Case-finding and improving patient outcomes for chronic obstructive pulmonary disease in primary care: the BLISS research programme including cluster RCT

Work package 1: clinical trial to evaluate case-finding; TargetCOPD trial

The aim of this WP was to ascertain the most clinically effective and cost-effective approach to identifying undiagnosed COPD. Initially, this was considered in terms of yield (WP1i), but with additional follow-up (in a variation to the contract) in terms of clinical outcomes (WP1v and vi). A Markov model was also developed to estimate the long-term cost-effectiveness of a systematic screening programme for undiagnosed COPD. Finally, we explored the views of patients on the process and outcomes of case-finding and the perspective of staff in primary care on the concept of case-finding and percieved implications for their practice.

The objectives were to:

• ascertain the clinical effectiveness and cost-effectiveness of targeted case-finding (opportunistic or active) compared with routine care

• develop a Markov model to compare the cost-effectiveness of systematic case-finding with current practice

• explore the views of patients and primary care staff on COPD case-finding

• describe the clinical management of screen-detected COPD patients in primary care

• assess the long-term effectiveness of COPD case-finding on respiratory hospitalisations and mortality

• compare outcomes among screen-detected COPD patients who were adequately managed by their general practitioner (GP) with those among patients who were not.

Work package 2: Birmingham primary care chronic obstructive pulmonary disease cohort

The aim of this WP was to develop a cohort of more than 2000 people with COPD who would be representative of those in a primary care setting, including more people with mild/moderate disease than included in previous cohorts. Participants were recruited from COPD registers in general practices as well as from participants in the TargetCOPD trial in WP1. This included those who were identified through case-finding and those who took part in the case-finding trial and reported respiratory symptoms but did not have COPD based on spirometry.

The participants were followed up every 6 months for around 3 years (2012–16) and their data were linked to routine data on hospitalisation and mortality obtained from NHS Digital. We used these data to externally validate an established prognostic model [the age, dyspnoea, airflow obstruction (ADO) prognostic score] in predicting mortality. In addition, we developed a new prognostic model [the Birmingham Lung Improvement StudieS (BLISS) index] to predict the risk of respiratory hospitalisation in this primary care population.

A sample of 26 cohort participants were also invited to attend focus groups to explore barriers to and facilitators of undertaking physical activity (PA), which is one of the most important components of treatment for people with COPD.

The establishment of the cohort also provides an opportunity for testing novel interventions in future.

The objectives were to:

• recruit a primary care cohort of 2000 new and existing COPD patients

• test the validity of existing COPD prognostic models in a primary care COPD population

• develop a prognostic model (BLISS index) to predict respiratory hospitalisations suitable for a primary care population

• explore barriers to and facilitators of PA participation among people with COPD.

Work package 3: chronic obstructive pulmonary disease and occupational performance

The aims of this WP were to (1) examine the relationship between clinical characteristics and severity of COPD and occupational outcomes, including employment, work absenteeism and presenteeism (working while unwell); (2) examine whether or not disease progression over time is associated with occupational outcomes; and (3) assess the feasibility and benefits of offering formal occupational health (OH) assessment and subsequent recommendations aimed at improving work-based indices. Participants in this WP were a subsample of the larger cohort recruited in WP2.

The objectives were to:

• examine factors associated with occupational performance (employment, absenteeism and presenteeism) among COPD patients of working age

• examine how disease progression (lung function decline, exacerbation) over time is associated with occupational performance (employment, absenteeism and presenteeism) among COPD patients in employment

• assess the feasibility of offering OH assessment with recommendations to people with COPD.

Rationale

Several reports from the UK government and health charities22,23 have highlighted the burden of COPD, the extent of underdiagnosis and the variation in access to and participation in relevant services. The Department of Health and Social Care published an outcomes strategy for people with COPD in 2011 that recommended opportunist and the systematic case-finding to minimise late diagnosis. 22 However, the most clinically effective and cost-effective approach for identifying undiagnosed cases was not known. Although some published studies reported the yield from either active24–26 or opportunistic27–29 approaches to case-finding using spirometry, these were limited because of a lack of comparison groups, the restricted number and range of participants, a lack of follow-up and different target populations (e.g. specific age groups; all current smokers or people who have smoked; whether or not symptoms were considered prior to spirometry). Furthermore, general population screening using spirometry was not recommended16 because it would identify many people without clinically important disease, for whom there is little evidence of effective interventions. 30 We therefore sought to evaluate different approaches to case-finding for undiagnosed COPD, focusing on yield and cost-effectiveness (from a health-care perspective). To our knowledge, this was the first major trial to compare different targeted recruitment approaches with case-finding with routine care.

What we did

We recruited 54 GP practices across the West Midlands (10 August 2012–22 June 2013) to take part in the 12-month trial and randomised these clusters to case-finding or to continue with routine care. All patients on the practice registers aged 40–79 years who had no existing diagnosis of COPD and had ever smoked (based on GP electronic records) were eligible for and included in the trial. Using a computer-generated randomisation sequence, we randomised practices (balanced on practice characteristics) and in the intervention case-finding arm we randomised individual households of patients to receive one of two approaches to case-finding: opportunistic (offered screening opportunistically when they visited the practice for any reason) or active (additionally offered screening through a postal invitation). Screening was carried out in two stages, with an initial questionnaire (see Appendix 2) followed by an invitation to attend for diagnostic spirometry for those who reported relevant chronic respiratory symptoms. Our study case definition of 1-year incident COPD was either (1) a new diagnosis of COPD by the GP (based on new entry on practice COPD register) or (2) the presence of airflow obstruction on screening [defined as forced expiratory volume in 1 second (FEV₁)/forced vital capacity (FVC) < 0.7 post bronchodilator, in line with recommendations from the guidelines31 produced by the National Institute for Health and Care Excellence (NICE)] in patients with chronic respiratory symptoms. Sample sizes were based on estimates from our published model (see published paper21 for full details) and required 27,768 patients per group. Primary outcomes were analysed using appropriate regression analyses adjusted for practice-level deprivation, ethnicity and age. The active versus opportunistic comparison required logistic regression with fixed effects and the targeted versus routine comparison required multilevel models with random effects. We undertook a within-trial cost-effectiveness analysis to calculate the cost per additional case detected of both strategies versus routine care, and we undertook a detailed cost analysis of the screening processes using standard NHS32,33 and trial-specific costs34 (2013 prices), including set-up costs and training costs. Equipment and training costs were amortised over 3–5 years using a discount rate of 3.5%. Sensitivity analyses considered patient costs, alternative case-finding scenarios and models of care (GP, community or secondary care led).

What we found

A flow diagram of participants is available in the published paper. 21 We found that very few new cases of COPD were diagnosed in the routine care practices (n = 337; 0.8%). The odds of finding new cases were seven times higher using the targeted approach than with routine care [n = 1278 (4%); adjusted odds ratio (OR) 7.45, 95% confidence interval (CI) 4.80 to 11.55; p < 0.0001], and active case-finding (combining opportunistic and active postal invitation to screening) was twice as effective as opportunistic-only [n = 822 (5%) vs. n = 370 (2%); adjusted OR 2.34, 95% CI 2.06 to 2.66; p < 0.0001]. Active case-finding may also be more cost-effective than the opportunistic approach (£333 vs. £376 per additional case detected, respectively). The incremental cost-effectiveness ratio (ICER) for active case-finding was £573 per additional case detected compared with opportunistic case-finding. Sensitivity analyses made little difference, although a secondary care-led service was more expensive.

Strengths and limitations

To our knowledge, this was the first RCT to compare yield of undiagnosed COPD from a comparison of different screening approaches with routine practice and to estimate their cost-effectiveness. However, the efficiency of screening could have been improved by using different screening tests, or pre-screening algorithms applied to electronic health records (EHRs) to target screening invitations to those at highest risk. Even with the most intensive screening approach, only 38% of those invited for screening responded, and not all those who responded and had symptoms attended for diagnostic spirometry. Further research should focus on how to maximise screening coverage and uptake and improve the efficiency of the screening process.

Rationale

The TargetCOPD trial confirmed that active approaches to screening result in a higher proportion of undiagnosed COPD patients being identified compared with routine care. People with screen-detected COPD are expected to benefit from treatment resulting in improved quality of life, increased survival and reduction in hospital admissions. To provide data for policy-makers to consider the longer term benefits of screening for COPD in relation to investment in other health services, a cost-utility model is needed. Published economic evaluations in COPD have primarily considered interventions for the disease rather than for those who are screen detected,36 and others have concentrated on the costs of COPD in burden of illness studies. 37 No trial-based economic evaluation had considered case-finding. NICE guidelines38 included a simple decision tree-based modelling to determine the cost-effectiveness of opportunistic case-finding among people aged > 35 years who have smoked and have a chronic cough. However, the model was simplistic and included many assumptions for which evidence was limited. We developed a model-based economic evaluation of the long-term costs and benefits of screening for undiagnosed COPD.

What we did

We used costs (using 2015 prices39) and outcome data from the TargetCOPD trial, in combination with the best available published data, and additional information from our linked Birmingham COPD cohort study (WP2) to develop a Markov decision-analytic model to address this objective. The model compared the cost-effectiveness of a 3-yearly systematic case-finding programme aimed at people aged > 50 years who have smoked with current practice because the yield of new cases was very small in younger age groups. The model had a time cycle of 3 months, which was short enough to capture important COPD-related events, and a time horizon of 50 years, assuming a maximum age of 100 years. Patient-level data on case-finding pathways were obtained from our TargetCOPD RCT. The model outcome was cost per quality-adjusted life-year (QALY) gained, from a health service perspective. Discounting was applied to costs and outcomes at 3.5%. Multiple one-way sensitivity analyses assessed what impact modification of key parameters had on the results. We considered varying the starting age, screening interval and time horizon as well as the screening processes (questionnaire response, spirometry attendance rate), treatment initiation rates and effectiveness of treatments.

What we found

We estimated the ICER of systematic case-finding compared with routine care to be £16,596 per additional QALY gained. Using the commonly used willingness-to-pay threshold in the UK (£20,000 per QALY), we estimated there was 78% probability of cost-effectiveness. The estimate was robust to sensitivity analyses, with the main cost-driver being uptake of screening. The most cost-effective age to begin screening was around 60 years. Better ascertainment of treatment effectiveness will help improve precision but, using the best current estimates from the literature,40,41 screening is likely to be cost-effective provided that at least 12% respond to a screening questionnaire, > 26% attend spirometry and > 8% of screen-detected patients are adequately treated and managed.

Strengths and limitations

To our knowledge, this is the first economic model to evaluate the long-term cost-effectiveness of a COPD case-finding strategy, using contemporaneous data sources to inform estimates and using multiple sensitivity analyses to test the robustness of the model. However, the validity of some of the assumptions underlying the model is unknown. In particular, there is uncertainty around the effect of treatment on progression and the natural history of COPD. Future research should refine the model based on data from studies that provide more accurate estimates of effectiveness and consider additional costs, such as those to the health service, of pathways to deal with a larger number of identified cases.

Rationale

Although there has been much research examining the yield from different case-finding activities, few previous studies have examined other aspects related to the development of a screening programme. One important aspect is the acceptability of screening and understanding the perspective of both patients and those who provide screening. We undertook two studies: one to explore the views of patients who had been invited for screening as part of the TargetCOPD trial, about the screening process and outcomes, and another to understand the process from the perspective of primary care staff who would manage those who are case-found.

What we did

For the patient perspective, we invited for interview people who had been invited for screening in either the active or opportunistic arm as part of the TargetCOPD trial (i.e. adults aged ≥ 40 years with a smoking history who had been considered eligible for the trial by their GP). We invited four groups of patients: those who (1) were invited and consented to take part in screening, (2) were invited and declined, (3) attended screening but did not have COPD and (4) attended and had abnormal lung function suggesting COPD. We sought their views on the screening process and their reflections on the outcomes of screening.

For the primary care staff perspective, we invited 20 staff, including GPs, nurses and managers in practices that had taken part in the TargetCOPD trial. Participants were invited to share their views on COPD case-finding, including their perceptions of the benefits, harms, and barriers to and facilitators of implementing a screening programme in primary care.

For both studies, interviews were transcribed and analysed using the framework approach.

What we found

Forty-three patients and 20 health-care staff were interviewed. Patients generally considered screening to be a good thing, and the presence of symptoms on prompting facilitated their attendance. The importance of ensuring that the screening process is convenient was highlighted, and patients worried that GPs did not have the time to follow up after screening.

Barriers to attending screening included psychological and practical factors. The former related to denial and failure to recognise symptoms, fear of the ‘test’ and perceiving lung disease as less important within the hierarchy of their health problems. Practical barriers included lack of time, inability to access GP appointments and having caring and other responsibilities that were considered more important.

Among primary care staff, although they also generally supported screening for undiagnosed COPD, they also commented on concerns around potential negative consequences, including an increase in workload for GPs and overdiagnosis in patients. Some commented that, currently, diagnosed patients were not being adequately treated. Perceived barriers to implementing screening included lack of resources and limited access to diagnostic services. However, potential solutions, including better support from secondary care, an increase in specialist COPD nurses and better community respiratory service provision, were also discussed. Poor knowledge of COPD in terms of recognising symptoms and how to manage those with the disease was also highlighted as a problem that needs to be addressed.

For a screening programme to be implemented and have high uptake, it is important to raise patient awareness of COPD risk factors and symptoms and provide training and additional resources for primary care. In particular, it is important to ensure that management pathways for diagnosed COPD patients are optimised before further cases are identified.

Strengths and limitations

To our knowledge, this was the first study to explore perceptions of patients and health-care providers on different stages of case-finding. However, we did not explore experiences of the approach to recruitment for screening (questionnaire at GP surgery or by post) and no participant commented on this. In addition, we did not invite and have not captured the views of people who reported no chronic respiratory symptoms as part of screening. Furthermore, inviting patients in the context of research may not reflect views on screening in practice. For health-care staff, those who participated are likely to be more engaged in case-finding and their views may represent those who are more proactive in management of case-found COPD.

Rationale

From a public health perspective, screening is more than a screening test. To have an impact on clinical outcomes, a number of criteria need to be fulfilled and, in the UK, the National Screening Committee considers these carefully before recommending commencement of a population screening programme. The criteria relate to the condition, the screening test, the treatments available for those who are screen detected and the characteristics of the programme and its implementation. In relation to the programme implementation, it is important to ensure that resources are available and pathways are in place to manage screen-detected individuals. In the UK, NICE has set out a pathway for the diagnosis and management of people with COPD. 31 As part of the TargetCOPD trial, patients with chronic respiratory symptoms who attended spirometry screening had their results fed back to their GP, with a note for them to follow NICE guidelines for further management. However, there are few studies that have examined how patients diagnosed with COPD are subsequently managed and whether or not primary care staff follow recommended pathways for managing these patients. We therefore obtained data from GP EHRs and self-reported data from case-found patients (those with airflow obstruction on spirometry who fulfil the NICE recommended criteria for diagnosing COPD) to describe the clinical management of case-found COPD patients (identified through the TargetCOPD trial) and compare this with that of patients newly diagnosed with COPD through routine care. In addition, we compared characteristics and management for those who were or were not entered on to the practice COPD register.

What we did

We identified patients who had been newly diagnosed with COPD (case-finding, n = 857; routinely diagnosed, n = 764) during the period August 2012 to June 2014 in the 54 GP practices that took part in the TargetCOPD trial. Data on demographic and clinical characteristics, as well as a range of clinical assessments and interventions recommended by NICE for people with COPD, were extracted from EHRs for a subset of patients covering the period April 2011 to September 2017. In addition, patients who had been identified through case-finding were invited to complete a health questionnaire around 5 years after their first diagnosis, in March 2018, with a reminder 2 months later.

For all patients, we determined whether or not they had been added to the practice COPD register used in reporting for the Quality and Outcomes Framework (QOF) by the end of the TargetCOPD trial period. The number of COPD-related clinical assessments and interventions were summed to form a clinical management score. Multilevel logistic regression was used to assess for associations between participant characteristics and the likelihood of being added to a disease register, comparing those who were identified through case-finding with those routinely diagnosed. Multilevel linear regression was used to assess associations between participant characteristics, COPD disease registration and the clinical management score.

What we found

Figure 2 shows a summary of participants included in these analyses. The primary analysis showed that just over one-fifth (182/857; 21.2%) of case-found patients, but almost all of the routinely diagnosed patients (708/764; 92.7%), had been added to the QOF COPD register within 12 months of assessment [median time from trial spirometry assessment to COPD registration 152 days, interquartile range (IQR) 72–258]. Factors associated with a higher likelihood of COPD registration among case-found patients were current and former smoking (adjusted OR 8.68, 95% CI 2.53 to 29.82, vs. OR 6.32, 95% CI 1.88 to 21.29, respectively) and lower percentage of predicted FEV₁ (OR 0.96, 95% CI 0.95 to 0.98).

FIGURE 2.

Flow of participants contributing to analyses for this study. a, Included in current study.

Electronic health record data were available for 532 out of 1629 patients (identified through case-finding, n = 344; identified through usual care, n = 188) (Table 1). The characteristics of participants with and without EHR data were broadly similar. Factors associated with a higher clinical management score were being on the COPD register (adjusted β 5.06, 95% CI 4.36 to 5.75, which means that the score was on average 5 units higher for those on the COPD register than for those not on the register) and a higher number of comorbidities (adjusted β 0.38, 95% CI 0.11 to 0.65, which means that the score increased by 0.38 units for each additional comorbidity). Although of only borderline statistical significance, there was also a negative association with being case-found rather than routinely diagnosed (adjusted β –0.69, 95% CI –1.44 to 0.07).

Self-reported questionnaire data were available for 375 out of 857 case-found patients. Only one-fifth of these patients were on the COPD register and, overall, one-third were aware of their diagnosis through their GP (88.5% of those on the COPD register vs. 17.5% of those not on the register). Around 45% had attended a COPD annual review, with the proportion being higher for those on the COPD register (83.3% vs. 34.7%). Factors associated with a higher clinical management score in this group were having a larger number of comorbidities (adjusted β 0.38,95% CI 0.10 to 0.65), higher COPD Assessment Test (CAT) score (adjusted β 0.05, 95% CI 0.01 to 0.08) and lower percentage of predicted FEV₁ (adjusted β –0.03, 95% CI –0.04 to –0.01). Being on the COPD register was also significantly associated with a higher management score (adjusted β 3.48, 95% CI 2.81 to 4.15).

A proportionately high number of case-found patients with COPD were not added to practice COPD registers and these patients were less likely to receive recommended effective treatments for their condition. Overall, even those who were on the COPD register did not receive all recommended interventions, including smoking cessation advice or referral to pulmonary rehabilitation.

Strengths and limitations

This is one of the largest studies to evaluate the clinical management of screen-detected COPD patients in primary care. The lack of availability of EHR data on all trial participants was a limitation, and the validity of the EHR data is dependent on the clinical coding practices used. Nevertheless, the missing data are likely to be random, based on similarity of characteristics between those with and without data. Furthermore, the self-reported data from those who responded to questionnaires broadly verified the findings from health records.

Rationale

Since starting our programme of work, the UK National Screening Committee undertook a review to consider screening for COPD44,45 and the USPSTF updated their review. 46 Both recommended against screening for the time being, citing the need to establish evidence on clinical effectiveness of early identification before recommending systematic programmes for screening. The benefits of case-finding in improving health-related quality of life (HRQoL) and QALY gains has also not been previously studied.

Through our original trial, the infrastructure was in place to assess whether or not screening and early detection of COPD benefited patients in the longer term. We therefore sought a variation to the contract to extend follow-up and to link data on all patients who were part of the original trial with routinely available data. We obtained data on all eligible participants from NHS Digital on hospitalisations [through Hospital Episode Statistics (HES) data] and mortality, from the start of the trial until the last date available. We also sought approval from the National Research Ethics Committee and legal approval from the Confidentiality Advisory Group to obtain relevant patient identifiable data from GP practice records through an opt-out process and hold these temporarily to allow data linkage with the data obtained from NHS Digital. We were then able to compare outcomes among those who were in practices where active screening took place with those in the routine care arm practices (WP1v).

Additionally, to improve the input into the health economic model developed previously, we sought to obtain additional data on quality of life among screen-detected COPD patients some years after diagnosis and to compare outcomes among those who were managed in accordance with NICE guidelines with outcomes among those who were not.

What we did

Data on mortality and hospitalisations (all-cause and respiratory) were obtained for all eligible patients (who were alive at the start of the intervention, n = 74,693) from 54 participating practices in the TargetCOPD trial. Patient demographic data and information on whether or not the patient was on the practice QOF COPD register within 12 months of the trial were obtained from GP records. We also administered a questionnaire in 2017/18 to all case-found patients identified through the TargetCOPD trial to invite them to respond to questions on quality of life as well as their health and treatments received for COPD. Cox proportional hazard models, using random effects to account for clusters and adjusted for potential confounding factors, were used to model the time to event outcomes (death, first all-cause hospital admission and first respiratory hospital admission) in the intervention (active or targeted) and routine care arms. The time to event was censored at the death (for admission outcomes) or data extraction (30 September 2017 for hospital admissions and 13 October 2017 for deaths) if no event occurred. We also compared outcomes for the two intervention arms (active and opportunistic). Finally, for screen-detected cases, we compared mortality and hospitalisation as well as HRQoL measures [EuroQol-5 Dimensions (EQ-5D) and CAT scores] among those who were or were not added to the practice QOF register. Based on data from objective v, we used addition to the practice COPD register as a proxy measure for being better managed and treated for COPD. Analyses were adjusted for baseline values as well as for a range of potential confounders, including age, sex, ethnicity and baseline values for lung function, comorbidities and smoking status.

What we found

Among the 32,743 participants in the case-finding arms, 1557 had a respiratory hospitalisation compared with 1899 among the 41,950 participants in the routine arm over a mean follow-up period of 4.3 years [adjusted hazard ratio (HR) 1.04, 95% CI 0.73 to 1.47]. The corresponding HR for all-cause hospitalisation and mortality were 1.06 (95% CI 0.66 to 1.71) and 1.15 (95% CI 0.82 to 1.61), respectively, suggesting that, overall, there was no significant difference in risk of first hospitalisation and mortality between those who were in the screening arm compared with those in the routine care arm of the trial.

Within the two intervention groups in the case-finding arm there was no statistically significant difference between groups in terms of overall hospitalisations and mortality (adjusted HR 1.01, 95% CI 0.98 to 1.04, and adjusted HR 1.08, 95% CI 0.96 to 1.20, respectively). The adjusted HR for respiratory hospitalisation in the active group compared with the opportunistic group was 1.14 (95% CI 1.02 to 1.27), indicating an increased hazard of respiratory admissions in the former group (where yield from screening was higher).

Comparison of outcomes for screen-detected patients who were on the QOF COPD register with those who were not also showed no statistically significant difference in relation to all-cause hospitalisation (adjusted HR 0.86, 95% CI 0.66 to 1.11), respiratory hospitalisation (adjusted HR 0.95, 95% CI 0.52 to 1.73) and mortality (adjusted HR 1.06, 95% CI 0.53 to 2.12).

Thus, despite screening resulting in a higher yield of undiagnosed cases of COPD, there was no difference between those who were in practices with or without screening in terms of clinical outcomes at 4 years. The poor clinical management of COPD generally, and very low addition of case-found patients (particularly those with less severe disease) to the practice COPD register, may be an explanation for the findings. Given these results, we did not undertake a health economic analysis to examine cost per hospital admission avoided or cost per life-year saved.

In the adjusted analyses examining HRQoL, there was no statistically significant difference in EQ-5D scores between the two groups (adjusted mean difference –0.006, 95% CI –0.048 to 0.036). The adjusted mean CAT score was statistically significantly and clinically higher in those who were on the COPD register than in those who were not (mean difference 2.317, 95% CI 0.481 to 4.153), indicating a greater impact of COPD on their life and poorer quality of life. This difference is likely to reflect the more severe disease and characteristics of those who are added to the COPD register compared with those who are not, rather than being a result of how they were managed.

Strengths and limitations

To our knowledge, this is the first trial to report clinical outcomes from a large trial of screening for undiagnosed COPD. The trial was not powered to detect clinical outcomes because that was not the primary aim, but, nevertheless, it has provided reasonable effect estimates. The poor clinical management of people with screen-detected COPD limits the ability of detecting any potential benefits and thus the interpretation of findings. The low proportion of screen-detected patients being entered on the COPD register may explain the observed lack of effectiveness of screening.

Jordan et al.47

This was the protocol for the TargetCOPD trial, outlining the rationale, methods and analysis plan.

Miller et al.48

In this analysis we used data from the TargetCOPD trial and compared how the application of two different definitions of airflow obstruction would impact on the clinical characteristics of the population who would be labelled as having COPD. The definition used in the trial was based on the ratio of FEV₁/FVC < 0.7 [the fixed ratio (FR)], which is recommended by NICE. The second definition was based on using the lower limit of normal (LLN) that is increasingly being recommended. We found that, among 2607 people who attended for spirometry, around one-third had airflow obstruction using the FR definition compared with 20% using the LLN definition. There was overlap between the two groups. However, those identified by the FR and not the LLN definition were older, had better lung function and fewer respiratory symptoms, but had a higher rate of heart disease. Overall, we demonstrated that using the FR rather than LLN identifies a greater proportion of individuals with cardiac, rather than respiratory, clinical characteristics.

Haroon et al.49

This analysis used data from the TargetCOPD trial to develop a validated algorithm and risk score to target case-finding on those at highest risk of undiagnosed COPD and thus improve the efficiency of any future case-finding process. Although other COPD risk scores have been developed, this was the first that was based on identifying case-found COPD (rather than incident clinical COPD diagnosed through routine care) and is therefore more useful in the context of case-finding in primary care.

Haroon et al.50

In this analysis we used a case–control study design to match incident COPD cases from 340 GP practice registers (using the UK Clinical Practice Research Datalink) to two controls (based on age, sex and practice). Predictive risk factors for COPD were identified from practice records and used to develop a clinical risk score. The risk score was validated using a sample from a further 20 practices. The model, including smoking status, history of asthma and lower respiratory tract infections, and prescription of salbutamol in the previous 3 years, resulted in reasonable prediction (c-statistic 0.85, 95% CI 0.83 to 0.86).

Haroon et al.51

This systematic review (based on studies from 1997 to 2013) summarised the uptake and yield from different approaches to screening for undiagnosed COPD in primary care. Data from three RCTs, one non-randomised trial and 35 uncontrolled studies showed that all approaches result in identification of new undiagnosed cases. The review suggested that targeting higher-risk individuals (e.g. smokers) and using questionnaires or handheld flow meters prior to diagnostic screening was likely to increase yield. However, it also highlighted the need for well-conducted RCTs.

Haroon et al.52

This review compared the diagnostic accuracy of different COPD screening tests in primary care. A total of 10 studies were identified from 1997 to 2013 and included use of screening questionnaires [mainly the COPD Diagnostic Questionnaire (CDQ)], handheld flow meters [e.g. the copd-6 (Vitalograph Ltd, Buckingham, UK)] or a combination. Handheld flow meters demonstrated higher test accuracy than questionnaires but the review highlighted the need for high-quality evaluation of comparative screening strategies.

Rationale

The natural history of COPD is still not understood, and several expert reviews have highlighted a need to further investigate both old and new longitudinal data. 5,15 Prior to this programme, a number of relevant COPD disease cohorts had been established,54–56 but these included patients with more advanced disease from secondary care settings, with short duration of follow-up, and were mainly of small size. Other large population cohorts have also been used to address questions relevant to COPD. 57–60 However, because these were not specifically set up to address COPD, not all relevant measures were undertaken and the quality of lung function was not always prioritised. There were no UK primary care COPD cohorts with patients representing the range of disease severity, particularly including people with mild/moderate disease, or a diverse socioeconomic mix. Furthermore, existing cohort studies included neither people with COPD who were identified through case-finding nor patients reporting respiratory symptoms but who had normal lung function [former Global Initiative for Chronic Obstructive Lung Disease (GOLD) stage 0]. 61 The evidence on progression to COPD in the latter group is limited and contradictory,62–64 and methods for assessing symptoms are inconsistent. 62,65 The clinical relevance and natural history for this patient group and screen-detected cases are unclear.

What we did

We recruited 2305 patients aged ≥ 40 years from 71 practices across the West Midlands (Figure 3), comprising 1564 patients with previously diagnosed COPD, 330 previously undiagnosed patients with respiratory symptoms and airflow obstruction confirmed by spirometry (case-found COPD) and 411 symptomatic patients with normal lung function confirmed by spirometry. 53

FIGURE 3.

Flow chart of patient recruitment and assessment for the Birmingham COPD cohort study.

Baseline assessments were undertaken by trained researchers using standardised protocols between 2012 and 2014 (Figure 4). Assessments included high-quality pre- and post-bronchodilator spirometry using an ndd EasyOne^® spirometer (ndd Medizintechnik AG, Zurich, Switzerland). Other measurements included height and weight, body fat percentage estimation using the Tanita BC-420SMA Body Composition Analyser (Tanita Europe BV, Amsterdam, the Netherlands), assessment of grip strength using a Saehan Hydraulic Hand Dynamometer (Saehan Corp., Masan, Republic of Korea) and assessment of exercise capacity using the sit-to-stand test. In addition, trained researchers used face-to-face interviews to obtain occupational history. Information on skill content of occupations was used to assign a four-digit standard occupational classification (SOC2010)66 code for current or main occupation using the CASCOT (computer assisted structured coding tool) software (online version, Office for National Statistics, Newport, UK).

FIGURE 4.

Cumulative number of cohort participants recruited and having baseline assessments by month.

Participants were also asked to complete questionnaires that sought data on demographic characteristics, lifestyle (smoking history and exercise habits), symptoms, exacerbation history, general health, diagnosed medical conditions, health-care usage and the home environment (see Appendix 3). HRQoL was assessed using disease-specific (CAT)67 and generic (EQ5D)68 instruments, and the Patient Health Questionnaire-9 (PHQ-9)69 instrument was used to screen for depression. Work productivity was assessed through questions on work absence and presenteeism using the Stanford Presenteeism Scale (SPS-6)70 and the Work Productivity and Activity Impairment questionnaire (WPAI). 71

At 6-monthly intervals, patients were sent follow-up questionnaires by post. All follow-up questionnaires included items on employment, general health, lung health, exacerbations, new diagnoses, attendance at pulmonary rehabilitation, health-care utilisation, smoking history, medications, depression and HRQoL. Some questionnaires included additional items, which are summarised in Table 2.

Patients were invited for a final assessment visit around 3 years after baseline (2015–16). In addition to post-bronchodilator spirometry, other baseline assessment measures and questionnaires were repeated. Cohort participants’ routine data on comorbidities and medications were extracted from GP records. Linked data on hospital episodes and mortality were also obtained from NHS Digital for the period 1 April 2012 to 31 March 2016.

Establishment of this cohort allowed important questions of relevance to patient benefit to be addressed (WP2i).

What we found

Follow-up data were available for 2250 patients (97.6%), and almost two-thirds (1469, 63.7%) returned for face-to-face follow-up assessment. Six-monthly questionnaires were completed by around two-thirds of patients (62.8–73.0%; see Table 2). In the initial 2 years of follow-up, 267 (12%) patients had at least one respiratory-related hospital admission, based on HES data. Over the entire period of follow-up (minimum 1.8 years, maximum 3.8 years), 382 patients (17%) had at least one respiratory hospital admission and 170 (7%) had died at the time data were obtained from NHS Digital.

Strengths and limitations

We established one of the largest primary care COPD cohorts, which is novel in that it includes case-found patients. A limitation was that fewer than one-third of case-found patients agreed to join the cohort. Our recruitment strategy resulted in an over-representation of patients with less severe disease because patients had to be ambulatory. Despite attempts to include patients from diverse ethnic backgrounds, the majority were white.

Rationale

A better understanding of factors predicting prognosis and the development of a prognostic model can facilitate doctor–patient consultations and inform management decisions and health service planning. For COPD, a simple measure of lung function, FEV₁, has historically been used to grade severity. However, there is increasing recognition that this measure is not a good predictor of clinical outcomes. Alternative measures of lung function may improve diagnostic73 and prognostic ability. 74,75

A number of studies have also described a range of factors other than lung function that are associated with COPD progression, deriving prognostic indices. The first [BODE (body mass index, airflow obstruction, dyspnoea, and exercise) index] was developed to predict mortality risk in those with COPD. 56 However, the method of development was not clear, its validity has not always been confirmed and not all measures are practical in non-specialist settings. 76 Since then, several other prognostic models have been developed and, since we started the programme, a number of systematic reviews have been undertaken to summarise these. 77–79 The reviews show that existing prognostic models are heterogeneous in terms of the number and type of predictors, the prognostic outcome, time horizon and statistical approach. Most focus on predicting mortality risk,56,76,80–82 although others were developed to predict additional outcomes such as exacerbations,83,84 COPD-related hospitalisation,85 respiratory hospital attendance/admission86 and exacerbation or hospitalisation. 87,88 Only three indices84,87,88 were derived from primary care populations, despite this being where most COPD patients are managed, and most included patients with more severe established disease. No models were developed in populations that included case-found patients. Few studies were validated or used recommended statistical approaches for deriving the model.

The most recent review showed that the ADO index was most discriminatory in predicting mortality. 78 For a prognostic model to be used by clinicians, it needs to be simple and capture the required patient data with minimum resource;89 the ADO index fulfils these criteria. We therefore undertook validation of the ADO index for mortality within our cohort (WP2, objective ii).

However, in relation to predicting respiratory hospitalisation, which is an important outcome to consider, current models have moderate discriminative ability. This suggests that other relevant predictors are missing from these prognostic models. We therefore derived a new COPD prognostic model, the BLISS index, for use in a primary care population (with 2-year respiratory hospitalisation as primary outcome and all-cause hospitalisation, exacerbations, primary care consultations and mortality as secondary outcomes), using recommended statistical techniques (WP2, objective iii, draft paper; see Appendix 4).

What we did (objective ii)

We validated the ADO index using data from 1701 patients from the Birmingham COPD cohort study (case-found or on the practice COPD register) who had complete data for the variables required for this study. We externally validated the ADO index for predicting 3-year mortality, with 1- and 2-year mortality as secondary end points. Discrimination was calculated using area under the curve (AUC), also known as the c-statistic, and calibration was assessed using a calibration plot with locally weighted scatterplot smoothing (LOWESS) and measures such as the calibration slope. In sensitivity analyses, we included only patients with existing COPD and those with complete data.

What we found (objective ii)

The ADO index was discriminatory for predicting 3-year mortality (c-statistic 0.74, 95% CI 0.69 to 0.79), with similar discriminatory ability for 1- and 2-year mortality (c-statistic 0.73, 95% CI 0.66 to 0.80, and c-statistic 0.72, 95% CI 0.67 to 0.76, respectively). The ADO index overpredicted mortality at each time point, which was more pronounced at 1- and 2-year mortality time points (calibration slopes 0.95, 0.79 and 0.79 for 3-, 2- and 1-year mortality, respectively) and in those with higher baseline ADO scores. Thus, although the ADO index shows promising discrimination in a primary care population, the model may need to be recalibrated if the ADO index is used to provide well-calibrated risk predictions for 1- or 2-year mortality. Discrimination and calibration were similar in sensitivity analyses.

What we did (objective iii)

To develop and internally validate a new prognostic model in primary care to predict respiratory hospital admissions within 2 years, we linked self-reported and clinical data for all patients with COPD from the Birmingham COPD cohort (331 case-found and 1558 previously diagnosed) with routine HES obtained through NHS Digital. The primary outcome for the prognostic model was the occurrence of a respiratory-related hospital admission (using primary diagnostic codes) from entry to the cohort study up to 2 years (May 2012 to June 2014). Secondary analysis considered outcomes during the full period until the NHS Digital data were obtained (1 April 2016). The maximum follow-up time was around 4 years.

A list of 23 candidate variables was drawn up based on those included in other models for COPD, along with other variables identified by a consensus panel comprising study investigators, clinicians and patients. The degree of airflow obstruction was deemed clinically important but, owing to the documented statistical problems with the commonly used FEV₁% predicted (FEV₁ as a percentage of what would be predicted as normal),75 the best variable to be included in the model was not clear. We therefore tested three variables as candidate predictors in the model [FEV₁% predicted, forced expiratory volume in 1 second quotient (FEV₁Q), and FEV₁/height2]. With 267 events for the primary outcome, up to 26 candidate variables could be used, based on the rule of thumb of 10 events per candidate variable. 90

The model was developed using backward elimination with p < 0.157 for retention. Fractional polynomials were considered and multiple imputation using chained equations was used for missing data. Discrimination was assessed using the c-statistic and calibration was also assessed. Bootstrapping was used for internal validation and the optimum-adjusted performance statistics were presented. A sensitivity analysis was conducted using only those with previously diagnosed COPD.

What we found (objective iii)

Over a median follow-up of 2.9 years (range 1.8 to 3.8 years), 382 participants (16%) had a respiratory admission and 267 (12%) had a respiratory admission in the primary 2-year period. Participants with hospitalisations were more likely to be older (70.3 vs. 67.0 years; p < 0.001), be male (66% vs. 59%; p = 0.017), be more deprived [median Index of Multiple Deprivation (IMD) score of 30.3 vs. 24.4; p = 0.0025], have a lower body mass index (BMI) (mean 28.2 vs. 28.8 kg/m²; p = 0.108), have more severe airflow obstruction (mean FEV₁% predicted 56.5% vs. 75.2%), have worse dyspnoea [Medical Research Council (MRC) dyspnoea grade 3 57.5% vs. 51%; p < 0.001] and have worse quality-of-life scores (median CAT score 24 vs. 17; p < 0.001). They were more likely to report previous exacerbations based on the use of antibiotics or steroids (60% vs. 43%; p < 0.001) and to report previous hospitalisations in the previous 12 months (18.4% vs. 2.6%; p < 0.001). Participants with hospitalisations were also more likely to report a higher rate of vapours, gases, dusts or fumes (VGDF) (71% vs. 62%; p = 0.004), to report exposure to smoking (31% vs. 27% current smokers; p = 0.019) and to have diabetes (24% vs. 15%; p = 0.001) and cardiovascular disease (20% vs. 15% with coronary heart disease; p = 0.049).

Using a pragmatic approach to model development, and including only variables that are widely available or feasible to obtain in primary care, six variables were retained in the final developed model: age, CAT score, respiratory admissions in the previous 12 months, BMI, diabetes and FEV₁% predicted. After adjustment for optimism, the primary model performed well in discriminating between those who will and those who will not have 2-year respiratory admissions (c-statistic 0.75, 95% CI 0.72 to 0.79). Four further variables were included in the secondary analysis but had similar score performance. Sensitivity analysis with prevalent COPD cases resulted in an identical apparent c-statistic but included smoking status in addition to the six variables in the primary model. Overall, the BLISS score may perform better in predicting respiratory admissions than the scores currently available, but further research is required to compare this model with existing ones in other data sets. Important next steps are external validation, proposing and evaluating a model of use to guide patient management, and exploration of the best ways to implement such a score in primary care practice.

Strengths and limitations

We used recommended and up-to-date approaches for our validation study and model development, overcoming limitations of previous studies. Using data from a research cohort (the Birmingham primary care COPD cohort) meant that measurements were of high quality and undertaken at prescribed time points. However, the cohort population may not be fully generalisable to primary care because patients with more severe disease who were housebound were excluded. Including screen-detected patients was a strength and limitation but sensitivity analyses excluding these patients did not substantially alter the findings. For model development, although the variable components for the score are relatively simple, these may not be routinely collected or available, and calculation of the score requires software.

Rationale

Although the majority of people with COPD who are likely to be detected through case-finding could be offered evidence-based interventions, there are few effective interventions for those with milder disease. One intervention that has received increasing interest is exercise. Observational studies have reported an association between higher PA levels and lower morbidity92–94 across the full range of COPD severity. Exercise is the cornerstone of pulmonary rehabilitation programmes (PRPs), which have been shown to have a positive impact on COPD symptoms and prognosis. 95 However, PRP provision is limited and uptake is low. 96 To better understand the motivation for PA engagement among people with COPD in the community, we explored perceived barriers and facilitators among people with COPD using the framework from social cognitive theory.

What we did

A purposive sample of 26 patients (age range 50–89 years; men, n = 15) from the Birmingham COPD cohort study, with a range of COPD severity, was recruited to participate in one of four focus groups. Thematic analysis was undertaken to identify key concepts related to their self-efficacy beliefs.

What we found

Several barriers to and enablers of PA closely related to self-efficacy beliefs and symptom severity were identified. The main barriers were health-related (fatigue, mobility problems, breathing issues caused by the weather), psychological (embarrassment, fear, frustration/disappointment), attitudinal (lack of feeling in control of their condition, disregard of PA benefits, older age perception) and motivational. The main enabling factors were related to motivation (PA as part of caring duties, deriving enjoyment from activity or the social aspects), attitudes (positive view of PA), self-regulation (e.g. keeping to a routine) and performance accomplishments (sense of achievement in fulfilling personal goals). This information can help to tailor management of people with COPD.

Strengths and limitations

The use of social cognitive theory in this study was novel, and allowed the identification of personal barriers related to perceptions, motivation and attitudes towards physical activity, which went beyond the external barriers identified in previous studies. This understanding can inform interventions that have the potential to improve attendance and adherence. Furthermore, by including distinct subgroups of patients, we identified context-specific factors, such as barriers specific to those who are in paid employment, thereby informing the tailoring of future interventions.

However, the study participants predominantly had mild to moderate COPD and so the findings may not reflect the views of those with more severe disease. Furthermore, the voluntary nature of participation may have meant that the views of some who were less interested in PA were not included. The emergence of themes may also have been influenced by the use of social cognitive theory, and potentially different themes might have dominated had a different theoretical framework been used.

Cohort data used for analyses leading to a PhD thesis: Buni97

Collaboration with Professor Mike Thomas from the University of Southampton: Brien et al.100

This analysis explored demographic factors, lung function/COPD-related symptoms and psychosocial/behavioural factors associated with quality-of-life impairment (using COPD CAT scores) in people with COPD. In a multivariable model, we showed that dyspnoea, illness perception, dysfunctional breathing symptoms and depression explained most of the impairment in quality of life. Thus, interventions targeting psychological factors could improve outcomes in people with COPD.

Linked trial funded through the NIHR National School of Primary Care Research: Jolly et al.101

This trial was a modification of a WP in the original programme grant, with additional funding. Overall, 577 people with earlier-stage COPD (MRC dyspnoea grade 1 or 2) were recruited from general practices (2014–15). Participants were randomised to a nurse-delivered telephone health coaching intervention (smoking cessation, increasing PA, medication management and action-planning) or usual care. Compared with usual care, participants in the intervention group reported significantly greater PA at 6 months.

Dickens et al.102

We obtained additional funding from the National Institute for Health Research (NIHR) School of Primary Care Research to undertake a linked study to assess the accuracy of microspirometry as a screening tool for undiagnosed COPD. The relevant measurements were undertaken during cohort participants’ visits. We compared lung function measures obtained from the Vitalograph^® (Vitalograph Ltd, Buckingham, UK) lung monitor with post-bronchodilator confirmatory spirometry. We found that the optimal cut-off point for the lung monitor was a FEV₁/FEV₆ of < 0.78, resulting in sensitivity of 82.8% (95% CI 78.3% to 86.7%) and specificity of 85.0% (95% CI 79.4% to 89.6%).

Cohort data used by Master of Public Health students: Khan et al.103

In this analysis, the extent of self-management behaviour and support reported by cohort participants was described. The majority of 1078 responders reported taking medications as instructed and receiving annual influenza vaccinations. However, only 40% had self-management plans and half reported never having received advice on diet/exercise. Fewer than half of current smokers had been offered help to quit in the previous year. Having a self-management plan was associated with better medication adherence and better disease knowledge.

Cohort data used as part of a PhD thesis: Kosteli104

One chapter is dedicated to the focus groups exploring the views of COPD patients on PA.

Rationale

Among those with COPD in the UK, approximately 40% are below retirement age; of these, 25% are not able to work. 107 Among those who continue to work, COPD is likely to affect work capability through sickness absence (9% of all certified absences) and working while unwell (presenteeism). 108 Data from other countries suggest that people with COPD (including undiagnosed109 and mild disease110) have a poorer employment history and retire earlier than people with normal lung function,111 but there were no data quantifying this in the UK, and no studies to examine presenteeism or productivity among working adults with COPD. Indirect societal costs attributable to COPD (largely owing to absenteeism) are high. Studies based on other conditions suggest that presenteeism costs may exceed those associated with absenteeism. 112 Few studies have examined which factors among people with COPD are associated with lower employment and work productivity. This information could inform future interventions, which could, in turn, improve patients’ work experience, thereby reducing the burden and societal costs related to COPD.

What we did

We undertook cross-sectional analysis of baseline data of patients from the Birmingham COPD cohort who were of working age. We compared the characteristics of those who were in paid employment with those who were not. Logistic regression analysis was used to assess the effects of sociodemographic, clinical and occupational characteristics on the likelihood of being employed. Using the subsample in paid employment, we examined characteristics associated with absenteeism (defined by self-report over the previous 12 months) and presenteeism (assessed using the Stanford Presenteeism Scale).

What we found

Among the 1889 people in the cohort who had COPD, 608 were of working age, of whom 248 (40.8%) were in work. Older age (60–64 years vs. 30–49 years: OR 0.28, 95% CI 0.12 to 0.65), lower educational level (no formal qualification vs. degree/higher level: OR 0.43, 95% CI 0.19 to 0.97), poorer prognostic score [highest vs. lowest quartile of modified BODE score: OR 0.10, 95% CI 0.03 to 0.33] and history of high occupational exposure to VGDF (high VGDF vs. no VGDF exposure: OR 0.32, 95% CI 0.12 to 0.85) were associated with a lower probability of being employed. Only the degree of breathlessness component within the BODE score was significantly associated with employment. Among those who were in paid employment, degree of breathlessness was the only factor associated with both absenteeism (high absenteeism in severe vs. mild dyspnoea: OR 1.84, 95% CI 0.54 to 6.27; p < 0.01) and presenteeism (working while unwell in severe vs. mild dyspnoea: OR 18.11, 95% CI 2.93 to 112.21; p < 0.01). Additionally, increasing history of occupational exposure to VGDF was independently associated with presenteeism (poor presenteeism in medium/high exposure vs. no exposure: OR 4.34, 95% CI 1.26 to 14.93; p < 0.01).

Based on these findings, future interventions should focus on managing breathlessness and reducing occupational exposures to VGDF to improve work capability among those with COPD.

Strengths and limitations

The inclusion of a wide range of patients with COPD from primary care, including case-found patients, was novel. The assessment of occupation in detail and linking with a job exposure matrix to estimate VGDF exposure was a strength. However, overall, the sample of participants in paid employment was small and the wide CIs for several estimates suggest that there was insufficient power to clarify associations. We did not have objective measures of absenteeism and some other measures were also based on self-report, which may introduce errors.

Rationale

The burden of COPD on the working population is high. The relationship between sickness and disability and unemployment is poorly understood and could be better informed by longitudinal follow-up. A number of factors, including sociodemographic characteristics, the general economic environment and the severity of chronic disease, have an impact on employment and an individual’s ability to work. We undertook longitudinal analysis to examine how disease progression is associated with occupational outcomes, adjusting for clinical, sociodemographic, occupational and labour market factors.

What we did

We used data collected during the follow-up period for those with COPD in the Birmingham COPD cohort study. Participants completed a series of questionnaires at baseline, providing information on their demographics, socioeconomic circumstances, health, lifestyle and occupation. At 6-monthly intervals they completed further questionnaires, reporting on changes in employment and, for those in paid employment, completed questions on presenteeism and absenteeism. Trained research assistants collected clinical information at two face-to-face assessments (baseline and the final follow-up), which included spirometry. We undertook longitudinal analyses, including participants who were in paid employment at baseline.

Four markers of disease progression were assessed: FEV₁ (no or limited decline vs. greater decline), number of respiratory-related hospital admissions, breathlessness (no increase vs. increase in MRC score) and symptom impact (no increase vs. increase in CAT score).

For the primary analyses, decline in FEV₁ was based on the following thresholds (comparing baseline spirometry values with those at the final follow-up): > 113.3 ml per year in men and > 90.1 ml per year in women. These were based on the upper limit of normal decline rates in FEV₁ described in healthy adults. 113 Respiratory-related hospital admissions was calculated as the number of admissions during the study follow-up (continuous data). The minimal clinically important difference (MCID) for the MRC respiratory questionnaire (1-point increase)114 was used to define worsening breathlessness, where MRC scores were compared at two time points (baseline and final follow-up for each outcome measure). Additionally, the MCID value for the CAT score was used to define worsening symptom impact. However, because this varies in the current literature,115 two analyses were conducted. The primary analysis was based on an increase of < 2 or ≥ 2 points at the final follow-up.

Multivariable regression analyses were conducted to examine the effects of disease progression on each of the outcomes, adjusting for age, sex, educational attainment, social deprivation (using the IMD derived from participant home postcode), MRC score, GOLD staging, occupational exposure to VGDF in current job at baseline visit (paid employment and presenteeism analyses) and number of hours worked at baseline (absenteeism analysis).

What we found

Among the 248 participants with COPD who were in paid employment at baseline, follow-up data were available for 174 (70.2%). Among those who were followed up, 144 (82.8%) remained in paid employment and 30 (17.2%) who had initially been in paid employment became unemployed. The mean length of follow-up was 25.8 months [standard deviation (SD) 5.8 months].

Our point estimates were suggestive of an association between increasing number of respiratory-related hospital admissions (OR 0.32, 95% CI 0.09 to 1.14; p = 0.08), decline in FEV₁ (OR 0.64, 95% CI 0.11 to 3.71; p = 0.62) and worsening MRC dyspnoea score (OR 0.62, 95% CI 0.19 to 2.06; p = 0.44) and reduced probability of remaining in paid employment, but CIs were very wide and no firm conclusions were possible. We found no associations between worsening symptom impact (CAT score) and reduced probability of remaining in paid employment.

Prospective absenteeism data were available for 113 (59.8%) participants, with a mean length of follow-up of 19.5 months (SD 5.3 months). Among this group, 63 (55.8%) reported taking > 1 day off over the follow-up period. Absenteeism ranged from 0.5 to 180.0 days per year [mean 16.3 days off per year (SD 28.9 days off per year); median 6.0 days off per year (IQR 2.7–5.5 days off per year)]. In the total cohort population, the mean and median days off per year were 9.0 (SD 23.0) and 1.5 (IQR 0.0–7.0), respectively.

Worsening breathlessness [incidence rate ratio (IRR) 3.06, 95% CI 1.29 to 7.26; p = 0.01] and respiratory hospital admissions (IRR 2.01, 95% CI 1.09 to 3.69; p = 0.03) were associated with an increased risk of sickness absence duration, and point estimates suggested that worsening symptom impact (CAT score) might also have an effect, although CIs were wide. Associations between FEV₁ decline and sickness absence duration were not observed.

Follow-up data on presenteeism were available for 163 (86.2%) participants, among whom 43 (26.4%) had worsening presenteeism. Worsening presenteeism was significantly associated with worsening CAT score (OR 5.74, 95% CI 1.18 to 27.83; p = 0.03) and may be associated with worsening MRC dyspnoea score (OR 2.25, 95% CI 0.76 to 6.68; p = 0.14). No strong evidence of any patterns were observed between FEV₁ decline or respiratory-related admissions and worsening presenteeism.

In summary, disease progression, characterised by a greater number of respiratory hospital admissions (proxy for severe exacerbations) and worsening symptoms, may be associated with poorer work productivity. We did not find any associations between physiological decline, measured by increase in airflow obstruction, and occupational outcomes. Given the wide CIs, further research is needed to increase power to detect genuine associations.

Strengths and limitations

Although participants were drawn from a population with a wide range of sociodemographic, clinical and occupational characteristics, the analyses were based on a small sample size, resulting in low power and lack of precision for many estimates. The observed associations raise hypotheses for future research and have to be interpreted with caution. The length of follow-up (2.5 years on average) may be insufficient for observing a decline in lung function, and cut-off points for abnormal decline are not agreed. The study included an older working population and, therefore, a healthy worker survivor effect might apply.

Rationale

A UK government report focused on the need to support individuals of working age to remain in or to return to employment from sickness absence. 117 The report made a number of recommendations, including the need for early workplace interventions, improved access to OH services and changes in sickness certification from ‘sick’ to ‘fit’ notes. 118 The feasibility and effectiveness of early workplace interventions to support people with COPD with poor work performance has not been assessed.

What we did

Within the Birmingham COPD cohort, we invited all those who were in paid employment at baseline to a tailored assessment with an OH practitioner to explore and identify workplace factors that might negatively affect their work performance or exacerbate their condition, and to recommend appropriate modifications. Participants’ self-management practices were also assessed. Recommendations were sent to the participant and, with their permission, to their GP and their employer. We examined acceptability (uptake of intervention and recommendations and exploration of participant views) and feasibility (proportion with recommendations and uptake) of the intervention.

What we found

Only 35 (11.3%) eligible patients agreed to take part; 109 (35.3%) declined and 153 (49.5%) did not respond. The main reasons for declining to take part included perceived lack of need (n = 54; 49.5%), had already made workplace adjustments (n = 8; 7.3%) and concern about employer involvement (n = 5; 4.6%). Most of those who took part (n = 28; 80.0%) required at least one OH recommendation and all required and received self-management recommendations. The most common OH recommendations were to modify working practices and to seek advice from the workplace OH department (or GP for those with limited access to OH services) about their respiratory symptoms in the work environment. However, only 28 out of 75 (37.3%) recommendations were reported as implementable by the interviewed participants.

Overall, the very low uptake rates for the intervention and low implementation of recommendations suggest that, in its current format, this is not a worthwhile intervention. In particular, participants were hesitant about employer involvement. Nevertheless, the finding that modifiable workplace adaptations and self-management actions were identified for almost all participants suggests that there may be benefit from such assessments to be undertaken in a different context.

Strengths and limitations

Although this was a novel intervention, the main limitation was the small sample size. Therefore, any patterns assessed among those who received recommendations should be interpreted with caution. Although the absence of randomisation and a control group makes it difficult to draw conclusions about the impact of the intervention on patients, the purpose was to explore feasibility. It was not possible to involve any employers in the study. Therefore, uptake of recommendations from the employer’s perspective is unknown.

Patient advisory group

Patient involvement has been a central part of this programme from the planning stage. From the start, we involved a patient with COPD (Michael Darby) who was the former chairperson of the Birmingham ‘Breathe Easy’ patient group and has advised on previous research studies. In discussion with him, we planned to set up a small panel of patients [the patient advisory group (PAG)] chaired by MD. MD was also invited to the external programme steering committee meetings and wider investigator meetings.

Prior to submission of the proposal, MD contributed to the programme plans by advising on the following aspects:

• the need for clear and simple patient information leaflets, with a suggestion that these are read through and commented on by the PAG

• the potential problem of incomplete response rates to mailed questionnaires and the suggestion of obtaining data through additional sources if possible (leading to us supplementing data collection by use of EHRs)

• highlighting potential ill health among the PAG and suggesting a deputy chairperson to support the role.

The PAG was formed once the programme started and five individuals with COPD were appointed. Over the course of the programme, the group met five times, with additional support in between to comment on documents and meetings with other researchers to discuss new project ideas. Each meeting was attended by at least three PAG members and lasted for around 2 to 3 hours. A charter with terms of reference was drawn up and agreed, and PAG members were provided with some initial training at the first meeting.

The PAG provided input to the following aspects of the programme:

• commenting on and modifying wording for the patient information leaflets

• suggestions for improving the wording and flow of questionnaires and how these were grouped

• piloting completion of questionnaires to provide an estimate of timing

• piloting the patient assessment process and advising on how to organise the assessments

• advising on practical issues that patients would face when attending assessments and how to support patient attendance

• ensuring that we included an opt-out process for patients who may not want their data accessed through the practice as part of the TargetCOPD follow-up study.

At each meeting, PAG members received an update on the programme and the findings so far and had an opportunity to reflect and comment. Mostly, this resulted in general approval and support of the findings, with no modifications.

The PAG has been consulted about the dissemination of findings and has suggested that an opportunity to invite patients to an open evening with interesting speakers, reports on the findings and a social element (preferably in spring) would be welcome.

Multistory

During the course of the programme we were approached by an arts-based charity, Multistory (URL: https://multistory.org.uk; accessed 31 August), who had commissioned an artist to undertake a project (Black Country Lungs) to describe the story of people with COPD in the Black Country. 120 The charity approached us based on the BLISS research programme from a website search. They used a discussion on the various themes in the programme and our findings as a basis for exploring ideas with patients who took part in their project.

In addition to their main exhibition, a preview of artwork was displayed at the University of Birmingham Science and Art Exhibition, where members of the public and participants from our research study were invited. Our research group was involved in a panel discussion at which we summarised the findings from our programme. 121

Screening for undiagnosed chronic obstructive pulmonary disease

Underdiagnosis of COPD is well recognised worldwide, with at least half of patients thought to be undiagnosed. 122 Although there is some evidence that, overall, these individuals consume health services to an equivalent degree to those with diagnosed disease,123 there is insufficient evidence that screening for undiagnosed COPD is worthwhile. 46,124,125 One common uncertainty is the lack of trial evidence that the early identification of COPD leads to clinical benefits.

Our systematic reviews of studies of case-finding have shown that targeting screening to a higher-risk group is important126 and identified which of the currently tested screening strategies is/are most effective in increasing yield. 51 The TargetCOPD trial demonstrated that case-finding for undiagnosed COPD is a cost-effective process for increasing the number of new COPD cases identified with much higher yield than routine practice, particularly if using an active, targeted approach. 21 Furthermore, we have used data from the trial to develop alternative algorithms for case-finding that may improve the efficiency of the process further by reducing the number needed to screen. 49

The majority of newly identified cases in the TargetCOPD trial had potential to benefit from evidence-based recommended treatments. Using the best available evidence from the literature for our assumptions, our economic model demonstrated that a systematic, 3-yearly screening programme is likely to be cost-effective, with favourable cost/QALY gained in all scenarios (paper under review). We also found that case-finding was generally acceptable to patients42 and primary care staff,43 although a lack of awareness in relation to symptoms (among patients) and approaches to effective management of early disease (among health-care staff), as well as limited resources to deal with increasing numbers of people with COPD, were highlighted by both groups.

Longer-term follow-up of patients who were case-found through our trial demonstrated that just over one in five were added to the practice COPD register and, even among these patients, guideline-recommended management was rarely administered. This may be related to the concerns raised in our qualitative study around low awareness of management of early COPD, lack of resources and perceived lack of effective strategies.

After 4 years’ follow-up, we demonstrated no evidence that screening had an effect on clinical outcomes. Hospitalisation and mortality rates did not significantly differ between patients in the case-finding or routine care practices. Although respiratory hospitalisation rates were significantly higher in the active than in the opportunistic case-finding arm, there was no significant difference between groups in overall hospitalisation or mortality. This may be related to more hospitalisations being attributed to a respiratory cause in these patients who had a COPD diagnosis.

This finding may seem to contrast with the findings from our Markov model, which suggest that estimates of cost-effectiveness are robust as long as 8% of screen-detected patients are optimally treated. However, considering the small number of case-found patients added to the COPD register, the low levels of implementation of treatment and the probably reduced benefits of such treatment among more mildly affected patients could take the effective benefit of case-finding in our patient population below 8%.

Multidimensional prognostic model for chronic obstructive pulmonary disease

People with COPD have high risk of respiratory infections resulting in hospitalisation and also a higher risk of premature mortality than those without COPD. However, COPD is heterogeneous and not all patients progress in the same way. Traditionally, lung function has been used to grade COPD severity, but there is increasing evidence that prognosis is determined by an inter-related set of factors, which has led to the development of multidimensional prognostic models. 127 Existing models have predominantly been developed in secondary care populations and are mainly developed to predict mortality risk; few have been developed using high-quality statistical approaches or been externally validated, and they often do not perform better than measures of lung function alone. 78,127,128 The most recent NICE guidelines31 highlight the need for prognostic tools that are validated in UK primary care COPD populations and that examine outcomes wider than just mortality.

We externally validated the ADO score, which was identified as the most discriminatory for 3-year mortality in a systematic review, in a UK primary care population. 78 Although we found that it has promising discrimination, the model needs to be recalibrated if used to predict risk of mortality within 1 to 2 years.

We also developed a new prognostic index to predict 2-year risk of hospitalisation for people with COPD in primary care. Our new BLISS model, which includes variables that are easily available in primary care settings, has promising discrimination and was adjusted for overfitting to help ensure calibration is more reliable in case-found individuals.

Further work is needed to externally validate the BLISS prognostic index. Further research is also needed to evaluate the use of the index to classify people with COPD into higher- and lower-risk groups to aid management decisions.

Occupational outcomes in chronic obstructive pulmonary disease

A high proportion of people with COPD are of working age, but they have poorer rates of employment and poor work productivity compared with those without COPD. To our knowledge, this was the first study to examine which factors are associated with occupational outcomes among people with COPD. We found that increasing breathlessness, disease progression (increasing number of respiratory hospital admissions rather than lung function decline) and greater occupational exposure to VGDF were associated with poorer work productivity. These disease-related factors were more likely to be associated with poor work productivity than sociodemographic and lifestyle factors. Our findings suggest that there may be a continuum from presenteeism to absenteeism to loss of employment.

Although our OH-focused intervention was not feasible, modifiable workplace adaptations and self-management actions were identified for almost all participants, suggesting possible benefit from such assessments in a different context.

Further research is needed to examine how presenteeism and absenteeism are related to better understand presenteeism in relation to health and to assess whether or not interventions to modify the course of COPD have an impact on occupational outcomes.

Screening

1. Development and evaluation of interventions to better implement effective treatments for COPD in primary care, including pathways to manage case-found COPD.

2. Evaluation of existing interventions in case-found COPD to determine if the interventions have similar effectiveness.

3. Long-term follow-up of TargetCOPD participants to establish whether or not clinical benefits might occur, given that people with case-found disease had milder symptoms and better lung function than those already on GP practice COPD registers. Development and evaluation of approaches for increasing uptake of invitation for screening for undiagnosed COPD.

4. Development of more efficient approaches to case-finding and identifying which screening test or strategy has the best performance (in terms of sensitivity and specificity).

5. Development and evaluation of different models for delivery of quality-assured diagnostic spirometry screening services, considering workforce implications and how this could be incorporated into the new early diagnostic hubs for primary care networks proposed in the long-term plan.

Prognosis

1. Description of natural history and prognosis of case-found patients and those with indicative symptoms but normal lung function, describing any heterogeneity, and establishing whether or not there are phenotypic characteristics that are associated with progression.

2. Exploration of how prognostic models might be used by primary care staff in managing COPD patients.

3. Evaluation of the BLISS prognostic model in directing patient management.

4. Validation of the BLISS case-finding algorithm.

5. External validation of the impact of the BLISS prognostic model.

6. Consideration of the development of new prognostic scores for primary care COPD patients that predict all-cause (rather than respiratory) hospitalisation, given the multimorbid nature of the condition.

Work-related impacts

1. Examination of the relationship between presenteeism and absenteeism and whether or not presenteeism has an impact on health-related outcomes.

2. Development and evaluation of interventions to reduce dyspnoea and VGDF exposure in occupational outcomes in people with COPD.

3. Codevelopment and evaluation of an OH intervention to improve work productivity among people with COPD who are in paid employment.