Discussion
In this nationwide, registry-based cohort study, we found that use of RAS-acting agents was associated with decreased risk of severe exacerbations of COPD or death 12 months after initiation of treatment, both in the primary active comparator analysis and in the propensity matched analysis. Similar results were evident in the adjusted Cox proportional hazards regression.
The association between the use of RAS-acting agents and the risk of severe exacerbations has not previously been systematically explored in patients with severe/very severe COPD. Previous studies have investigated the effect of RASi in combination with statins,14 15 or only specifically ARB,16 29 with mortality and myocardial infarction as primary outcomes and a shorter follow-up period. Furthermore, the statistical power in these studies was presumably low due to small sample sizes.
Only one study investigated AECOPD as outcome, and they found results similar to ours. However, this study had an older population (>65 years), their primary outcome was all-cause mortality and myocardial infarction as well as exacerbations, and they had a longer follow-up period.15
There is an ongoing discussion on whether the use of statins play a role in the prevalence of AECOPD.30 31 However, in this study, there was only approximately 13% absolute difference in statin user prevalence between the RASi group and the bendroflumethiazide group in the main analysis (table 1). The relative risk reduction of statins would thus not be able to drive a difference in our results. Neither the active comparator analysis nor the adjusted Cox analysis was adjusted for the use of statins. Meanwhile, the signal on the protective effect of RASi treatment was strongest in the active comparator analysis, which indicates that adjusting for the use of statins would not make a significant difference in the results.
To our knowledge, this study is the first large observational nationwide registry study that investigates the impact of RASi treatment on severe and moderate exacerbations of COPD in a real-life outpatient cohort with 12 months follow-up. The study had a large population size with almost 40 000 patients meeting all inclusion criteria, of whom more than 10 000 were exposed to the drugs of interest. All patients were registered in national Danish registries, which allowed us to control for important confounders and thereby reduce the risk of bias and ensure no loss to follow-up. Furthermore, all patients with cardiac failure were excluded from the study before the primary analysis was conducted, which eliminated an important source of confounding in patients with COPD. DrCOPD enabled us to ensure a correct COPD diagnosis, made by a respiratory specialist and verified at annual outpatient visits with spirometries, thereby reducing the risk of misclassification bias. For our primary analysis, we used bendroflumethiazide as an active comparator, which reduces the risk of unmeasured ‘new user’ bias. For the same reason, we propensity-score matched the sensitivity analysis on several known predictors of the outcome and the survival analysis was adjusted for the same variables. Of note, the matching, both in the primary active comparator analysis and in the propensity-score-matched cohort, seemed quite balanced on key confounders. In the primary analysis, the effect was very similar on AECOPD alone as it was on AECOPD and all-cause mortality, which validates our findings and substantiates their relevance to pulmonary disease.
However, there are also limitations to this study. First, despite having adjusted the analyses for known confounders, the survival analysis suggested an increased risk of AECOPD with the use of LABA, LAMA and ICS, which indicates some residual bias by indication, since at least LAMA and ICS are documented to reduce AECOPD.32–34 Second, it was not possible, qua the registry design, to monitor the patients’ adherence, as exposure to the drug was defined by a collected prescription. This leaves some uncertainty as to whether patients used the medication, which is an inborn error in the design of the retrospective registry study. However, the uncertainty is limited because we know the prescriptions are collected, which makes it very likely that the drugs were also used by the patients. Non-adherence would lead to a conservative bias since actual non-users would be classified as users. Third, the IRs implied an increased risk of AECOPD after initiating treatment with RASi compared with before exposure to the drug. But a similar result was evident with exposure to bendroflumethiazide, and the explanation is probably the increased risk of events in a new user of a drug as well as the fact that the risk of a patient with COPD having an exacerbation simply increases over time as the disease progresses. This is the main reason for performing the active comparator analysis. Fourth, we acknowledge that finding the perfect drug to use as active comparator is very challenging, as no two treatment groups are completely alike. We have accommodated this issue by using two drugs that are both first-line drugs in treatment for hypertension. Furthermore, all patients that could have other indications for the treatment (renal failure and heart failure) were excluded from the study before performing the primary analysis.
In conclusion, we observed a consistent signal suggesting an association between treatment with an RASi and a reduced risk of acute exacerbations and death in patients with COPD. Our data thus support the safety of such very frequently used drugs among people with COPD. This finding is biologically plausible; however, it should be interpreted with caution because of the observational nature of the data. Randomised controlled trials (RCTs) can clarify the degree of residual confounding and whether the use of RASi treatment is actually causative for such a protective effect. This remains an unanswered hypothesis, but it seems very unlikely that treatment with drugs inhibiting the RAS should increase the risk of AECOPD.
Since both COPD and treatment with RASi are very frequent, we argue that the size of potential benefit is large enough that such RCTs should be performed.