OBJECTIVES: Finite mixture regression models have been used to estimate skewed distributions. However, mixture models have not been broadly considered in outcomes research. This study examined whether mixture regressions gave a better fit to the data than regressions with single distribution when estimating prescription drug utilization and prescription drug expenditures with high positive skewness. METHODS: The Medical Expenditure Panel Survey which is a nationally representative survey with comprehensive information on health care use and spending was used to estimate prescription drug utilization and prescription drug expenditures of adult patients (≥20 years old) with RA in 2005. Poisson distributions to estimate prescription drug utilization and gamma distributions to estimate prescription drug expenditures were considered. Bayesian information criteria (BIC) were used to compare regression models. RESULTS: A total of 4,546 patients with a diagnosis of RA (mean age: 60.3 years, female: 65.0%) were included in the study. Mean number of prescriptions was 31 (median: 21). Skewness and Kurtosis in number of prescriptions were 1.9 and 8.2, respectively. Mean expenditure on prescription drugs was $2050.80 (median: $1210.50). Skewness and Kurtosis in expenditure on prescription drug were 8.4 and 170.4, respectively. After controlling demographic and clinical variables, a mixture model with two Poisson distributions (BIC: 28,735) gave a better fit to the drug utilization data than a model with a Poisson distribution (BIC: 114,339) and a negative binomial model (BIC: 38,559). A mixture model with two gamma distributions (BIC: 35,806) gave a better fit to the drug expenditure data than a model with a gamma distribution (BIC: 36,208). CONCLUSION: This study showed that mixture models provided better fits to the data with high positive skewness than regression models with single distribution when estimating the number of prescriptions and the prescription drug expenditures of patients with RA.