OBJECTIVE: This study was designed to assess the cost-effectiveness of cyclooxygenase-2 specific (COX-2) inhibitors (rofecoxib and celecoxib) over non-selective non-steroidal anti-inflammatory drugs (NSAIDs) in high-risk arthritis patients from the perspective of the Veterans Health Administration (VA). METHODS: This literature-based economic analysis compared rofecoxib and celecoxib to NSAIDS in two arthritis patient populations considered at higher risk of developing clinically significant upper gastrointestinal events (CSUGIEs): 1) patients of any age with previous medical history of perforation/ulcer/bleed (PUB), and 2) patients 65 years and older (regardless of history of PUB). Two outcomes measures were reported 1) incremental cost per CSUGIE averted over 1 year, and 2) incremental cost per quality-adjusted life-year (QALY) gained, considering both the mortality and morbidity associated with gastrointestinal (including CSUGIEs) and cardiovascular-related adverse events. When possible, costs were modeled to reflect the VA perspective. Sensitivity analyses were conducted to test the robustness of the analysis. RESULTS: Compared to NSAIDS, rofecoxib and celecoxib increased costs but reduced the incidence of CSUGIE. Cost per CSUGIE avoided were $7,476 and $16,379 (in patients with a PUB history) and $14,294 and $18,376 (in patients aged > 65 years) for celecoxib and rofecoxib, respectively. In both populations, celecoxib was associated with a cost per QALY less than $50,000. In contrast, rofecoxib was found to cost more and result in a net QALYs loss, due in particular to the increase in the risk of cardiovascular complications, and was therefore considered cost-ineffective. Results were most dependent on assumptions about the incidence of cardiovascular events and CSUGIE and the COX-2 inhibitors acquisition price. CONCLUSIONS: This analysis suggests that the COX-2 inhibitors may be cost-effective from the perspective of the VA. However, cost-effectiveness appears to depend less on the specific characteristics of the high-risk target population considered but more on the agent evaluated.