OBJECTIVE: To compare results from two approaches - propensity score binning or matching - for reduction of selection bias when comparing costs and utilization between insulin lispro and regular human insulin users. METHODS: A retrospective analysis of medical and pharmacy claims was conducted among users of insulin lispro or regular insulin during the identification period, March 1, 2000 to February 28, 2001. Propensity scores (PS) were estimated using age, gender, comorbidities, use of oral antidiabetic medications, prescription copayment, and baseline period diabetes-related costs and utilization. Our binning analyses classified all patients into five PS strata (quintiles). Overall cost and utilization differences during the 12-month follow-up period were then calculated using weights inversely proportional to variances of within bin differences. In our matched analyses, all subjects who could not be "paired" (1:1 lispro to regular insulin) with PS estimates agreeing within ±0.0001 were excluded. RESULTS: Of 6436 subjects, 1972 (30.6%) received insulin lispro and 4464 (69.4%) received regular insulin. At baseline, lispro subjects were younger, had fewer comorbidities, and were less likely to use oral antidiabetic medications than regular insulin users. Within PS quintiles, there was no significant imbalance on baseline characteristics. When matching, only 969 well-matched subject pairs could be retained. Baseline characteristics of these well-matched subjects were similar to those of subjects in quintiles 3 and 4, indicating that matching had tended to exclude subjects with extreme combinations of characteristics (quintiles 1 and 5). CONCLUSIONS: Relative to matching, the binning approach to adjustment for treatment selection bias appears to provide an answer that is more easily generalized to our full diabetic population. The binning approach uses all available outcome data to estimate overall treatment differences. Our analyses suggest that restrictions on insulin lispro availability to save pharmacy costs may not be economically justifiable from a more comprehensive payer's perspective.