Past experience with a drug may modify the risk of adverse event associated with current use of this drug. This effect was investigated empirically with a study on non-steroidal anti-inflammatory drugs (NSAIDs)-gastropathy. A hospital-based case-control study was conducted with 244 cases of upper gastrointestinal bleeding (UGIB) age 68 and over and 615 matched controls. Data on all medications dispensed to the study patients during the 3 years preceding admission were obtained from the Quebec universal prescription program automated database. Recent use (within 30 days prior to admission) of non-aspirin NSAIDs increased the risk of UGIB. The estimate of relative risk (RR) was 3.4 (CI, 2.1-5.5). Use of NSAIDs in the previous 3 years was associated with a lower risk of UGIB; the estimate of RR was 0.7 (CI, 0.4-1.0). The estimate of RR for first-time users was 22.7 (2.8-200.0) vs 3.0 (1.9-4.7) for those who had used the drugs at least once in the past 3 years. These results provide empirical evidence of a depletion of susceptibles effect whereby patients who remain on the drugs are those who can tolerate them while those who are susceptible select themselves out of the population at risk. Thus, past use should be considered as a potential risk modifier in non-experimental risk assessment of events associated with drug use.