The risk of acute myocardial infarction associated with non-steroidal anti-inflammatory drugs users: Impact of additional confounding control for variables collected from self-reported data

Background: Several observational studies have employed electronic health databases to study the association between non-steroidal anti-inflammatory drugs (NSAIDs) and myocardial infarction. Because some important potential confounders might not be routinely collected in such data sources, patients' reports could be utilized additionally. Objectives: This study evaluated the impact of using additional information from patients' reports when assessing the association between use of NSAIDs and the risk of acute myocardial infarction (AMI). Methods: A case-control study was conducted among adult patients with hypertension and/or hypercholesterolemia in the Utrecht Cardiovascular Pharmacogenetics study. Information was collected from the Dutch PHARMO Database Network (Pharmacy and hospitalization records) and patients' questionnaires (body mass index, alcohol use, smoking, physical activity, and familial history of cardiovascular diseases). For each case, up to 13 controls were matched based on age and gender at the date cases were hospitalized (index date). Conditional logistic regression analysis was applied to estimate odd ratios (ORs) and 95% confidence intervals (95% CI). Results: We identified 970 AMI cases and 2,974 controls during 1985-2005. Of all cases, 140 patients (14.4%) were exposed to conventional NSAIDs and 9 patients (1.0%) were exposed to selective COX-2 inhibitors at the index date. Compared to nonuse, neither conventional NSAIDs [(Adj. OR 0.98, 95% CI: 0.91-1.06) nor selective COX-2 inhibitors (Adj. OR 1.00, 95% CI: 0.74- 1.36) were associated with an increased risk of AMI after adjustment for confounders routinely collected in pharmacy records. Additional adjustment for confounders collected from patients' reports did not change the risk estimates [(Adj. OR 0.97, 95% CI: 0.90-1.05) and (Adj. OR 1.01, 95% CI: 0.75- 1.35)], respectively. Conclusions: This study showed that additional potential confounders collected from patients' reports did not significantly change the risk estimates.