Adherence to statin therapy: the key to survival?

Adherence to drug therapy is a critical issue in ensuring optimal benefits of treatment. It is usually measured as the proportion of prescribed or recommended doses of a drug that was actually taken by the patient over a certain time period (1). Low drug adherence may thus represent a pattern of irregular or interrupted drug intake or complete drug discontinuation during the respective time period. In recent years, several observational studies have analysed the pattern of drug adherence of different important classes of drugs in clinical practice, and often reported a disappointingly low persistence to preventive drug therapy in patients with chronic conditions (1), with the greatest drop often occurring during the first 3–6 months of treatment (2). Studies on the clinical consequences of different levels of drug adherence, however, repeatedly revealed surprisingly pronounced benefits of good drug adherence. For example, several observational studies on the real-world effectiveness of statins estimated survival benefits attributed to statin therapy by comparing patients with good adherence vs. those with low adherence (Table 1). Most studies revealed a substantially higher reduction in all-cause mortality than meta-analyses of randomised controlled trials (RCTs) (3–7) on statins vs. placebo (Table 1, Figure 1). There are several reasons why one should interpret results such as these with scepticism (8–10) rather than enthusiasm: Forest plot of the effects of statins on all-cause mortality in meta-analyses of RCTs and observational studies on adherence to statins First, RCTs are designed to demonstrate drug efficacy. They are usually performed in a highly controlled environment (e.g. strict inclusion and exclusion criteria, high frequency and intensity of follow-up investigations, high completeness of follow up, methods to increase drug compliance, etc.) in which the detection of specific drug effects is much easier than in the complex real-world situation. If observational studies report higher drug benefits anyhow, plausible reasons for this discrepancy should be provided. Second, in the study of drug effects, one would like to see some kind of specificity. For instance, the effects of statins should be more pronounced with respect to specific endpoints (e.g. coronary revascularisation; cardiovascular events and cardiovascular deaths) than with respect to non-specific ones such as all-cause mortality. Two of the studies listed in Table 1 reported the effects of good adherence to statins not only on mortality but also on more specific endpoints. In the study by Ho et al. (11), the effects of statins were less pronounced in specific endpoints, such as coronary revascularisation (risk reduction of 10%) or cardiovascular hospitalisation (risk reduction of 26%), compared with the risk reduction observed for the unspecific endpoint all-cause mortality (risk reduction of 46%). Similarly, McGinnis et al. (12) reported no effect of adherence to statins on non-fatal cardiac events, but a 56% risk reduction with respect to all-cause mortality. Patterns of drug effectiveness such as these seem implausible. Third, one should note that there are several different reasons of death, especially in heterogeneous populations. If it is reported that a single class of drugs such as statins can prevent approximately 50% of these deaths, this should trigger scepticism. For instance, Shalev et al. (13) reported that in primary prevention of coronary heart disease, patients adherent to statin treatment had a 45% lower risk of death compared with patients with poor adherence. To reach this level of effectiveness, statins would have to prevent not only all deaths resulting from cardiovascular causes (which is already implausible) but also a substantial number of deaths from other causes (8). An observational study on the effectiveness of influenza vaccination in elderly subjects (14) can serve as another example that certain levels of all-cause mortality reduction are not plausible. The study revealed an approximately 48% reduction in all-cause mortality attributable to influenza vaccination. It has been reported that in elderly subjects, the fraction of deaths attributable to influenza is, however, at the most 10% (15). Braun et al. (16) excellently summed up the main concern: ‘How can influenza vaccine prevent a much larger percentage of deaths than are caused by the disease that the vaccine is supposed to prevent?’ Why did most observational studies on adherence to statins seem to overestimate the effect on all-cause mortality? It is known for more than 25 years that adherence itself is a strong predictor of morbidity and mortality. The randomised and placebo-controlled Coronary Drug Project evaluated the efficacy and safety of several lipid-modifying drugs, including clofibrate, in the long-term treatment of patients with coronary heart disease. Although the main analyses revealed no mortality differences between patients treated with clofibrate or placebo (20.0% vs. 20.9%; p = 0.55), patients who had a good adherence to clofibrate (i.e. took ≥ 80% of the prescribed amount of tablets) had a substantially lower 5-year mortality than did poor adherers to clofibrate (15.0% vs. 24.6%; p < 0.001) (17). One might initially tend to explain this finding with the hypothesis that clofibrate only had an impact on mortality if it was taken as intended by the protocol. However, a very similar finding was also noted in the placebo group. Patients with good adherence to placebo therapy had a substantially lower mortality than patients with poor adherence (15.1% vs. 28.3%; p < 0.001) (17). This astonishing effect of good adherence to placebo was replicated in analyses of other RCTs (18,19). One important reason for this ‘healthy adherer’ effect is probably that adherers and non-adherers differ in several important patient characteristics. In the Coronary Drug Project, patients with high placebo adherence had a poorer health status than patients with low adherence (17). In cohort studies performed in the USA and Canada, predictors of poor adherence were black and other non-white race, low socioeconomic status, a high number of co-prescriptions, age of 75 years and older and the presence of depression or dementia (2,20,21). An observational study revealed that patients with high adherence to statins were more likely to seek out preventive health services such as cancer screening, influenza vaccinations and pneumococcal vaccinations (22), thereby suggesting that ‘good adherence’ might be a marker for a generally healthier lifestyle. Another observational study (23) revealed that patients with good adherence to statins had a lower risk of motor vehicle accidents (RR = 0.75; 95% CI 0.72–0.79), workplace accidents (RR = 0.77; 95% CI 0.74–0.81) and developing other diseases unlikely to be related to a biological effect of statins (e.g. dental problems or drug dependency; RR = 0.87; 95% CI 0.86–0.89). The authors interpreted this as additional evidence for the fact that patients who adhere to statins are systematically more health seeking than patients who do not remain adherent. Several studies, including the aforementioned analysis of placebo adherence in the Coronary Drug Project, tried to adjust for important baseline confounders but failed to eliminate the bias (17,24). One explanation might be that there were unmeasured confounders (such as data on socioeconomic status) that have not been included in the analysis. In addition, it is rarely considered that some diseases may act as time-dependent confounders, i.e. diseases occurring after the start of statin treatment change the probability of exposure (staying adherent) and outcome (survival). For instance, it has been reported that 53% of patients who developed lung cancer during statin treatment discontinued statins before their date of death (25). Most of them would have been classified as poor adherers to statins according to the definitions of drug adherence used in the observational studies. Other factors that possibly lead to statin discontinuation are the occurrence of cardiovascular events (2) or of adverse effects (26). Usually, studies on the effects of drug adherence only adjust for baseline differences and thus do not cover these changes in health status during follow up. Simply including such changes during follow up as new time-dependent covariates into the multivariate model might, however, also lead to bias if they are not only confounders but also intermediate variables on the causal pathway between statin treatment and survival. The use of marginal structural models has been proposed as a possible solution (27–29). In these models, it is possible to adjust for time-varying covariates that are simultaneously confounders and intermediate variables. However, there are only few experiences with this rather new analytical approach. Randomised, controlled trials will remain the gold standard for assessing efficacy of a drug, but are limited in several important aspects. They are costly, involve only a limited number of participants, often include only a highly selected patient population, usually only compare active treatment vs. placebo (and not other therapeutic alternatives), are of short duration and are often performed in a highly controlled environment not comparable with routine clinical practice (30,31). Given the number and the importance of unanswered questions after drug approval, there is often a strong need for post-marketing observational studies of intended drug effects. Established and newer epidemiological and statistical methods to address the methodological challenges of such studies are available (31–35) and should be implemented. Comparisons between observational and non-randomised trials revealed that well-designed observational studies can provide valid estimates of treatment effects (36,37). Non-randomised studies on the impact of drug adherence showing amazingly high but implausible drug benefits, however, neither contribute to the evaluation of drug effects nor increase the reputation of this area of research. No conflicts of interest to declare.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.