Improved outcomes in ex-smokers with COPD: a UK primary care observational cohort study

Discussion

Our objectives were twofold: to quantify smoking status in a large and representative clinically defined UK primary care COPD cohort and to explore the relationship between smoking status and clinical outcomes.

We were able to define smoking status in 91.3% of our cohort and have shown that over one-third continue to smoke, with the highest prevalence of active smokers among younger patients, consistent with other reports [20]. After adjusting for confounding factors (including age, disease severity and inhaled medication) ex-smokers had a significantly reduced risk of all three adverse outcomes: death, unplanned respiratory-cause hospital admissions and ED attendances, with an estimated reduction in these outcomes of 14.6%, 11.8% and 14.6% respectively if all smokers quit. In never-smokers, mortality risk was reduced, as others have reported [21], but for hospitalisation the adjusted effect estimates were consistent in size and direction with those in ex-smokers, providing weaker evidence of reduced risk. Failure to reach significance at the 5% level is probably because the number of never-smokers was small, but might also be because they represent an especially heterogeneous group.

We acknowledge the limitations and biases inherent in observational studies using routine clinical data that have not been collected to answer specific research questions and where factors motivating data recording vary. We report our findings in accordance with the RECORD statement [15]. Although not all Hampshire practices submit data to HHRA, missing practices are dispersed across the catchment area, with varying rural/urban classification, socioeconomic deprivation and patient composition. We are not aware of any systematic differences between missing practices and those whose data are represented.

There are problems in accurately identifying diseases using electronic health records, since diagnostic codes are used inconsistently in primary care, without quality control at the point of data entry [22]. However, use of electronic templates, common in UK general practice (and widely used in COPD), reduces missing data and encourages correct code use. Quint et al. [23] have shown that COPD patients can be reliably identified from routine health records using diagnostic codes, with little additional benefit from including spirometry. All our code lists were created by clinicians familiar with the Read code system in routine practice. Our cohort-defining codes included designated codes in the UK “Quality and Outcome Framework” (QoF) (www.hscic.gov.uk/qof) incentive scheme that define COPD for performance-related pay calculations and have been widely used since the scheme began in 2004. COPD is underdiagnosed, but the size of our prevalent cohort (1.2% of HHRA population) is consistent with reported prevalence rates in England [24], so it is likely to be representative of the UK diagnosed COPD population.

Smoking status is now widely recorded in UK primary care [25]. Inconsistent recording may occur, relating to coding, to different interpretations of the terms “current non-smoker”, “never”, “trivial” and “ex”-smoker in routine practice and because reports of quitting are often exaggerated [26]. We used the last smoking code prior to the study start, to improve the validity of any relationship with outcomes, and have demonstrated that these data were recent. A limitation of our data source is that we were unable to quantify smoking exposure (e.g. pack-years) in most of our patients, or the time since quitting, factors shown to be important in affecting outcomes [27]. Pack-year data were not included in our primary analysis plan, because we expected these to be infrequently and inconsistently recorded in routine primary care. A supplementary “post hoc” analysis supports this view, in that pack-year data were only recorded in 15% of patients and showed no relationship to outcomes (see supplementary material). Reported ex-smoker outcomes also depend on how long subjects are required to have not smoked to be considered ex-smokers [28]. Our data do not allow calculation of time since quitting and estimates of a minimum quit time would be problematic in relation to time-to-event analyses, because estimates would include left-censored covariates. Nevertheless, despite these limitations, we have demonstrated significant and clinically important differences in outcomes between active and ex-smokers.

We may have underestimated the size of the association between smoking status and outcomes because of selective survival bias in our prevalent cohort. Active smokers in the cohort may be healthier than those who died before cohort entry; however, ex-smokers (the main comparator group) might also exhibit this effect, although perhaps to a lesser degree. The extent of survival bias because of differing disease duration may only be modest, since the median time from COPD diagnosis to study start for ex-smokers was only 13 months more than for active smokers. Greater smoking exposure and reduced time since quitting might both increase mortality prior to cohort entry. We were limited in not having smoking cessation date or pack-year history, with which we could have introduced time-varying covariate information to reduce the extent of potential survivorship bias.

Missing data are an important limitation of any study using routinely collected health data. We used multiple imputation to impute missing data among the adjustment variables used in our Cox regression analyses. This depends on the assumption that imputed data are “missing at random” (MAR), conditional upon the presence of certain variables. We have attempted to satisfy this condition in our imputation model, but acknowledge that MAR is an essentially untestable assumption. Baseline data used to characterise our cohort reflect the most recent values recorded prior to the study start. Later data were not included but, by the end of the study, it was noted that missing data for lung function and BMI had halved and smoking status was missing in only 2.6%, suggesting better recording within primary care.

The disparity between our univariate and multivariate outcome analyses highlights the fact that observational data need to be interpreted with care, because of potential confounding factors. Unadjusted outcomes in ex-smokers were confounded by factors such as age and severity but, in adjusted analyses, smoking cessation was independently associated with both reduced mortality and hospitalisation. Our findings are consistent with a recent Italian study (using life tables and smoking questionnaire data) which showed that quitting smoking confers mortality benefits at all ages, with a one-third reduction in the risk of dying from COPD in those aged 70–74 years [29]. Similarly, a large Australian study showed reduced hospitalisation because of COPD within 5–14 years of stopping smoking, even at older ages [27]. Reduced hospitalisation has been demonstrated in both early and sustained quitters [11], but other studies report increased admission rates in recent quitters (less than 5 years) but reduced respiratory symptoms in sustained quitters [30] and a reducing risk of exacerbations with longer periods of abstinence [9].

The prevalence of never-smokers in our clinically defined cohort is considerably lower than in population-based cross-sectional surveys of subjects whose spirometry met criteria for COPD [31, 32]. Many factors cause COPD worldwide [33], but clinical experience in the UK suggests that COPD in a non-smoker is uncommon and indicates the need for diagnostic review and consideration of asthma or rare diseases such as α₁-antitrypsin deficiency. However, one explanation for low numbers of never-smokers on practice COPD registers is that the diagnosis is less likely to be considered, because of the disease's clear association with smoking. In this study, we re-classified, as ex-smoker, those patients whose latest code was “never-smoker”, but whose earlier codes indicated a past history of smoking. We believe this strategy improved the overall reliability of our never-smoker definition, though potentially overcompensating if earlier smoking history was incorrect.

A strength of HHRA is that it links routinely collected patient-anonymised data from primary and secondary care, providing more reliable estimates of secondary care contacts than most UK primary care datasets, in which hospitalisation data rely on coding in primary care records after discharge [34]. We were able to use hospital-coded discharge data for the entire cohort and used “primary” diagnoses to quantify respiratory-cause admissions. A systematic review has shown primary diagnosis coding accuracy of 96.0% (IQR 89.3–96.3), concluding that “routinely collected data are sufficiently robust to support their use for research and managerial decision-making” [35]. In quantifying hospitalisations, we have not made any distinction between patients with one or more episodes, mainly because of the low numbers of patients with more than two. We report all-cause mortality, as we do not have access to certified cause of death.

We believe that our results are generalisable to a wider UK population. HHRA holds data for over 1.4 million patients, representing >75% of Hampshire's population. COPD prevalence, healthcare and outcomes (hospitalisation and mortality) appear to be “average” when Hampshire is compared with the rest of England, although considerable variation has been demonstrated within Hampshire [24]. Hampshire may not be representative of large inner cities with marked ethnic diversity; indeed, 89% of its inhabitants were White British in the 2011 census [36]. However, there is heterogeneity across the county, with urban areas having greater ethnic diversity, especially Southampton, where the proportion of White British is 77.7%, compared with a national average of 80.5%. Ethnicity was poorly recorded in our dataset but our cohort encompassed a broad socioeconomic spectrum, as reflected in the range of IMD scores (IQR 3–8, median 6) when these weighted standardised measures of socioeconomic status were ranked according to national deciles.

In conclusion, we used a UK primary–secondary care linked database to define a cohort of >16 000 patients with a clinical diagnosis of COPD in whom we assessed smoking status and its relationship to major clinical outcomes over 3 years. Our findings suggest that 14.6% of deaths and respiratory-cause ED attendances and 11.8% of respiratory-cause admissions could be avoided if all smokers quit. Over one-third of this cohort continues to smoke, as do the majority of younger patients, who have the most to gain by stopping. Our findings could be used to encourage patients to quit. Smoking cessation should be top priority in COPD [37, 38], but worrying evidence suggests that fewer people are seeking specialist services [39], despite advances in the science and delivery of smoking cessation support [40, 41]. While funding cuts have led to reduced provision of these services, our findings emphasise the need for continued investment to improve patient care and reduce pressures on health services.