OBJECTIVES: To compare health care service utilization and medication adherence in an elderly (age = 65 years) population newly starting thiazolidinedione therapy with patients starting insulin or other oral antidiabetic medications in the same period. METHODS: This was a retrospective data analysis using a Medicare HMO database. We compared adherence, health care service utilitzation, and cost impact of starting thiazolidinediones for 12 months after initiation of therapy. A total of 165 patients starting thiazolidinedione therapy between July 1999 and December 2001 were compared with patients starting insulin (n = 82) and other oral antidiabetic medication (n = 288). Propensity scores were created by using variables from a comprehensive health status assessment and health care claims data in the year prior to beginning new antidiabetic medication and utilized in multivariate analyses examining predictors of health care costs and diabetes medication adherence in the year following start of therapy. RESULTS: Patients starting thiazolidinedione therapy had 56% higher medication adherence than did patients starting insulin (P <0.01) and had adherence comparable with that in patients starting other oral antidiabetic medications (after confounder adjustment). There were no health care cost differences between patients starting either drug therapy. CONCLUSIONS: The introduction of thiazolidinedione therapy in an elderly type 2 diabetic population was associated with improved adherence compared with insulin and with no increases in health care costs. The usefulness of propensity scores in the bias reduction and balanced sample selection while determining outcomes associated with introduction of new pharmacotherapies is highlighted through these findings.