Studies performed for drug registration provide little insight into the long-term use and effectiveness of drugs in "real world" populations and settings. To obtain such insight, we used 10 years of electronic medical-record data from Kaiser Permanente Northwest Division, a large, group-model health maintenance organization in the United States, to study drug transitions, lapses in drug therapy, and mortality among 693 persons with newly diagnosed type 2 diabetes mellitus in 1988. We also studied an equivalently defined cohort of 1071 persons with new diagnoses in 1994, for whom the availability of laboratory results via electronic data permitted additional analyses. Cumulative mortality in the 1988 cohort increased steadily to 207 of 571 patients (36%) by 1997 (year 10). In 1988, 548 of 693 patients (79%) received initial monotherapy with a sulfonylurea. Insulin use rose as the use of sulfonylureas declined. Over this period, 504 of 693 patients (73%) discontinued or added drug therapy. Eight percent to 10% of both sulfonylurea users and insulin users discontinued drug use during the study period. In the 1994 cohort, two thirds of the subjects who discontinued therapy and were tested for glycosylated hemoglobin (Hb A1c) (n = 86) maintained good-to-excellent glycemic control. However, 78 discontinuers (38%) were not tested for Hb A1c, and, among this subset, 32% failed to visit a primary care clinician. The results of this study suggest that 5% to 10% of persons with type 2 diabetes mellitus avoid contact with the medical care system. Avoidance persists for at least the first 10 years after diagnosis but is more common in the first year after diagnosis. In addition, secondary failure of sulfonylureas begins within 1 year of diagnosis and continues at a steady pace. Almost 80% of patients initially treated with sulfonylureas added or switched to metformin or insulin within 10 years of diagnosis.