A system of postmarketing surveillance of antirheumatic drugs employing prospective protocol based consecutive patient cohorts is described, together with use of recursive partitioning techniques for statistical adjustment for potentially confounding variables. An analysis of 2 side effects (purpura and upper abdominal pain) is presented. Purpura was found to be associated with age, sex, disease duration, and amount of disability. The combination of aspirin and prednisone was associated with the highest prevalence of purpura. Upper abdominal pain also varied across drug classes. Within the nonsteroidal antiinflammatory drug category, there were clinically important differences in the relative prevalence of upper gastrointestinal pain between specific drugs.