PURPOSE: To compare hospitalization rates for serious upper and lower gastrointestinal (GI) events between chronic and acute users of a traditional non-steroidal anti-inflammatory drugs (tNSAID) + proton pump inhibitor (PPI) and users of a COX-2 seective inhibitor (Coxib). METHODS: The PHARMO Record Linkage System, including linked drug-dispensing and hospital records of approximately 3 million individuals in the Netherlands was used. We selected new Coxib or tNSAID users (01/01/2000-31/12/2004) with > or =1 year history before the first NSAID dispensing and > or =1 year follow-up ending at the first hospitalization for GI event (the outcome), last dispensing, or end of the study period. Chronic users were patients who used any NSAIDs for > or =60 days during the first year (n = 58 770); others were acute users (n = 538 420). Multivariate analysis was performed by Poisson regression adjusted for gender, age, and duration of follow-up, tNSAID and Coxib dose, NSAID/PPI adherence, use of other gastroprotective agents, anticoagulants, acetaminophen, corticosteroids, and cardiovascular disease. RESULTS: The cohort included 23 999 new tNSAIDs + PPI users and 25 977 new Coxib users, with main characteristics: mean +/- SD age 58.1 +/- 15.5 vs. 56.7 +/-17.5; female 55.3% vs. 62.2%; duration of treatment (days): 137 +/- 217 vs. 138 +/- 179, respectively. Among acute users, adjusted hazard ratios (95% Confidence Interval) were 0.21 (0.14-0.32) for upper and 0.26 (0.16-0.42) for lower GI events, for Coxib versus tNSAIDs + PPI users. Among chronic users, these were 0.35 (0.22-0.55) for upper GI and 0.43 (0.25-0.75) for lower GI events. CONCLUSIONS: Coxib users had significantly lower rates of GI events. Further research should elucidate the possible impact of selection bias.