 395 providers. CIs reported in parentheses. Asterisks represent statistically significant differences, with ***p<0.01, **p<0.05, *p<0.1.

37.4*** Chest X-ray 31.9 (27.3 to 36.7) 57.1 (38.2 to 76.1) 28.3 (13.6 to 42.9) 29.6 (23.9 to 35.3) 28.9** 27.5*** TB test 2.3 (1.0 to 4.3) 2.9 (−2.8 to 8.5) 4.3 (−1.7 to 10.4) 1.9 (0.4 to 3.4) −1.5 1.0 Diagnosis Gave any diagnosis 92.4 (89.3 to 94.8) 97.1 (91.3 to 103) 95.7 (89.4 to 101.9) 91.4 (88.3 to 94.5) 1.5 5.7* Correct diagnosis 60.0 (55.0 to 64.9) 91.4 (81.8 to 101) 73.9 (60.8 to 87) 54.5 (48.8 to 60.1) 17.5** 37.0*** Correct diagnosis, if any 64.9 (59.8 to 69.8) 94.1 (86.1 to 102.2) 77.3 (64.4 to 90.2) 59.6 (53.6 to 65.6) 16.8** 34.5*** Treatment Gave any treatment 66.6 (61.7 to 71.2) 82.9 (71.0 to 94.8) 80.4 (68.1 to 92.8) 62.7 (57.2 to 68.3) 2.4 20.1*** Correct treatment 14.4 (11.1 to 18.3) 45.7 (26.0 to 65.4) 19.6 (8.6 to 30.6) 10.2 (6.8 to 13.6) 26.2** 35.5*** Correct treatment, if any 21.7 (16.8 to 27.1) 55.2 (33.4 to 76.9) 24.3 (11.2 to 37.5) 16.2 (11.1 to 21.4) 30.9** 38.9*** Others Recommend referral 48.9 (43.8 to 53.9) 77.1 (63.3 to 91.0) 63.0 (48.8 to 77.3) 43.6 (37.4 to 49.9) 14.1 33.5*** Knowledge score −1.0 (−1.2 to −0.73) 0.0 (−0.6 to 0.6) −1.3 (−2.1 to −0.45) −1.0 (−1.3 to −0.7) 1.3** 1.0*** Values are percentage except for the competence score variable. Source: Vignette survey. Observations from 395 providers. CIs reported in parentheses. Asterisks represent statistically significant differences, with ***p<0.01, **p<0.05, *p<0.1. Few providers asked for results of common diagnostic tests, though MBBS doctors were significantly more likely to ask than other types of providers (table 2). Overall, 31.9% of providers asked for a chest X-ray, 20% asked for results from a Sputum test and 12.7% asked for a Mantoux tuberculin skin test. (See vignette in online supplementary appendix for positive results reported for each of the tests.) While 57.1% of MBBS providers asked for a chest X-ray, only 29.6% of unqualified providers and 28.3% of BA/H/UMS providers did so. Similarly, 16.9% of unqualified providers and 15.2% of BA/H/UMS providers asked for sputum test results relative to 54.3% of MBBS providers. The average number of questions asked per provider was 2.66 and the average number of examinations performed was 1.75, with no significant differences across provider types.

d so. Similarly, 16.9% of unqualified providers and 15.2% of BA/H/UMS providers asked for sputum test results relative to 54.3% of MBBS providers. The average number of questions asked per provider was 2.66 and the average number of examinations performed was 1.75, with no significant differences across provider types. Although almost all providers (92.4%) reported a diagnosis on the vignette interview, unqualified providers and providers without MBBS degrees were more likely to provide an incorrect diagnosis. Among the providers who gave a diagnosis, 94.1% of MBBS providers gave a correct diagnosis, compared with 77.3% of BA/H/UMS providers and 59.6% of unqualified providers. Two-thirds of all providers (66.6%) prescribed treatment on the vignette interviews, but only 14.4% of them prescribed correct treatment. Among the ones who prescribed any treatment, 55.2% of MBBS providers gave the correct TB treatment. In comparison, only 24.3% of BA/H/UMS providers and 16.2% of the unqualified providers prescribed the correct treatment. When we restricted the sample to providers who claimed to treat TB in their practice, the share of providers who gave the correct treatment was 19.5%, compared with 14.4% among all providers. Almost all prescriptions recommended also included additional medicines such as multivitamin syrups, cough medicines and antipyretics. Further, a substantial proportion of providers recommended referring patients to larger facilities for treatment (48.9%), with MBBS providers being statistically more likely to refer relative to other types of providers.

d also included additional medicines such as multivitamin syrups, cough medicines and antipyretics. Further, a substantial proportion of providers recommended referring patients to larger facilities for treatment (48.9%), with MBBS providers being statistically more likely to refer relative to other types of providers. The majority of providers (58.6%) prescribed other drugs that did not include any medicines that are part of the WHO multidrug treatment regimen for new TB patients (see online supplementary appendix table A-1). Further, only 31.9% of providers prescribed any combination of isoniazid (H) and rifampicin (R) and only 21.7% prescribed it for 6 months (180 days) or more (per the WHO guidelines).

ude any medicines that are part of the WHO multidrug treatment regimen for new TB patients (see online supplementary appendix table A-1). Further, only 31.9% of providers prescribed any combination of isoniazid (H) and rifampicin (R) and only 21.7% prescribed it for 6 months (180 days) or more (per the WHO guidelines). Knowledge scores and provider characteristics Summarising the knowledge reflected in diagnostic workup, diagnosis and treatment using IRT scores, MBBS doctors showed significantly higher levels of TB diagnostic and treatment knowledge than other provider types (table 2 and figure 2) based on the Kolmogorov-Smirnov (KS) tests of equality of distributions (p values of 0.07 and 0.02 compared with BA/H/UMS and unqualified providers, respectively). However, there is no statistically significant difference in the distribution of knowledge scores when restricted to the sample of providers who claim to treat TB (figure 3). The KS test of equality of distributions is not significant when comparing providers who claim to treat TB (p value of 0.34 and 0.36, respectively). The KS tests do not, however, account for clustering, which would typically result in higher p values by making estimates less precise. Figure 2 Knowledge distribution by qualification. Figure 3 Knowledge distribution by qualification, among providers who report treating tuberculosis.

Knowledge scores and provider characteristics Summarising the knowledge reflected in diagnostic workup, diagnosis and treatment using IRT scores, MBBS doctors showed significantly higher levels of TB diagnostic and treatment knowledge than other provider types (table 2 and figure 2) based on the Kolmogorov-Smirnov (KS) tests of equality of distributions (p values of 0.07 and 0.02 compared with BA/H/UMS and unqualified providers, respectively). However, there is no statistically significant difference in the distribution of knowledge scores when restricted to the sample of providers who claim to treat TB (figure 3). The KS test of equality of distributions is not significant when comparing providers who claim to treat TB (p value of 0.34 and 0.36, respectively). The KS tests do not, however, account for clustering, which would typically result in higher p values by making estimates less precise. Figure 2 Knowledge distribution by qualification. Figure 3 Knowledge distribution by qualification, among providers who report treating tuberculosis. We are interested in whether patients might be able to choose providers with higher levels of TB diagnostic and treatment knowledge by observing certain provider characteristics. Both in the parsimonious model in column 1 of table 3 as well as in column 2, which controls for additional observable characteristics, age and experience are not associated with knowledge scores. Overall, the observable characteristics only explain 15.9% of the variation in knowledge measured on the vignettes, suggesting that patients would have a very difficult time assessing the knowledge of providers by using observable characteristics. Relative to providers with an MBBS degree, those with BA/H/UMS qualifications as well those with other (including non-medically trained) qualifications have knowledge scores that are half an SD lower (−0.55 for BA/H/UMS and −0.46 for other). Hence, patients could in principle select to go to an MBBS provider should one be available in their area if they desired higher knowledge levels. However, as seen in column 2, controlling for type of medical qualifications, practicing Homeopathic/Ayurvedic type of medicine is associated with over a third of an SD (0.356) higher knowledge score. In column 3, after controlling for additional characteristics, including whether the provider reported treating TB, medical qualification is no longer associated with knowledge score. Controlling for medical qualification and self-reported TB treatment and all other characteristics, practicing Ayurveda/Homoeopathy/Unani continues to be significantly associated with a 0.29 SD higher knowledge score and an increase of about 10 working hours/week is associated with a 0.07 SD increase in the knowledge score. Column 3 also indicates that the knowledge score among providers who claim to treat TB is about 0.46 SDs higher compared with those who did not report this.

antly associated with a 0.29 SD higher knowledge score and an increase of about 10 working hours/week is associated with a 0.07 SD increase in the knowledge score. Column 3 also indicates that the knowledge score among providers who claim to treat TB is about 0.46 SDs higher compared with those who did not report this. Table 3 Knowledge score and providers characteristics

antly associated with a 0.29 SD higher knowledge score and an increase of about 10 working hours/week is associated with a 0.07 SD increase in the knowledge score. Column 3 also indicates that the knowledge score among providers who claim to treat TB is about 0.46 SDs higher compared with those who did not report this. Table 3 Knowledge score and providers characteristics Variables Estimated effect (95% CI) [1] [2] [3] Age 0.03 (−0.04 to 0.09) 0.02 (−0.04 to 0.09) 0.02 (−0.05 to 0.08) Age2 −0.00 (−0.00 to 0.00) −0.00 (−0.00 to 0.00) −0.00 (−0.00 to 0.00) Experience (years) 0.00 (−0.02 to 0.02) −0.01 (−0.03 to 0.02) −0.01 (−0.03 to 0.01) Medical qualification: BA/H/UMS −0.55 (−1.00 to −0.10) −0.59 (−1.08 to −0.10) −0.20 (−0.66 to 0.25) Medical qualification: other −0.46 (−0.75 to −0.17) −0.39 (−0.77 to −0.00) 0.27 (−0.09 to 0.62) Practice allopathy −0.01 (−0.35 to 0.33) −0.09 (−0.42 to 0.24) Practice Ayur/Homoeo/Unani 0.36 (0.13 to 0.58) 0.29 (0.07 to 0.51) Working hours (/week) 0.01 (−0.00 to 0.01) 0.01 (0.00 to 0.01) Average patient caseload (day) 0.00 (−0.01 to 0.01) −0.00 (−0.01 to 0.01) Run camps 0.14 (−0.24 to 0.52) −0.03 (−0.41 to 0.35) Public health facility 0.35 (−0.13 to 0.83) 0.28 (−0.36 to 0.92) Treat TB 0.46 (0.28 to 0.63) Sell drug −0.20 (−0.40 to 0.01) Infrastructure index 0.08 (0.01 to 0.15) Consultation fee (Rs) 0.00 (−0.00 to 0.01) Observations 395 395 395 R2 0.035 0.072 0.159 Columns 1–3 show OLS regression estimates of association between knowledge score and providers characteristics. The reference group for Medical Qualification is MBBS. The first column includes variables related mainly with sociodemographic characteristics of the providers. Column 2 includes variables related to effort/case load and type of facility. The last column includes variables related with the provider's capability (self-reported ability to treat TB), whether they sell drugs, reported fees and infrastructure index. The infrastructure index was computed according to the following variables: electricity, power backup, number of consulting rooms, number of bed for day observation, provision of tests, provision of X-rays and computer system. SEs in parentheses, clustered at the level of study cluster. Sources: Vignette survey and Provider Questionnaire.

ndex was computed according to the following variables: electricity, power backup, number of consulting rooms, number of bed for day observation, provision of tests, provision of X-rays and computer system. SEs in parentheses, clustered at the level of study cluster. Sources: Vignette survey and Provider Questionnaire. The regression results in online supplementary appendix table A-2 show the analogous analyses from table 3 restricted to the sample of providers who claimed to treat TB. The results indicate that none of the observable characteristics (such as medical qualification, type of medicine practiced, average patient caseload and infrastructure index) are significantly associated with the knowledge score in this sample. These results indicate that patients are likely to find it especially difficult to discern the knowledge of providers among those who treat TB, and the dominant marker of provider knowledge is whether they claim to treat TB or not. Unfortunately, this marker might not be readily observable to patients, especially ones who might not know what their illness is.

likely to find it especially difficult to discern the knowledge of providers among those who treat TB, and the dominant marker of provider knowledge is whether they claim to treat TB or not. Unfortunately, this marker might not be readily observable to patients, especially ones who might not know what their illness is. The results reported in table 4, where we conduct OLS regressions of consultation fees on the set of observable characteristics, further suggest that patients might not be able to assess providers' knowledge based on observed prices in the market. In the parsimonious model in column 1 of table 4, the coefficient on medical qualification indicates that qualified providers charge on average INR 24.1 more than BA/H/UMS providers and INR 46.9 more than unqualified providers. Also a 1 SD increase in the knowledge score, controlling for qualifications, age and experience, is associated with a higher consultation fee of INR 3.4. Even after controlling for additional observable characteristics in columns 2 and 3, the coefficients on medical qualification and the knowledge score remain significant and within a similar range. Table 4 Consultation fee and providers characteristics

The results reported in table 4, where we conduct OLS regressions of consultation fees on the set of observable characteristics, further suggest that patients might not be able to assess providers' knowledge based on observed prices in the market. In the parsimonious model in column 1 of table 4, the coefficient on medical qualification indicates that qualified providers charge on average INR 24.1 more than BA/H/UMS providers and INR 46.9 more than unqualified providers. Also a 1 SD increase in the knowledge score, controlling for qualifications, age and experience, is associated with a higher consultation fee of INR 3.4. Even after controlling for additional observable characteristics in columns 2 and 3, the coefficients on medical qualification and the knowledge score remain significant and within a similar range. Table 4 Consultation fee and providers characteristics Variables Estimated effect (95% CI) [1] [2] [3] Age −0.73 (−2.93 to 1.47) −0.79 (−2.94 to 1.36) −0.83 (−2.91 to 1.26) Age2 0.01 (−0.01 to 0.03) 0.01 (−0.01 to 0.03) 0.01 (−0.01 to 0.03) Experience (years) 0.21 (−0.30 to 0.72) 0.14 (−0.28 to 0.56) 0.11 (−0.31 to 0.53) Medical qualification: BA/H/U/MS −24.07 (−45.10 to −3.04) −34.00 (−53.00 to −15.00) −28.12 (−48.40 to −7.84) Medical qualification: other −46.95 (−68.49 to −25.42) −58.38 (−78.42 to −38.33) −47.97 (−69.85 to −26.08) Knowledge score 3.40 (1.42 to 5.39) 4.06 (1.97 to 6.15) 2.28 (−0.09 to 4.65) Practice allopathy 14.71 (5.54 to 23.87) 12.97 (3.24 to 22.69) Practice Ayur/Homoeo/Unani 0.33 (−6.33 to 6.99) −0.15 (−6.36 to 6.06) Working hours (/week) −0.09 (−0.26 to 0.08) −0.07 (−0.24 to 0.10) Average patient caseload (day) 0.35 (−0.08 to 0.77) 0.25 (−0.18 to 0.68) Run camps 2.69 (−14.50 to 19.88) −0.97 (−18.65 to 16.72) Public health facility −77.43 (−105.07 to −49.79) −87.04 (−113.93 to −60.15) Treat TB 7.57 (0.98 to 14.16) Sell drug −7.18 (−12.85 to −1.51) Infrastructure index 2.42 (−0.68 to 5.52) Observations 395 395 395 R2 0.245 0.363 0.396 Columns 1–3 show OLS regression estimates of association between consultation fee and providers characteristics. The reference group for Medical Qualification is MBBS. The first column includes variables related mainly with sociodemographic characteristics of the providers. Column 2 includes variables related with effort/case load and type of facility. The last column includes variables related with the provider's capability (self-reported ability to treat TB), whether they sell drugs, reported fees and infrastructure index. The infrastructure index was computed according to the following variables: electricity, power backup, number of consulting rooms, number of bed for day observation, provision of tests, provision of X-rays and computer system. SEs in parentheses, clustered at the level of study cluster. Sources: Vignette survey and Provider Questionnaire.

ndex was computed according to the following variables: electricity, power backup, number of consulting rooms, number of bed for day observation, provision of tests, provision of X-rays and computer system. SEs in parentheses, clustered at the level of study cluster. Sources: Vignette survey and Provider Questionnaire. After controlling for the full set of observable characteristics, including treating TB and provider knowledge score, MBBS providers charge an average of INR 28.1 more than BA/H/UMS providers and INR 47.9 more than unqualified providers. This suggests a large premium charged by MBBS providers given that, as we saw in table 2, there is no statistically significant difference in the knowledge score of qualified versus non-qualified providers once we control for treating TB. The regression-adjusted difference in the knowledge score between MBBS and BA/H/UMS providers is 0.202 (from table 3), which would only predict an increase of INR 0.46 (0.202×2.282=0.76) instead of INR 28.1. If we do not control for TB treatment, the predicted fee would be only INR 2.40 higher (0.591×4.061). In a framework where providers with higher knowledge could charge higher fees, one possible explanation for our finding that knowledge is not adequately priced could be due to substantial asymmetric information: patients cannot directly assess providers' knowledge and cannot infer it from outcomes (either theirs or of other patients) due to variations in case-mix and infrequent experience. However, another explanation might be that knowledge might be only very loosely related to performance,13 which is what ultimately matters for the patient.

directly assess providers' knowledge and cannot infer it from outcomes (either theirs or of other patients) due to variations in case-mix and infrequent experience. However, another explanation might be that knowledge might be only very loosely related to performance,13 which is what ultimately matters for the patient. Additionally, the regressions in column 3 of table 4 indicate that the practice of allopathic medicine is associated with INR 12.97 higher consultation fee even after controlling for qualifications. This probably explains why so many BA/H/UMS providers end up adopting allopathic therapeutic methods. Public health facilities though, charge about INR 87.04 lower. Providers who also sell medicines as part of their practice charge INR 7.18 lower than those who do not sell drugs on their premises, presumably because profits from drug sales could offset the fees. In online supplementary appendix table A-3, which shows the results from the same regressions as in table 4 but run using a sample restricted to providers who treat TB, the MBBS qualification is still significantly associated with higher consultation fee. MBBS providers are likely to charge INR 41.33 more than providers who are not qualified, and INR 14.65 more than BA/H/UMS providers (although the latter is not statistically significant), suggesting a substantial premium for the MBBS qualification.

ication is still significantly associated with higher consultation fee. MBBS providers are likely to charge INR 41.33 more than providers who are not qualified, and INR 14.65 more than BA/H/UMS providers (although the latter is not statistically significant), suggesting a substantial premium for the MBBS qualification. Correct diagnosis and treatment Next, we turn to the relationship between provider characteristics and making a correct diagnosis. As columns 1 and 2 of table 5 show, unqualified providers are significantly less likely to make a correct diagnosis relative to MBBS providers (∼40%), and BA/H/UMS providers do not show statistically significant differences relative to MBBS providers. The knowledge score is significantly and positively associated with making the correct diagnosis. A 1 SD increase in the knowledge score is associated with a 13 percentage point increase (columns 1 and 2) in the probability of making a correct diagnosis. The significant effects of medical qualification and knowledge scores persist even after controlling for other observable characteristics as seen in column 3 of table 5. Table 5 Correct diagnostic (if any)—marginal effects

Correct diagnosis and treatment Next, we turn to the relationship between provider characteristics and making a correct diagnosis. As columns 1 and 2 of table 5 show, unqualified providers are significantly less likely to make a correct diagnosis relative to MBBS providers (∼40%), and BA/H/UMS providers do not show statistically significant differences relative to MBBS providers. The knowledge score is significantly and positively associated with making the correct diagnosis. A 1 SD increase in the knowledge score is associated with a 13 percentage point increase (columns 1 and 2) in the probability of making a correct diagnosis. The significant effects of medical qualification and knowledge scores persist even after controlling for other observable characteristics as seen in column 3 of table 5. Table 5 Correct diagnostic (if any)—marginal effects Variables Estimated effect (95% CI) [1] [2] [3] Age 0.01 (−0.02 to 0.04) 0.01 (−0.01 to 0.04) 0.01 (−0.01 to 0.04) Age2 −0.00 (−0.00 to 0.00) −0.00 (−0.00 to 0.00) −0.00 (−0.00 to 0.00) Experience (years) 0.00 (−0.01 to 0.01) 0.00 (−0.01 to 0.01) 0.00 (−0.01 to 0.01) Medical qualification: BA/H/UMS −0.19 (−0.46 to 0.07) −0.22 (−0.50 to 0.07) −0.12 (−0.39 to 0.15) Medical qualification: other −0.40 (−0.63 to −0.17) −0.43 (−0.68 to −0.17) −0.27 (−0.53 to −0.00) Knowledge score 0.13 (0.09 to 0.17) 0.13 (0.09 to 0.18) 0.11 (0.07 to 0.16) Practice allopathy 0.08 (−0.09 to 0.25) 0.05 (−0.12 to 0.22) Practice Ayurv/Homoeo/Unani −0.01 (−0.12 to 0.11) −0.02 (−0.13 to 0.10) Working hours (/week) −0.00 (−0.00 to 0.00) −0.00 (−0.00 to 0.00) Average patient caseload (day) −0.00 (−0.01 to 0.00) −0.00 (−0.01 to 0.00) Run camps −0.07 (−0.24 to 0.11) −0.11 (−0.28 to 0.07) Public health facility 0.10 (−0.30 to 0.50) 0.07 (−0.40 to 0.55) Treat TB 0.08 (−0.02 to 0.18) Sell drug −0.07 (−0.16 to 0.02) Infrastructure index 0.02 (−0.02 to 0.07) Consultation fee (Rs) 0.00 (−0.00 to 0.00) Observations 365 365 365 Columns 1–3 show marginal effects of probit regression estimates of association between correct diagnosis variable for tuberculosis case and providers characteristics. The reference group for medical qualification is MBBS. The first column includes variables related mainly with sociodemographic characteristics of the providers. Column 2 includes variables related with effort/case load and type of facility. The last column includes variables related with the provider's capability (self-reported ability to treat TB), whether they sell drugs, reported fees and infrastructure index. The infrastructure index was computed according to the following variables: electricity, power backup, number of consulting rooms, number of bed for day observation, provision of tests, provision of X-rays and computer system. SEs in parentheses, clustered at the level of study cluster. Sources: Vignette Survey and Provider Questionnaire.

ndex was computed according to the following variables: electricity, power backup, number of consulting rooms, number of bed for day observation, provision of tests, provision of X-rays and computer system. SEs in parentheses, clustered at the level of study cluster. Sources: Vignette Survey and Provider Questionnaire. We see similar results in online supplementary appendix table A-4 when the regression is run on the restricted sample of providers who stated that they treat TB. Again, MBBS qualifications (relative to unqualified providers) and knowledge scores are significantly positively associated with the probability of making a correct diagnosis. In this restricted sample, we also see that the probability of making a correct diagnosis is lower by 16.1 percentage points if a provider also sells drugs. The parsimonious models in columns 1 and 2 in table 6 show that providers with BA/H/UMS or no qualifications are 20–30 percentage points less likely to prescribe correct treatment relative to MBBS providers. Table 6 Correct treatment (if any)—marginal effects

We see similar results in online supplementary appendix table A-4 when the regression is run on the restricted sample of providers who stated that they treat TB. Again, MBBS qualifications (relative to unqualified providers) and knowledge scores are significantly positively associated with the probability of making a correct diagnosis. In this restricted sample, we also see that the probability of making a correct diagnosis is lower by 16.1 percentage points if a provider also sells drugs. The parsimonious models in columns 1 and 2 in table 6 show that providers with BA/H/UMS or no qualifications are 20–30 percentage points less likely to prescribe correct treatment relative to MBBS providers. Table 6 Correct treatment (if any)—marginal effects Estimated effect (95% CI) Variables [1] [2] [3] Age −0.00 (−0.04 to 0.04) −0.00 (−0.04 to 0.03) −0.00 (−0.04 to 0.03) Age2 −0.00 (−0.00 to 0.00) 0.00 (−0.00 to 0.00) 0.00 (−0.00 to 0.00) Experience (years) 0.00 (−0.01 to 0.01) 0.00 (−0.01 to 0.01) −0.00 (−0.01 to 0.01) Medical qualification: BA/H/UMS −0.20 (−0.39 to −0.02) −0.20 (−0.41 to 0.00) −0.14 (−0.34 to 0.07) Medical qualification: other −0.30 (−0.44 to −0.16) −0.29 (−0.46 to −0.13) −0.11 (−0.31 to 0.10) Knowledge score 0.04 (−0.02 to 0.10) 0.04 (−0.02 to 0.11) 0.02 (−0.04 to 0.08) Practice allopathy 0.23 (0.04 to 0.42) 0.20 (0.02 to 0.38) Practice Ayurv/Homoeo/Unani 0.01 (−0.11 to 0.12) 0.00 (−0.11 to 0.11) Working hours (/week) −0.00 (−0.00 to 0.00) −0.00 (−0.00 to 0.00) Average patient caseload (day) −0.00 (−0.01 to 0.00) −0.00 (−0.01 to 0.00) Run camps 0.08 (−0.08 to 0.23) 0.04 (−0.11 to 0.19) Public health facility −0.10 (−0.38 to 0.18) 0.01 (−0.38 to 0.40) Treat TB 0.14 (0.03 to 0.24) Sell drug −0.03 (−0.12 to 0.07) Infrastructure index 0.01 (−0.03 to 0.05) Consultation fee (Rs) 0.00 (0.00 to 0.00) Observations 263 263 263 Columns 1–3 show marginal effect of probit regression estimates of association between correct diagnosis variable for tuberculosis case and providers characteristics. The reference group for medical qualification is MBBS. The first column includes variables related mainly with sociodemographic characteristics of the providers. Column 2 includes variables related with effort/case load and type of facility. The last column includes variables related with the provider's capability (self-reported ability to treat TB), whether they sell drugs, reported fees and infrastructure index. The infrastructure index was computed according to the following variables: electricity, power backup, number of consulting rooms, number of bed for day observation, provision of tests, provision of X-rays and computer system. SEs in parentheses, clustered at the level of study cluster. Sources: Vignette Survey and Provider Questionnaire.

ndex was computed according to the following variables: electricity, power backup, number of consulting rooms, number of bed for day observation, provision of tests, provision of X-rays and computer system. SEs in parentheses, clustered at the level of study cluster. Sources: Vignette Survey and Provider Questionnaire. However, neither medical qualifications nor knowledge scores are associated with the prescription of correct treatment after controlling for the full set of characteristics in column 3. Being a provider who treats TB though is associated with a 13.6 percentage point higher likelihood of prescribing the correct treatment. We see similar results in online supplementary appendix table A-5 when the regression is run on the sample restricted to providers who state that they treat TB. Referral The likelihood of making a referral is significantly associated with medical qualifications. Compared with those with MBBS degrees, unqualified and BA/H/UMS providers are less likely to refer the TB case even after controlling for all observable characteristics (column 3 of table 7). BA/H/UMS providers were 23.2 percentage points less likely, and unqualified were 37.7 percentage points less likely. While the knowledge score is significantly and positively associated with referrals in parsimonious models, it does not have a significant association after controlling for the full set of observable characteristics. Table 7 Recommend referral—marginal effects

Referral The likelihood of making a referral is significantly associated with medical qualifications. Compared with those with MBBS degrees, unqualified and BA/H/UMS providers are less likely to refer the TB case even after controlling for all observable characteristics (column 3 of table 7). BA/H/UMS providers were 23.2 percentage points less likely, and unqualified were 37.7 percentage points less likely. While the knowledge score is significantly and positively associated with referrals in parsimonious models, it does not have a significant association after controlling for the full set of observable characteristics. Table 7 Recommend referral—marginal effects Variables Estimated effect (95% CI) [1] [2] [3] Age 0.01 (−0.02 to 0.04) 0.02 (−0.02 to 0.05) 0.01 (−0.02 to 0.05) Age2 −0.00 (−0.00 to 0.00) −0.00 (−0.00 to 0.00) −0.00 (−0.00 to 0.00) Experience (years) 0.00 (−0.00 to 0.01) 0.00 (−0.00 to 0.01) 0.00 (−0.00 to 0.01) Medical qualification: BA/H/UMS −0.13 (−0.34 to 0.08) −0.30 (−0.56 to −0.04) −0.23 (−0.50 to 0.03) Medical qualification: other −0.32 (−0.48 to −0.16) −0.50 (−0.71 to −0.28) −0.38 (−0.63 to −0.12) Knowledge score 0.04 (−0.00 to 0.09) 0.05 (−0.00 to 0.09) 0.03 (−0.02 to 0.08) Practice allopathy −0.06 (−0.25 to 0.14) −0.07 (−0.27 to 0.13) Practice Ayurv/Homoeo/Unani −0.00 (−0.12 to 0.11) −0.02 (−0.13 to 0.10) Working hours (/week) −0.00 (−0.00 to 0.00) −0.00 (−0.00 to 0.00) Average patient caseload (day) −0.01 (−0.01 to −0.00) −0.01 (−0.01 to −0.00) Run camps −0.11 (−0.31 to 0.08) −0.15 (−0.34 to 0.03) Public health facility −0.15 (−0.57 to 0.26) −0.27 (−0.74 to 0.20) Treat TB 0.10 (−0.03 to 0.22) Sell drug −0.01 (−0.11 to 0.09) Infrastructure index 0.03 (−0.01 to 0.07) Consultation fee (Rs) 0.00 (−0.00 to 0.00) Observations 395 395 395 Columns 1–3 show marginal effects of probit regression estimates of association between recommend referral variable for tuberculosis case and providers characteristics. The reference group for medical qualification is MBBS. The first column includes variables related mainly with sociodemographic characteristics of the providers. Column 2 includes variables related with effort/case load and type of facility. The last column includes variables related with the provider's capability (self-reported ability to treat TB), whether they sell drugs, reported fees and infrastructure index. The infrastructure index was computed according to the following variables: electricity, power backup, number of consulting rooms, number of bed for day observation, provision of tests, provision of X-rays and computer system. SEs in parentheses, clustered at the level of study cluster. Sources: Vignette Survey and Provider Questionnaire.

ndex was computed according to the following variables: electricity, power backup, number of consulting rooms, number of bed for day observation, provision of tests, provision of X-rays and computer system. SEs in parentheses, clustered at the level of study cluster. Sources: Vignette Survey and Provider Questionnaire. In online supplementary appendix table A-6, among providers who claim to treat TB, unqualified providers are significantly less (39.4 percentage points) likely to make a referral. Providers who sell drugs also have 16.6 percentage points lower probability of making a referral.

ndex was computed according to the following variables: electricity, power backup, number of consulting rooms, number of bed for day observation, provision of tests, provision of X-rays and computer system. SEs in parentheses, clustered at the level of study cluster. Sources: Vignette Survey and Provider Questionnaire. In online supplementary appendix table A-6, among providers who claim to treat TB, unqualified providers are significantly less (39.4 percentage points) likely to make a referral. Providers who sell drugs also have 16.6 percentage points lower probability of making a referral. Discussion In rural Bihar, as in much of other parts of India, TB continues to be a major public health challenge. This study provides further evidence that provider knowledge of how to diagnose and treat a case of TB, as measured by clinical vignettes, is low. This is the case for providers who lack formal medical qualifications (who provide most of the care in rural areas9) as well as those providers who have formal medical qualifications. Providers ask few diagnostic questions (only 32% asked about duration of cough for TB, for example) and seldom seek diagnostic test information (only 20% asked for sputum test for TB). Provider performance on making a correct diagnosis is inadequate but not as low as provider performance on prescribing appropriate treatment. Over 60% of providers arrived at the correct diagnosis; but while more than 66% of providers prescribed a treatment, only 21.7% of those were correct according to TB treatment guidelines. Low rates of correct treatment have significant implications for problems of TB drug resistance.

prescribing appropriate treatment. Over 60% of providers arrived at the correct diagnosis; but while more than 66% of providers prescribed a treatment, only 21.7% of those were correct according to TB treatment guidelines. Low rates of correct treatment have significant implications for problems of TB drug resistance. The problem of poor TB diagnostic and treatment accuracy is unlikely to be solved by shifting patients towards MBBS providers in the area. While MBBS providers had higher average scores on diagnostic workup, diagnostic accuracy and prescribing correct treatment than other provider types, only 45.7% MBBS doctors still only prescribed correct treatment. This pattern of MBBS providers performing marginally better, but still at a level that is unacceptably low is consistent with evidence from recent studies in urban India using standardised patients (SPs), which provide information on actual practice.27 Second, after controlling for whether the provider offers treatment for TB, there are no statistically significant differences in terms of provider knowledge between MBBS and other types of difference. It is important to bear in mind that 82.9% of MBBS providers offer treatment for TB, relative to 80.4% of BA/H/UMS providers and 62.74% of providers with no formal training. However, since the informal sector providers are ubiquitous in the rural health landscape (almost 10 times as many as MBBS providers in our study sample), it is salient that after controlling for offering TB treatment, these providers are comparable in knowledge and correct treatment. A more disheartening view of this finding is that providers with formal medical degrees do not perform much better than those without formal training on vignette-based assessments of knowledge of diagnosing and treating TB. However, MBBS providers are able to command a premium in terms of consulting fees that is far higher than what a higher knowledge score might indicate.

ders with formal medical degrees do not perform much better than those without formal training on vignette-based assessments of knowledge of diagnosing and treating TB. However, MBBS providers are able to command a premium in terms of consulting fees that is far higher than what a higher knowledge score might indicate. We also find that prescription of incorrect treatments is related to the practice of selling drugs as part of their medical practice. Lower rates of selling drugs among those with formal qualifications reflect the fact that in the informal health sector, there is a missing market for consultations. Informal providers typically tend to charge a fee for ‘treatment’ that includes the drugs provided as opposed to a consultant service, which ends with diagnosis and prescription.

rugs among those with formal qualifications reflect the fact that in the informal health sector, there is a missing market for consultations. Informal providers typically tend to charge a fee for ‘treatment’ that includes the drugs provided as opposed to a consultant service, which ends with diagnosis and prescription. Our study faces limitations related to using data from vignette-based interviews. The main limitation of the vignette method is that it does not capture the actual level of quality of care provided. As documented in recent research on ‘know-do gaps’,7 28 the quality of care provided to patients might in fact be considerably lower than what is reported and measured on vignettes. In fact, the know-do gap for TB care has been reported in India, using data from SPs.28 While some of the measures we report (such as rates of sputum test recommendations) are comparable with those reported from previous research with SPs,28 we note that measures such as correct diagnosis rates and appropriate referral are higher with data from vignettes. As a result, the (low) level of knowledge that we report represents the upper bound of what providers might actually provide. Further, our estimates of provider knowledge in rural Bihar are representative of similar socioeconomic and geographic areas in developing countries like India, but might not be widely generalisable.

result, the (low) level of knowledge that we report represents the upper bound of what providers might actually provide. Further, our estimates of provider knowledge in rural Bihar are representative of similar socioeconomic and geographic areas in developing countries like India, but might not be widely generalisable. Low levels of provider knowledge of TB diagnosis and treatment could not only hamstring ongoing efforts to control TB, but also make it worse by contributing to multidrug resistance.29 Recent experimental evidence on improvement on provider knowledge and adherence to protocol from intensive training programmes offered to informal sector providers in West Bengal provides a potential solution to addressing this challenge.30 Medical organisations in India have typically opposed proposals to offer training or improving capacity of informal sector providers, advocating instead for policies to increase MBBS trained providers. The challenge, however, is that trained MBBS/MD graduates are unlikely to choose to practice and live in rural areas. While the number of medical training institutions in India have increased dramatically in the past few decades,23 India's rural population continues to receive healthcare primarily from informal sector providers (Das et al. 2015. Forthcoming). Policymakers in India, and elsewhere, might want to prioritise strategies such as training, incentives, task shifting and regulation to improve knowledge and performance of existing providers in the healthcare system.

ntinues to receive healthcare primarily from informal sector providers (Das et al. 2015. Forthcoming). Policymakers in India, and elsewhere, might want to prioritise strategies such as training, incentives, task shifting and regulation to improve knowledge and performance of existing providers in the healthcare system. This research was made possible by funding from the Bill and Melinda Gates Foundation (grant OPP1025880). The authors are grateful to Bhartendu Trivedi and Margaret Pendzich for project management, and to Sambodhi Research and Communications Pvt and to Institute of Socio-Economic Research on Development and Democracy (ISERDD) for fieldwork and data collection. Divya Guru Rajan provided excellent research assistance. The authors are grateful to Madhukar Pai for helpful comments. Handling editor: Seye Abimbola. Contributors: MM, JDG-F and MV-H contributed to study design. MM, JDG-F and MV-H contributed to instrument development. MM, SG, JDG-F and MV-H contributed to data collection, cleaning and analysis. All coauthors contributed to data interpretation and manuscript writing and revision. Funding: This study was funded by the Bill and Melinda Gates Foundation (Grant number OPP1025880). Disclaimer: The funders had no role in study design, data collection, analysis, interpretation, writing of the manuscript or decision to submit the paper for publication. Competing interests: None declared. Ethics approval: This study, as part of the BEST study protocol, was approved by Duke University (29755) and India's Health Ministry Steering Committee (number 12/2008/30-HMSC/4).

Disclaimer: The funders had no role in study design, data collection, analysis, interpretation, writing of the manuscript or decision to submit the paper for publication. Competing interests: None declared. Ethics approval: This study, as part of the BEST study protocol, was approved by Duke University (29755) and India's Health Ministry Steering Committee (number 12/2008/30-HMSC/4). Provenance and peer review: Not commissioned; externally peer reviewed. Data sharing statement: All our protocols and instruments are made available publicly at http://cohesiveindia.org/publications-downloads.html. We also plan to submit our complete data set with documentation to the Harvard Dataverse data depository for public access. i Using the average 2011 exchange rate of INR 49.12 to US$1. ii Based on evidence that informal and formally qualified doctors provide low quality of care, we examine provider characteristics to understand if other factors such as experience, qualifications or volume are correlated with provider quality. iii This recommended treatment for new TB cases (fourth edition published in 2010) has been in place at least since the publication of the third edition of the WHO tuberculosis treatment guidelines in 2003.

ii Based on evidence that informal and formally qualified doctors provide low quality of care, we examine provider characteristics to understand if other factors such as experience, qualifications or volume are correlated with provider quality. iii This recommended treatment for new TB cases (fourth edition published in 2010) has been in place at least since the publication of the third edition of the WHO tuberculosis treatment guidelines in 2003. iv Evidence from recent studies from India and other developing countries shows that providers with qualifications and those without qualifications provide care that is of very low quality.17 Hence, our analysis of quality and provider characteristics helps understand if there are other observable provider characteristics such as age, experience and patient volume that might serve as useful signals of quality to patients. v All analyses were conducted in STATA V.14.0. vi We do not have information on whether informal providers have had any TB training, but since the TB programme in India is run primarily through the formal public sector, we do not expect that these providers would have received training.

Key questions What is already known about this topic? Out-of-pocket (OOP) medical payments can lead to catastrophic health expenditure and impoverishment. Studies on household healthcare cost of pneumonia and diarrhoeal disease among children under five are scarce in sub-Saharan African countries and are non-existent in Ethiopia. What are the new findings? Better estimates of the current household OOP medical payments for pneumonia and diarrhoea treatment in developing countries allow for more precision in estimating the expected poverty impact of health interventions such as vaccines, independent of the interventions' health impact. Recommendations for policy The study is on OOP medical expenses that may not have a direct impact on clinical practice. However, given the fact that OOP payments for the treatment of pneumonia or diarrhoea are high and the major cost driver being medication might influence the choice of generic drugs over brands. Introduction In low-income and middle-income countries, diarrhoea and respiratory infections are the most common causes of childhood illnesses and healthcare visits. Similarly, severe cases of diarrhoea and pneumonia are among the most common reasons for hospital admission of children. Childhood pneumonia and diarrhoea are the leading causes of death globally and in Ethiopia.1 2

hoea and respiratory infections are the most common causes of childhood illnesses and healthcare visits. Similarly, severe cases of diarrhoea and pneumonia are among the most common reasons for hospital admission of children. Childhood pneumonia and diarrhoea are the leading causes of death globally and in Ethiopia.1 2 Illnesses impose a huge economic burden on individuals and families. Direct payments for healthcare can have negative consequences for families, including pushing families into poverty or further into deeper poverty. User fees exacerbate inequity, as poor people are more likely to reduce service usage and become impoverished from the effects of catastrophic health expenditures (CHE)—defined as household's financial contributions to the health system exceeding 40% of income remaining after subsistence needs have been met.3 The 2005 healthcare financing reform in Ethiopia allowed public health facilities to collect, retain and use the revenues and user fees that they generate from different sources, as an addition to the government budget, for improving the quality of health services.4 The retained revenues generated from user fees covered 56% of the total health budget for health centres in the year 2011/2012.5 A system of fee waivers and exemptions was part of the reform. Though preventive services (eg, immunisation, prenatal care, etc) are delivered freely at public health facilities, curative child health services are not provided free of charge in public health centres and public hospitals.

centres in the year 2011/2012.5 A system of fee waivers and exemptions was part of the reform. Though preventive services (eg, immunisation, prenatal care, etc) are delivered freely at public health facilities, curative child health services are not provided free of charge in public health centres and public hospitals. Attainment of universal health coverage (UHC) is a central theme of the Sustainable Development Goals.6 To achieve UHC, countries should address all three dimensions of the cube: (1) Whom to include first? (2) Which services to cover? (3) Proportion of the costs covered. The World Health Report identifies continued reliance on direct payments, including user fees, as by far the greatest obstacle to the attainment of UHC.7 Despite fee waivers for preventive health services, the OOP expenditures for curative care for children are a burden in Ethiopia, accounting for close to 50% of total child healthcare expenditures in 2010/2011.8

reliance on direct payments, including user fees, as by far the greatest obstacle to the attainment of UHC.7 Despite fee waivers for preventive health services, the OOP expenditures for curative care for children are a burden in Ethiopia, accounting for close to 50% of total child healthcare expenditures in 2010/2011.8 Studies on household medical expenses of pneumonia and diarrhoea among children under five are scarce in sub-Saharan African countries. Two studies, one in Kenya and another in Zambia, have examined medical costs of pneumonia treatment.9 10 The study in Kenya was hospital based while the study in Zambia involved one health centre, so neither study was representative of nationwide disease costs of inpatient and outpatient pneumonia treatment. Rheingans et al,11 in their study in three African countries, concluded that diarrhoea episodes resulted in substantial economic cost. To date, no studies have measured healthcare costs of pneumonia or diarrhoea in Ethiopia. Ethiopia has recently introduced pneumococcal conjugate vaccine and rotavirus vaccine as part of the basic vaccine programme.12 Reduction in new cases of pneumonia and diarrhoea may offer protection against impoverishment and OOP expenditures for such diseases. Better estimates of the current household OOP expenses allow for more precision in estimating the expected poverty impact of these new vaccines, independent of the interventions' health benefits.13 14

ew cases of pneumonia and diarrhoea may offer protection against impoverishment and OOP expenditures for such diseases. Better estimates of the current household OOP expenses allow for more precision in estimating the expected poverty impact of these new vaccines, independent of the interventions' health benefits.13 14 The objectives of this study are to: (1) estimate and characterise household OOP expenses for an episode of childhood diarrhoea and pneumonia by type and level of care; (2) assess the extent to which OOP expenses for diarrhoea and pneumonia contribute to household impoverishment and (3) examine the distribution of household OOP expenses across wealth quintiles and by place of residence. Methods We conducted a descriptive facility-based cost study of diarrhoeal disease and all-cause pneumonia in children under five in Ethiopia from the household (patient) perspective. OOP expenses were measured in terms of local currency and converted to US dollars (US$). The average 2013 exchange rate of 18.6 Ethiopian Birr (ETB) to US$1 was used.15

ve facility-based cost study of diarrhoeal disease and all-cause pneumonia in children under five in Ethiopia from the household (patient) perspective. OOP expenses were measured in terms of local currency and converted to US dollars (US$). The average 2013 exchange rate of 18.6 Ethiopian Birr (ETB) to US$1 was used.15 Study area and population Ethiopia is the second most populous country in Africa with an estimated 94 million inhabitants.16 A majority of the Ethiopian population lives in rural areas (84%), and the population pyramid remains quite young: 44% are under 15 years of age.17 At present, Ethiopia is administratively structured into nine national regional states—Oromia, Amhara, Southern Nations Nationalities and People Region (SNNPR), Tigray, Benishangul-Gumuz, Gambella, Afar, Somali and Harari—and two city administrations, that is, Addis Ababa City Administration and Dire Dawa City Council. In spite of rapid economic development in the last decade, at an average annual growth rate of 11% per year, Ethiopia remains one of the poorest countries in Africa with annual per capita earnings of about US$550, which is well below the sub-Saharan African average of US$1640.18 19

istration and Dire Dawa City Council. In spite of rapid economic development in the last decade, at an average annual growth rate of 11% per year, Ethiopia remains one of the poorest countries in Africa with annual per capita earnings of about US$550, which is well below the sub-Saharan African average of US$1640.18 19 The study population was individuals seeking treatment at health facilities in four major regions (Oromia, Amhara, SNNPR and Tigray) and Addis Ababa city administration (the capital city) in Ethiopia (figure 1). These regions were selected because they are home to 90% of the Ethiopian population and are ethnically and culturally diverse. Furthermore, 90% of the health centres and 81% of the public hospitals are located in these regions and the capital city.20 Public healthcare delivery in Ethiopia consists of a three-tier system.21 The primary healthcare (PHC) unit is the first level which is composed of a Health Center, five satellite Health Posts and Primary Hospital serving an average population of 100 000. The secondary care level comprised general hospitals. Each general hospital provides inpatient and ambulatory services to an average of 1 million people. Tertiary care is provided in specialised hospitals each serving an average of 5 million people. Among those who sought care in health facilities for a complaint of cough or diarrhoea, 25% and 23%, respectively, visited the private health sector (Demographic and Health Survey (DHS), 2011).22 Figure 1 Distribution of health facilities included in the study.

The study population was individuals seeking treatment at health facilities in four major regions (Oromia, Amhara, SNNPR and Tigray) and Addis Ababa city administration (the capital city) in Ethiopia (figure 1). These regions were selected because they are home to 90% of the Ethiopian population and are ethnically and culturally diverse. Furthermore, 90% of the health centres and 81% of the public hospitals are located in these regions and the capital city.20 Public healthcare delivery in Ethiopia consists of a three-tier system.21 The primary healthcare (PHC) unit is the first level which is composed of a Health Center, five satellite Health Posts and Primary Hospital serving an average population of 100 000. The secondary care level comprised general hospitals. Each general hospital provides inpatient and ambulatory services to an average of 1 million people. Tertiary care is provided in specialised hospitals each serving an average of 5 million people. Among those who sought care in health facilities for a complaint of cough or diarrhoea, 25% and 23%, respectively, visited the private health sector (Demographic and Health Survey (DHS), 2011).22 Figure 1 Distribution of health facilities included in the study. Study sites and sample selection Data were collected from individuals seeking services from a sample of 6 public hospitals, 15 public health centres, 9 health posts and 5 private health facilities (a total of 35 health facilities).

The study population was individuals seeking treatment at health facilities in four major regions (Oromia, Amhara, SNNPR and Tigray) and Addis Ababa city administration (the capital city) in Ethiopia (figure 1). These regions were selected because they are home to 90% of the Ethiopian population and are ethnically and culturally diverse. Furthermore, 90% of the health centres and 81% of the public hospitals are located in these regions and the capital city.20 Public healthcare delivery in Ethiopia consists of a three-tier system.21 The primary healthcare (PHC) unit is the first level which is composed of a Health Center, five satellite Health Posts and Primary Hospital serving an average population of 100 000. The secondary care level comprised general hospitals. Each general hospital provides inpatient and ambulatory services to an average of 1 million people. Tertiary care is provided in specialised hospitals each serving an average of 5 million people. Among those who sought care in health facilities for a complaint of cough or diarrhoea, 25% and 23%, respectively, visited the private health sector (Demographic and Health Survey (DHS), 2011).22 Figure 1 Distribution of health facilities included in the study. Study sites and sample selection Data were collected from individuals seeking services from a sample of 6 public hospitals, 15 public health centres, 9 health posts and 5 private health facilities (a total of 35 health facilities). We used convenience sampling to select facilities after stratifying them based on level of care (primary to tertiary), urban/rurali location23 and implementation of the integrated management of childhood illnesses (IMCI) strategy. DHS 2011 disaggregates the type of facilities visited for cases of acute respiratory infection or diarrhoea. We used these data as a reference to allocate the number of cases enrolled in the study by type of facility (table 1).

on23 and implementation of the integrated management of childhood illnesses (IMCI) strategy. DHS 2011 disaggregates the type of facilities visited for cases of acute respiratory infection or diarrhoea. We used these data as a reference to allocate the number of cases enrolled in the study by type of facility (table 1). Table 1 Distribution of cases by type of facility visited in the five regions included in the study Regions Oromia Amhara SNNPR Tigray Addis Ababa Type of health facility No. of health facilities No. of cases No. of health facilities No. of cases No. of health facilities No. of cases No. of health facilities No. of cases No. of health facilities No. of cases Public hospital 2 51 1 25 1 30 1 26 1 28 Health centre 4 95 3 72 3 45 3 90 2 66 Health post 3 29 2 20 2 20 2 20 – – Private clinic/hospital 1 21 1 8 1 10 1 10 1 20 Total 10 196 7 125 7 105 7 146 4 114 SNNPR, Southern Nations Nationalities and People Region.

facilities No. of cases No. of health facilities No. of cases Public hospital 2 51 1 25 1 30 1 26 1 28 Health centre 4 95 3 72 3 45 3 90 2 66 Health post 3 29 2 20 2 20 2 20 – – Private clinic/hospital 1 21 1 8 1 10 1 10 1 20 Total 10 196 7 125 7 105 7 146 4 114 SNNPR, Southern Nations Nationalities and People Region. We included children 0–59 months of age with a clinical diagnosis of pneumonia or diarrhoea but without other illnesses. On the basis of two previous cost studies,24 25 we calculated that 65 patients in each wealth quintile would allow reporting of results, suggesting a mean difference of at least 3.0 ETB across successive wealth quintiles with a SD of 6.1 ETB at 95% level of confidence and a power of 80. Hence, we aimed to collect data from a sample consisting of 375 patients (325 plus 15% non-response) with a diagnosis of pneumonia or diarrhoea. We planned to include 33 severe pneumonia and 33 severe diarrhoea cases (10% of diarrhoea and 10% of pneumonia cases) admitted for inpatient care in hospitals. Outpatient cases were enrolled consecutively when an IMCI-trained clinician identified them as having either diarrhoea or pneumonia until the sample size quota was obtained. Similarly, severe cases of pneumonia or diarrhoea were consecutively enrolled from paediatric inpatient units after the physician in charge had confirmed the diagnosis of either severe pneumonia or severe diarrhoea.

nician identified them as having either diarrhoea or pneumonia until the sample size quota was obtained. Similarly, severe cases of pneumonia or diarrhoea were consecutively enrolled from paediatric inpatient units after the physician in charge had confirmed the diagnosis of either severe pneumonia or severe diarrhoea. Data collection This study employs a mix of retrospective and prospective primary household data collection. Data on direct medical costs (registration, diagnostic workup, medications and hospital bed), direct non-medical costs (transportation, food and drinks, lodging, etc) and parents’ time loss were collected through exit interview using a retrospective structured questionnaire. Furthermore, parents were asked whether they had used over-the-counter medications and/or had a visit to traditional healers before visiting the formal private or public sector. In order to ascertain recovery and estimate additional costs (families may incur additional healthcare expenses in relation to the current illness after leaving the facility), a prospective follow-up interview was conducted at the household level within 2 weeks of initial interview or discharge. The additional expenses may relate to having another visit (because they failed to improve or for follow-up) or costs related to injections or other costs. If additional costs were incurred, we included these costs in the calculation of total medical expenditures. To collect OOP expenses for the current illness episode, we used a reference period of 2 weeks (the time between initial and follow-up interviews), since pneumonia and diarrhoea episodes are usually acute and were likely to be resolved in the period.

ncluded these costs in the calculation of total medical expenditures. To collect OOP expenses for the current illness episode, we used a reference period of 2 weeks (the time between initial and follow-up interviews), since pneumonia and diarrhoea episodes are usually acute and were likely to be resolved in the period. Household consumption expenditure data were collected by asking caretakers for monthly estimates of amounts spent on food, housing, fuel, electricity, water, education and healthcare for the month preceding the survey. Whenever possible, household heads were involved in eliciting expenditures on specific items. We derived an estimate of annual household expenditures based on the monthly survey data. Households were also asked about the availability of durable consumer goods such as radio, television, refrigerator, bicycle, car/truck, motorbike, farm equipment and agricultural land. Caretakers’ time loss was estimated by adding the time spent seeking healthcare prior to outpatient consultation and/or admission and the duration of outpatient and/or inpatient stay.

of durable consumer goods such as radio, television, refrigerator, bicycle, car/truck, motorbike, farm equipment and agricultural land. Caretakers’ time loss was estimated by adding the time spent seeking healthcare prior to outpatient consultation and/or admission and the duration of outpatient and/or inpatient stay. An investigator visited each site, identified an IMCI-trained nurse (in hospitals and health centres) or IMCI-trained health extension worker (in health posts) and provided training on the use of data collection tools to ensure that data were accurate, complete and consistent across sites. In hospitals and health centres, the investigator observed the data collection process on at least one patient with either pneumonia or diarrhoea. Owing to the low case load at health post level, exit interviews could not yield the required results in a reasonable time. We therefore identified households that had accessed care from registers at the health posts after which data were collected through a visit to their dwellings. Data were collected for the period August–December, 2013. All study participants gave a written informed consent.

ld the required results in a reasonable time. We therefore identified households that had accessed care from registers at the health posts after which data were collected through a visit to their dwellings. Data were collected for the period August–December, 2013. All study participants gave a written informed consent. Data analysis To obtain direct medical expenses per case, we added up OOP payments for registration, diagnostic work-up, medications and hospital stay. Similarly, direct non-medical expenses per case were calculated by summing the OOP payments for transportation, food, lodging and other costs incurred in relation to treatment services sought and received. Total OOP expenditure per case was calculated as the sum of the direct medical and non-medical expenses. We did not estimate the economic value of productivity losses associated with caregiver's transport and health seeking time. The two accepted approaches to value time loss (human capital and friction cost approaches) use gross wages, which is less meaningful in an economy that is largely subsistence.26

edical expenses. We did not estimate the economic value of productivity losses associated with caregiver's transport and health seeking time. The two accepted approaches to value time loss (human capital and friction cost approaches) use gross wages, which is less meaningful in an economy that is largely subsistence.26 We examined how household economic status, type of health facility, region and geographic locations (urban vs rural) were associated with OOP expenses incurred by households. We used the logarithmic transformation of OOP expenses because of the skew in the natural distribution of costs. We used linear regression model (after log transformation of OOP costs) to assess the effect of the predictor variables on the mean household OOP expenses. Logistic regression was used to identify the variables that were major drivers of differences in the rate of catastrophic head count among different wealth quintiles, by type of health facility visited and place of residence. P values of 0.05 or lower were deemed to be significant.

s on the mean household OOP expenses. Logistic regression was used to identify the variables that were major drivers of differences in the rate of catastrophic head count among different wealth quintiles, by type of health facility visited and place of residence. P values of 0.05 or lower were deemed to be significant. CHE to households associated with healthcare OOP expenses for pneumonia or diarrhoea was calculated by computing OOP expenditure incurred minus any reimbursements from third-party payers divided by annual household non-food expenditure (capacity to pay defined as effective income net of subsistence spending), following the WHO definition of CHE.3 More specifically, we defined capacity to pay (non-food expenditure) as total household expenditure net of food spending. One can better distinguish between the rich and the poor by using non-food expenditures than total expenditure.

s effective income net of subsistence spending), following the WHO definition of CHE.3 More specifically, we defined capacity to pay (non-food expenditure) as total household expenditure net of food spending. One can better distinguish between the rich and the poor by using non-food expenditures than total expenditure. We measured the incidence (catastrophic payment head count) of catastrophic expenditures.27 The measurement of this parameter is as follows: let P be OOP healthcare payment, x be total household expenditure and y be food expenditure, therefore x−y is the capacity to pay. Then, a household is said to have incurred catastrophic payments if P/(x−y), exceeds a specified threshold, z. The threshold represents the point at which families will have severe disruptions to their living standards due to healthcare spending. We used the WHO CHE threshold of healthcare payments of at least 40% of a household's capacity to pay. As childhood diarrhoea and pneumonia are usually acute conditions with shorter durations of illnesses, we opted to examine short-term and long-term impact of OOP healthcare costs on households for a single illness episode. To assess short-term impact, we used monthly capacity to pay as the denominator in the computation of CHE, while the annual estimate for capacity to pay was used as denominator to assess long-term impact.

pted to examine short-term and long-term impact of OOP healthcare costs on households for a single illness episode. To assess short-term impact, we used monthly capacity to pay as the denominator in the computation of CHE, while the annual estimate for capacity to pay was used as denominator to assess long-term impact. To measure catastrophic head count in relation to capacity to pay, let us define an indicator T, which equals 1 if Pi/(xi−yi)>z and zero otherwise. Then, an estimate of the catastrophic head count (H) measured at the household level (i) is given by 1 where N is the sample size. Medical impoverishment was measured as the expected number of households that fell below the poverty threshold of US$1.25 due to OOP spending on healthcare. Poverty head count is the fraction of people living in poverty (fraction below the poverty line (PL)). First, we constructed a PL=3180 ETB using a PPP in 2013 of 6.97.28 Then, we computed the poverty head count as follows: let wi be the per capita consumption expenditure of household i. An estimate of the poverty head count ratio without health payment deduction is 2 where =1 if wi<PL and is 0 otherwise, ni is the number of individuals in the household and N is the number of households in the sample. Then (the poverty head count after deducting healthcare payment from the per capita consumption expenditure) is computed as =1 if (wi−Pi)<PL and is 0 otherwise.

payment deduction is 2 where =1 if wi<PL and is 0 otherwise, ni is the number of individuals in the household and N is the number of households in the sample. Then (the poverty head count after deducting healthcare payment from the per capita consumption expenditure) is computed as =1 if (wi−Pi)<PL and is 0 otherwise. The total household consumption expenditure and an adult equivalent (AE)ii score (calculated based on the number and ages of household members) for each household were used to identify the economic quintile to which each study household belonged.29 Data were analysed using the statistical software package STATA (V.13). Ethical clearance The study was approved by Regional committees for medical and health research ethics (REK) in Norway and Ethiopian Health and Nutrition Research Institute (EHNRI) scientific and ethical review committee. Results Sample characteristics Of the 686 patients enrolled in the study (91% response rate), 303, 42, 309 and 32 were pneumonia, severe pneumonia, diarrhoea and severe diarrhoea cases, respectively. The mean age of patients was 1.7 years (95% CI 1.6 to 1.8 years). Details of sample characteristics are presented in table 2. Table 2 Sample characteristics, by diagnosis

Results Sample characteristics Of the 686 patients enrolled in the study (91% response rate), 303, 42, 309 and 32 were pneumonia, severe pneumonia, diarrhoea and severe diarrhoea cases, respectively. The mean age of patients was 1.7 years (95% CI 1.6 to 1.8 years). Details of sample characteristics are presented in table 2. Table 2 Sample characteristics, by diagnosis Pneumonia Severe pneumonia with inpatient care Diarrhoea Severe diarrhoea with inpatient care No. of observations 303 (44%) 42 (6%) 309 (45%) 32 (5%) Mean age in years (95% CI) 1.7 (1.5–1.8) 1.6 (1.0–2.1) 1.8 (1.7–2.0) 1.8 (1.3–2.4) Sex distribution (% female) 48% 31% 51% 65% Mean days of hospitalisation – 4 – 3 Percentage of rural residents 37% 44% 41% 38% Mean family size 4.89 5 4.82 4.88 Respondent (mother) 77% 44% 79% 50% Respondent (father) 20% 56% 20% 50% Mean age of the respondent in years (95% CI) 30 (30–31) 32 (29–35) 30 (29–31) 34.7 (30–39) Respondents education (% with some secondary education) 34% 33% 29% 33% Respondent's employment status (% in full time work) 38% 55% 37% 46% Respondent's employment status (housewife) 50% 33% 51% 38% Time spent by the respondent in relation to facility visit (hours) 8 96 6 78 Costs to the household Among the 686 patients enrolled in the study, 631 had complete data on costs incurred for the treatment of their current illness and on household consumption expenditures. Data on household consumption expenditure were missing for 55 study participants. We were able to reach 530 households for follow-up interviews to ascertain and record additional expenses incurred. We assumed that no additional expenses were incurred for the unreached households. We used data on these 631 cases for further cost analysis.

onsumption expenditure were missing for 55 study participants. We were able to reach 530 households for follow-up interviews to ascertain and record additional expenses incurred. We assumed that no additional expenses were incurred for the unreached households. We used data on these 631 cases for further cost analysis. The mean OOP direct medical expenses (in 2013 US$) were US$6 and US$5 for outpatient pneumonia and diarrhoea services, respectively. Average OOP expenses were higher for inpatient services at US$51 for severe pneumonia and US$59 for severe diarrhoea. Medication costs accounted for the major share (60%) of direct medical costs. For inpatient care, the second largest expense was the bed charge, constituting 28% of direct medical costs. Diagnostic investigations covered 16% of direct medical costs. The average associated direct non-medical expenses (mainly transport costs) for pneumonia, diarrhoea, severe pneumonia and severe diarrhoea were US$2, US$2, US$13 and US$20, respectively. A breakdown of the direct medical and non-medical costs incurred by households is detailed in table 3. Table 3 Mean (SD) medical expenditure in US$ per disease episode by cost type and diagnosis

The mean OOP direct medical expenses (in 2013 US$) were US$6 and US$5 for outpatient pneumonia and diarrhoea services, respectively. Average OOP expenses were higher for inpatient services at US$51 for severe pneumonia and US$59 for severe diarrhoea. Medication costs accounted for the major share (60%) of direct medical costs. For inpatient care, the second largest expense was the bed charge, constituting 28% of direct medical costs. Diagnostic investigations covered 16% of direct medical costs. The average associated direct non-medical expenses (mainly transport costs) for pneumonia, diarrhoea, severe pneumonia and severe diarrhoea were US$2, US$2, US$13 and US$20, respectively. A breakdown of the direct medical and non-medical costs incurred by households is detailed in table 3. Table 3 Mean (SD) medical expenditure in US$ per disease episode by cost type and diagnosis Diagnosis Cost type Pneumonia Diarrhoea Severe pneumonia with inpatient care Severe diarrhoea with inpatient care Transportation 0.97 (2.22) 0.99 (3.30) 6.25 (7.66) 9.64 (11.02) Registration/consultation 0.82 (1.76) 0.71 (1.45) 2.15 (2.72) 2.18 (2.64) Laboratory 1.20 (3.48) 0.88 (2.36) 7.34 (12.94) 10.22 (17.05) Medicines and supplies 4.27 (6.42) 3.02 (5.28) 28.53 (30.78) 28.89 (33.86) Hospital bed – – 12.69 (15.37) 17.62 (32.96) Traditional healer visit* 0.11 (0.56) 0.12 (0.90) – 1.15 (4.47) Other† 0.60 (2.31) 0.48 (2.09) 6.81 (6.35) 9.58 (12.61) DMC‡ 6.30 (10.51) 4.65 (8.43) 50.70 (52.38) 58.9 (68.95) DNMC‡ 1.68 (3.85) 1.59 (4.92) 13.05 (10.48) 20.37 (21.44) Total medical expenditure§ 7.98 (12.83) 6.24 (11.88) 63.76 (54.26) 79.27 (74.38) *Among 345 pneumonia cases who visited health facilities 18 had had a visit to a traditional healer with a mean (SD) cost of 1.72 (1.49). Among 341 diarrhoea cases who visited health facilities 16 had had a visit to a traditional healer with a mean (SD) cost of 2.74 (3.39).

.83) 6.24 (11.88) 63.76 (54.26) 79.27 (74.38) *Among 345 pneumonia cases who visited health facilities 18 had had a visit to a traditional healer with a mean (SD) cost of 1.72 (1.49). Among 341 diarrhoea cases who visited health facilities 16 had had a visit to a traditional healer with a mean (SD) cost of 2.74 (3.39). †Other costs include expenses incurred for food, lodging, etc. ‡DMC includes registration fee, medicines, laboratory and diagnostics and bed charges while DNMC includes transport, lodging, traditional healer, etc. §Total medical expenditure is the sum of DMC and DNMC. DMC, direct medical costs; DNMC, direct non-medical costs. The mean total medical expenditures for an episode of pneumonia, diarrhoea, severe pneumonia or severe diarrhoea were 2.3–3.8 times higher in private facilities than at government hospitals (table 4). Type of health facility visited was the main predictor of a difference in the mean total medical expenditure for each disease category. Child healthcare services were not entirely free of charge at public PHC facilities. At health posts, though consultation fees were not paid, parents were obliged to buy medication from private outlets because of a lack of drug stock at health posts. In most of the health centres, parents paid fees for consultation and medications. Table 4 Average total medical expenditure per disease episode in US$ by type of health facility visited

The mean total medical expenditures for an episode of pneumonia, diarrhoea, severe pneumonia or severe diarrhoea were 2.3–3.8 times higher in private facilities than at government hospitals (table 4). Type of health facility visited was the main predictor of a difference in the mean total medical expenditure for each disease category. Child healthcare services were not entirely free of charge at public PHC facilities. At health posts, though consultation fees were not paid, parents were obliged to buy medication from private outlets because of a lack of drug stock at health posts. In most of the health centres, parents paid fees for consultation and medications. Table 4 Average total medical expenditure per disease episode in US$ by type of health facility visited Diagnosis Type of health facility No. of cases (%) Mean cost (SD) Pneumonia* Health post (HP) 42 (14%) 1.61 (2.71) Health centre (HC) 181 (60%) 4.06 (5.91) Government hospital 57 (19%) 12.08 (12.05) Private clinic/hospital 23 (7%) 28.12 (8.85) Diarrhoea* Health post 47 (15%) 0.97 (1.97) Health centre 183 (59%) 3.89 (6.13) Government hospital 57 (19%) 5.66 (5.97) Private clinic/hospital 22 (7%) 21.41 (11.17) Severe pneumonia with inpatient care Health post 0 – Health centre 3 (7%) 12.13 (8.80) Government hospital 26 (62%) 47.89 (28.81) Private clinic/hospital 13 (31%) 139.66 (71.97) Severe diarrhoea with inpatient care Health post 0 – Health centre 1 (3%) 15.59 Government hospital 20 (63%) 55.92 (58.96) Private clinic/hospital 11 (34%) 151.86 (84.33) *For both conditions, medical costs per episode were five to seven times greater in private facilities compared with health centres. The differences by facility type were statistically significant for both conditions (p<0.001).

9 Government hospital 20 (63%) 55.92 (58.96) Private clinic/hospital 11 (34%) 151.86 (84.33) *For both conditions, medical costs per episode were five to seven times greater in private facilities compared with health centres. The differences by facility type were statistically significant for both conditions (p<0.001). There were marked variations in total medical expenditures by wealth quintile, place of residence and region (tables 5 and 6). Table 5 shows the distribution of total medical expenses for diarrhoea and pneumonia by wealth quintile. The wealthiest households spent six times more on treatment as compared to the poorest households. The mean total medical expenditure was 1.7–2 times higher in urban than in rural households. Urban households and wealthier quintiles were more likely to visit private facilities or public hospitals than PHC facilities. 33% of urban households and 16% of rural households had outpatient visits in either private facilities or public hospitals. Similarly, 39% of the wealthiest two quintiles and 13% of the poorest two quintiles had outpatient visits in either private facilities or public hospitals. Number of severe pneumonia and severe diarrhoea cases was disproportionately high in Addis Ababa, accounting for 27% and 38% of all reported severe cases, respectively. Table 5 Mean monthly consumption expenditure and total medical expenditure per disease episode in US$ by wealth quintile

There were marked variations in total medical expenditures by wealth quintile, place of residence and region (tables 5 and 6). Table 5 shows the distribution of total medical expenses for diarrhoea and pneumonia by wealth quintile. The wealthiest households spent six times more on treatment as compared to the poorest households. The mean total medical expenditure was 1.7–2 times higher in urban than in rural households. Urban households and wealthier quintiles were more likely to visit private facilities or public hospitals than PHC facilities. 33% of urban households and 16% of rural households had outpatient visits in either private facilities or public hospitals. Similarly, 39% of the wealthiest two quintiles and 13% of the poorest two quintiles had outpatient visits in either private facilities or public hospitals. Number of severe pneumonia and severe diarrhoea cases was disproportionately high in Addis Ababa, accounting for 27% and 38% of all reported severe cases, respectively. Table 5 Mean monthly consumption expenditure and total medical expenditure per disease episode in US$ by wealth quintile Pneumonia* Diarrhoea* Wealth quintile Mean monthly consumption expenditure (US$) Mean total medical expenditure (US$) Mean monthly consumption expenditure (US$) Mean total medical expenditure (US$) I 56 3.17 (6%) 48 3.18 (7%) II 90 4.71 (5%) 89 4.58 (5%) III 107 9.13 (9%) 115 4.84 (4%) IV 125 8.20 (7%) 126 6.45 (5%) V 209 15.11 (7%) 195 13.43 (7%) *For both conditions, medical costs per episode were four to five times greater in the highest wealth quintile compared with the lowest. The difference by wealth quintile was statistically significant for both conditions (p<0.001). The numbers in parentheses denote the mean total medical expenditure divided by the mean monthly consumption expenditure.

ts per episode were four to five times greater in the highest wealth quintile compared with the lowest. The difference by wealth quintile was statistically significant for both conditions (p<0.001). The numbers in parentheses denote the mean total medical expenditure divided by the mean monthly consumption expenditure. Table 6 Total medical expenditure (mean and SD) per disease episode in US$ by place of residence and region Place of residence Region Diagnosis Urban Rural Amhara SNNPR Oromia Tigray Addis Ababa No. of observations 411 (60%) 274 (40%) 125 (18%) 105 (15%) 196 (29%) 146 (21%) 114 (17%) Pneumonia* 8.66 (12.58) 4.36 (6.35) 2.07 (1.60) 5.26 (4.08) 7.23 (10.58) 9.21 (10.53) 12.30 (18.74) Diarrhoea* 6.51 (9.81) 2.99 (3.94) 2.36 (3.71) 3.73 (3.24) 4.55 (6.98) 6.65 (8.83) 7.97 (13.56) Severe pneumonia 91.01 (75.52) 48.35 (39.06) 34.11 (12.85) 25.47 (9.02) 53.34 (35.83) 45.86 (15.66) 126.03 (77.34) Severe diarrhoea 98.81 (83.40) 59.08 (78.75) – 16.30 (3.37) 58.51 (54.60) 10.75 146.59 (82.53) *For both conditions, medical costs per episode were three to six times greater in Addis Ababa compared with Amhara region. The regional differences were statistically significant for both conditions (p<0.002). SNNPR, Southern Nations Nationalities and People Region.

Place of residence Region Diagnosis Urban Rural Amhara SNNPR Oromia Tigray Addis Ababa No. of observations 411 (60%) 274 (40%) 125 (18%) 105 (15%) 196 (29%) 146 (21%) 114 (17%) Pneumonia* 8.66 (12.58) 4.36 (6.35) 2.07 (1.60) 5.26 (4.08) 7.23 (10.58) 9.21 (10.53) 12.30 (18.74) Diarrhoea* 6.51 (9.81) 2.99 (3.94) 2.36 (3.71) 3.73 (3.24) 4.55 (6.98) 6.65 (8.83) 7.97 (13.56) Severe pneumonia 91.01 (75.52) 48.35 (39.06) 34.11 (12.85) 25.47 (9.02) 53.34 (35.83) 45.86 (15.66) 126.03 (77.34) Severe diarrhoea 98.81 (83.40) 59.08 (78.75) – 16.30 (3.37) 58.51 (54.60) 10.75 146.59 (82.53) *For both conditions, medical costs per episode were three to six times greater in Addis Ababa compared with Amhara region. The regional differences were statistically significant for both conditions (p<0.002). SNNPR, Southern Nations Nationalities and People Region. Catastrophic health expenditures and impoverishment Household annual mean total expenditures and mean non-food expenditures were US$1320 and US$349, respectively. For outpatient care, 0.3–0.6% of households incurred CHE at 40% annual capacity to pay threshold level (table 7). The figure rises to 21–24% when we used the 40% monthly capacity to pay threshold level as the denominator. The incidence of CHE was higher for severe cases of pneumonia and diarrhoea. Disaggregation of CHE by place of residence and wealth quintile revealed that rural and poor households were less able to cope with any given level of health expenditure than urban and wealthier households (table 7). For outpatient pneumonia or diarrhoea episodes, 0.3% of households were pushed into extreme poverty due to OOP payments. The figures were much higher for inpatient care, where 7% and 6% of the households with severe pneumonia and severe diarrhoea cases, respectively, were pushed below the extreme PL.

seholds (table 7). For outpatient pneumonia or diarrhoea episodes, 0.3% of households were pushed into extreme poverty due to OOP payments. The figures were much higher for inpatient care, where 7% and 6% of the households with severe pneumonia and severe diarrhoea cases, respectively, were pushed below the extreme PL. Table 7 Incidence of Catastrophic Health Payments per disease episode, defined with respect to capacity to pay in Ethiopia, 2013 Out-of-pocket health spending as share of CTP, at 40% threshold budget share Annual CTP Monthly CTP Diagnosis Average Average Rural Urban The bottom half quintile The upper half quintile Private facilities Public facilities Both* All categories 1.6% (631) 31% (631) 36% (286) 27% (345)† 35% (341) 26% (290)† 78% (64) 25% (567)† Outpatient Pneumonia 0.3% (277) 24% (277) 27% (103) 23% (174) 28% (159) 17% (118)† 83% (21) 19% (256)† Diarrhoea 0.6% (280) 21% (280) 31% (157) 15% (123)† 29% (150) 13% (130)† 53% (19) 19% (261)† Inpatient* Severe pneumonia or diarrhoea 11% (74) 96% (74) 100% (26) 94% (48) 100% (32) 93% (42) 96% (24) 96% (50) *Both include outpatient and inpatient cases. †The rate of catastrophic head count varied significantly by place of residence, type of health facility visited or wealth quintile. The numbers in parentheses are the number of observations. CTP, capacity to pay.

Out-of-pocket health spending as share of CTP, at 40% threshold budget share Annual CTP Monthly CTP Diagnosis Average Average Rural Urban The bottom half quintile The upper half quintile Private facilities Public facilities Both* All categories 1.6% (631) 31% (631) 36% (286) 27% (345)† 35% (341) 26% (290)† 78% (64) 25% (567)† Outpatient Pneumonia 0.3% (277) 24% (277) 27% (103) 23% (174) 28% (159) 17% (118)† 83% (21) 19% (256)† Diarrhoea 0.6% (280) 21% (280) 31% (157) 15% (123)† 29% (150) 13% (130)† 53% (19) 19% (261)† Inpatient* Severe pneumonia or diarrhoea 11% (74) 96% (74) 100% (26) 94% (48) 100% (32) 93% (42) 96% (24) 96% (50) *Both include outpatient and inpatient cases. †The rate of catastrophic head count varied significantly by place of residence, type of health facility visited or wealth quintile. The numbers in parentheses are the number of observations. CTP, capacity to pay. Discussion and conclusions Our study documented OOP payments and time loss for the two most common causes of morbidity and mortality in children 0–59 months in Ethiopia. The findings demonstrate that OOP expenditures associated with diarrheal illness or pneumonia can be a substantial economic burden for households. Most of the total medical expenditures (ranging from 74% to 80%) consist of direct medical costs. Medications were the major contributor to direct medical costs for outpatient and inpatient visits, followed by bed charges for inpatient care. Several previous studies conducted elsewhere reported comparable estimates of total household medical expenditures, as well as identifying direct medical costs and medications as the major drivers of total medical expenditures.9–11 30–32 Among the direct non-medical costs, transportation costs presented families with a significant financial hurdle even before accessing needed formal care.

tes of total household medical expenditures, as well as identifying direct medical costs and medications as the major drivers of total medical expenditures.9–11 30–32 Among the direct non-medical costs, transportation costs presented families with a significant financial hurdle even before accessing needed formal care. OOP expenses varied depending on the facility visited, families spending significantly higher costs in private health facilities. The average OOP expenses for treating pneumonia and diarrhoea in private facilities were US$28 and US$21 per case, respectively. Households incurred the least costs at public PHC facilities, where the mean total medical expenditures at health centres for outpatient care of pneumonia or diarrhoea were US$4.1 and US$3.9, respectively.

P expenses for treating pneumonia and diarrhoea in private facilities were US$28 and US$21 per case, respectively. Households incurred the least costs at public PHC facilities, where the mean total medical expenditures at health centres for outpatient care of pneumonia or diarrhoea were US$4.1 and US$3.9, respectively. Our study shows that wealthier households have greater demand and access to health services and a wider range of choices to select from. Urban and wealthier households were more likely to visit private facilities or public hospitals where the perceived quality of care is superior. At the same time, the poor have lower overall OOP expenses, especially for outpatient services, reflecting their more limited access to health services. Barnet and Tefera reported a preference among poor households in Ethiopia for higher-level health facilities because the quality and quantity of services available at PHC facilities were perceived as inferior.33 Despite such perceptions, poor households were less likely to visit facilities where they were more likely to incur higher expenditures, possibly a function of households' inability to absorb medical payments. User fees at public health facilities are associated with decreased service usage, even more so for marginalised segments of the population such as women, children and the poor.7 34 35 Evidence from similar settings in Africa also suggests that abolition of user fees results in increased service usage in all population groups.36 User fees could hamper the Ethiopian government's efforts to make essential priority services universally accessible.37 One of the fundamental impediments to UHC is over reliance on direct payments at the time people need care.38 39