Hypothesis formulation from subgroup analyses: Nonadherence or nonsteroidal anti-inflammatory drug use explains the lack of clinical benefit of aspirin on first myocardial infarction attributed to “aspirin resistance”

Abstract

"Aspirin resistance" has been defined as the occurrence of cardiovascular events despite regular intake of aspirin. One major analytic study suggesting that "aspirin resistance" is a clinical reality was unable to control for confounding by nonadherence or nonsteroidal anti-inflammatory drugs (NSAIDs). We formulated a hypothesis from subgroup analyses in the Physicians' Health Study, a randomized double-blind placebo-controlled trial testing 325 mg of aspirin every other day among 22,071 apparently healthy US male physicians. We classified participants by nonadherence or NSAIDs and used time-varying Cox proportional hazard models to adjust for confounding. After 5 years, the blinded aspirin component was terminated early based on the unanimous recommendation of the Data and Safety Monitoring Board. Of 378 confirmed first myocardial infarctions (139 aspirin and 239 placebo), the relative risk (RR) was 0.56 (95% CI 0.45-0.70, P < .00001). There was no statistically significant reduction among aspirin <150/180 pills/y (RR = 0.91, 95% CI 0.61-1.35, P = .62) or NSAID users >60 days per year (RR = 1.54, 95% CI 0.68-3.47, P = .31). There was a statistically significant reduction among aspirin >150/180 pills/y and NSAID users <60 days/y (RR = 0.55, 95% CI 0.44-0.70, P < or = .0001) and an increase among aspirin <150/180 pills/y and NSAID users >60 days/y (RR of 3.43, 95% CI 1.41-8.33, P = .007). In subgroup analyses useful to formulate hypotheses from a large randomized trial in apparently healthy men, aspirin nonadherence or NSAID use explained the lack of clinical benefit of aspirin on first myocardial infarction that has been attributed to "aspirin resistance." Direct randomized comparisons are necessary in trials designed a priori to test this hypothesis.