OBJECTIVES The risk of acute myocardial infarction (AMI) with COX-2 inhibitors may offset their gastrointestinal (GI) benefit compared with non-selective (NS) non-steroidal anti-inflammatory drugs (NSAIDs). We aimed to compare the risks of hospitalization for AMI and GI bleeding among elderly patients using COX-2 inhibitors, NS-NSAIDs and acetaminophen. METHODS We conducted a retrospective cohort study using administrative data of patients > or =65 years of age who filled a prescription for NSAID or acetaminophen during 1999-2002. Outcomes were compared using Cox regression models with time-dependent exposures. RESULTS Person-years of exposure among non-users of aspirin were: 75,761 to acetaminophen, 42,671 to rofecoxib 65,860 to celecoxib, and 37,495 to NS-NSAIDs. Among users of aspirin, they were: 14,671 to rofecoxib, 22,875 to celecoxib, 9,832 to NS-NSAIDs and 38,048 to acetaminophen. Among non-users of aspirin, the adjusted hazard ratios (95% confidence interval) of hospitalization for AMI/GI vs the acetaminophen (with no aspirin) group were: rofecoxib 1.27 (1.13, 1.42), celecoxib 0.93 (0.83, 1.03), naproxen 1.59 (1.31, 1.93), diclofenac 1.17 (0.99, 1.38) and ibuprofen 1.05 (0.74, 1.51). Among users of aspirin, they were: rofecoxib 1.73 (1.52, 1.98), celecoxib 1.34 (1.19, 1.52), ibuprofen 1.51 (0.95, 2.41), diclofenac 1.69 (1.35, 2.10), naproxen 1.35 (0.97, 1.88) and acetaminophen 1.29 (1.17, 1.42). CONCLUSION Among non-users of aspirin, naproxen seemed to carry the highest risk for AMI/GI bleeding. The AMI/GI toxicity of celecoxib was similar to that of acetaminophen and seemed to be better than those of rofecoxib and NS-NSAIDs. Among users of aspirin, both celecoxib and naproxen seemed to be the least toxic.