OBJECTIVES: Our objectives were to examine patterns of medication switching, treatment discontinuation, and associated costs in subjects treated with COX-2 specific inhibitors or non-selective NSAIDs in a managed care setting. METHODS: We conducted a retrospective claims analysis in 19 managed care health plans in the United States to evaluate three cohort groups treated with celecoxib, rofecoxib, or commonly prescribed non-selective NSAIDs between July 1, 1999 and June 30, 2000. Medication utilization, switching, discontinuation, and costs associated with medication switching and discontinuation were evaluated. We used propensity score analysis to match cohorts to reduce possible sampling selection bias. RESULTS: 164,596 subjects were assigned to the celecoxib (n=9,475), rofecoxib (n=7,734) or non-selective NSAID groups (n=147,387). COX-2 specific inhibitor users were significantly less likely to switch therapy (23% less likely by odds ratio test [p<0.001]) and had a lower rate of discontinuation compared to NSAID users (7.2 vs. 18.7 discontinuations per 1000). Cox proportional hazards analysis demonstrated that persistency on therapy is higher for COX-2 users (57% less likely to discontinue) versus NSAID users. Patients remained longer on celecoxib before discontinuation compared to subjects in any other cohort (celecoxib 63.7 days, rofecoxib 58.9 days, ibuprofen 21.1 days, naproxen 27.8 days, and diclofenac 34.4 days). Age was also associated with longer time on therapy, with older age decreasing the likelihood of discontinuation by 2%. Switching therapy increased total costs by 21.9% (p<0.01). Outpatient, inpatient, and emergency room costs increased in subjects who switched therapy compared to those who remained on therapy, when COX-2 specific inhibitor and NSAID matched cohorts were compared. CONCLUSIONS: Less medication switching and longer persistency were seen in patients treated with COX-2 specific inhibitors. Medication switching was associated with increased treatment costs. The increased resource use may be ascribed to adverse events, which may be higher in the non-selective NSAID cohorts.