Background: Safety concerns regarding severe cardiovascular events associated with the use of selective cyclooxygenase-2 (COX-2) inhibitors resulted in the market withdrawal of rofecoxib in September 2004. Objective: Using Taiwan's National Health Insurance 2003-;2004 claims database, this population-based retrospective cohort study assessed changes in physicians' practice of prescribing a COX-2 inhibitor after market withdrawal of rofecoxib. Methods: Patients with rheumatoid arthritis (RA) or osteoarthritis (OA) who were chronic users of COX-2 inhibitors before market withdrawal of rofecoxib were identified. Eligible chronic users of COX-2 inhibitors were patients who received ≥3 consecutive prescriptions for celecoxib or rofecoxib to treat RA or OA between April 1 and September 30, 2004. The main outcome evaluated in this study was the volume of celecoxib prescribing for each patient by his or her prescribing physician at the first postwithdrawal outpatient visit related to RA or OA. For each matched physician, we calculated the prescribing practice indexes before and after the withdrawal of rofecoxib to assess changes in prescribing practice. A higher value of the index represented less potential that patients with a history of cardiovascular events who were seen by a physician received a prescription for a COX-2 inhibitor. A 2-stage model analysis was used to assess changes, in physicians' prescribing practice after rofecoxib's withdrawal, in the volume of prescribing a COX-2 inhibitor for each matched patient by his or her prescribing physician. Results: Of the chronic users of COX-2 inhibitors identified, 13,101 were taking celecoxib and 8763 were taking rofecoxib before rofecoxib was withdrawn from the market. Concerns about safety after the mar- ket withdrawal of rofecoxib reduced physicians' volume of prescribing a COX-2 inhibitor, particularly in patients who had previously received rofecoxib. After rofecoxib's withdrawal, 72.50% of previous rofecoxib users (n = 6353) and 49.50% of previous celecoxib users (n = 6485) stopped taking COX-2 inhibitors. Only 27.50% of the rofecoxib users (n = 2410) were switched to celecoxib after rofecoxib's withdrawal. Overall, physicians' prescribing practice for celecoxib increased, from baseline to 3 months after rofecoxib's withdrawal, from 0.66 to 0.79 for academic medical center physicians, 0.71 to 0.82 for metropolitan hospital physicians, 0.77 to 0.88 for local community hospital physicians, and 0.86 to 0.92 for primary care clinic physicians. After controlling for patients' demographics, the linear regression analysis revealed that changes in physicians' prescribing practice after the withdrawal of rofecoxib affected the volume of prescribing a COX-2 inhibitor (P < 0.001). Conclusions: The volume of celecoxib prescribing was greatly reduced after rofecoxib was withdrawn from the market. Physicians' prescribing practice for COX-2 inhibitors significantly changed after the withdrawal, and it had a significant impact on the postwithdrawal volume of celecoxib prescribing.