Characterising undiagnosed chronic obstructive pulmonary disease: a systematic review and meta-analysis

Background A significant proportion of patients with chronic obstructive pulmonary disease (COPD) remain undiagnosed. Characterising these patients can increase our understanding of the ‘hidden’ burden of COPD and the effectiveness of case detection interventions. Methods We conducted a systematic review and meta-analysis to compare patient and disease risk factors between patients with undiagnosed persistent airflow limitation and those with diagnosed COPD. We searched MEDLINE and EMBASE for observational studies of adult patients meeting accepted spirometric definitions of COPD. We extracted and pooled summary data on the proportion or mean of each risk factor among diagnosed and undiagnosed patients (unadjusted analysis), and coefficients for the adjusted association between risk factors and diagnosis status (adjusted analysis). This protocol is registered with PROSPERO (CRD42017058235). Findings 2,083 records were identified through database searching and 16 articles were used in the meta-analyses. Diagnosed patients were less likely to have mild (v. moderate to very severe) COPD (odds ratio [OR] 0·30, 95% CI 0·24-0·37, 6 studies) in unadjusted analysis. This association remained significant but its strength was attenuated in the adjusted analysis (OR 0·72, 95% CI 0·58-0·89, 2 studies). Diagnosed patients were more likely to report respiratory symptoms such as wheezing (OR 3·51, 95% CI 2·19-5·63, 3 studies) and phlegm (OR 2·16, 95% CI 1·38-3·38, 3 studies), had more severe dyspnoea (modified Medical Research Council scale mean difference 0·52, 95% CI 0·40-0·64, 3 studies) and slightly greater smoking history than undiagnosed patients. Patient age, sex, current smoking status, and the presence of coughing were not associated with a previous diagnosis. Interpretation Patients with undiagnosed persistent airflow limitation had less severe airflow obstruction and fewer respiratory symptoms than diagnosed patients. This indicates that there is lower disease burden among undiagnosed patients compared to those with diagnosed COPD, which may significantly delay the diagnosis of COPD. Funding Canadian Institutes of Health Research. Declaration of interests We declare no competing interests. Author Contributions MS, SB, and KJ formulated the study idea and designed the study. KJ and SG performed all data analyses and MS, SB and DS contributed to interpretation of findings. KJ wrote the first draft of the manuscript. All authors critically commented on the manuscript and approved the final version. MS is the guarantor of the manuscript.

Evidence before this study 3 6 Many cross-sectional prevalence studies have compared the characteristics of patients 3 7 with persistent airflow limitation but no prior diagnosis of COPD ('undiagnosed') to 3 8 those with persistent airflow limitation and a diagnosis of COPD ('diagnosed'). We 3 9 searched MEDLINE and EMBASE for observational studies published in English 4 0 between January 1, 1980 and April 11, 2017 that assessed diagnosis status among adult 4 1 patients with spirometrically defined persistent airflow limitation. We used search terms Introduction 8 2 Chronic Obstructive Pulmonary Disease (COPD) is an inflammatory lung disorder that is 8 3 characterised by persistent airflow limitation 1 and associated with symptoms of shortness 8 4 of breath, cough and sputum production. 2 Patients with COPD generally seek medical their risk factors, respiratory symptoms, and disease stage influence the likelihood of We conducted a systematic review and meta-analysis to compare patient characteristics, analysis of individual data. We did not include conference abstracts unless they met the 1 1 4 inclusion criteria and provided the required information, and we did not assess grey 1 1 5 literature. We extracted summary data from the eligible articles and contacted the authors 1 1 6 to obtain additional information when required (one author group provided us with 1 1 7 additional information). Title and abstract screening were initially performed, followed 1 1 8 by full-text analysis to determine article eligibility. We extracted data using a customized 1 1 9 Excel spreadsheet after the eligible articles had been compiled. KJ initially performed the 1 3 8 prior diagnosis of COPD or an obstructive lung disease (emphysema, chronic bronchitis, 1 3 9 asthma) from a health-care professional were considered to have 'diagnosed' COPD, whereas those with persistent airflow limitation but without a prior health professional diseases were excluded. We included studies that sampled patients from any population 1 4 3 or health-care setting. Given the exploratory nature of the observational studies included in this review, we used 1 4 5 a broad definition of risk factors that included any observable factor that could be 1 4 6 associated with the probability of having received a diagnosis of COPD. Risk factors 1 4 7 included patient-reported respiratory symptoms (cough, wheeze, phlegm, dyspnoea), sex, We extracted summary data from each eligible article, which included study 1 5 3 characteristics, the definition of persistent airflow limitation that was employed in each of 1 5 4 the studies, the method of COPD diagnosis, and sample size. We also extracted the 1 5 5 proportion or mean of risk factors between the diagnosed and undiagnosed groups, as 1 5 6 well as the odds ratios (ORs) and their confidence intervals in studies that used regression 1 5 7 modelling to assess the independent impact of the risk factors on diagnosis status. The 1 5 8 protocol for this study is registered on the PROSPERO register of systematic reviews 1 5 9 (CRD42017058235). 10 Data analysis 1 6 1 We used data extracted from articles measuring categorical data to generate ORs and standard errors for the association between risk factors and the probability of having Otherwise, we used random-effects models. We quantified heterogeneity between studies using the I 2 statistic. 14  largest sample size. We conducted a sensitivity analysis to determine the association 1 7 6 between the risk factors and COPD diagnosis only among population-based studies 1 7 7 (those based on random sampling of the general population as opposed to convenience  Role of the funding source The funder of this study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author and co-authors had full 1 8 2 access to the data in the study and take responsibility for the integrity of the data, the  The search resulted in 1,857 references after excluding duplicates. 1,788 references were excluded by screening their titles and abstracts, and 69 remained for full text review to 1 8 7 determine eligibility. A total of 18 articles met the inclusion criteria following the 1 8 8 screening process, but only 16 articles were included in quantitative synthesis ( Figure 1).

8 9
The overall agreement between reviewers was high (90%). A summary of the 18 eligible articles is presented in Table 1. However, two eligible articles were excluded from the meta-analysis because they were missing the necessary majority of the 18 eligible articles were cross-sectional (n=16), and were population- based (n=10). Other studies sampled patients from primary care clinics (n=4), 1 9 6 hospitalized patients (n=3), or participants in a smoking cessation program (n=1). Studies 1 9 7 originated from Latin America (n=6), Europe (n=6), Canada (n=2), and Asia (n=2). Data Burden of Obstructive Lung Disease (BOLD), 4,22 were used in three, four, and two  The quality of the 18 eligible articles was variable. Half of the studies were assigned a were deemed poor in quality. Studies that were not assigned a 'good' quality rating generally had a primary study focus that was not our question of interest. For example, tangentially in five studies, and disease severity was the only factor that was compared studies that used regression modelling, the risk factors that were adjusted for varied 2 1 7 substantially.

1 8
Unadjusted analysis 2 1 9 Comparisons of the characteristics of diagnosed and undiagnosed patients with persistent pooled results from studies that used this definition are reported in the main text and symptoms, current smoking status, smoking history, and COPD severity among patients  Patients with 'diagnosed' COPD were more likely to be experiencing respiratory likely to have mild (grade I) COPD than moderate to very severe COPD (grade II-IV) as between studies was relatively low (I 2 <35.0% for wheeze, phlegm, dyspnoea, any symptoms, and COPD severity); however, the I 2 statistics should be interpreted smoking status, and smoking history were not associated with 'diagnosed' COPD. Having a cough was also not significantly associated with diagnosis status, however  results (Appendix, Figure A2); however, cough was marginally associated with diagnosis 2 4 4 status in this analysis (OR 1·65, 95% CI 1·02-2·66). Similarly, patients with 'diagnosed' COPD (fixed ratio definition) were more impaired by -12·85%, 95% CI -15·26% to -10·44%, 4 studies) than undiagnosed patients. Patients with 'diagnosed' COPD also had a slightly greater smoking history (pack-years MD 8·39, 95% CI 0·68-16·44, 4 studies); however there was high variability between the study 2 5 2 means (I 2 84·2%). There was no difference in mean age between diagnosed and 2 5 3 undiagnosed patients. Adjusted analysis 2 5 6 Articles using regression modelling to assess the independent impact of risk factors on COPD diagnosis ('adjusted analysis') were pooled by risk factor type, and the results are presented in Figure 4 for the fixed ratio definition of persistent airflow limitation (5 a weak independent association with the diagnosis of COPD (OR 1·16, 95% CI 1·00- 1·35, 2 studies) using the fixed ratio definition. The presence of wheezing (OR 1·20, 95% independently associated with a diagnosis. In contrast, mild COPD (GOLD grade I OR 2 6 5 0·72, 95% CI 0·58-0·80) or moderate COPD (GOLD grade II, OR 0·71, 95% CI 0·58- 0·86), were independently associated with a lower likelihood of diagnosis, compared with severe or very severe (reference GOLD grades III-IV). Sex and the presence of cough did number of studies were small and heterogeneity in the effect estimates between studies was very high (I 2 >70.0% for all risk factors except sex). Three risk factors were pooled in our assessment of studies using adjusted analysis based  (I 2 75·2%). Patient sex and the presence of cough had no independent effect. The presence of respiratory symptoms and GOLD 3 or 4 disease severity was strongly (based on contingency tables), and mean percent predicted FEV 1 was 13% lower in 2 8 7 diagnosed than undiagnosed patients. Disease severity was also the only risk factor that modelling) analyses. In the adjusted analysis, patients with moderate COPD were 29% 2 9 0 less likely to have received a diagnosis than patients with severe or very severe COPD. Respiratory symptoms were another group of risk factors that were correlated with a 2 9 2 COPD diagnosis. Among respiratory symptoms, the presence of dyspnoea was the most 'diagnosed' COPD scored 0·52 points higher on the mMRC dyspnoea scale. However, 2 9 5 there was only one study 25 in which the mean score on the mMRC scale could have been 2 9 6 used to distinguish undiagnosed from diagnosed patients using commonly accepted Following dyspnoea, the presence of wheeze, and phlegm was also strongly associated 2 9 9 with 'diagnosed' COPD in the unadjusted analysis. However, in the adjusted analysis, (discussed in detail below), our findings suggest a strong association between the with respiratory symptoms being more likely to seek care, current guidelines now Patient characteristics such as sex and age were not associated with an increased 3 1 0 likelihood of having received a diagnosis in any of the pooled analyses. There was some 3 1 1 indication that patients with 'diagnosed' COPD had a greater pack-year smoking history, 3 1 2 although current smoking status and smoking history were not statistically significant when they were assessed as the presence of former smoking and never smoking. The effects of risk factors on the likelihood of being diagnosed were weaker in the expected to reduce the effects sizes compared to odds ratios derived from contingency results that received disproportionate weighting. In contrast to all other studies in this review, Herrera et al. 23 found that respiratory symptoms were not associated with the were pooled with one other study 17 , which found that the presence of respiratory symptoms strongly impacted the likelihood of receiving a diagnosis. This discrepancy 3 2 4 between studies may be due to differences in the population that was sampled (primary observed smaller differences between undiagnosed and diagnosed patients because they sampled from a subset of patients that were prompted to seek care because of a symptom burden. Our systematic review has several strengths. First, we used data from a total of 16 articles 3 3 0 in the meta-analysis, and these articles were mostly population-based studies that scored 3 3 1 high in quality. Second, there were a robust number of studies for many risk factors; 3 3 2 patient sex was assessed in 10 studies in total, followed by disease severity in 9 studies, 3 3 3 and respiratory symptoms and smoking history in 8 studies each. The methods used to 3 3 4 measure disease severity, respiratory symptoms, and smoking history were relatively several pooled analyses to assess the sensitivity of our findings to alternate definitions of COPD (fixed ratio and LLN) and analysis methods (unadjusted and adjusted analyses).

8
Except for one study, 23 our findings were consistent. However, our systematic review also has several limitations. First, half of the pooled  findings that vary across settings. Second, although the total number of studies for each 3 4 4 risk factor was robust, the number of studies assessing each risk factor within pooled 3 4 5 analyses tended to be small. This was partly because separate articles using the same other respiratory symptoms in the pooled analyses were measured as binary variables diagnosis, a more nuanced assessment of their severity might result in an even greater 3 5 2 ability to distinguish between undiagnosed and diagnosed patients. In addition, because 3 5 3 respiratory symptoms were self-reported in all studies, reporting bias might have 3 5 4 exaggerated the difference in symptoms between the undiagnosed and diagnosed groups.
The findings from this systematic review have important implications for research and 3 5 6 policy around COPD diagnosis, for example, in estimating the return on investment in 3 5 7 screening and case detection strategies for COPD. The true burden of COPD is the sum of the disease burden in diagnosed and undiagnosed patients, and our results indicate that 3 5 9 undiagnosed patients generally have milder disease and therefore a lower disease burden. On one hand, this indicates that strategies aiming to reduce the underdiagnosis problem diagnosis among patients that have already developed symptoms. Given the potential for 3 6 5 disease modification at early stages of COPD, reducing this delay could be associated 3 6 6 with substantial improvement in long-term patient outcomes and a reduction in mortality 3 6 7 and costs. Acknowledgements: This study was funded by the Canadian Institutes of Health Research (application number 142238).    Regression models were adjusted for age (