To study exposure to NSAIDs in relation to the index date (the etiologically relevant exposure time window) while simultaneously controlling for the potentially confounding effects of calendar time, we used a time-matched, nested case–control analysis (28–29). The index date of each case-patient was used to define the risk sets from which controls were chosen. For each case-patient, we randomly selected 20 controls matched on month and year of cohort entry and age (±1 year) and assigned them the case-patient's index date. Consequently, follow-up time was identical for case-patients and controls within each risk set. We compared the risk for acute MI associated with the current use of various NSAIDs with that of “nonusers” in the year preceding the index date. Thus, we could disentangle the independent effect of individual COX inhibitors from those of naproxen and other NSAIDs since all exposure groups were compared with nonusers. We estimated crude and adjusted rate ratios (RRs) for these associations using conditional logistic regression (30–31). These measures of association are equal to the hazard ratios that would be estimated from the corresponding Cox proportional hazards regression. All rate ratios were adjusted for the potentially confounding effects of well-established conventional risk factors, including age; sex; hypertension; coronary artery disease; cerebrovascular disease; peripheral vascular disease; congestive heart failure; diabetes; and use of antilipemic agents, anticoagulants, and aspirin. In addition, the presence of respiratory illness, gastrointestinal ulcer disease, thyroid disorders, depression or psychiatric illness, use of oral corticosteroids, 3 measures of health care utilization (the number of hospitalizations, medical outpatient visits, and visits to a cardiologist), as well as 3 measures of comorbidity (the chronic disease score , the number of distinct drugs dispensed , and the Charlson index ), were evaluated as possible confounders using the change-in-estimate method (35). Covariates were retained in the final model if the risk estimate changed by 10% or more. We report risk estimates adjusted for all aforementioned covariates, including nonconfounders, because a parsimonious modeling approach provided no important gain in precision. With the exception of health care utilization and indices of comorbidity, which were assessed in the year preceding the index date, all other covariates were assessed in the year before cohort entry (at baseline). We identified medications using the prescription drug database and identified comorbid conditions using both discharge diagnosis codes and specific corresponding drug treatments.