Browse the corpus

Key questions What is the key question? What is the prevalence of frailty in stable COPD, and does frailty affect the completion and outcomes of pulmonary rehabilitation? What is the bottom line? Frailty affects one in every four patients with COPD entering pulmonary rehabilitation, is associated with favourable outcomes, but is also strong risk factor for non-completion. Why read on? This is the first characterisation of the frailty phenotype in stable COPD and demonstrates that physical frailty is amenable to treatment with pulmonary rehabilitation. Introduction Frailty describes a clinical syndrome characterised by multisystem decline that leads to reduced functional reserve and increased vulnerability to dependency or mortality following minor stressor events.1 It affects an estimated 1 in every 10 people aged over 65 years2 and is consistently associated with increased risk of falls, disability, hospitalisation and death.3 Although frailty is conventionally considered secondary to age-related decline, chronic disease(s) can accelerate the rate of decline and precipitate a frail state. In COPD, extrapulmonary manifestations include physical inactivity, muscle weakness, anorexia, osteoporosis and fatigue.4 Each of these systemic impacts of the disease is frequently observed in physical frailty.

to age-related decline, chronic disease(s) can accelerate the rate of decline and precipitate a frail state. In COPD, extrapulmonary manifestations include physical inactivity, muscle weakness, anorexia, osteoporosis and fatigue.4 Each of these systemic impacts of the disease is frequently observed in physical frailty. The relevance of frailty to chronic respiratory disease has not been fully dissected. In retrospective cohort studies, self-reported frailty is more common in older people with COPD than without it, and markers of frailty have identified those at increased risk of subsequent hospital admission or death.5 6 Few studies have examined the prevalence of frailty in respiratory disease using validated definitions. One notable exception is a report restricted to lung transplant candidates, among whom frailty was associated with increased risk of delisting or death.7 Identifying frailty earlier in the course of disease is important, as interventions may then be introduced to prevent functional decline, hospital admissions and/or death in those at high risk. Frailty may prove a valuable way of stratifying patients with COPD for future management as it accounts for multiple deficits that influence disease prognosis, for example, muscle weakness or physical inactivity,8 including deficits not considered by other syndromes or comorbidity indices.9 10

th in those at high risk. Frailty may prove a valuable way of stratifying patients with COPD for future management as it accounts for multiple deficits that influence disease prognosis, for example, muscle weakness or physical inactivity,8 including deficits not considered by other syndromes or comorbidity indices.9 10 Another important but unstudied topic is the interplay between frailty and pulmonary rehabilitation. Pulmonary rehabilitation is highly effective at improving symptom burden, physical function and health status, although patient response is heterogeneous.11 12 Conceptually, pulmonary rehabilitation targets many components of frailty, including slowness,13 fatigue,14 weakness15 and physical inactivity,11 and provides a holistic approach to encourage self-management, disease education and behaviours to improve overall health. Pulmonary rehabilitation also targets dyspnoea, which may be a contributing factor to the development of frailty in people with chronic respiratory disease. Finally, some programmes incorporate falls prevention strategies, through balance training and education,16 again focusing on a frailty-related outcome. However, in the same way that frailty affects planned surgical management,7 the syndrome may prevent patients from engaging in pulmonary rehabilitation—a mainstay of disease management.11 The factors for heterogeneous adherence to pulmonary rehabilitation are widely debated,17 18 but physical frailty is a plausible candidate given the close relationship with adverse outcomes in COPD6 7 and other long-term conditions.3 If frailty does hinder completion of pulmonary rehabilitation, this would suggest a need to support patients who are frail with alternative or supplementary rehabilitation strategies.

physical frailty is a plausible candidate given the close relationship with adverse outcomes in COPD6 7 and other long-term conditions.3 If frailty does hinder completion of pulmonary rehabilitation, this would suggest a need to support patients who are frail with alternative or supplementary rehabilitation strategies. Study objectives were to determine the prevalence of frailty among patients with stable COPD, describe the clinical characteristics of the COPD frailty phenotype and examine whether frailty affects completion and clinical outcomes of pulmonary rehabilitation. We hypothesised that frailty would independently predict non-completion of pulmonary rehabilitation.

revalence of frailty among patients with stable COPD, describe the clinical characteristics of the COPD frailty phenotype and examine whether frailty affects completion and clinical outcomes of pulmonary rehabilitation. We hypothesised that frailty would independently predict non-completion of pulmonary rehabilitation. Methods Participants and design Participants were recruited to this prospective cohort study from respiratory outpatient and pulmonary rehabilitation clinics at Harefield Hospital (Middlesex, UK) between November 2011 and January 2015. Eligible patients were aged 35 years or above, with a physician diagnosis of COPD consistent with the Global Initiative for Chronic Obstructive Lung Disease (GOLD) criteria19 and appropriate for a pulmonary rehabilitation referral. Exclusion criteria were an exacerbation in the previous 4 weeks that required a change in medication, a condition that might make moderate-intensity exercise unsafe, for example, unstable cardiac disease or a predominant neurological disability. Treating clinicians identified potentially eligible patients and offered a written information sheet. Referral criteria for pulmonary rehabilitation were able to walk at least 5 m, any degree of functional impairment secondary to dyspnoea (typically Medical Research Council (MRC) dyspnoea score 2 or more), no previous supervised pulmonary rehabilitation in previous 12 months and without unstable cardiovascular disease, in line with British Thoracic Society Quality Standards.20 Data for some participants (528/816, 64.7%) relating to sarcopenia, but not frailty, have been reported previously.15 All participants gave written informed consent in accordance with the principles of Good Clinical Practice and the Declaration of Helsinki, and the study was approved by the West London (11/H0707/2) and London Camberwell St Giles (11/LO/1780) Research Ethics Committees.

but not frailty, have been reported previously.15 All participants gave written informed consent in accordance with the principles of Good Clinical Practice and the Declaration of Helsinki, and the study was approved by the West London (11/H0707/2) and London Camberwell St Giles (11/LO/1780) Research Ethics Committees. Frailty assessment Frailty was defined using the Fried phenotype model,9 which is well established and validated in large epidemiological studies.3 This model comprises five characteristics that reduce physiological reserve and precipitate a vulnerable state; shrinking (unintentional weight loss), exhaustion, low physical activity, slowness and weakness.9 Characteristics were assessed by weight loss history, two questions from the Center for Epidemiological Studies-Depression (CES-D) questionnaire (exhaustion),21 weekly self-reported energy expenditure using the modified Minnesota Leisure-Time Physical Activity Questionnaire (low physical activity),22 4-m gait speed (4MGS, slowness)23 and handgrip dynamometry (weakness). Standardised criteria, derived from the original reference cohort,9 were used to define each characteristic as either present or absent for each patient (see online supplementary table S1), providing an ordinal score ranging 0–5. Patients with no criteria present were considered not-frail/robust, those meeting 1–2 criteria were considered prefrail and those with ≥3 criteria present were considered frail. 10.1136/thoraxjnl-2016-208460.supp1Supplementary data

Frailty assessment Frailty was defined using the Fried phenotype model,9 which is well established and validated in large epidemiological studies.3 This model comprises five characteristics that reduce physiological reserve and precipitate a vulnerable state; shrinking (unintentional weight loss), exhaustion, low physical activity, slowness and weakness.9 Characteristics were assessed by weight loss history, two questions from the Center for Epidemiological Studies-Depression (CES-D) questionnaire (exhaustion),21 weekly self-reported energy expenditure using the modified Minnesota Leisure-Time Physical Activity Questionnaire (low physical activity),22 4-m gait speed (4MGS, slowness)23 and handgrip dynamometry (weakness). Standardised criteria, derived from the original reference cohort,9 were used to define each characteristic as either present or absent for each patient (see online supplementary table S1), providing an ordinal score ranging 0–5. Patients with no criteria present were considered not-frail/robust, those meeting 1–2 criteria were considered prefrail and those with ≥3 criteria present were considered frail. 10.1136/thoraxjnl-2016-208460.supp1Supplementary data Additional assessments Bioelectrical impedance analysis (Quadscan 4000, Bodystat, Isle of Man, UK) was used to estimate skeletal muscle mass index (SMI).24 The presence of sarcopenia was defined according to the consensus European Working Group on Sarcopenia in Older People criteria.25 Quadriceps maximum voluntary contraction (QMVC) was measured using a fixed strain gauge,26 with weakness diagnosed according to healthy predicted values27 and sex-specific functionally relevant cut-points.28 Exercise performance was assessed using the incremental shuttle walk test (ISWT).29 Further measurements included evaluation of respiratory disability by MRC dyspnoea score, comorbidity burden using the age-adjusted Charlson Index,30 help with activities of daily care using the Katz Index,31 composite disease severity by the age, dyspnoea, and airflow obstruction (ADO) index,32 Hospital Anxiety and Depression Scale (HADS) and health status using the self-reported Chronic Respiratory Questionnaire (CRQ)33 and COPD Assessment Test (CAT).34

n Index,30 help with activities of daily care using the Katz Index,31 composite disease severity by the age, dyspnoea, and airflow obstruction (ADO) index,32 Hospital Anxiety and Depression Scale (HADS) and health status using the self-reported Chronic Respiratory Questionnaire (CRQ)33 and COPD Assessment Test (CAT).34 Pulmonary rehabilitation Pulmonary rehabilitation was an 8-week outpatient exercise and multidisciplinary education programme, comprising two supervised and at least one additional home-based session each week and organised according to the British Thoracic Society Quality Standards for Pulmonary Rehabilitation.20 Supervised sessions comprised 1 hour of exercise and 45 min of education. Exercise training was individualised and progressive. Initial walking speed prescription was at 80% of predicted peak oxygen consumption based on ISWT performance,35 while initial endurance cycling was set at a workload with the aim of patients completing 10 min of continuous training. Lower limb resistance training comprised 2 sets of 10 seated leg press repetitions, performed with an initial training load of 60% one-repetition maximum, as well as sit-to-stand, knee extension, hip flexion and hip abduction exercises with ankle weights. Upper limb resistance training comprised biceps curls, shoulder press and upright row with free weights. Education was delivered by a multidisciplinary team. Topics were chosen to develop patients' understanding and holistic management of their disease and included physical activity and exercise, medication use, diet, smoking cessation, coping strategies, as well as managing infections through early recognition, rescue medication and appropriate general practice/hospital presentation.

chosen to develop patients' understanding and holistic management of their disease and included physical activity and exercise, medication use, diet, smoking cessation, coping strategies, as well as managing infections through early recognition, rescue medication and appropriate general practice/hospital presentation. Statistical analysis Our sample size was based on the precision to which the overall prevalence of frailty could be estimated. Assuming it was within the range 10%–60%, prevalence could be estimated to within ±3.5% using 800 participants with a large sample normal approximation (nQuery Advisor V.6.0). Analyses were performed using SPSS (V.22, IBM, New York, USA) and graphs produced using Prism 5 (GraphPad Software, San Diego, California, USA). Data were presented as proportions (95% CIs) or mean (SD).

be estimated to within ±3.5% using 800 participants with a large sample normal approximation (nQuery Advisor V.6.0). Analyses were performed using SPSS (V.22, IBM, New York, USA) and graphs produced using Prism 5 (GraphPad Software, San Diego, California, USA). Data were presented as proportions (95% CIs) or mean (SD). The prevalence of frailty was determined overall and then compared across groups according to age, GOLD spirometry stage, MRC dyspnoea score and age-adjusted Charlson comorbidity score using χ2 for trend. Baseline characteristics as well as pulmonary rehabilitation uptake, attendance and completion rates were compared across groups (not frail/robust, pre-frail, frail) using one-way analysis of variance or χ2 for trend with a Bonferroni correction applied to post hoc pairwise comparisons. Uptake was defined as the proportion of assessed patients who attended the first supervised session, adherence was assessed using the number of supervised sessions attended and completion was defined as the proportion of patients who had documented attendance at a minimum of eight supervised sessions, representing 50% attendance.36

fined as the proportion of assessed patients who attended the first supervised session, adherence was assessed using the number of supervised sessions attended and completion was defined as the proportion of patients who had documented attendance at a minimum of eight supervised sessions, representing 50% attendance.36 Univariate logistic regression was used to assess the relationships between non-completion, frailty status (not or prefrail/frail) and candidate explanatory variables informed from existing literature and clinical judgement.17 18 Variables associated with non-completion (p<0.15) were considered in a multivariate model. After checking for collinearity (r<0.75), we used a backwards conditional approach to retain variables in the model (p<0.10). Outcomes of rehabilitation were summarised as change pre-to-post rehabilitation for patients who completed as per the above definition. Outcomes were then compared across groups using analysis of covariance adjusting for age and sex. To control for Type I errors in view of multiple testing, we applied a Bonferroni correction to a significance level of 0.05 when comparing baseline patient characteristics and rehabilitation outcomes.

he above definition. Outcomes were then compared across groups using analysis of covariance adjusting for age and sex. To control for Type I errors in view of multiple testing, we applied a Bonferroni correction to a significance level of 0.05 when comparing baseline patient characteristics and rehabilitation outcomes. Results Prevalence of frailty Eight-hundred and sixteen patients took part (484 men, mean (SD) age 70 (10) years, FEV1 48.9 (21.0) % predicted, MRC score 3.3 (1.0)) (table 1). The overall prevalence of frailty was 25.6% (95% CI 22.7% to 28.7%), while 10.0% (95% CI 8.2% to 12.3%) of patients did not meet any of the frailty criteria and were considered robust (figure 1). Of the frailty characteristics, the exhaustion criterion was met by the largest proportion of patients (65.3%) followed by weakness, low physical activity and slowness. The unintentional weight loss criterion was met by fewest patients (14.2%) (table 1). Frailty tended to be more common among women than men (29.7% vs 22.8%, p=0.08). Prevalence increased statistically with increasing age (p<0.001), GOLD stage (p=0.01), MRC score (p<0.001) and comorbidity burden (p=0.004). There was a twofold increase in frailty prevalence among patients with GOLD stage IV as compared with stage I disease (34.7% vs 17.9%, p<0.001) and a threefold increase among patients with an MRC score of 5 as compared with a score of 3 (62.1% vs 21.0%, p<0.001; figure 2). Table 1 Baseline characteristics and progression through pulmonary rehabilitation for the total cohort and stratified by frailty status

Results Prevalence of frailty Eight-hundred and sixteen patients took part (484 men, mean (SD) age 70 (10) years, FEV1 48.9 (21.0) % predicted, MRC score 3.3 (1.0)) (table 1). The overall prevalence of frailty was 25.6% (95% CI 22.7% to 28.7%), while 10.0% (95% CI 8.2% to 12.3%) of patients did not meet any of the frailty criteria and were considered robust (figure 1). Of the frailty characteristics, the exhaustion criterion was met by the largest proportion of patients (65.3%) followed by weakness, low physical activity and slowness. The unintentional weight loss criterion was met by fewest patients (14.2%) (table 1). Frailty tended to be more common among women than men (29.7% vs 22.8%, p=0.08). Prevalence increased statistically with increasing age (p<0.001), GOLD stage (p=0.01), MRC score (p<0.001) and comorbidity burden (p=0.004). There was a twofold increase in frailty prevalence among patients with GOLD stage IV as compared with stage I disease (34.7% vs 17.9%, p<0.001) and a threefold increase among patients with an MRC score of 5 as compared with a score of 3 (62.1% vs 21.0%, p<0.001; figure 2). Table 1 Baseline characteristics and progression through pulmonary rehabilitation for the total cohort and stratified by frailty status All (n=816) Not frail (n=82) Prefrail (n=525) Frail (n=209) p Value Age (years) 69.8 (9.7) 67.4 (8.1) 69.0 (9.6) 72.6 (10.0)*† <0.001 Males, n (%) 484 (59.3) 49 (59.8) 324 (61.7) 110 (52.6) 0.08 Smoking status Current:former:never (%) 17.9:76.5:6.6 9.8:84.1:6.1 18.7:75.0:6.3 15.3:77.0:7.7 0.45 FEV1 % predicted 48.9 (21.0) 57.0 (22.4) 49.0 (20.8) 46.3 (20.1)* <0.001 MRC score 3.3 (1.1) 2.4 (0.8) 3.2 (1.0) 4.0 (0.9)*† <0.001 Age-adjusted Charlson Index 4.3 (1.6) 4.4 (1.6) 4.3 (1.6) 4.4 (1.6) 0.57 ADO score 4.9 (1.7) 3.6 (1.3) 4.7 (1.6) 6.0 (1.5)*† <0.001 BMI (kg/m) 27.8 (6.7) 27.2 (5.2) 27.8 (6.5) 27.9 (7.6) 0.71 SMI (kg/m2) 8.47 (1.87) 8.6 (1.9) 8.6 (1.9) 8.1 (1.8)*† 0.002 Sarcopenia (%) 12.4 (10.3, 14.8) 1.2 (0.2, 6.6) 9.5 (7.3, 12.3) 23.9 (18.6, 30.1) <0.001 Handgrip (kg) 27.0 (9.9) 33.0 (8.9) 28.3 (9.6) 21.3 (8.2)*† <0.001 Peak QMVC (kg) 26.2 (10.0) 31.0 (10.1) 27.3 (9.6) 21.0 (9.0)*† <0.001 QMVC % predicted 59.1 (20.0) 67.2 (20.6) 60.6 (20.0) 51.1 (16.9)*† <0.001 Below QMVC cut-point (%) 25.9 (22.8, 29.2) 6.4 (2.8, 14.1) 21.4 (18.0, 25.3) 44.0 (36.8, 51.6)*† <0.001 4MGS (m/s) 0.90 (0.24) 1.11 (0.21) 0.95 (0.19) 0.66 (0.20)*† <0·001 ISWT (m) 222.3 (151.3) 375.4 (168.3) 245.1 (135.4) 105.0 (89.5)*† <0.001 CRQ dyspnoea score 13.9 (5.6) 16.4 (5.9) 14.0 (5.7) 12.7 (5.0)*† <0.001 CRQ fatigue score 13.4 (5.3) 17.6 (4.3) 13.8 (5.2) 10.9 (4.6)*† <0.001 CRQ emotional score 30.6 (9.5) 36.1 (7.7) 31.3 (9.1) 26.6 (9.5)*† <0.001 CRQ mastery score 17.6 (6.0) 21.7 (4.6) 18.0 (5.8) 15.0 (5.7)*† <0.001 Self-reported weekly energy expenditure (kcal) 698.9 (1478.1) 1878.4 (1631.0) 1110.5 (1549.7) 257.2 (450.0)*† <0.001 Self-reported time in moderate activity (min/week) 279.9 (431.1) 532.8 (491.7) 322.6 (465.0) 73.6 (1292)*† <0.001 CAT score 20.7 (8.3) 13.3 (5.6) 20.2 (7.8) 25.0 (7.9)*† <0.001 Katz score 5.7 (0.7) 5.9 (0.2) 5.8 (0.5) 5.4 (1.0)*† <0.001 HADS anxiety 7.1 (4.6) 5.2 (3.4) 6.8 (4.4) 8.3 (5.2)*† <0.001 HADS depression 6.6 (3.8) 4.1 (2.7) 6.3 (3.7) 8.2 (4.0)*† <0.001 Self-reported hospital admission in previous year, n (%) 337 (41.3) 28 (34.1) 215 (41.0) 94 (45.0)*† <0.001 Self-report number of exacerbations in previous year 2.8 (3.6) 2.1 (2.3) 2.8 (3.9) 3.1 (3.0)* <0.0

7.1 (4.6) 5.2 (3.4) 6.8 (4.4) 8.3 (5.2)*† <0.001 HADS depression 6.6 (3.8) 4.1 (2.7) 6.3 (3.7) 8.2 (4.0)*† <0.001 Self-reported hospital admission in previous year, n (%) 337 (41.3) 28 (34.1) 215 (41.0) 94 (45.0)*† <0.001 Self-report number of exacerbations in previous year 2.8 (3.6) 2.1 (2.3) 2.8 (3.9) 3.1 (3.0)* <0.0 01 Frailty characteristic (% meeting criteria) Unintentional weight loss 14.2 (12.0 to 16.8) 0 11.8 (9.3 to 14.9) 25.4 (19.9 to 31.7) <0.001 Exhaustion 65.3 (62.0 to 68.5) 0 68.4 (64.3 to 72.2) 83.3 (77.6 to 87.7) <0.001 Low physical activity 35.9 (32.7 to 39.3) 0 23.8 (20.4 to 27.6) 80.4 (74.5 to 85.2) <0.001 Slow gait speed 24.4 (21.6 to 27.4) 0 9.1 (7.0 to 11.9) 72.2 (65.8 to 77.9) <0.001 Weak handgrip strength 43.6 (40.3 to 47.1) 0 35.8 (31.8 to 40.0) 80.4 (74.5 to 85.2) <0.001 Pulmonary rehabilitation Started (% of referred) 84.7 (82.0 to 87.0) 80.4 (94.1) 87.4 (84.3 to 90.0) 76.1 (69.9 to 81.4) <0.001 Number of sessions attended 11.4 (4.2) 13.2 (3.0) 11.6 (4.1) 10.2 (4.7) <0.001 Completed (% of starters) 83.1 (80.1 to 85.7) 94.5 (86.7 to 97.8) 82.8 (79.1 to 86.0) 72.3 (64.9 to 78.7) <0.001 Completed (% of referred) 70.3 (67.1 to 73.4) 84.1 (74.7 to 90.5) 74.3 (70.4 to 77.8) 55.0 (48.2 to 61.6) <0.001 Values are mean (SD) or proportions (95% CI) unless stated. *Statistically different to non-frail group. †Statistically different to prefrail group.

01 Frailty characteristic (% meeting criteria) Unintentional weight loss 14.2 (12.0 to 16.8) 0 11.8 (9.3 to 14.9) 25.4 (19.9 to 31.7) <0.001 Exhaustion 65.3 (62.0 to 68.5) 0 68.4 (64.3 to 72.2) 83.3 (77.6 to 87.7) <0.001 Low physical activity 35.9 (32.7 to 39.3) 0 23.8 (20.4 to 27.6) 80.4 (74.5 to 85.2) <0.001 Slow gait speed 24.4 (21.6 to 27.4) 0 9.1 (7.0 to 11.9) 72.2 (65.8 to 77.9) <0.001 Weak handgrip strength 43.6 (40.3 to 47.1) 0 35.8 (31.8 to 40.0) 80.4 (74.5 to 85.2) <0.001 Pulmonary rehabilitation Started (% of referred) 84.7 (82.0 to 87.0) 80.4 (94.1) 87.4 (84.3 to 90.0) 76.1 (69.9 to 81.4) <0.001 Number of sessions attended 11.4 (4.2) 13.2 (3.0) 11.6 (4.1) 10.2 (4.7) <0.001 Completed (% of starters) 83.1 (80.1 to 85.7) 94.5 (86.7 to 97.8) 82.8 (79.1 to 86.0) 72.3 (64.9 to 78.7) <0.001 Completed (% of referred) 70.3 (67.1 to 73.4) 84.1 (74.7 to 90.5) 74.3 (70.4 to 77.8) 55.0 (48.2 to 61.6) <0.001 Values are mean (SD) or proportions (95% CI) unless stated. *Statistically different to non-frail group. †Statistically different to prefrail group. 4MGS, 4-m gait speed; ADO, age, dyspnoea, and airflow obstruction; BMI, body mass index; CAT, COPD Assessment Test; CRQ, Chronic Respiratory Disease Questionnaire; HADS, Hospital Anxiety and Depression scale; ISWT, incremental shuttle walk test; kcal, kilocalorie; MRC, Medical Research Council; QMVC, quadriceps maximum voluntary contraction; SMI, skeletal muscle mass index. Figure 1 Profile showing recruitment, frailty status and flow of patients through the trial with reasons for non-uptake or non-completion of pulmonary rehabilitation.

4MGS, 4-m gait speed; ADO, age, dyspnoea, and airflow obstruction; BMI, body mass index; CAT, COPD Assessment Test; CRQ, Chronic Respiratory Disease Questionnaire; HADS, Hospital Anxiety and Depression scale; ISWT, incremental shuttle walk test; kcal, kilocalorie; MRC, Medical Research Council; QMVC, quadriceps maximum voluntary contraction; SMI, skeletal muscle mass index. Figure 1 Profile showing recruitment, frailty status and flow of patients through the trial with reasons for non-uptake or non-completion of pulmonary rehabilitation. Figure 2 Prevalence of frailty in COPD according to age (A), GOLD spirometric stage (B), Medical Research Council (MRC) Dyspnoea score (C) and co-morbidity burden (D) (n=816). Between-group differences (p<0.01) compared with base group (far left) denoted by asterisk.

Figure 1 Profile showing recruitment, frailty status and flow of patients through the trial with reasons for non-uptake or non-completion of pulmonary rehabilitation. Figure 2 Prevalence of frailty in COPD according to age (A), GOLD spirometric stage (B), Medical Research Council (MRC) Dyspnoea score (C) and co-morbidity burden (D) (n=816). Between-group differences (p<0.01) compared with base group (far left) denoted by asterisk. The COPD frailty phenotype Patients who were frail with COPD had significantly reduced SMI but not body mass index as compared with prefrail or robust patients (table 1). Reduced physical function was evident beyond the characteristics used to define frailty, with reduced QMVC and ISWT performance, and a greater prevalence of sarcopenia (23.9% vs 9.5% in the prefrail group and 1.2% in the robust group, p<0.001). Patients who were frail reported a worse health status across all instruments and domains, and higher levels of anxiety and depression as compared with patients who were not-frail and prefrail (table 1). Among patients with QMVC measurements (n=707), almost three-quarters (73.1%) of patients who were frail had concurrent quadriceps weakness,20 while only 25.0% had concurrent sarcopenia. Similar proportions of patients who were frail had both (14.4%) or neither (16.3%) of these phenotypes.

-frail and prefrail (table 1). Among patients with QMVC measurements (n=707), almost three-quarters (73.1%) of patients who were frail had concurrent quadriceps weakness,20 while only 25.0% had concurrent sarcopenia. Similar proportions of patients who were frail had both (14.4%) or neither (16.3%) of these phenotypes. Engagement in pulmonary rehabilitation Overall rates of programme uptake and completion for the cohort were 84.7% (95% CI 82.0% to 87.0%) and 70.3% (95% CI 67.1% to 73.4%), respectively, and mean (SD) attendance was 11.4 (4.2) of 16 supervised sessions (table 1). Rates were lowest in the frail group such that 55.0% of candidates completed rehabilitation, as compared with 74.5% of prefrail and 84.1% of not frail/robust candidates (p<0.001). In the univariate regression age, MRC score, FEV1% predicted, Charlson Index, ISWT, QMVC, ADO, CAT score, HADS anxiety and depression, and frailty status were associated with non-completion. Each of the frailty characteristics was individually associated with non-completion though the relationship was strongest when frailty status was considered overall (table 2). In the multivariate regression, frailty status, age, ISWT and CAT score were retained in the final model. Frailty was a strong independent predictor and being frail was associated with double the odds of non-completion; adjusted OR 2.20 (95% CI 1.39% to 3.46%), p=0.001 (table 2). When examining reasons for not taking up or completing a programme, proportionally more frail patients experienced deterioration in their condition or were admitted to hospital (figure 1). Among patients who were frail, baseline MRC score was higher in non-completers as compared with completers (4.3 vs 3.9, p=0.001) and there was weak evidence of additional functional impairment in non-completers (see online supplementary table S2).

eterioration in their condition or were admitted to hospital (figure 1). Among patients who were frail, baseline MRC score was higher in non-completers as compared with completers (4.3 vs 3.9, p=0.001) and there was weak evidence of additional functional impairment in non-completers (see online supplementary table S2). Table 2 Univariate and multivariate logistic regression for variables associated with non-completion of pulmonary rehabilitation in patients with COPD (n=816)

eterioration in their condition or were admitted to hospital (figure 1). Among patients who were frail, baseline MRC score was higher in non-completers as compared with completers (4.3 vs 3.9, p=0.001) and there was weak evidence of additional functional impairment in non-completers (see online supplementary table S2). Table 2 Univariate and multivariate logistic regression for variables associated with non-completion of pulmonary rehabilitation in patients with COPD (n=816) Univariate OR (95% CI) p Value Age 0.983 (0.967 to 0.998) 0.029 Sex 0.855 (0.628 to 1.164) 0.32 Current smoker (yes/no) 1.000 (1.000 to 1.000) 0.52 MRC 1.619 (1.388 to 1.887) <0.001 FEV1 % predicted 0.995 (0.987 to 1.002) 0.15 GOLD stage 1.000 (1.000 to 1.000) 0.52 Age-adjusted Charlson Index 0.926 (0.849 to 1.019) 0.11 ISWT distance 0.996 (0.995 to 0.998) <0.001 QMVC 0.986 (0.970 to 1.003) 0.12 ADO 1.122 (1.022 to 1.231) 0.015 CAT score 1.059 (1.038 to 1.080) <0.001 HAD anxiety 1.061 (1.027 to 1.096) <0.001 HAD depression 1.093 (1.051 to 1.137) <0.001 Frailty characteristic Unintentional weight loss 1.578 (1.0044 to 2.385) 0.030 Exhaustion 1.615 (1.154 to 2.261) <0.001 Low physical activity 2.334 (1.708 to 3.189) <0.001 Slow gait speed 2.317 (1.655 to 3.244) <0.001 Weak handgrip strength 1.258 (0.927 to 1.709) 0.141 Frailty status (not or pre-frail/frail) 2.699 (1.936 to 3.762) <0.001 Multivariate Adjusted OR (95% CI) p Value Age 0.964 (0.944 to 0.984) 0.001 ISWT distance 0.998 (0.997 to 1.000) 0.026 CAT score 1.024 (0.998 to 1.051) 0.07 Frailty status, yes/no 2.195 (1.392 to 3.463) 0.001 ADO, age, dyspnoea, and airflow obstruction; CAT, COPD Assessment Test; HADS, Hospital Anxiety and Depression scale; ISWT, incremental shuttle walk test; MRC, Medical Research Council; QMVC, quadriceps maximum voluntary contraction.

T score 1.024 (0.998 to 1.051) 0.07 Frailty status, yes/no 2.195 (1.392 to 3.463) 0.001 ADO, age, dyspnoea, and airflow obstruction; CAT, COPD Assessment Test; HADS, Hospital Anxiety and Depression scale; ISWT, incremental shuttle walk test; MRC, Medical Research Council; QMVC, quadriceps maximum voluntary contraction. Outcomes of pulmonary rehabilitation Following completion of rehabilitation, significant improvements were observed across all groups for SMI, QMVC, CRQ dyspnoea, fatigue, emotional and mastery domains, and physical activity (table 3) and in prefrail and frail groups for MRC score, handgrip strength, ISWT, CAT score and HADS domains (table 3). Adjusting for age and sex, a gradient of treatment response in favour of patients who were frail was evident for MRC score, handgrip strength, ISWT, CRQ fatigue, emotional and mastery domains, CAT score and HADS scores (table 3). Outcomes related to frailty characteristics (handgrip strength, 4MGS and physical activity) and responses to CES-D exhaustion questions also improved, such that postrehabilitation, fewer patients met case criteria for frailty (figure 3). Among the 115 completers who were frail prior to pulmonary rehabilitation, 71 (61.7%) were prefrail (64, 55.6%) or robust (7, 6.1%) following it. A small minority of completers, 13/390 (3.3%), had moved from a prefrail to a frail state (figure 3). Table 3 Comparison of clinical outcomes following pulmonary rehabilitation according to frailty status

Outcomes of pulmonary rehabilitation Following completion of rehabilitation, significant improvements were observed across all groups for SMI, QMVC, CRQ dyspnoea, fatigue, emotional and mastery domains, and physical activity (table 3) and in prefrail and frail groups for MRC score, handgrip strength, ISWT, CAT score and HADS domains (table 3). Adjusting for age and sex, a gradient of treatment response in favour of patients who were frail was evident for MRC score, handgrip strength, ISWT, CRQ fatigue, emotional and mastery domains, CAT score and HADS scores (table 3). Outcomes related to frailty characteristics (handgrip strength, 4MGS and physical activity) and responses to CES-D exhaustion questions also improved, such that postrehabilitation, fewer patients met case criteria for frailty (figure 3). Among the 115 completers who were frail prior to pulmonary rehabilitation, 71 (61.7%) were prefrail (64, 55.6%) or robust (7, 6.1%) following it. A small minority of completers, 13/390 (3.3%), had moved from a prefrail to a frail state (figure 3). Table 3 Comparison of clinical outcomes following pulmonary rehabilitation according to frailty status Not frail (n=69) Prefrail (n=390) Frail (n=115) p Value MRC 0.1 (−0.3 to 0.5) −0.5 (−0.7 to −0.4) −1.4 (−1.1 to −1.7)*† <0.001 SMI (kg/m2) 0.6 (0.5 to 1.1) 0.5 (0.2 to 0.7) 0.5 (0.1 to 0.8) 0.90 Handgrip (kg) −0.2 (−1.2 to 0.9) 1.2 (0.8 to 1.5) 1.6 (1.0 to 2.3)* 0.002 Peak QMVC (kg) 2.7 (1.1 to 4.3) 1.9 (1.2 to 2.5) 1.8 (0.8 to 2.7) 0.55 Below QMVC cut-point (%) 6.4 (−1.4 to 14.1) −21.74 (−17.7 to −25.3) −36.6 (−24.8 to −46.9)*† <0.001 4MGS (m/s) 0.08 (0.05 to 0.12) 0.07 (0.05 to 0.08) 0.11 (0.09 to 0.14) 0.004 ISWT (m) 17.8 (−21.7 to 57.3) 51.8 (24.4 to 79.2) 145.9 (108.6 to 183.2)*† <0.001 CRQ dyspnoea score 3.8 (1.4 to 6.2) 4.4 (3.3 to 5.4) 6.8 (5.0 to 8.5) 0.006 CRQ fatigue score −0.8 (−3.2 to 1.5) 3.1 (2.1 to 4.0) 6.1 (4.6 to 7.7)*† <0.001 CRQ emotional score −0.5 (−2.6 to 3.7) 4.0 (2.4 to 5.6) 8.6 (5.6 to 11.5)*† <0.001 CRQ mastery score 0.7 (−1.1 to 2.4) 3.1 (2.1 to 4.1) 5.2 (3.4 to 6.9)*† <0.001 Self-reported weekly energy expenditure (kcal) 1276.0 (714.1 to 1838.0) 606.2 (390.0 to 822.5) 767.1 (546.4 to 987.8) 0.08 Self-reported time in moderate activity (min/week) 417.5 (184.7 to 650.4) 137.0 (75.2 to 198.9) 190.3 (127.4 to 253.3) 0.006 CAT score 0.4 (−1.4 to 2.1) −1.3 (−2.7 to 0.2) −7.3 (−9.7 to −4.8)*† <0.001 Katz score 0.0 (−0.1 to 0.1) 0.0 (−0.1 to 0.1) 0.1 (−0.1 to 0.3) 0.73 HADS anxiety −0.3 (−2.0 to 1.4) −1.0 (−1.7 to −0.3) −2.8 (−4.4 to −1.2)* <0.001 HADS depression 0.9 (−0.2 to 2.1) −0.8 (−1.4 to −0.1) −2.9 (−4.0 to −1.7)*† <0.001 Values are mean change (95% CI) pre-to-post rehabilitation.

7.3 (−9.7 to −4.8)*† <0.001 Katz score 0.0 (−0.1 to 0.1) 0.0 (−0.1 to 0.1) 0.1 (−0.1 to 0.3) 0.73 HADS anxiety −0.3 (−2.0 to 1.4) −1.0 (−1.7 to −0.3) −2.8 (−4.4 to −1.2)* <0.001 HADS depression 0.9 (−0.2 to 2.1) −0.8 (−1.4 to −0.1) −2.9 (−4.0 to −1.7)*† <0.001 Values are mean change (95% CI) pre-to-post rehabilitation. *Statistically different to non-frail group. †Statistically different to prefrail group—tested if ANCOVA p value <0.003 following Bonferroni adjustment for multiple testing. 4MGS, 4-m gait speed; ANCOVA, analysis of covariance; BMI, body mass index; CAT, COPD Assessment Test; CRQ, Chronic Respiratory Disease Questionnaire; HADS, Hospital Anxiety and Depression scale; ISWT, incremental shuttle walk test; kcal, kilocalorie; MRC, Medical Research Council; QMVC, quadriceps maximum voluntary contraction; SMI, skeletal muscle mass index. Figure 3 Patients with COPD grouped according to Fried's frailty criteria before and after pulmonary rehabilitation (n=574). Overall, rehabilitation led to a shift away from physical frailty towards a more robust state.

4MGS, 4-m gait speed; ANCOVA, analysis of covariance; BMI, body mass index; CAT, COPD Assessment Test; CRQ, Chronic Respiratory Disease Questionnaire; HADS, Hospital Anxiety and Depression scale; ISWT, incremental shuttle walk test; kcal, kilocalorie; MRC, Medical Research Council; QMVC, quadriceps maximum voluntary contraction; SMI, skeletal muscle mass index. Figure 3 Patients with COPD grouped according to Fried's frailty criteria before and after pulmonary rehabilitation (n=574). Overall, rehabilitation led to a shift away from physical frailty towards a more robust state. Discussion In this prospective cohort study, we identified that one-quarter (25.6%) of patients with stable COPD referred for pulmonary rehabilitation were frail according to the Fried phenotype criteria. Patients who were frail demonstrated high levels of impairment as compared with prefrail or robust patients, one consequence of which was more difficulty engaging in pulmonary rehabilitation as a mainstay of their disease management. Adjusting for all known confounders, being frail was associated with over double the odds of programme non-completion. Nonetheless, in those who completed rehabilitation there was a gradient of treatment response, in favour of patients who were frail, for outcomes relating to symptom burden, exercise performance and health status. Furthermore, in a large proportion of those who completed pulmonary rehabilitation, the frailty phenotype was reversed, at least in the short term.

ilitation there was a gradient of treatment response, in favour of patients who were frail, for outcomes relating to symptom burden, exercise performance and health status. Furthermore, in a large proportion of those who completed pulmonary rehabilitation, the frailty phenotype was reversed, at least in the short term. Our point prevalence estimate for frailty is within the range previously reported in COPD (10.2% to 28.0%),5 7 37 and over twice the 9.9% (95% CI 9.6% to 10.2%) observed among older people living in the community.2 Prevalence in our cohort, with a mean age of 70 years, is similar to that found among those aged 85 and above in the general population.2 The only estimate to suggest frailty prevalence is not increased in respiratory disease arose from a retrospective analysis of an older-persons cohort and included patients with very mild spirometric disease (mean FEV1 79.6 (25.2) % predicted) and minimal evidence of functional impairment (only 2.5% of patients with COPD had a slow gait speed).5 While frailty was more common in patients with spirometrically advanced disease, prevalence among patients in the GOLD I category was almost 20% emphasising the importance of multidimensional assessment, even in early disease.

l evidence of functional impairment (only 2.5% of patients with COPD had a slow gait speed).5 While frailty was more common in patients with spirometrically advanced disease, prevalence among patients in the GOLD I category was almost 20% emphasising the importance of multidimensional assessment, even in early disease. The detailed phenotypic data highlight the extent to which frailty adversely impacts patients with chronic respiratory disease. Others have associated frailty with poor exercise capacity and physical disability.7 38 We extend these findings by demonstrating that frailty relates to reduced physical function (lower limb muscle strength, exercise capacity, physical activity), dyspnoea, dependency, increased anxiety and depression, and worse emotional distress and health status. These deficits have also been linked to sarcopenia,15 which is considered a component of frailty.39 Frailty is more multifaceted and can occur without skeletal muscle dysfunction, for example, due to ventilatory impairment, as confirmed by the only partial overlap we observed between these two syndromes.40

alth status. These deficits have also been linked to sarcopenia,15 which is considered a component of frailty.39 Frailty is more multifaceted and can occur without skeletal muscle dysfunction, for example, due to ventilatory impairment, as confirmed by the only partial overlap we observed between these two syndromes.40 Irrespective of cause, frailty denotes an increased state of risk related to falls, disability, hospitalisation and mortality.3 The prognostic utility of frailty in chronic respiratory disease for mortality and hospitalisation is supported by earlier studies,5 7 41 and its adverse impact on the clinical management of patients is emerging.7 Here, we highlight a new example of frailty altering COPD management in that it prevented patients from fully engaging in pulmonary rehabilitation, an internationally recommended standard of care.11 Patients who were frail experienced frequent episodes of clinical deterioration and/or hospitalisation, which restricted their participation in rehabilitation. Although previous studies have identified risk factors for pulmonary rehabilitation non-completion, for example, smoking status, breathlessness or low mood,17 18 frailty was comparatively a far stronger independent predictor in our cohort.

and/or hospitalisation, which restricted their participation in rehabilitation. Although previous studies have identified risk factors for pulmonary rehabilitation non-completion, for example, smoking status, breathlessness or low mood,17 18 frailty was comparatively a far stronger independent predictor in our cohort. The impressive outcomes seen following programme completion provide strong grounds to explore how better to support patients who are frail through rehabilitation, potentially through organisation changes or by supplementing supervised exercise with novel strategies, for example, neuromuscular electrical stimulation.42 Improvements across physical, psychological and global health were observed after rehabilitation, often with an apparent gradient of treatment response in favour of patients who were frail (table 2). The magnitude of effect among patients who were frail is noteworthy and many improvements far exceeded minimum important differences (MID); mean change (95% CI) ISWT 146 m (109 to 183)/MID 47.5 m43 and CAT score −7.3 (−9.7 to −4.8) MID −2.0.34 This was reflected in the shift away from frailty after rehabilitation shown in figure 3. In part, this reflects the working of the Fried phenotype model, which uses cut-points to define frailty, therefore in some patients a subtle improvement would ‘declassify’ their frail state. Nonetheless, our data validate the contemporary view of frailty as a condition that is amenable to treatment.3 39 Treatments with evidence of efficacy in frailty management include exercise, nutritional support, self-management strategies and reduction of polypharmacy.39 Many of these components have already been operationalised to be delivered as pulmonary rehabilitation, which raises the idea the model could be adapted to support frail people outside of the respiratory specialty. Indeed, tailored frailty programmes for older adults are being piloted within geriatric medicine—the aims, structure and content of which are similar to pulmonary rehabilitation.44 We believe there is an opportunity to share learning, skills and component interventions to benefit patients across settings.

ry specialty. Indeed, tailored frailty programmes for older adults are being piloted within geriatric medicine—the aims, structure and content of which are similar to pulmonary rehabilitation.44 We believe there is an opportunity to share learning, skills and component interventions to benefit patients across settings. There are limitations to consider. We purposefully selected the Fried model which mainly reflects physical frailty and incorporates measures that are effort dependant or rely on patient recall. However, the model has proven construct and predictive validity and is the most established criterion measure of frailty.3 7 9 Other frailty assessments may capture the broader experience of the syndrome, for example, cognitive, social or environmental.2 We only enrolled patients who attended for pulmonary rehabilitation assessment, and there will likely be patients with ‘hidden’ frailty who were referred but unable to attend due to poor mobility or cognition. Our prevalence estimate does not take this subgroup into account, but this may be counterbalanced by the omission of asymptomatic patients with COPD, who may not have been referred to hospital outpatient or pulmonary rehabilitation clinics. We did not obtain outcomes on participants declining or dropping out of rehabilitation; therefore, our findings concerning clinical response to rehabilitation should not be generalised beyond those completing a programme. As patients who were frail had the greatest levels of impairment at baseline, regression to the mean may partially accounts for the preferential response. However, this bias is likely to be small and withholding rehabilitation from a control group could be considered unethical. The differences in some outcomes between patients who were frail and not-frail were, in part, a product of the poor response for the 10% of the cohort considered robust. This was most notable for exercise capacity and health-related quality of life, which may reflect that this fitter group had better preserved exercise capacity and health status at enrolment, and therefore had alternative targets for pulmonary rehabilitation. An example might be behavioural outcomes such as daily physical activity, which increased significantly in this group and to a greater extent to the prefrail and frail groups. Finally, outcomes were obtained immediately following rehabilitation and may reflect a transient change in frailty status.

onary rehabilitation. An example might be behavioural outcomes such as daily physical activity, which increased significantly in this group and to a greater extent to the prefrail and frail groups. Finally, outcomes were obtained immediately following rehabilitation and may reflect a transient change in frailty status. The value of these changes in COPD with respect to long-term outcomes, for example, admissions and mortality, will emerge in due course. In conclusion, frailty affects one-quarter of patients with stable COPD assessed for pulmonary rehabilitation. Frailty is an independent risk factor for programme non-completion but appears to result in favourable rehabilitation outcomes. Future research should identify how best to support patients who are frail through pulmonary rehabilitation. The authors acknowledge the Pulmonary Rehabilitation Unit at Harefield Hospital for their assistance in collecting data for this manuscript. Contributors: MM and WD-CM contributed to the conception and design of the study. SSCK, JLC, SEJ and CMN recruited patients and conducted the study. MM, SSCK, AL, MIP and WD-CM contributed to data analysis and interpretation. All authors contributed to data interpretation and drafting the manuscript for important intellectual content.

and WD-CM contributed to the conception and design of the study. SSCK, JLC, SEJ and CMN recruited patients and conducted the study. MM, SSCK, AL, MIP and WD-CM contributed to data analysis and interpretation. All authors contributed to data interpretation and drafting the manuscript for important intellectual content. Funding: This work was supported by the NIHR Respiratory Biomedical Research Unit at the Royal Brompton and Harefield NHS Foundation Trust and Imperial College, London, UK, who partly fund MIP's salary. MM is supported by the NIHR Collaboration for Leadership in Applied Health Research and Care (CLAHRC) for South London and by Cicely Saunders International. CMN is supported by a NIHR Doctoral Fellowship. WD-CM was supported by a NIHR Clinician Scientist Award, Medical Research Council (UK) New Investigator Research Grant, NIHR Clinical Trials Fellowship and by the NIHR CLAHRC for Northwest London. Competing interests: MIP has received personal reimbursement for lecturing or consultancy regarding muscle function in COPD from Novartis and Philips Respironics. He discloses institutional reimbursement for consultancy from GSK, Novartis, Regneron, Lilly, Biomarin and BI and institutional agreements to conduct research with GSK, Novartis, AZ and Philips Respironics. Ethics approval: West London (11/H0707/2) and London Camberwell St Giles (11/LO/1780) Research Ethics Committees. Provenance and peer review: Not commissioned; externally peer reviewed.

Key messages What is the key question? What is the role of laryngeal narrowing in the regulation of ventilation during exercise in patients with COPD? What is the bottom line? Dynamic laryngeal narrowing during expiration is prevalent in subjects with moderate or severe COPD and relates to lung emptying and dynamic hyperinflation during exercise, suggesting this could be a newly identified mechanism by which patients offset intrinsic positive end-expiratory pressure. Why read on? This is the first study to describe the active role of the larynx in regulating lung emptying and ventilation during exercise in COPD and highlights potential implications for exercise intolerance. Introduction Dynamic hyperinflation (DH) and in particular the loss of inspiratory reserve volume are recognised features in the aetiology of the dyspnoea and exercise limitation in COPD.1–3 Strategies which reduce DH, including drugs4 and lung volume reduction5 improve exercise performance in COPD. Pursed-lip breathing is a strategy which patients can employ to create intrinsic positive end-expiratory pressure (PEEPi).6 Although the mechanics of this manoeuvre have not been fully studied, pursed lip breathing at rest is associated with a prolonged expiratory time, an increased tidal volume (VT) and reduced end expiratory chest wall volumes, presumably because the creation of PEEPi allows greater expiratory flow.7 The adoption of a pursed lip breathing strategy may also be associated with improved exercise tolerance.8

athing at rest is associated with a prolonged expiratory time, an increased tidal volume (VT) and reduced end expiratory chest wall volumes, presumably because the creation of PEEPi allows greater expiratory flow.7 The adoption of a pursed lip breathing strategy may also be associated with improved exercise tolerance.8 In health, the functional neural control of the larynx and respiratory musculature are closely aligned, so that diaphragmatic contraction and laryngeal abduction occur in virtual synchrony. Nevertheless, even in healthy subjects, there is a minor reduction in glottic aperture during passive expiration9–11 and a tight coupling between the movement of the vocal cords and the phase and pattern of volume change within the respiratory cycle.9 Little is known about the behaviour of the larynx in patients with COPD; Higenbottam and Payne12 studied 34 patients with obstructive lung disease and observed expiratory narrowing of the glottis at rest, while Campbell et al13 hypothesised a similar mechanism based on measures of upper airway resistance. The laryngeal response to physical exertion in patients with COPD has never been studied. However, the occurrence of laryngeal narrowing during the expiratory phase of respiration may be relevant and, by increasing PEEPi, serves to ‘splint’ the airway12 14 15; this may be relevant in the presence or absence of pursed lip breathing, that is, glottic narrowing is a possible additional mechanism acting to optimise operational lung mechanics in COPD.

rowing during the expiratory phase of respiration may be relevant and, by increasing PEEPi, serves to ‘splint’ the airway12 14 15; this may be relevant in the presence or absence of pursed lip breathing, that is, glottic narrowing is a possible additional mechanism acting to optimise operational lung mechanics in COPD. Technical advances now permit accurate continuous recording of laryngeal movement during exercise. More specifically, a technique termed the continuous laryngoscopy during exercise (CLE) test provides a feasible and safe method with which to obtain continuous acquisition of laryngeal movement from a nasendoscope view in a fixed position.16 This technique allows a comparative assessment of dynamic laryngeal movement in response to exposure to physiological stress, for example, exercise. The aim of this study was to evaluate dynamic movements of the larynx in patients with COPD and to evaluate relationships with static and dynamic lung function measures and exercise tolerance using a CLE technique. We hypothesised that laryngeal narrowing, during the expiratory phase of respiration, would be more prevalent in patients with COPD than healthy age-matched controls and relate to airflow obstruction. We also hypothesised that this laryngeal narrowing would occur predominantly at the vocal cord level and become more pronounced during exercise, relating to indices of DH.

expiratory phase of respiration, would be more prevalent in patients with COPD than healthy age-matched controls and relate to airflow obstruction. We also hypothesised that this laryngeal narrowing would occur predominantly at the vocal cord level and become more pronounced during exercise, relating to indices of DH. Methods Study population Subjects with confirmed COPD on the basis of a positive smoking history, symptoms and lung function17 were recruited from the Royal Brompton Hospital at a time of clinical stability; that is, no exacerbation in the month prior to study. Subjects with a present or past history of cardiac disease, laryngeal disease or use of maintenance oxygen or ventilatory support were excluded. In addition, healthy age-matched volunteers, free from significant respiratory (FEV1>80% predicted), cardiovascular or metabolic disease were recruited from a departmental database. Subjects provided written informed consent and the study was approved by the Local Research Ethics Committee (11/LO/1404). Study design and procedures Subjects attended the laboratory on one occasion for clinical assessment including COPD assessment test,18 lung function measures and a CLE test. Anthropometric characteristics were recorded and spirometry was performed on the day in all patients and controls in accordance with recommendations19 (Microlab 3300 Spirometer, Micro Medical, Rochester, UK). Static lung volumes and gas transfer measures were used if they had been taken within 1 year of the assessment date.

opometric characteristics were recorded and spirometry was performed on the day in all patients and controls in accordance with recommendations19 (Microlab 3300 Spirometer, Micro Medical, Rochester, UK). Static lung volumes and gas transfer measures were used if they had been taken within 1 year of the assessment date. Exercise procedures: laryngoscopy during exercise Continuous laryngoscopy testing was performed based on previous methodology.16 In brief, the larynx was visualised by placing a fibreoptic nasenodoscope (Olympus ENF-VQ, Olympus, Japan) in the posterior nasopharynx and securing it using specialist headgear (see figure E1 in the online data supplement). Video images were thereafter continuously recorded (MediCap USB 200, Medicapture, USA). Subjects then performed an incremental maximal cycle-ergometer cardiopulmonary exercise test with cardiorespiratory measurement according to accepted definitions.20 Operational lung volumes were determined from inspiratory capacity (IC) manoeuvres performed every 2 min, to calculate the dynamic hyperinflation index (DHI).21

cts then performed an incremental maximal cycle-ergometer cardiopulmonary exercise test with cardiorespiratory measurement according to accepted definitions.20 Operational lung volumes were determined from inspiratory capacity (IC) manoeuvres performed every 2 min, to calculate the dynamic hyperinflation index (DHI).21 Laryngeal movement analysis Image stills were taken from the continuous recording, at end inspiration and end expiration. Breathing cycle was verified by corresponding pneumotachograph flow data. An expiratory narrowing ratio was calculated for the glottic (distance between vocal folds at the mid-point of their length, glottic narrowing ratio, GNR) and supra-glottic apertures (distance between the medial margin of the arytaenoid cartilage, SGNR) as equal to 1-aperture measurements at end expiration divided by the width at end inspiration (figure 1); whereby a narrowing ratio of 1=complete narrowing and 0=no change. This ratio provides a measure of the narrowing of the laryngeal structures in relative terms during one breathing cycle, which is independent of the distance between the laryngoscope and the glottis. Peak exercise images were taken from a breath sequence considered representative of the last 30 s of exercise and free of artefact (eg, coughing). Figure 1 Example cases illustrating dynamic expiratory laryngeal narrowing and method for calculation of glottic and supra-glottic narrowing ratio. Narrowing scores were determined as the mean of scores reported independently by two observers (AMG and GH), blinded to subject characteristics.

Laryngeal movement analysis Image stills were taken from the continuous recording, at end inspiration and end expiration. Breathing cycle was verified by corresponding pneumotachograph flow data. An expiratory narrowing ratio was calculated for the glottic (distance between vocal folds at the mid-point of their length, glottic narrowing ratio, GNR) and supra-glottic apertures (distance between the medial margin of the arytaenoid cartilage, SGNR) as equal to 1-aperture measurements at end expiration divided by the width at end inspiration (figure 1); whereby a narrowing ratio of 1=complete narrowing and 0=no change. This ratio provides a measure of the narrowing of the laryngeal structures in relative terms during one breathing cycle, which is independent of the distance between the laryngoscope and the glottis. Peak exercise images were taken from a breath sequence considered representative of the last 30 s of exercise and free of artefact (eg, coughing). Figure 1 Example cases illustrating dynamic expiratory laryngeal narrowing and method for calculation of glottic and supra-glottic narrowing ratio. Narrowing scores were determined as the mean of scores reported independently by two observers (AMG and GH), blinded to subject characteristics. Statistical analysis Data are expressed as mean (SEM) unless otherwise stated. Differences in continuous variables were assessed using Student's t test and/or one-way analysis of variance with Tukey post hoc analysis or the non-parametric equivalent. Groups were analysed as healthy, mild–moderate COPD (Global Initiative for Chronic Obstructive Lung Disease (GOLD) 1–2) and severe COPD (GOLD 3–4). Correlation was assessed by Pearson or Spearman's rank methods. Statistical calculations were made with GraphPad Prism V.5.0 (GraphPad Software, San Diego, California, USA) and SPSS V.21 (SPSS Inc., Chicago, Illinois, USA). Significance is reported at p<0.05.

tructive Lung Disease (GOLD) 1–2) and severe COPD (GOLD 3–4). Correlation was assessed by Pearson or Spearman's rank methods. Statistical calculations were made with GraphPad Prism V.5.0 (GraphPad Software, San Diego, California, USA) and SPSS V.21 (SPSS Inc., Chicago, Illinois, USA). Significance is reported at p<0.05. Results Subject characteristics and resting measurements Thirty-four subjects were recruited, however technically adequate data were only obtained for 30 subjects (n=11 healthy; n=8 mild–moderate COPD; n=11 severe COPD) (table 1). Reasons for inadequate data were poor visualisation secondary to secretions (n=2) and inability to complete CLE (n=2). All patients were prescribed regular inhaled β-2 agonist therapy and 79% were prescribed long-acting β agonist/inhaled corticosteroid and long-acting muscarinic antagonist maintenance therapy. Patients had a median (range) 35 (15–60) pack-year smoking history. Table 1 Subject characteristics and pulmonary function

Results Subject characteristics and resting measurements Thirty-four subjects were recruited, however technically adequate data were only obtained for 30 subjects (n=11 healthy; n=8 mild–moderate COPD; n=11 severe COPD) (table 1). Reasons for inadequate data were poor visualisation secondary to secretions (n=2) and inability to complete CLE (n=2). All patients were prescribed regular inhaled β-2 agonist therapy and 79% were prescribed long-acting β agonist/inhaled corticosteroid and long-acting muscarinic antagonist maintenance therapy. Patients had a median (range) 35 (15–60) pack-year smoking history. Table 1 Subject characteristics and pulmonary function Healthy (n=11) Mild–mod COPD (n=8) Severe COPD (n=11) Age (years) 60 (1) 61 (2) 60 (3) Sex (M:F) 6:5 3:5 6:5 Height (cm) 172 (3) 166 (3) 168 (3) Weight (kg) 75 (5) 84 (6) 69 (3) BMI (kg/m2) 25 (1) 30± (2) 24 (1)† CAT (of/40) 4 (0–9) 12 (4–24)* 21 (11–31)**† Pulmonary function FEV1 (L) 3.1 (0.3) 1.9 (0.2)* 0.9 (0.1)**† FEV1 (%pred) 104 (5) 72 (3)** 32 (2)**†† FVC (L) 4.2 (0.3) 3.1 (0.4) 2.7 (0.2)* FVC (%pred) 114 (4) 96 (4)* 79 (5)** FEV1/FVC 74 (2) 62 (3)* 32 (2)**†† TLC (L) 6.9 (0.5) 6.7 (0.8) 7.8 (0.3) TLC (%pred) 110 (3) 113 (5) 138 (3)**† RV (L) 2.4 (0.2) 3.0 (0.3) 4.4 (0.3)**† RV (%pred) 109 (5) 137 (13) 203 (7.5)**†† RV/TLC (%) 35 (2) 46 (4) 56 (3)** TLCOc (%pred) 98 (6) 74 (5)* 43 (4)**† Data shown as mean (SEM) or median (range). *p<0.05, **p<0.01 from healthy, †p<0.05, ††p<0.01 from mild–mod COPD. Note, lung volume and gas transfer data available in 21 subjects.

2) 3.0 (0.3) 4.4 (0.3)**† RV (%pred) 109 (5) 137 (13) 203 (7.5)**†† RV/TLC (%) 35 (2) 46 (4) 56 (3)** TLCOc (%pred) 98 (6) 74 (5)* 43 (4)**† Data shown as mean (SEM) or median (range). *p<0.05, **p<0.01 from healthy, †p<0.05, ††p<0.01 from mild–mod COPD. Note, lung volume and gas transfer data available in 21 subjects. BMI, body mass index; CAT, COPD assessment tool; RV, residual volume; TLC, total lung capacity; TLCOc, transfer coefficient for carbon monoxide corrected for haemoglobin. Patients and controls were well matched for age, sex, height and body mass index (BMI, all p>0.05). However, as expected, patients reported a heightened symptom burden; median (range) CAT score in healthy subjects, 4 (0–9) and patients 19 (4–31) (p<0.001). They also had impaired pulmonary function (p=0.0086) and baseline oxygen saturation (p=0.013) compared with control subjects (table 1). Exercise measurements As expected, patients demonstrated a reduced peak work rate and oxygen uptake compared with healthy controls: 60±8 W and 13.7±0.8 mL/min/kg and 133±15 W and 24.6±1.8 mL/kg/min respectively (both p<0.001). The majority of patients (79%) stopped exercise because of dyspnoea and exhibited evidence of ventilatory limitation to exercise capacity.

emonstrated a reduced peak work rate and oxygen uptake compared with healthy controls: 60±8 W and 13.7±0.8 mL/min/kg and 133±15 W and 24.6±1.8 mL/kg/min respectively (both p<0.001). The majority of patients (79%) stopped exercise because of dyspnoea and exhibited evidence of ventilatory limitation to exercise capacity. IC manoeuvres during exercise were not satisfactory in two patients; the remainder demonstrated evidence of DH as indicated by a mean (SEM) reduction in IC of COPD group +0.35±0.07 L (p<0.01 compared with controls) and DHI, healthy −0.01±0.06 compared with all patients with COPD 0.75±0.18 (p<0.01) (table 2). DHI was directly related to FEV1 for all subjects (r=−0.46, p=0.024). Table 2 Selected parameters at peak exercise

IC manoeuvres during exercise were not satisfactory in two patients; the remainder demonstrated evidence of DH as indicated by a mean (SEM) reduction in IC of COPD group +0.35±0.07 L (p<0.01 compared with controls) and DHI, healthy −0.01±0.06 compared with all patients with COPD 0.75±0.18 (p<0.01) (table 2). DHI was directly related to FEV1 for all subjects (r=−0.46, p=0.024). Table 2 Selected parameters at peak exercise Healthy (n=11) Mild–mod COPD (n=8) Severe COPD (n=11) Stopped 2° breathing; n (%) 2 (18) 5 (63) 10 (91) Dyspnoea Borg (×/10) 3 (1–5) 4 (2–5) 4 (3–5) Load (W) 133 (15) 82 (13)* 44 (8)** VO2 (L/min) 1.85 (0.19) 1.31 (0.14)* 0.83 (0.15)** VO2 (mL/min/kg) 24.6 (1.8) 15.9 (1.7)** 12.0 (0.4)** VO2% predicted 100 (6) 77 (5)* 49 (4)**†† HR (b/m) 147 (7) 131 (7) 107 (4)**† HR reserve (%) 8 (5) 18 (4) 33 (3)**† VE (L/min) 66 (10) 46 (6) 28 (2)** BR (%) 49 (5) 41 (3) 16 (6)**†† SpO2 (%) 98 (0) 98 (1) 94 (1)**†† VT (L) 2.15 (0.24) 1.56 (0.22) 1.11 (0.08)** TI/TTot 0.45 (0.07) 0.41 (0.02)* 0.36 (0.01)**† IC (L) 3.44 (0.27) 2.52 (0.28)* 1.99 (0.15)** Delta IC (%IC rest) −0.35 (2.90) 12.10 (3.54) 15.26 (4.08)** DHI −0.01 (0.06) 0.46 (0.18) 0.96 (0.27)** Data shown as mean (SEM) or median (range). *p<0.05, **p<0.01 from healthy, †p<0.05, ††p<0.01 from mild–mod COPD. Operational lung volume data available in 7 patients with mild–moderate COPD and 10 patients with severe COPD.

(%IC rest) −0.35 (2.90) 12.10 (3.54) 15.26 (4.08)** DHI −0.01 (0.06) 0.46 (0.18) 0.96 (0.27)** Data shown as mean (SEM) or median (range). *p<0.05, **p<0.01 from healthy, †p<0.05, ††p<0.01 from mild–mod COPD. Operational lung volume data available in 7 patients with mild–moderate COPD and 10 patients with severe COPD. BR, breathing reserve; DHI, dynamic hyperinflation index; HR, heart rate; IC, inspiratory capacity; SpO2, peripheral oxygen saturation; TI/TTOT, respiratory duty cycle; VE, minute ventilation; VO2, oxygen consumption; VT, tidal volume. Laryngeal movement analysis Rest In healthy subjects, there was a median (range) percentage reduction in GNR during expiration of 10% (0–52%) (figures 1 and 2 and table 3). This narrowing was more pronounced in patients: 50% (4%–92%) (p<0.001) (table 3). A similar relationship was apparent at the supra-glottic level, although overall narrowing was less marked in all subjects; that is, predominant narrowing was at the glottic level. There was no relationship between either GNR or SGNR and age, height, BMI (all p>0.05). Likewise there was no difference in GNR or SGNR between male and female subjects. Table 3 Laryngeal narrowing ratios at rest and peak exercise

Laryngeal movement analysis Rest In healthy subjects, there was a median (range) percentage reduction in GNR during expiration of 10% (0–52%) (figures 1 and 2 and table 3). This narrowing was more pronounced in patients: 50% (4%–92%) (p<0.001) (table 3). A similar relationship was apparent at the supra-glottic level, although overall narrowing was less marked in all subjects; that is, predominant narrowing was at the glottic level. There was no relationship between either GNR or SGNR and age, height, BMI (all p>0.05). Likewise there was no difference in GNR or SGNR between male and female subjects. Table 3 Laryngeal narrowing ratios at rest and peak exercise Glottic narrowing ratio Supra-glottic narrowing ratio Healthy Mild–mod COPD Severe COPD Healthy Mild–mod COPD Severe COPD Rest 0.14 (0.05) 0.43 (0.09)* 0.63(0.07)** 0.14 (0.03) 0.31 (0.10) 0.36 (0.05)* † Exercise 0.16 (0.08) 0.35 (0.08) 0.64 (0.05)**† 0.10 (0.05) 0.20 (0.06) 0.43 (0.07)**† Change 0.02 (0.09) −0.09 (0.05) −0.01 (0.07) −0.04 (0.07) −0.11 (0.08) −0.07 (0.07) Data shown as mean (SEM). *p<0.05, **p<0.01 from healthy, †p<0.05, ††p<0.01 from mild–mod COPD. Figure 2 Laryngeal narrowing ratio at rest and peak exercise at glottic (A) and supraglottic (B) level. There was a correlation between the resting GNR and the FEV1 in the whole cohort (r=−0.71, p<0.001) and the patients with COPD considered alone (r=−0.53, p=0.018) (figure 3A). In all subjects a correlation was seen between markers of gas trapping and GNR, however this was not seen for patients with COPD alone.

Figure 2 Laryngeal narrowing ratio at rest and peak exercise at glottic (A) and supraglottic (B) level. There was a correlation between the resting GNR and the FEV1 in the whole cohort (r=−0.71, p<0.001) and the patients with COPD considered alone (r=−0.53, p=0.018) (figure 3A). In all subjects a correlation was seen between markers of gas trapping and GNR, however this was not seen for patients with COPD alone. Figure 3 Relationship between FEV1% predicted and glottic narrowing ratio at rest (A) and peak exercise (B), and supra-glottic narrowing ratio at rest (C) and peak exercise (D). In patients, there was no relationship between GNR and SGNR and COPD assessment tool. Moreover, an analysis of a subgroup of patients with excessive resting laryngeal narrowing (ie, >GNR 0.5 based on mean +2 SD of control data) revealed no significant difference in clinical or physiological characteristics (data not shown). A high level of inter-observer agreement for GNR scoring measures (r=0.93, p<0.0001) was observed (see figure E2 in the online data supplement).

esting laryngeal narrowing (ie, >GNR 0.5 based on mean +2 SD of control data) revealed no significant difference in clinical or physiological characteristics (data not shown). A high level of inter-observer agreement for GNR scoring measures (r=0.93, p<0.0001) was observed (see figure E2 in the online data supplement). Exercise Exercise GNR and SGNR data are shown in table 3. There was no change in either GNR or SGNR between the resting and exercise phase in either healthy controls or patients (p>0.05) (table 3). Exercise GNR was inversely related to peak ventilation in all subjects (r=−0.55, p=0.0015) and patients (r=−0.71, p<0.001) and peak exercise VT (r=−0.58, p=0.0062 and r=−0.55, p=0.0076, respectively). Likewise exercise GNR was inversely related to peak VO2 (% predicted) in all subjects (r=−0.65, p<0.001) and patients alone (r=−0.58, p=0.014). A similar relationship was apparent for SGNR. Exercise inspiratory duty cycle related to exercise GNR for all subjects (r=−0.69, p<0.001) and patients (r=−0.62, p<0.001) (figure 4). Change in GNR rest-peak and change in TI/TTOT were not related. Figure 4 Relationship between respiratory duty cycle (TI/TTOT) and glottic narrowing ratio at rest (A) and peak exercise (B).

Exercise inspiratory duty cycle related to exercise GNR for all subjects (r=−0.69, p<0.001) and patients (r=−0.62, p<0.001) (figure 4). Change in GNR rest-peak and change in TI/TTOT were not related. Figure 4 Relationship between respiratory duty cycle (TI/TTOT) and glottic narrowing ratio at rest (A) and peak exercise (B). Patients with COPD and a significant exercise GNR (ie, >0.5) were no more likely to stop secondary to dyspnoea or report a higher peak exercise dyspnoea BORG score (p>0.05). Peak exercise GNR related to DHI if all subjects were considered together (r=0.45, p=0.018) but not if patients with COPD were considered alone (r=0.24, p=0.35) (figure 5). There was no difference in peak exercise GNR or SGNR between patients who demonstrated DH based on reduction in IC. There was no relationship between peak GNR and oxygen saturation and no significant difference in peak GNR between patients who de-saturated, that is, SpO2<94% (n=6). Figure 5 Relationship between peak glottic narrowing ratio (GNR) and dynamic hyperinflation index (DHI) in controls (•), patients with mild–moderate COPD (□) and patients with severe COPD (▪). Data are presented as mean values with error bars representing 1 SEM. *p<0.05 from controls.

Patients with COPD and a significant exercise GNR (ie, >0.5) were no more likely to stop secondary to dyspnoea or report a higher peak exercise dyspnoea BORG score (p>0.05). Peak exercise GNR related to DHI if all subjects were considered together (r=0.45, p=0.018) but not if patients with COPD were considered alone (r=0.24, p=0.35) (figure 5). There was no difference in peak exercise GNR or SGNR between patients who demonstrated DH based on reduction in IC. There was no relationship between peak GNR and oxygen saturation and no significant difference in peak GNR between patients who de-saturated, that is, SpO2<94% (n=6). Figure 5 Relationship between peak glottic narrowing ratio (GNR) and dynamic hyperinflation index (DHI) in controls (•), patients with mild–moderate COPD (□) and patients with severe COPD (▪). Data are presented as mean values with error bars representing 1 SEM. *p<0.05 from controls. Discussion Dynamic laryngeal narrowing during expiration is prevalent in patients with COPD and is directly related to disease severity and exercise capacity. The reduction in laryngeal aperture arises predominantly from narrowing at the glottic level (ie, vocal cord narrowing) and was evident during quiet breathing at rest. During exercise, glottic narrowing relates to indices of DH and interestingly the relationship between GNR and TI/TTOT became closer during exercise suggesting that glottic narrowing exerts functional consequences which may be beneficial if associated with increased PEEPi.

nd was evident during quiet breathing at rest. During exercise, glottic narrowing relates to indices of DH and interestingly the relationship between GNR and TI/TTOT became closer during exercise suggesting that glottic narrowing exerts functional consequences which may be beneficial if associated with increased PEEPi. Critique of the method It has been posited that close neurophysiological coupling between anatomical components of the airway tract and the expiratory muscles functions to optimise airflow and respiratory loading at rest and during exercise.9 22 23 Moreover it has been proposed that, because of its variable resistance, the larynx may aid the post-inspiratory activity of the diaphragm to control end-expiratory lung volume.24 In animal studies the recurrent laryngeal nerve demonstrates impulse traffic during expiration and the laryngeal adductors or the arytaenoid cartilages show heightened expiratory phase activity.22 Of note in patients with COPD, expiratory muscle recruitment is heterogeneous, perhaps explaining observed variance in glottic narrowing.25 26

he recurrent laryngeal nerve demonstrates impulse traffic during expiration and the laryngeal adductors or the arytaenoid cartilages show heightened expiratory phase activity.22 Of note in patients with COPD, expiratory muscle recruitment is heterogeneous, perhaps explaining observed variance in glottic narrowing.25 26 Our observation that glottic narrowing did not worsen with exercise was unexpected. Our patients knew that they were about to exercise and we therefore acknowledge that this observation may also be explained by the fact that there is early neural activation with glottic adaptation prior to exercise onset. However, we suspect more likely that, for patients with severe disease, it is possible that the larynx is providing optimal flow control consistent with balancing the need to expire air whilst simultaneously creating PEEPi even at rest. This concept would also explain why, on exertion, some patients use pursed-lip breathing to provide additional resistance. It would be informative to evaluate laryngeal narrowing indices during steady-state exercise at iso-ventilation and in addition during walking exercise. Laryngoscopic examination without a mouthpiece in situ would also be informative to evaluate the relationship and interaction between any development of pursed lip breathing and glottic narrowing.

aluate laryngeal narrowing indices during steady-state exercise at iso-ventilation and in addition during walking exercise. Laryngoscopic examination without a mouthpiece in situ would also be informative to evaluate the relationship and interaction between any development of pursed lip breathing and glottic narrowing. Previous studies in healthy adults specifically evaluating the influence of lung volume change indicate that glottic aperture increases in line with lung inflation,11 27 28 so we do not think this would explain the current findings. Our patients with severe COPD had evidence of hyperinflation on their static lung function tests; it is unclear what influence this has over glottic function. Blood gas tensions could potentially influence glottic function. England et al29 evaluated the role of altered oxygen and carbon dioxide tensions on laryngeal function in healthy humans. They concluded that hypoxic stimulation of peripheral chemoreceptors precipitated expiratory glottic narrowing and resulted in a relatively high laryngeal airflow resistance. In contrast, hypercapnia was accompanied by a low expiratory laryngeal resistance. Although we did not obtain arterial gas tensions in the current study, oxygen saturation was assessed as part of the cardiopulmonary exercise testing performed and we found no relationship between exercise oxygen saturation and glottic narrowing.

st, hypercapnia was accompanied by a low expiratory laryngeal resistance. Although we did not obtain arterial gas tensions in the current study, oxygen saturation was assessed as part of the cardiopulmonary exercise testing performed and we found no relationship between exercise oxygen saturation and glottic narrowing. It is conceivable that the testing methodology employed (ie, direct laryngoscopy during exercise) may have played a role in precipitating the development of the laryngeal narrowing, although prior studies in which laryngeal resistance has been measured would argue against this, and would not explain observed differences between patients and controls. Moreover, it is possible that use of a flow turbine mouthpiece during exercise testing and the requirement to perform periodic IC manoeuvres influenced laryngeal function and respiratory-laryngeal neural recruitment. To address this consideration, we obtained video recording images in patients with and without the mouthpiece in situ and found no significant difference in the degree or magnitude of glottic narrowing in these individuals, either at rest or near to peak exercise. Moreover, in contrast to prior studies,14 every care was taken to ensure local anaesthetic was only applied to the nares. Our analysis was based on the GNR and SGNR as detailed above. We also undertook analysis expressing the inspiratory and expiratory distances as a percentage reduction in aperture (ie, {(b–a)/a}×100) as opposed to (1–b/a) but our conclusions were unchanged (data not shown).

It is conceivable that the testing methodology employed (ie, direct laryngoscopy during exercise) may have played a role in precipitating the development of the laryngeal narrowing, although prior studies in which laryngeal resistance has been measured would argue against this, and would not explain observed differences between patients and controls. Moreover, it is possible that use of a flow turbine mouthpiece during exercise testing and the requirement to perform periodic IC manoeuvres influenced laryngeal function and respiratory-laryngeal neural recruitment. To address this consideration, we obtained video recording images in patients with and without the mouthpiece in situ and found no significant difference in the degree or magnitude of glottic narrowing in these individuals, either at rest or near to peak exercise. Moreover, in contrast to prior studies,14 every care was taken to ensure local anaesthetic was only applied to the nares. Our analysis was based on the GNR and SGNR as detailed above. We also undertook analysis expressing the inspiratory and expiratory distances as a percentage reduction in aperture (ie, {(b–a)/a}×100) as opposed to (1–b/a) but our conclusions were unchanged (data not shown). Significance of the findings Prior research has evaluated respiratory movement of the glottis, using either direct visualisation techniques or indirectly inferring laryngeal narrowing from alterations in upper airway resistance measures,12 13 30 31 which provide no information regarding the level at which obstruction occurs. Specifically, Higenbottam and Payne12 demonstrated a relationship between spirometric indices of airflow and glottic narrowing using a bronchoscopic technique in patients, but at rest in a semi-recumbent position with laryngeal anaesthetic applied.

provide no information regarding the level at which obstruction occurs. Specifically, Higenbottam and Payne12 demonstrated a relationship between spirometric indices of airflow and glottic narrowing using a bronchoscopic technique in patients, but at rest in a semi-recumbent position with laryngeal anaesthetic applied. Our findings, in a clinically pragmatic setting, specifically the seated position with no sedation or laryngeal anaesthesia, reveal dynamic laryngeal narrowing to be present both at rest and during exercise in a well characterised population of patients with COPD. A direct relationship was found between glottic narrowing and disease severity, as classified by FEV1, lung volumes or gas transfer. Moreover, this is the first study to provide a description of the anatomical location of glottic narrowing. This is pertinent given that in the condition of exercise-induced laryngeal obstruction, it is recognised that narrowing occurs predominantly at the supraglottic level (ie, arytaenoid level).16 In contrast, in our patients, it is apparent that while expiratory narrowing occurs at both the glottic and supraglottic levels, glottic narrowing predominates.

e condition of exercise-induced laryngeal obstruction, it is recognised that narrowing occurs predominantly at the supraglottic level (ie, arytaenoid level).16 In contrast, in our patients, it is apparent that while expiratory narrowing occurs at both the glottic and supraglottic levels, glottic narrowing predominates. The current study is also the first to describe laryngeal performance in patients with COPD during exercise. Exercise hyperpnoea results not only in increased minute ventilation but has sequelae for airway turbulence, pulmonary receptor activation and respiratory muscle loading, as well as upstream neural activation relevant to the control of breathing.32 Several important functional adaptations are recognised to occur in the larynx in the context of exercise notably increased activity of the laryngeal abductor muscles, acting to minimise turbulence and airways resistance.33

ry muscle loading, as well as upstream neural activation relevant to the control of breathing.32 Several important functional adaptations are recognised to occur in the larynx in the context of exercise notably increased activity of the laryngeal abductor muscles, acting to minimise turbulence and airways resistance.33 We hypothesised that in COPD the laryngeal aperture would close further in response to exercise in patients with COPD to counter the development of DH. However, in contrast, we found no significant change in glottic dimensions during exercise and certainly little further reduction in glottic aperture in patients at peak exercise. Despite the absence of a macroscopic change in glottic aperture it is noteworthy that the relationship between GNR and both FEV1% predicted and TI/TTOT is closer during exercise than at rest (figure 4B). We speculate that this represents fine-tuning of the glottic aperture to preserve a balance between optimal expiratory flow and PEEPi. In a prior study in six healthy subjects and three with mild airflow obstruction, Pellegrino et al34 found evidence of flow limitation during tidal breathing. Moreover they observed that the addition of an expiratory threshold load resulted in a reduction in end-expiratory lung volume; likely relating to an impact on the time constraints of lung emptying. The laryngeal narrowing observed in the current study would be consistent with the hypothesis that the glottis acts similarly to modulate loading. In the future we suggest this could be explored by studying the relationship between PEEPi measured using an oesophageal balloon and applied levels of external PEEP during CLE. This is in contrast to findings during hyperpnoea in young healthy subjects in which this provocation is associated with glottic dilatation.10

In the future we suggest this could be explored by studying the relationship between PEEPi measured using an oesophageal balloon and applied levels of external PEEP during CLE. This is in contrast to findings during hyperpnoea in young healthy subjects in which this provocation is associated with glottic dilatation.10 In a model, which in some respects simulates COPD, Brancatisano et al24 previously demonstrated that breathing with an expiratory resistive load also enhanced glottic narrowing during expiration in healthy subjects. The authors commented that the expiratory glottic narrowing in response to expiratory resistive loads was in proportion to the prolongation in expiration (as in our patients) and that this effect appears due to a reflex modulated by change in the rate of lung emptying. The role of expiratory loading on glottic function in COPD is yet to be studied. Non-invasive ventilation has an established role in the treatment of patients with COPD, both in terms of acute and chronic hypercapnic respiratory failure but also to improve exercise tolerance. Most modern devices enforce some expiratory positive airway pressure (EPAP); the present data suggest that the relationship between glottic aperture and the optimal applied EPAP requires further study.

with COPD, both in terms of acute and chronic hypercapnic respiratory failure but also to improve exercise tolerance. Most modern devices enforce some expiratory positive airway pressure (EPAP); the present data suggest that the relationship between glottic aperture and the optimal applied EPAP requires further study. It has been previously commented that glottic narrowing may contribute to reduced indices on effort dependent measures.12 In healthy subjects, glottic widening has been observed during a forced manoeuver,35 indicating that simultaneous recruitment of the expiratory muscles and laryngeal abductors. Forced expiratory manoeuvres are associated with activation of the posterior cricoarytenoid and laryngeal adductors. In patients with COPD, the use of breathing techniques that increase positive expiratory pressure such as ‘pursed-lip’ breathing has been recognised for decades, however the precise benefit of this technique remains debated.36 Garrod et al37 found no significant difference in 6-min walking distance in those employing pursed-lip breathing, but noted that pursed-lip breathing during exercise and recovery resulted in a lower respiratory rate after exercise and results in a quicker resolution of exercise-induced breathlessness compared with exercise without pursed-lip breathing. One would hypothesise that pursed-lip breathing would be less useful in patients with substantial glottic narrowing, however we speculate that variability in pursed-lip breathing may represent heterogeneity of underlying glottic function.

rcise-induced breathlessness compared with exercise without pursed-lip breathing. One would hypothesise that pursed-lip breathing would be less useful in patients with substantial glottic narrowing, however we speculate that variability in pursed-lip breathing may represent heterogeneity of underlying glottic function. Conclusion Dynamic laryngeal narrowing during respiration, to some degree, appears to be almost ubiquitous in subjects with moderate or severe COPD. This expiratory phase narrowing occurs at the glottic level predominantly and changes little between rest and exercise. Further work should focus on establishing the mechanisms underlying this glottic narrowing and in particular its relationship with flow limitation and PEEPi. We believe that understanding laryngeal physiology better will allow the development of new therapies, and permit optimal application of existing therapies, in particular non-invasive ventilation. Supplementary Material Web supplement We wish to thank the Lung Function Department Staff at the Royal Brompton Hospital for their assistance. Contributors: All authors contributed to the design of the study, analysis of data and preparation of the final manuscript. MB, MIP and JHH conceived the idea and act as guarantors of the paper, taking responsibility for the integrity of the work as a whole, from inception to published article.

Supplementary Material Web supplement We wish to thank the Lung Function Department Staff at the Royal Brompton Hospital for their assistance. Contributors: All authors contributed to the design of the study, analysis of data and preparation of the final manuscript. MB, MIP and JHH conceived the idea and act as guarantors of the paper, taking responsibility for the integrity of the work as a whole, from inception to published article. Funding: The work was supported by the NIHR Respiratory Biomedical Research Unit at the Royal Brompton and Harefield NHS Foundation Trust and Imperial College, London UK who part fund the salary of MIP. MB was supported by an ERS Long Term Research Fellowship (2012). Competing interests: None. Ethics approval: NRES London-Fulham Research Ethics Committee. Provenance and peer review: Not commissioned; externally peer reviewed.

Key messages What is the key question? What is the validity, responsiveness and minimum important difference of the EQ-5D-5L, a generic health status questionnaire that is widely used in health economic evaluation, in patients with COPD? What is the bottom line? This is the first study to demonstrate that the EQ-5D-5L utility index and visual analogue score are valid and responsive in stable COPD and provides estimates of the minimum clinically important difference. Why read on? This data will help in the design of clinical intervention trials, particularly with regard to assessment of cost-effectiveness. Introduction The EQ-5D is a simple, generic health-related quality of life (HRQoL) instrument that is self-administered and is widely used as a patient-reported outcome measure. It comprises five health dimensions (mobility, self-care, usual activities, pain/discomfort and anxiety/depression): the most commonly used version of the questionnaire, the EQ-5D-3L, has three levels of severity for each dimension.1 The EQ-5D is widely used in health economic evaluation—a utility index (UI) can be calculated by applying ‘social tariffs’, which are used to estimate health benefits in terms of quality-adjusted life-years (QALYs). It is one of only a few measures recommended for use in cost-effectiveness analyses by the Washington Panel on Cost Effectiveness in Health & Medicine, while the United Kingdom National Institute for Health and Care Excellence has recommended the EQ-5D to be the preferred HRQoL instrument to generate QALYs.2

s (QALYs). It is one of only a few measures recommended for use in cost-effectiveness analyses by the Washington Panel on Cost Effectiveness in Health & Medicine, while the United Kingdom National Institute for Health and Care Excellence has recommended the EQ-5D to be the preferred HRQoL instrument to generate QALYs.2 Other advantages for using generic instruments include the comparison of HRQoL across different diseases, and the potential for capturing aspects of HRQoL that may not be addressed by disease-specific questionnaires. For example, in patients with COPD, the EQ-5D may better reflect side effects of extrapulmonary manifestations such as cardiac comorbidity3 or sarcopenia.4 The EQ-5D-3L is simple and quick to use with high patient completion rates in general and COPD-specific populations,5 6 and has been reported in some trials of patients with COPD.7–9 However, investigators have questioned the ability of the EQ-5D-3L to differentiate small changes in health status, and therefore, it may be less responsive than disease-specific HRQoL questionnaires in COPD.6 10 Furthermore, the EQ-5D is well recognised to have a significant ceiling effect (ie, scores recording perfect health) in both general and disease-specific populations,6 11 leaving less room for improvement in response to an intervention.

, it may be less responsive than disease-specific HRQoL questionnaires in COPD.6 10 Furthermore, the EQ-5D is well recognised to have a significant ceiling effect (ie, scores recording perfect health) in both general and disease-specific populations,6 11 leaving less room for improvement in response to an intervention. To address these issues, the EQ-5D-5L was developed in 201112 with the levels of severity for each dimension increased to a choice of five, thus allowing the description of 3125 different health states, in comparison to the 243 health states possible in the EQ-5D-3L. However, studies examining the psychometric properties of the EQ-5D-5L are limited. Furthermore, previous studies have only estimated the UI of the EQ-5D-5L as a ‘crosswalk’ value by mapping to the EQ-5D-3L.13 Recently, the EQ-5D-5L value set for England, derived from 1000 individuals selected at random from the adult general population of England, was published, thus allowing the UI to be directly calculated.14 In addition to the UI, the EQ-5D-5L (like the 3L) includes a visual analogue scale (EQ-VAS).

ng to the EQ-5D-3L.13 Recently, the EQ-5D-5L value set for England, derived from 1000 individuals selected at random from the adult general population of England, was published, thus allowing the UI to be directly calculated.14 In addition to the UI, the EQ-5D-5L (like the 3L) includes a visual analogue scale (EQ-VAS). The aim of the current study was to assess the validity of the EQ-5D-5L UI and EQ-VAS in a COPD-specific outpatient population by comparing with well-established disease-specific HRQoL questionnaires and other indices of disease severity. The responsiveness of the UI and EQ-VAS was also tested in a separate COPD cohort undergoing pulmonary rehabilitation. Finally, the minimum important difference (MID)—the smallest change in score that patients perceive as beneficial or detrimental—of the UI and EQ-VAS were estimated using a range of anchor-based and distribution-based methods. We hypothesised that (1) the EQ-5D-5L would correlate significantly with COPD-specific HRQoL questionnaires and be able to distinguish different levels of disease severity; (2) the EQ-5D-5L would improve with pulmonary rehabilitation; and (3) that change in EQ-5D-5L would correlate significantly with change in COPD-specific HRQoL questionnaires with pulmonary rehabilitation.

nificantly with COPD-specific HRQoL questionnaires and be able to distinguish different levels of disease severity; (2) the EQ-5D-5L would improve with pulmonary rehabilitation; and (3) that change in EQ-5D-5L would correlate significantly with change in COPD-specific HRQoL questionnaires with pulmonary rehabilitation. Methods Participants All participants had a diagnosis of COPD according to the global initiative for chronic obstructive lung disease (GOLD) criteria.15 This study was a secondary analysis of data from two cohorts of patients with COPD recruited in order to determine whether the presence of sarcopenia and frailty impacts upon prognosis in COPD.4 Study 1: validity of the EQ-5D-5L in outpatients with COPD This was a cross-sectional cohort study that took place between April 2012 and October 2014. The EQ-5D-5L,12 COPD assessment test (CAT),16 St George's respiratory questionnaire (SGRQ),17 the self-report chronic respiratory questionnaire (CRQ)18 and the clinical COPD questionnaire (CCQ)19 were measured in 616 outpatients attending respiratory clinics at Harefield Hospital. Spirometry20 and the Medical Research Council Dyspnoea Scale (MRC)21 were also recorded. The age dyspnoea obstruction (ADO) index, a validated composite prognostic score in COPD22 and surrogate marker of disease severity, was calculated.22

measured in 616 outpatients attending respiratory clinics at Harefield Hospital. Spirometry20 and the Medical Research Council Dyspnoea Scale (MRC)21 were also recorded. The age dyspnoea obstruction (ADO) index, a validated composite prognostic score in COPD22 and surrogate marker of disease severity, was calculated.22 Study 2: response of the EQ-5D-5L to pulmonary rehabilitation Between August 2013 and October 2014, 400 participants were recruited from pulmonary rehabilitation clinics at Harefield Hospital to this prospective cohort study. Additional inclusion criteria were an ability to walk 5 m without assistance and no contraindication to aerobic exercise. The EQ-5D-5L, CAT, SGRQ and CRQ were prospectively measured at baseline, and following an 8-week outpatient PR programme, comprising twice-weekly supervised exercise and education sessions.4 In addition to questionnaires, the incremental shuttle walk, the five-repetition sit-to-stand and the 4 m gait speed were measured to assess change in physical performance.23–25 Participants, blinded to the results of their postpulmonary rehabilitation assessments, rated their overall change in health status following rehabilitation using an adapted five-point global rating of change questionnaire26 ‘1: much better’; ‘2: a little better’; ‘3: no change’; ‘4: a little worse’ and ‘5: much worse’.

articipants, blinded to the results of their postpulmonary rehabilitation assessments, rated their overall change in health status following rehabilitation using an adapted five-point global rating of change questionnaire26 ‘1: much better’; ‘2: a little better’; ‘3: no change’; ‘4: a little worse’ and ‘5: much worse’. The EQ-5D-5L and disease-specific HRQoL questionnaires The scoring of the EQ-5D-5L (UI and EQ-VAS) and the disease-specific questionnaires (CAT, SGRQ, CCQ and CRQ) are detailed in the online supplementary material. To summarise, the EQ-5D-5L comprises two components: the UI and the EQ-VAS. The UI is calculated from patient scoring of five dimensions (mobility, self-care, usual activities, pain/discomfort, anxiety/depression). For each dimension, participants are asked to mark between 1: ‘no problems’ to 5: ‘unable to/extreme problems’. The responses are combined to produce a five-digit number describing the participant's health status (ranging from 11111 to 55555). This is converted to a UI based on the EQ-5D-5L value set for England14 (see online supplementary figure S1). The UI ranges from −0.208 (worst possible health) to 1.000 (best possible health). For the EQ-VAS, participants are asked to record their self-rated health on a vertical VAS with the end points ‘The worst health you can imagine’ and ‘The best health you can imagine’ at the bottom (‘0’) and top of the scale (‘100’), respectively. Hence, an improvement in HRQoL is associated with an increase in UI and EQ-VAS. 10.1136/thoraxjnl-2015-207782.supp1Supplementary data

The EQ-5D-5L and disease-specific HRQoL questionnaires The scoring of the EQ-5D-5L (UI and EQ-VAS) and the disease-specific questionnaires (CAT, SGRQ, CCQ and CRQ) are detailed in the online supplementary material. To summarise, the EQ-5D-5L comprises two components: the UI and the EQ-VAS. The UI is calculated from patient scoring of five dimensions (mobility, self-care, usual activities, pain/discomfort, anxiety/depression). For each dimension, participants are asked to mark between 1: ‘no problems’ to 5: ‘unable to/extreme problems’. The responses are combined to produce a five-digit number describing the participant's health status (ranging from 11111 to 55555). This is converted to a UI based on the EQ-5D-5L value set for England14 (see online supplementary figure S1). The UI ranges from −0.208 (worst possible health) to 1.000 (best possible health). For the EQ-VAS, participants are asked to record their self-rated health on a vertical VAS with the end points ‘The worst health you can imagine’ and ‘The best health you can imagine’ at the bottom (‘0’) and top of the scale (‘100’), respectively. Hence, an improvement in HRQoL is associated with an increase in UI and EQ-VAS. 10.1136/thoraxjnl-2015-207782.supp1Supplementary data The CAT was reported as a single score (0–40), the SGRQ was reported as individual domain (symptoms, activity, impact) and total scores (0–100),17 and the CCQ was reported as individual domain (symptoms, function and mental) and total scores (0–6).19 For these three questionnaires, a higher score equates to worse HRQoL. The CRQ was expressed as individual domain (dyspnoea (5–35), fatigue (4–28), emotion (7–49), mastery (4–28)) and total summed scores (20–140), with higher scores equating to better HRQoL.18

n (symptoms, function and mental) and total scores (0–6).19 For these three questionnaires, a higher score equates to worse HRQoL. The CRQ was expressed as individual domain (dyspnoea (5–35), fatigue (4–28), emotion (7–49), mastery (4–28)) and total summed scores (20–140), with higher scores equating to better HRQoL.18 Data analysis Data analyses and graphs were produced using SPSS V.21 (IBM, USA) and Prism 5 (GraphPad, USA). Baseline characteristics were presented as mean (SD). Pearson's r correlation coefficients (where the null hypothesis=no correlation) were used to report associations between EQ-5D-5L and other questionnaires. UI and EQ-VAS were reported in groups stratified according to GOLD spirometric stage, MRC dyspnoea scale and the ADO index to assess the association with disease severity. One-way analysis of variance (ANOVA) was used for multiple group comparisons. As there were few patients in GOLD spirometric stage 1 or with MRC 1, GOLD spirometric stages 1 and 2, and MRC 1 and 2 were grouped together for the purposes of analysis. Changes in outcomes before and after PR were compared using paired t tests. Responsiveness was expressed as standardised response means (mean change/SD of change).

few patients in GOLD spirometric stage 1 or with MRC 1, GOLD spirometric stages 1 and 2, and MRC 1 and 2 were grouped together for the purposes of analysis. Changes in outcomes before and after PR were compared using paired t tests. Responsiveness was expressed as standardised response means (mean change/SD of change). MID was estimated using distribution-based (half SD) and anchor-based methods (linear regression and receiver operating characteristic (ROC) plots).16 We calculated the mean (95% CI) change in UI and EQ-VAS in those reporting feeling ‘a little better’ with rehabilitation on the global rating of change questionnaire. Further details of the linear regression and ROC analysis are described in the online supplementary material. Results Study 1: validity of the EQ-5D-5L in outpatients with COPD Complete EQ-5D-5L data were obtained in 616 of 625 patients approached. The study flow chart is shown in the online supplementary material. Baseline characteristics of the cohort are presented in table 1. Figure 1 shows the distribution of responses to the EQ-5D-5L descriptive system. The mobility and usual activities dimensions showed the greatest self-reported impairment with 58% and 53% of the cohort reporting at least moderate problems, respectively. Table 1 Baseline characteristics and Pearson's correlation coefficients with EQ-5D-5L utility index and EQ-VAS (n=616)

Results Study 1: validity of the EQ-5D-5L in outpatients with COPD Complete EQ-5D-5L data were obtained in 616 of 625 patients approached. The study flow chart is shown in the online supplementary material. Baseline characteristics of the cohort are presented in table 1. Figure 1 shows the distribution of responses to the EQ-5D-5L descriptive system. The mobility and usual activities dimensions showed the greatest self-reported impairment with 58% and 53% of the cohort reporting at least moderate problems, respectively. Table 1 Baseline characteristics and Pearson's correlation coefficients with EQ-5D-5L utility index and EQ-VAS (n=616) Characteristic Mean (SD) Pearson's correlation coefficient with EQ-5D-5L Utility index EQ-VAS r p Value r p Value Age (years) 70.4 (9.3) 0.138 <0.001 0.039 0.330 Male (%) 59.7 Smoking pack years 43.4 (34.8) −0.123 0.003 −0.163 <0.001 BMI (kg/m2) 27.5 (6.8) −0.132 0.001 −0.104 0.010 FEV1 (% predicted) 46.1 (19.6) 0.156 0.002 0.112 0.006 MRC 3.4 (1.0) −0.490 <0.001 −0.376 <0.001 SGRQ symptoms 67.5 (26.4) −0.257 <0.001 −0.283 <0.001 SGRQ activities 69.2 (21.2) −0.603 <0.001 −0.409 <0.001 SGRQ impact 36.1 (19.9) −0.596 <0.001 −0.457 <0.001 SGRQ total 51.1 (18.2) −0.623 <0.001 −0.469 <0.001 CRQ dyspnoea 13.7 (5.4) 0.403 <0.001 0.346 <0.001 CRQ fatigue 13.8 (5.3) 0.572 <0.001 0.500 <0.001 CRQ emotion 30.4 (9.6) 0.593 <0.001 0.468 <0.001 CRQ mastery 17.4 (5.8) 0.578 <0.001 0.426 <0.001 CRQ total 75.2 (21.9) 0.704 <0.001 0.518 <0.001 CAT 20.7 (8.2) −0.528 <0.001 −0.428 <0.001 CCQ symptoms 2.8 (1.3) −0.483 <0.001 −0.406 <0.001 CCQ function 2.8 (1.5) −0.674 <0.001 −0.459 <0.001 CCQ mental 2.9 (1.8) −0.507 <0.001 −0.382 <0.001 CCQ total 2.9 (1.3) −0.626 <0.001 −0.483 <0.001 Utility index 0.68 (0.24) 0.538 <0.001 EQ-VAS 61.0 (20.6) 0.538 <0.001 Data expressed as mean (SD) or Pearson's r.

<0.001 CCQ symptoms 2.8 (1.3) −0.483 <0.001 −0.406 <0.001 CCQ function 2.8 (1.5) −0.674 <0.001 −0.459 <0.001 CCQ mental 2.9 (1.8) −0.507 <0.001 −0.382 <0.001 CCQ total 2.9 (1.3) −0.626 <0.001 −0.483 <0.001 Utility index 0.68 (0.24) 0.538 <0.001 EQ-VAS 61.0 (20.6) 0.538 <0.001 Data expressed as mean (SD) or Pearson's r. BMI, body mass index; CAT, COPD assessment test; CCQ, clinical COPD questionnaire; CRQ, chronic respiratory questionnaire; MRC, Medical Research Council Dyspnoea Score; SGRQ, St George's respiratory questionnaire. Figure 1 Distribution of responses to the descriptive system of the EQ-5D-5L, from which the utility index is derived. Mean (SD) UI was 0.681 (0.236) and ranged from −0.160 to 1.000, with 43 patients (7%) describing perfect health and five patients with a negative UI (health state worse than death). Mean (SD) EQ-VAS was 60.95 (20.62). Sixteen patients (2.6%) reported an EQ-VAS of 100 (best possible health) and two patients reported an EQ-VAS of 0 (worst possible health).

anged from −0.160 to 1.000, with 43 patients (7%) describing perfect health and five patients with a negative UI (health state worse than death). Mean (SD) EQ-VAS was 60.95 (20.62). Sixteen patients (2.6%) reported an EQ-VAS of 100 (best possible health) and two patients reported an EQ-VAS of 0 (worst possible health). Table 1 describes the relationships between UI and EQ-VAS with baseline characteristics and disease-specific HRQoL questionnaires. There were significant but weak correlations (r<0.2) between EQ-5D-5L variables and age, body mass index, FEV1 and smoking pack years. There were moderate-to-strong correlations between EQ-5D-5L variables and disease-specific HRQoL questionnaire total scores, with Pearson's r ranging from 0.47 to 0.72. In general, the correlations between disease-specific HRQoL questionnaires were stronger with the UI than with EQ-VAS (table 1). Figure 2 demonstrates the relationship between UI and EQ-VAS with CRQ total score. Figure 2 Association between baseline EQ-5D-5L utility index (UI) and visual analogue scale (VAS) with chronic respiratory disease questionnaire (CRQ) total score. p Values all <0.001. UI decreased (worsening HRQoL) with increasing GOLD stage (worsening FEV1) (ANOVA: p=0.004), increasing MRC (p<0.001) and increasing ADO index (p<0.001). EQ-VAS decreased (worsening HRQoL) with increasing GOLD stage (p=0.014), increasing MRC and ADO index (p both <0.001) (figure 3). On two group comparison, neither UI nor EQ-VAS was able to clearly differentiate between GOLD 1/2 from GOLD 3.

(ANOVA: p=0.004), increasing MRC (p<0.001) and increasing ADO index (p<0.001). EQ-VAS decreased (worsening HRQoL) with increasing GOLD stage (p=0.014), increasing MRC and ADO index (p both <0.001) (figure 3). On two group comparison, neither UI nor EQ-VAS was able to clearly differentiate between GOLD 1/2 from GOLD 3. Figure 3 Mean (95% CIs) EQ-5D-5L utility index (UI) and visual analogue scale (VAS) stratified according to global initiative for chronic obstructive lung disease (GOLD) spirometric stages, Medical Research Council (MRC) dyspnoea score and age dyspnoea obstruction (ADO) Index. p Values derived from one-way analysis of variance. Study 2 Response to pulmonary rehabilitation Complete pre-EQ-5D-5L and post-EQ-5D-5L data were recorded in 324 of 400 patients (81% completion rate; see study flow chart in online supplementary material). As expected, all measures of physical performance and HRQoL improved with pulmonary rehabilitation (table 2). With regard to ceiling effect, 19 (6%) and 36 patients (11%) reported a UI of 1.00 before and after rehabilitation, while 10 (3%) and 14 (4%) patients reported an EQ-VAS score of 100 before and after rehabilitation. The distribution of responses to the descriptive system of the EQ-5D-5L are shown in figure 4. Standardised response means were 0.39 and 0.44 for UI and EQ-VAS, respectively. Standardised response means were 0.51, 0.52 and 0.76 for the CAT, SGRQ total score and CRQ total score, respectively, and 0.85, 0.73 and 0.62 for shuttle walk, gait speed and sit-to-stand, respectively.

f the EQ-5D-5L are shown in figure 4. Standardised response means were 0.39 and 0.44 for UI and EQ-VAS, respectively. Standardised response means were 0.51, 0.52 and 0.76 for the CAT, SGRQ total score and CRQ total score, respectively, and 0.85, 0.73 and 0.62 for shuttle walk, gait speed and sit-to-stand, respectively. Table 2 Baseline characteristics and response to pulmonary rehabilitation (PR): n=324 Characteristic Baseline Change with PR Age (years) 70.2 (69.2 to 71.2) Male/female (n) 192/132 BMI (kg/m2) 28.3 (27.5 to 29.1) FEV1 (% predicted) 49.8 (47.5 to 52.0) MRC 3.3 (3.2 to 3.4) −0.67 (−0.77 to −0.58) ISW (m) 240 (222 to 258) 48.5 (42.1 to 54.8) 5STS (s) 14.7 (14.0 to 15.3) −2.9 (−3.4 to −2.3) 4MGS (/ms) 0.88 (0.86 to 0.91) 0.13 (0.11 to 0.15) SGRQ symptoms 66.1 (62.9 to 69.2) −4.0 (−6.6 to −1.4) SGRQ activities 69.3 (66.3 to 72.2) −6.4 (−8.9 to −3.9) SGRQ impact 35.8 (33.1 to 38.5) −4.9 (−6.8 to −3.0) SGRQ total 50.8 (48.3 to 53.3) −5.1 (−6.8 to −3.4) CRQ dyspnoea 14.2 (13.6 to 14.8) 4.5 (3.9 to 5.2) CRQ fatigue 14.0 (13.4 to 14.6) 2.9 (2.5 to 3.4) CRQ emotion 30.4 (29.4 to 31.4) 4.2 (3.4 to 5.1) CRQ mastery 17.6 (17.0 to 18.2) 2.8 (2.3 to 3.3)) CRQ total 76.1 (73.7 to 78.4) 14.5 (12.5 to 16.5) CAT 20.1 (18.9 to 21.3) −3.5 (−4.7 to −2.2) Utility index 0.697 (0.673 to 0.720) 0.065 (0.047 to 0.083) EQ-VAS 61.1 (58.9 to 63.3) 8.6 (6.5 to 10.7) Data expressed as mean (95% CI).

n 30.4 (29.4 to 31.4) 4.2 (3.4 to 5.1) CRQ mastery 17.6 (17.0 to 18.2) 2.8 (2.3 to 3.3)) CRQ total 76.1 (73.7 to 78.4) 14.5 (12.5 to 16.5) CAT 20.1 (18.9 to 21.3) −3.5 (−4.7 to −2.2) Utility index 0.697 (0.673 to 0.720) 0.065 (0.047 to 0.083) EQ-VAS 61.1 (58.9 to 63.3) 8.6 (6.5 to 10.7) Data expressed as mean (95% CI). 4MGS, 4 m gait speed; 5STS, five repetition sit to stand; BMI, body mass index; CAT, COPD assessment test; CRQ, chronic respiratory questionnaire; EQ-VAS, visual analogue scale; ISW, incremental shuttle walk; MRC, Medical Research Council Dyspnoea Score; SGRQ, St George's respiratory questionnaire. Figure 4 Distribution of responses to the descriptive system of the EQ-5D-5L pre-pulmonary and post-pulmonary rehabilitation. Estimation of the minimum important difference Using 0.5 SD, the distribution-based estimates for the MID of the UI and EQ-VAS were 0.109 and 10.1, respectively. Figure 5 demonstrates the mean (95% CI) changes in UI and EQ-VAS according to global rating of change questionnaire response. In total, 173 (53%) patients reported feeling much better, 124 (38%) patients reported feeling a little better, 20 (6%) patients reported no change and 7 (2%) reported feeling a little worse. No patient reported feeling ‘much worse’ following pulmonary rehabilitation. The mean (95% CI) changes in UI and EQ-VAS in those reporting feeling ‘a little better’ following rehabilitation were 0.054 (0.028 to 0.080) and 6.99 (3.78 to 10.20), respectively.

reported no change and 7 (2%) reported feeling a little worse. No patient reported feeling ‘much worse’ following pulmonary rehabilitation. The mean (95% CI) changes in UI and EQ-VAS in those reporting feeling ‘a little better’ following rehabilitation were 0.054 (0.028 to 0.080) and 6.99 (3.78 to 10.20), respectively. Figure 5 Mean (95% CIs) change (Δ) in EQ-5D-5L utility index (UI) and visual analogue scale (VAS) according to response to global rating of change questionnaire (GRCQ). 1=‘much better’; 2=‘a little better’; 3=‘the same’; 4=‘a little worse’; 5=‘much worse’. Responses 3–5 were combined due to small numbers—>90% reported feeling ‘much better’ or ‘a little better’ following pulmonary rehabilitation. There were significant but weak-to-moderate correlations between change in UI and EQ-VAS with change in disease-specific HRQoL questionnaires (table 3). The slope, y-intercept and correlation coefficient between change in UI or EQ-VAS with change in other outcome measures are shown in the online supplementary material. The UI and VAS were not correlated with the CAT or SGRQ with a correlation coefficient >0.3. Table 3 Correlation coefficients of change in EQ-5D-5L utility index and EQ-VAS with pulmonary rehabilitation against external anchors

There were significant but weak-to-moderate correlations between change in UI and EQ-VAS with change in disease-specific HRQoL questionnaires (table 3). The slope, y-intercept and correlation coefficient between change in UI or EQ-VAS with change in other outcome measures are shown in the online supplementary material. The UI and VAS were not correlated with the CAT or SGRQ with a correlation coefficient >0.3. Table 3 Correlation coefficients of change in EQ-5D-5L utility index and EQ-VAS with pulmonary rehabilitation against external anchors Variable r p Value Utility index ΔSGRQ symptoms −0.05 0.538 ΔSGRQ activities −0.12 0.185 ΔSGRQ impact −0.13 0.155 ΔSGRQ total −0.14 0.127 ΔCRQ dyspnoea 0.25 <0.001 ΔCRQ fatigue 0.29 <0.001 ΔCRQ emotion 0.39 <0.001 ΔCRQ mastery 0.31 <0.001 ΔCRQ total 0.40 <0.001 ΔCAT −0.14 0.111 EQ-VAS ΔSGRQ symptoms −0.15 0.084 ΔSGRQ activities −0.11 0.205 ΔSGRQ impact −-0.27 0.002 ΔSGRQ total −0.21 0.020 ΔCRQ dyspnoea 0.31 <0.001 ΔCRQ fatigue 0.32 <0.001 ΔCRQ emotion 0.30 <0.001 ΔCRQ mastery 0.30 <0.001 ΔCRQ total 0.38 <0.001 ΔCAT −0.28 0.001 CAT, COPD assessment test; CRQ, chronic respiratory questionnaire; SGRQ, St George's respiratory questionnaire; Δ, change.

0.205 ΔSGRQ impact −-0.27 0.002 ΔSGRQ total −0.21 0.020 ΔCRQ dyspnoea 0.31 <0.001 ΔCRQ fatigue 0.32 <0.001 ΔCRQ emotion 0.30 <0.001 ΔCRQ mastery 0.30 <0.001 ΔCRQ total 0.38 <0.001 ΔCAT −0.28 0.001 CAT, COPD assessment test; CRQ, chronic respiratory questionnaire; SGRQ, St George's respiratory questionnaire; Δ, change. For change in UI, changes in CRQ-emotion, CRQ-mastery and CRQ total were associated with a correlation coefficient >0.3—these were subsequently used as anchors to estimate the MID for the UI. Using linear regression and the established MID for each anchor, estimates of the MID for UI ranged from 0.059 and 0.062. Using the same anchors, ROC plots identified estimates for the UI between 0.037 to 0.046 with C-statistic ranging from 0.66 to 0.72 (see tables 4 and 5). Table 4 Receiver operating characteristic curves to identify estimates of EQ-5D-5L that best identified achievement of the minimum important difference of the anchor, giving equal weighting to sensitivity and specificity Anchor Cut-point Sensitivity (%) Specificity (%) AUC p Value Utility index CRQ emotion 0.046 66.3 66.5 0.72 <0.001 CRQ mastery 0.038 63.5 63.7 0.66 <0.001 CRQ total 0.037 65.6 66.4 0.69 <0.001 EQ-VAS CRQ dyspnoea 6.5 70.0 62.0 0.69 <0.001 CRQ fatigue 6.5 56.1 66.1 0.65 <0.001 CRQ emotion 6.5 61.5 68.0 0.68 <0.001 CRQ mastery 6.5 58.5 67.9 0.66 <0.001 CRQ total 6.5 59.1 69.8 0.67 <0.001 AUC, area under curve or C-statistic; CRQ, chronic respiratory questionnaire.

6 <0.001 CRQ total 0.037 65.6 66.4 0.69 <0.001 EQ-VAS CRQ dyspnoea 6.5 70.0 62.0 0.69 <0.001 CRQ fatigue 6.5 56.1 66.1 0.65 <0.001 CRQ emotion 6.5 61.5 68.0 0.68 <0.001 CRQ mastery 6.5 58.5 67.9 0.66 <0.001 CRQ total 6.5 59.1 69.8 0.67 <0.001 AUC, area under curve or C-statistic; CRQ, chronic respiratory questionnaire. Table 5 Anchor-based and distribution-based estimates of the minimum important difference (MID) of the EQ-5D-5L utility index and EQ-VAS Approach Anchor/method MID estimate Utility index Distribution 0.5 SD 0.109 Mean change GRCQ 0.054 Linear regression CRQ emotion 0.063 Linear regression CRQ mastery 0.062 Linear regression CRQ total 0.059 ROC CRQ emotion 0.046 ROC CRQ mastery 0.038 ROC CRQ total 0.037 EQ-VAS Distribution 0.5 SD 10.1 Mean change GRCQ 6.9 Linear regression CRQ dyspnoea 6.5 Linear regression CRQ fatigue 7.2 Linear regression CRQ emotion 8.0 Linear regression CRQ mastery 7.6 Linear regression CRQ total 6.7 ROC CRQ dyspnoea 6.5 ROC CRQ fatigue 6.5 ROC CRQ emotion 6.5 ROC CRQ mastery 6.5 ROC CRQ total 6.5 CRQ, chronic respiratory questionnaire; GRCQ, global rating of change questionnaire; ROC, receiver operating characteristic curves. For change in EQ-VAS, changes in all CRQ domain and total scores were associated with a correlation coefficient >0.3 and were subsequently used as anchors. For EQ-VAS, linear regression estimates of the MID ranged from 6.5 to 8.0 and ROC consistently identified a cut-off of 6.5 with area under curve (AUC) ranging from 0.65 and 0.69 (tables 4 and 5).

n all CRQ domain and total scores were associated with a correlation coefficient >0.3 and were subsequently used as anchors. For EQ-VAS, linear regression estimates of the MID ranged from 6.5 to 8.0 and ROC consistently identified a cut-off of 6.5 with area under curve (AUC) ranging from 0.65 and 0.69 (tables 4 and 5). All estimates of the MID for UI and EQ-VAS are outlined in table 5. Giving equal weighting to the anchor-derived estimates, the mean (range) estimates for the MID for UI and EQ-VAS were 0.051 (0.037–0.063) and 6.9 (6.5–8.0), respectively. If prioritising the global rating of change questionnaire, which measures patient assessment of improvement or decline directly, similar mean estimates for the UI and EQ-VAS were observed (0.054 and 6.99, respectively). Discussion This study is the first to demonstrate the validity of the EQ-5D-5L UI and EQ-VAS in patients with COPD by showing significant correlations with established disease-specific HRQoL questionnaires and an ability to differentiate between groups defined according to disease severity. Furthermore, we demonstrate that the EQ-5D-5L is responsive to change following pulmonary rehabilitation, and that change in EQ-5D-5L correlates significantly with change in disease-specific HRQoL measures. Furthermore, to our knowledge, this is the first study to prospectively and purposely estimate the MID for both the EQ-5D-5L directly calculated UI and EQ-VAS. Using anchors measuring similar construct, we estimated the minimum important improvement in UI and VAS to be approximately 0.05 and 7.0, respectively.

easures. Furthermore, to our knowledge, this is the first study to prospectively and purposely estimate the MID for both the EQ-5D-5L directly calculated UI and EQ-VAS. Using anchors measuring similar construct, we estimated the minimum important improvement in UI and VAS to be approximately 0.05 and 7.0, respectively. The generic format of the EQ-5D enables comparisons of health change to be made with other conditions. It has been used in national surveys to measure population-level health status, including the Health Survey for England, and is routinely used as a measure of organisational performance in delivering some common treatments in the UK.27 The breadth of dimensions included in the instrument enables comorbidities and adverse effects of treatment to be captured in a single measure. Furthermore, the availability of a utility value set enables its use in the cost-effectiveness analyses of treatments, which is accepted or recommended by several health technology assessment agencies.2 28 29 To our knowledge, this is the first study to directly calculate values for the UI, following the recent publication of the EQ-5D-5L Value Set for England.14 Previous studies have only estimated the UI by using a Crosswalk Index Value Calculator that maps scores from the 5L to the 3L.11 An example in the COPD literature is the study from Lin and colleagues.30 Convergent validation was against the PROMIS-43 short-form questionnaire, which itself has not been well validated in COPD.30

studies have only estimated the UI by using a Crosswalk Index Value Calculator that maps scores from the 5L to the 3L.11 An example in the COPD literature is the study from Lin and colleagues.30 Convergent validation was against the PROMIS-43 short-form questionnaire, which itself has not been well validated in COPD.30 Our analysis of the psychometric properties of the UI, derived from the Value Set for England, is likely to be of interest to investigators using the EQ-5D-5L in both patients with COPD and other populations. Our results were based on large sample sizes (616 patients for the assessment of validity, 324 patients for the assessment of responsiveness and MID). High response rates were achieved, with a 99% questionnaire completion rate in study 1 and 81% completion rate in the longitudinal study 2 (completion at both time points). We also used multiple well-established, validated disease-specific HRQoL measures, including the SGRQ, CRQ, CCQ and CAT. The findings for EQ-5D-5L were robust to the choice of comparator measure, providing some internal corroboration of our findings.

on rate in the longitudinal study 2 (completion at both time points). We also used multiple well-established, validated disease-specific HRQoL measures, including the SGRQ, CRQ, CCQ and CAT. The findings for EQ-5D-5L were robust to the choice of comparator measure, providing some internal corroboration of our findings. Previous studies relating to the psychometric properties of the three-level version of the EQ-5D in COPD have had mixed conclusions. Pickard et al31 identified 12 relevant studies and concluded that EQ-5D-3L was a reliable (test–retest) and valid measure of health status in people with COPD; however, they noted limited ability of EQ-5D-3L to differentiate between milder stages of disease defined using the GOLD criteria—a similar finding for the EQ-5D-5L was observed in our study. Although this may be construed as a weakness of the questionnaire, it is well recognised that the relationship between FEV1 and HRQoL is poor in COPD.32 Furthermore, in our study, both UI and EQ-VAS were able to differentiate categories of other validated measures of disease severity, including the MRC Dyspnoea Scale and the composite ADO index (figure 3). Petrillo et al33 demonstrated ceiling effects with 13% of all patients reporting no problems in all dimensions at discharge from hospital despite patients having severe or very severe COPD. In a previous study of severe or very severe patients with COPD undergoing pulmonary rehabilitation, Ringbaek et al observed that 12.7% reported ‘perfect’ health at baseline, increasing to 17.9% after rehabilitation. In contrast, despite our study cohort having milder spirometric abnormality, we observed a lower prevalence of ceiling effect. Also, 7% of study 1 and 6% (pre-rehabilitation) and 11% (post-rehabilitation) reported perfect health following. This provides evidence that the 5-level EQ-5D has a smaller ceiling effect than the three-level questionnaire in patients with COPD.

spirometric abnormality, we observed a lower prevalence of ceiling effect. Also, 7% of study 1 and 6% (pre-rehabilitation) and 11% (post-rehabilitation) reported perfect health following. This provides evidence that the 5-level EQ-5D has a smaller ceiling effect than the three-level questionnaire in patients with COPD. The responsiveness of the EQ-5D-3L has been reported previously.6 34 35 Ringbaek et al6 demonstrated that the 3L UI improved significantly with rehabilitation, but was less responsive than SGRQ or endurance shuttle walk time. The EQ-VAS showed no significant improvement with rehabilitation. In contrast, our study showed larger changes in 5L UI and EQ-VAS both in absolute terms and in terms of standardised response means. This could be accounted for by differences in the intervention or population, but could also reflect increased responsiveness of the EQ-5D-5L questionnaire. However, we still found the responsiveness of the EQ-5D-5L to be lower than the disease-specific HRQoL questionnaires or physical performance measures.

onse means. This could be accounted for by differences in the intervention or population, but could also reflect increased responsiveness of the EQ-5D-5L questionnaire. However, we still found the responsiveness of the EQ-5D-5L to be lower than the disease-specific HRQoL questionnaires or physical performance measures. Our study is the first to report the MID of the EQ-5D-5L UI. Walters and Brazier have previously reported estimates for the three-level version of the EQ-5D from eight longitudinal studies in 11 patient groups, including COPD.35 Based on a 0.5 SD approach, they report estimates of MID of 0.12 and 0.15, which are similar but slightly higher than our results for EQ-5D-5L of 0.11. However, our anchor-based estimates of MID differed substantially from those previously reported. Mean changes in 3L UI for patients with COPD reporting their health to be ‘somewhat better’ were widely divergent at 0.013 and −0.128 in the study by Walters and Brazier,35 although this was based on a very small sample size (n=9), explaining the wide CIs (including negative values) and lack of precision.35 In comparison, the mean change in EQ-5D-5L UI in our study was a more congruent 0.054 for patients reporting feeling ‘a little better’ in our study.

the study by Walters and Brazier,35 although this was based on a very small sample size (n=9), explaining the wide CIs (including negative values) and lack of precision.35 In comparison, the mean change in EQ-5D-5L UI in our study was a more congruent 0.054 for patients reporting feeling ‘a little better’ in our study. Although the determination of the MID remains controversial with no firm consensus on methodology,16 our study used both distribution-based and anchor-based methods and provided 8 and 12 estimates of the MID for the UI and EQ-VAS, respectively (table 3). The anchor-based estimates were broadly consistent, although it was noted that the relationship between change in EQ-5D-5L and change in anchor questionnaires was only modest. The MID of the 5L EQ-VAS has only previously been estimated in a retrospective study that evaluated the response of the EQ-VAS to a 3-week inpatient rehabilitation programme. In contrast to our study, the authors only used a single anchor (a breathlessness score, rather than a HRQoL questionnaire).36 Using an ROC plot, the cut-off identified was 8, which is higher than the estimates generated in our study. This may reflect differences in the cohort populations, intervention and choice of anchor.

ntrast to our study, the authors only used a single anchor (a breathlessness score, rather than a HRQoL questionnaire).36 Using an ROC plot, the cut-off identified was 8, which is higher than the estimates generated in our study. This may reflect differences in the cohort populations, intervention and choice of anchor. There were some limitations to this study. We did not explore test–retest reliability of the EQ-5D-5L, although this has been confirmed in non-COPD populations.37–39 The patients were recruited from secondary care or pulmonary rehabilitation clinics populated with symptomatic outpatients and so whether similar findings would be obtained in patients with milder (eg, in those managed exclusively in primary care setting) or more severe disease (eg, acutely hospitalised inpatients) is open to further study. In addition, participants completed the questionnaires in the clinic setting where health professionals were on hand to answer questions. It is possible that the extremely high response rates obtained here may not be replicated in studies using other modes of administration, for example, by post or online. Another limitation of the study is the lack of a gold standard measure of HRQoL with which to compare. However, we employed a range of measures of HRQoL and clinical indices in this study, all of which have been previously validated in patients with COPD. The overall results were robust to the choice of measure used, although the strongest relationships were observed for the ‘total’ scores of the COPD-specific measures that capture a variety of impacts on functioning and symptoms. Furthermore, although the EQ-5D-5L was validated against a variety of measures, the predictive ability of this questionnaire was not explored. Future longitudinal studies would be of interest as there is a paucity of information on this topic.

ific measures that capture a variety of impacts on functioning and symptoms. Furthermore, although the EQ-5D-5L was validated against a variety of measures, the predictive ability of this questionnaire was not explored. Future longitudinal studies would be of interest as there is a paucity of information on this topic. In summary, our findings demonstrate that the EQ-5D-5L is a valid and responsive measure of HRQoL in people with COPD. Although some ceiling effects and lack of responsiveness persist with the EQ-5D-5L, these appear to be reduced compared with results previously reported for the EQ-5D-3L.10 Given the importance of the EQ-5D-5L in health economic analyses, inclusion in clinical studies of COPD would provide useful additional cost-effectiveness data of interest to health technology agencies. Contributors: Concept and design of study: WDCM. Acquisition of data: CMN, JLC, SEJ and SSCK. Analysis of data: CMN, LL, JL, WDCM. Drafting of manuscript: CMN, LL, JL and WDCM. Revision of manuscript critically for important intellectual content and approval of final manuscript: all authors. Funding: This work was funded through a National Institute for Health Research (NIHR) Clinical Scientist award (CS/7/007), NIHR Clinical Trials Fellowship (NIHR-CTF-01-12-04) and Medical Research Council (MRC) New Investigator Grant (G1002113) awarded to WD-CM. Disclaimer: The views expressed in this publication are those of the authors and not necessarily those of the NHS, the NIHR nor the Department of Health.

Funding: This work was funded through a National Institute for Health Research (NIHR) Clinical Scientist award (CS/7/007), NIHR Clinical Trials Fellowship (NIHR-CTF-01-12-04) and Medical Research Council (MRC) New Investigator Grant (G1002113) awarded to WD-CM. Disclaimer: The views expressed in this publication are those of the authors and not necessarily those of the NHS, the NIHR nor the Department of Health. Competing interests: CMN is funded by a NIHR Doctoral Research Fellowship (DRF-2014-07-089). JLC and SEJ are funded by the NIHR Respiratory Biomedical Research Unit, Royal Brompton & Harefield NHS Foundation Trust and Imperial College. SSCK was funded by the MRC. WD-CM is part funded by the NIHR Collaboration for Leadership in Applied Health Research and Care (CLAHRC) for NW London. Patient consent: Obtained. Those with significant cognitive impairment or unable to read English were excluded. Ethics approval: London-Camberwell St. Giles Research Ethics Committee. Provenance and peer review: Not commissioned; externally peer reviewed.