OBJECTIVE: To examine the cardiovascular risk of selective cyclooxygenase-2 (COX-2) inhibitors compared to non-specific conventional nonsteroidal anti-inflammatory agents (NSAIDs) in a high-risk Medicaid population. METHODS: All medical and prescription claims for non-institutionalized Maryland Medicaid enrollees who received at least one prescription for a COX-2 or NSAID between January 2000 and June 2002 were retrieved. Selected for study were adults with no use in the first six months, and who claimed at least 60 total days of supply over the study period. Patients who took both a COX-2 and an NSAID were classified as COX-2 users. Those having any use of the NSAID naproxen were excluded, as studies have already indicated naproxen has some cardiovascular protective effect compared to other NSAIDs. The primary outcome variable was the combined incidence of cardiovascular, hemorrhagic, and unknown deaths, nonfatal myocardial infarctions, nonfatal strokes, and pulmonary and venous embolism/thrombosis. To adjust for confounding by indication, we developed a logistic model of propensity for COX-2 treatment and stratified patients by quintiles of their propensity score. Model variables included age, sex, race, COX-2 indications (history of gastrointestinal problems, osteoarthritis, rheumatoid arthritis, back pain, acute pain), and cardiovascular risk factors (hypertension, hyperlipidemia, obesity, diabetes, renal problems, alcohol/tobacco/drug abuse, and prior cardiovascular events). RESULTS: Sample COX-2 users numbered 1208 and non-naproxen NSAID users numbered 5274. Overall incidence of the primary outcome was 14%. Seventy percent were female, forty percent were African-American, and thirty percent were over 50 years old. The Mantel-Haenszel average of the propensity-adjusted odds-ratios showed no significant effect of COX-2 use on cardiovascular risk (OR, 1.11; 95% CI, 0.93-1.32; p = 0.259). CONCLUSIONS: We did not find that COX-2 inhibitors increased cardiovascular risk over non-naproxen NSAIDs in a high-risk Medicaid population. Study limitations include its restriction to the Medicaid population and the two-year maximum observation period.