WHAT IS ALREADY KNOWN ON THIS TOPICWHAT THIS STUDY ADDS

We show that a widely used CVD risk scoring tool, QRISK, substantially underestimates 10-year CVD risk in COPD. We show that such an underestimation of risk is especially pronounced in younger COPD patients.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICYIntroduction

Chronic obstructive pulmonary disease (COPD) is a common chronic respiratory condition that affects over 300 million persons worldwide, is responsible for 3 million deaths per year and is the leading cause of hospitalisations globally.1 2 In addition to respiratory symptoms, the course of COPD is characterised by the presence of various comorbidities, including cardiovascular disease (CVD). Several studies have demonstrated a higher burden of CVD in patients living with COPD compared with the general population of similar age/sex distribution.3–5 CVD is a significant cause of morbidity and mortality in COPD, especially in those with milder disease.6 7

Many known risk factors are shared between COPD and CVD, such as ageing and smoking, which partly explain the coexistence of these two conditions. However, there are at least two other mechanisms by which CVD risk can be increased in COPD. First, there are likely unobserved common risk factors. Examples include shared genetic predispositions, oxidative stress, inflammation, common environmental triggers such as air pollution and viral respiratory tract infections and complex socioeconomic factors such as poverty and poor diet.8 Yet an alternative pathway connecting the two conditions is the direct effect of COPD on CVD. For example, a recent systematic review showed that people living with COPD are highly susceptible to acute cardiovascular events during and shortly following severe exacerbations.9

To what extent the COPD–CVD association is mediated by unobserved risk factors or through direct influences is of clinical importance. Primary prevention of CVD is largely based on risk calculation, with most guidelines recommending preventive therapies (eg, with statins) for those with a 10-year risk ≥7.5%–10%.10 Risk stratification is based on published risk scoring tools that are endorsed by expert panels, and many are incorporated into electronic medical records and clinical decision support systems. An example is the QRISK scoring tool, which is widely used in the United Kingdom (UK) and other countries.11 The latest version of the tool, QRISK3, has been developed based on data from 7.89 million patients from 981 general practices in the UK and has been validated among 2.67 million patients from 328 separate practices.11 It includes up to 21 predictors relating to demographics, medical diagnoses, social deprivation, blood markers and health behaviour, making it one of the most comprehensive CVD risk assessment tools currently available. The influential UK’s National Institute for Health and Care Excellence recently updated its guidelines on CVD risk assessment and reduction, strongly emphasising the use of QRISK3 for risk stratification.12

However, QRISK and other CVD risk scoring tools are developed based on data from general population. As such, they will generate valid predictions in patients living with COPD only if the entirety of increased CVD risk is captured by risk factors included in the tool. Any significant violation of this assumption will result in the underestimation of CVD risk, leading to missed opportunities to mitigate the risk of CVD in patients with COPD. To what extent commonly known predictors in contemporary CVD risk scoring tools explain the added CVD risk in COPD is currently not well understood.

To address this knowledge gap, in this study, we estimated CVD risk in a population-based primary care cohort of patients with COPD and compared it with predicted risk based on QRISK3 as an exemplary CVD risk scoring tool.11 We hypothesised that due to the contribution of several unobserved risk factors and shared disease pathways, the CVD risk in patients living with COPD is higher than in the general population, and that QRISK3 will underestimate the true risk of CVD events in patients living with COPD.

MethodsStudy design and population

We conducted a retrospective cohort study of patients with COPD from the UK Clinical Practice Research Datalink (CPRD) GOLD primary care database between 1 January 1998 and 31 July 2018 (the study period). The CPRD consists of anonymous, longitudinal data from over 14 million individuals across ~700 general practices in the UK.13 It includes data on medical symptoms and diagnoses, demographic information, specialist referrals, test results, prescriptions and lifestyle details such as smoking and alcohol consumption. As QRISK is developed based on the UK primary care data, it is expected to provide accurate predictions in the general sample from this data set. Indeed, previous versions of QRISK have been validated in CPRD, with the authors concluding that the performance of the tool was comparable with its performance in the original development sample.14

Patients who met the following criteria were included: they were permanently registered with a primary care practice, provided data for at least 1 year and had their data linked to UK Hospital Episode Statistics discharge data, the Office of National Statistics death data, and Index of Multiple Deprivations (a measure of socioeconomic status) within the study period. The data custodian, the independent Scientific Advisory Committee of the UK General Practice Research Datalink, approved our study protocol for anonymised research purposes, which required no human consent to participate.

Using a validated case definition,15 we created an incidence cohort of COPD patients aged between 40 and 84 years. The incidence cohort approach was chosen because the time of COPD diagnosis is a natural vantage point for assessing CVD risk as recommended by guidelines.16 The case definition for COPD (details in online supplemental table 1) has a positive predictive value of 86.5% in CPRD.15 The upper age bound was placed to be aligned with the inclusion criteria of QRISK3 development sample, while the lower age bound was placed to reduce the risk of including patients with asthma. To be classified as an incident case, the initial COPD diagnosis must have been made after at least 1 year of data availability, with no COPD-related health resource utilisation during this period.

The index date was the date of COPD diagnosis. COPD patients were excluded if they had a missing Index of Multiple Deprivation, pre-existing CVD or were prescribed statins before the index date, in line with the intended use of QRISK (and the exclusion criteria of its development study). Follow-up was censored at the earliest recorded date of CVD, death from any cause, emigration outside a CPRD practice site or the end of the study period (31 July 2018), whichever occurred first.

Outcomes

CVD outcomes were defined following the development approach for the QRISK3 algorithm.11 They included the first recorded episode of fatal or non-fatal coronary heart disease, ischaemic stroke, or transient ischaemic attack (as a composite endpoint). These events were determined from the GP database using Read codes (see online supplemental table 2) and from hospitalisation and mortality outcomes based on International Classification of Diseases, 10th revision codes (see online supplemental table 3).

Table 3

Predicted (by QRISK3) versus observed 10-year CVD risk and observed/predicted ratio, in total and by sex and age groups

Predictors

All predictors were ascertained according to the methodology used in the original QRISK3 development procedure (predictors are listed in table 1).11 Accordingly, we obtained the most recent values of blood pressure and smoking status recorded before the index date and all comorbidities recorded at the baseline visit. We selected the closest value to cohort entry for laboratory test values such as total and high-density lipoprotein. The use of medications (corticosteroids and second-generation ‘atypical’ antipsychotics, which are predictors in QRISK3) at baseline was defined as at least two prescriptions, the most recent being not more than 28 days before the index date. Details on the evaluations of predictors and the measurement time windows are shown in online supplemental table 4.

Table 1

Baseline characteristics of the included sample

Statistical analyses

We calculated the raw incidence rates per 100 person-years as the total number of CVD events divided by the total person-years of follow-up, multiplied by 100. To compare the CVD incidence between people with COPD and the QRSIK3 development cohort, the baseline sample of the latter cohort was used as the reference to calculate age-standardised (with 1-year bins) incidence ratios (SIR). SIRs were reported separately by sex and younger (≤65) versus older adults (65+ years of age). The chosen age cut-off was employed to facilitate meaningful comparisons of SIRs across age groups and is the cut-off frequently used to separate younger and older adults with COPD.17 18

We calculated 10-year observed risk using Kaplan-Meier analysis, in the whole cohort and in the subgroups defined by sex and age groups. We assessed the accuracy of QRISK3 by calculating the observed versus predicted 10-year CVD risk in the entire cohort and across subgroups. Aside from the Deprivation Index for which missingness was an exclusion criterion, we used multiple imputations by chained equations to impute the value of other missing predictors. We created 10 independent data sets each with randomly imputed data, performed all the analyses separately within each data set and combined the results using Rubin’s rule.19

Cohort creation and predictor and outcome extraction were conducted in SAS (V.9.4, SAS Institute, Cary, North Carolina). Statistical analyses related to missing value imputation, SIR calculation and Kaplan-Meier analyses were performed in R (V.4.2.2).20 QRISK3 R package was used to calculate QRISK scores.21 The statistical code is available from https://github.com/resplab/papercode/tree/main/COPD_CVD.

Results

The original data set included 29 605 patients with obstructive airway disease, of whom 13 208 satisfied the case definition of COPD, had 1 year of data without COPD-related events before diagnosis, were free of a statin prescription on COPD diagnosis, had non-missing Deprivations Index and were without CVD at baseline. Figure 1 illustrates the cohort creation steps.

Figure 1

COPD patient selection and CVD outcome incidence. COPD, chronic obstructive pulmonary disease; CVD, cardiovascular disease; IMD, Index of Multiple Deprivation.

Table 1 summarises the characteristics of patients in the final data set. A slight majority (55%) were men; the average age at diagnosis was 64.9 years, with 48.7% younger than 65 years of age. The distribution of risk factors was generally similar between women and men (table 1).

Incidence

The incidence of CVD was 3.53 (95% CI 3.41 to 3.68) events per 100 person-years. This value was 74.6% higher in men than in women (4.01 (95% CI 3.81 to 4.21) vs 2.99 (95% CI 2.81 to 3.18)). In comparison, CVD incidence in the QRISK3 development population for individuals 40–84 years of age was 1.20 (95% CI 1.19 to 1.21) in total, 1.37 (95% CI 1.36 to 1.38) in men, and 1.02 (95% CI 1.01 to 1.03) in women.11 The incidence rates of CVD in patients living with COPD were higher than those in the general population across all sex and age groups (figure 2; detailed results in online supplemental table 5). Overall, for both the COPD and the general population, CVD incidence increased with age, but the association between age and CVD incidence was sharper among patients with COPD. CVD incidence was >3 times higher among COPD patients ≤65 years of age than their general population counterparts; this ratio was <1.5 among patients >65 years of age.

Figure 2

Incidence rates (95% CIs) by sex and age groups for COPD (blue) and the general population (orange). COPD, chronic obstructive pulmonary disease; CVD, cardiovascular disease; PY, person-years.

Standardised incidence ratios

When standardised to the QRISK3 development sample, the SIR was 1.62 (95% CI 1.54 to 1.64) among men and 1.71 (95% CI 1.61 to 1.75) among women. Again, there was a significant age effect in standardised CVD incidence between COPD and the general population. Among men, the SIR was 1.86 (95% CI 1.74 to 1.90) among those ≤65 years of age at diagnosis and 1.38 (95% CI 1.28 to 1.42) among older patients. Among women, the corresponding values were 2.13 (95% CI 1.94 to 2.19) and 1.46 (95% CI 1.33 to 1.50). Detailed results are provided in table 2.

Table 2

Standardised incidence ratio (SIR) between COPD and general primary care population

Predicted versus observed 10-year CVD risk

The predicted and observed CVD risk, in total and by sex and age groups, are provided in table 3. The average predicted 10-year CVD risk was 22.1% (95% CI 21.8% to 22.4%) among patients with COPD. In comparison, the observed 10-year CVD risk was 33.5% (95% CI 32.3% to 34.7%). As such, the observed risk was 1.52 times higher than the predicted risk (p<0.001). The discrepancy between the predicted and observed risks was particularly pronounced in younger patients, with an observed risk that was 1.82 times higher than the predicted (p<0.001; observed: 25.7% (95% CI 24.2% to 27.2%), predicted: 14.1% (95% CI 13.9% to 14.3%)). In comparison, the observed risk was 1.38 times higher than predicted in older patients (p<0.001; observed: 42.6% (95% CI 40.7% to 44.5%), predicted: 30.8% (95% CI 30.5% to 31.3%)). Average predicted 10-year risk in women was 18.9% (95% CI 18.5% to 19.2%), while the observed risk was 29.5% (95% CI 27.8 to 31.3%), corresponding to an observed/predicted ratio of 1.56 (p<0.001). Average predicted 10-year risk in men was 24.7% (95% CI 24.4% to 25.1%) and the observed risk was 36.7% (95% CI 35.0% to 38.4%), giving rise to an observed/predicted ratio of 1.49 (p<0.001).

Discussion

We used a representative sample of UK English primary care patients with COPD to evaluate the incidence and 10 year risk of CVD at the time of COPD diagnosis and determined to what extent QRISK3 can accurately predict CVD risk in people living with COPD. We showed that for both men and women, the incidence of CVD in people with COPD is substantially higher than in the general population, particularly among younger people with COPD. As an example, the annual incidence of CVD among 55-year-old individuals with COPD diagnosis was similar to the annual incidence among 70-year-olds in the general population. This gap narrowed by age, with the annual incidence of CVD among 70-year-old individuals on COPD diagnosis close to the annual incidence among 75-year-old members of the general population.

Critically, we showed that the QRISK3 score captures only part of the excess risk of CVD in patients with COPD. According to QRISK3, the 10-year CVD risk was, on average, 22.1% at the time of COPD diagnosis. However, at 33.5%, the observed 10-year risk was >50% higher than the predicted. Again, the discrepancy between the predicted and observed risk was particularly pronounced in younger individuals, with actual risk being >80% higher than predicted in adults 65 years of age or younger at the time of COPD diagnosis. One potential mechanism is that owing to genetic variations and/or environmental exposures, many individuals who develop COPD at a younger age have reduced peak lung growth. Reduced lung size has been causally linked with coronary artery disease and CVDs, though the exact mechanism(s) by which this occurs remain obscure.22 The QRISK scores do not consider genetic or many salient early life environmental factors, which may have outsized roles in the pathogenesis of COPD and CVD in younger persons.

Our findings of elevated CVD risk in patients with COPD are consistent with existing studies. A recent population-based study from the province of Ontario, Canada, documented a two-fold higher sex-standardised and age-standardised risk of major CVD events in COPD compared with the general population.23 After controlling for a large set of classical CVD risk factors, COPD was still associated with CVD, with an adjusted HR of 1.25 (95% CI 1.23 to 1.27). Another retrospective Canadian study found that the risk of hospitalisation due to various cardiovascular causes was more than twofold higher in the COPD group than in the comparison group.24 A meta-analysis of observational data has shown that adults with COPD are at a higher risk of developing various CVD events and cardiovascular risk factors such as hypertension and diabetes.5 Furthermore, a US-based study found that individuals with COPD had a significantly higher prevalence of CVD, with COPD increasing the odds of developing CVD by 170%.25 However, the utility of existing global CVD risk tools in predicting CVD events in patients with COPD has not been well studied. As such, our assessment of QRISK provides novel insight into the discrepancy between the predicted and observed CVD risk in COPD, with important clinical and research implications.

These results have important clinical implications. CVD risk assessment is often neglected in COPD patients.26 However, our results show that even when CVD risk is assessed, it underestimates the true risk, leading to potential missed opportunities for CVD risk modification through lifestyle interventions and therapeutics. This underestimation is particularly the case among individuals diagnosed with COPD earlier in their lives, who presumably stand to benefit the most from primary CVD prevention. Underestimated risks also hinder informed shared decision-making. Indeed, the magnitude of risk can be a motivator for behaviour change.27

Our results also have important implications for research. QRISK is not the only commonly used CVD risk scoring tool, and the validity of similar tools in COPD patients need to be rigorously examined. Other major scoring tools such as Framingham28 and ASCVD (AtheroSclerotic CardioVascular Disease)29 risk scores have fewer predictors and thus are potentially more prone to underestimating risk in COPD. More broadly, given the high prevalence of COPD and the high incidence of CVD, the inclusion of COPD as a predictor in CVD risk scoring tools should be seriously considered.

QRISK includes predictors related to other conditions such as rheumatoid arthritis, chronic kidney disease and migraine. Lack of consideration of COPD in contemporary CVD risk scoring tools probably reflects the general lack of awareness about the prevalence of airway diseases and an old but persistent notion that COPD is a disease of smokers and the inclusion of smoking as a risk factor is sufficient. We do not criticise the creators of CVD risk scoring tools, as it is ultimately the responsibility of the respiratory community to raise awareness around the burden and multifaceted pathophysiology of airway diseases. We also note that the choice of predictors in common CVD risk scoring tools is based on their discovered associations with CVD in the general population. Other predictors (eg, coronary artery calcium score30) might prove valuable in patients with COPD. However, given the significant underestimation of risk, a more immediate step while awaiting the next generation of CVD risk scoring tools is to update existing tools to generate calibrated CVD risk for patients with COPD. This has been undertaken for short-term CVD outcomes. For example, Rothnie et al reported that Global Registry of Acute Coronary Events underestimates 6-month mortality after acute coronary events in patients with COPD.31 They found that correcting the score by a multiplicative factor of 1.3 significantly improves its calibration in COPD.

The strengths of our study include access to a large, representative primary care cohort of people with COPD and a comprehensive assessment of CVD risk factors and outcomes, which was faithful to the definitions used to develop and validate QRISK. The relatively large sample size enabled us to investigate CVD incidence and predicted and observed risks within important patient subgroups with high statistical power, resulting in unequivocal interpretations about the performance of QRISK. However, our study also has limitations. This was not a dedicated validation study of QRISK. Such studies are required to follow specific methodology to evaluate the discrimination, calibration and clinical utility of a scoring tool, and if needed, propose modified versions of the tool that correct for miscalibration.32 33 The assessment of predicted and observed risk was not performed in several subgroups (eg, by ethnicity, socioeconomic status) as well as in prevalent COPD cases and those with previous CVD events. These can be performed as part of a future dedicated validation study. Similar to the QRISK development algorithm, we used multiple imputation to include individuals with missing predictor values, but the pattern of missingness might be different between individuals with COPD and the general population. Several predictors (eg, lab tests outside regular check-ups) are often ascertained based on clinical suspicion, and as such, the prevalence of abnormalities (associated with higher CVD risk) in the imputed predictor values might be higher than the true, unobserved values. However, any bias due to such overestimation is in the opposite direction of our results (as it will result in overestimating the QRISK score), and thus is unlikely to threaten the validity of our findings. We did not have a comparison group from the general CPRD population. As such, to draw comparisons, we relied on the data provided in the QRISK3 original study for the general population.11 While the QRISK development population and our study population are similar (both from electronic medical records of primary care UK practices), we cannot rule out differences due to different sampling methods in the two populations. However, this has little implications for the most important finding of our results: QRISK3 underestimates CVD risk in this representative primary care COPD sample.

In conclusion, increased CVD risk in COPD patients cannot simply be explained by shared risk factors such as age and smoking. CVD risk scoring tools developed for the general population capture a fraction of the added risk. This is particularly the case for younger patients whose actual CVD risk can be >80% higher than predicted values. Such discrepancy can result in missed opportunities for the prevention of CVD through risk factor modification, behavioural change and therapeutics. Current CVD risk scoring tools need to be validated and recalibrated to provide clinical utility for patients with COPD, and COPD should be considered as a distinct predictor in future CVD risk scoring tools.

Data availability statement

Data may be obtained from a third party and are not publicly available. This study is partly based on data from the Clinical Practice Research Datalink (CPRD-GOLD) obtained under licence from the UK Medicines and Healthcare Products Regulatory Agency. However, the interpretation and conclusions contained in this study are those of the authors alone. Data are available upon request from the data custodian, the CPRD.

Ethics statementsPatient consent for publication

Not applicable.

Ethics approval

Ethical approval was granted by the Health Research Ethics Board at Memorial University of Newfoundland (HREB #2017.024), and the Independent Scientific Advisory Committee overseeing CPRD approved our study protocol (Protocol 18_005RA3PAA).