Impact of time-varying and time-invariant measures of adherence to secondary prevention therapies post-acute myocardial infarction: An application of marginal structural models (MSMs)

The purpose of the study was to explore the relationship between patient adherence to secondary prevention therapies post an initial episode of acute myocardial infarction (AMI) and subsequent risk of cardiovascular events by using time-invariant and time-varying measures of adherence. The effectiveness of both measures in predicting a (i) recurrent AMI, and (ii) mortality using various mathematical models and statistical techniques was compared. Time dependent confounding was accounted for by using marginal structural models (MSMs).;A longitudinal cohort observational study design was employed using the retrospective Medicare 5% random national sample claims data from January 1st, 2006 to December 31st, 2008. The time-invariant measure of adherence was measured over a fixed one year period. Time-varying adherence was measured quarterly along with other time-varying confounders over a maximum follow-up of 11 quarters. Estimates of the effect of adherence from Cox regression models and MSMs were compared, along with model-fit-statistics.;Of the total 1,427 patients included in the study, cohort A (statin therapy) comprised of 1,091 patients, and cohorts B (beta-blocker therapy) and C (angiotensin converting enzyme inhibitor (ACEI)/angiotensin receptor blocker (ARB) therapy) included 1,021 and 1,025 patients, respectively. When accounting only for baseline covariates in a discrete-event time model the hazard for a recurrent AMI among statin adherent patients in cohort A was 63% of the hazard among non-adherent patients (Hazard Ratio (HR) = 0.63; 95% CI [0.40, 0.99]; p = 0.0471). When accounting for baseline covariates and time-varying covariates in a discrete-event time model the hazard for a recurrent AMI among statin adherent patients in cohort A was 61% of the hazard among non-adherent patients (HR = 0.61, 95% CI [0.38, 0.97]; p = 0.0366). The results for the effect of adherence to beta-blockers and ACEI/ARBs on subsequent cardiovascular events were not statistically significant. The stabilized weights used in estimation of the MSMs did not have optimum variability and the results from the MSMs were not statistically significant.;Further studies are required to understand if MSMs should be the preferred methodology when exploring the relationship between long-term medication adherence and health outcomes.