Rationale: With the recent increases in medical spending, employers have changed health care benefits to decrease health insurance costs by implementing increased cost sharing and decreasing the amount of covered services. It is unclear what the effect of changing plan design has on health care utilization and expenditures. Objective: To examine how a change in benefit design by a self-insured employer affected health care expenditures and utilization. Methods: This study takes advantage of a benefit change where the main medical benefit option was changed from full first dollar coverage of services to a plan which included relatively small copayments for visits to specialty care physicians and emergency rooms, and relatively low (10%) coinsurance for all other services. Primary care visits were not subject to copayments and preventive services were exempted from coinsurance. There was no gatekeeping before or after the benefit change and there was no change in prescription drug benefits. The benefit change was implemented in January 2004. The primary source of data for this analyses were the medical plan eligibility files, medical insurance claims, and institutional administrative databases. The primary outcomes of interest were total medical payments, specialty care visits, primary care visits, and number and payments for prescription drugs per member per year. We evaluate the outcomes using finite mixture models, which unlike standard specifications has the ability to distinguish between distinct classes of health care utilization (e.g. low users and high users). We estimate the expenditures using finite mixtures of Gamma densities and estimate the counts using finite mixtures of negative binomial densities. We compare the estimated mixture models to traditional expenditure (generalized linear models) and count (negative binomial) models to assess statistical fit. The explanatory variables in these models included demographic characteristics, plan type, employee status, type of coverage (family or individual), and health status. Standard errors for all analyses were bootstrapped and accounted for repeated observations across years. Results: Total number of members per year ranged from 63,000-65,000 over this time period. Based on statistical criteria, the finite mixture models provided a better fit for all models except the specialty care visit model, which was modeled using a hurdle-negative binomial specification. Total predicted annual payments per person decreased by 7.25% between 2003 and 2004, decreasing from $3,400 to $3,160. Total predicted annual payments for low users decreased by 22% between 2003 and 2004, while total predicted annual payments for high users increased by 7.6%. Specialty care visits per person per year decreased by 12% between 2003 and 2004. Total number of primary care visits increased and number of prescriptions and prescription drug expenditures per person decreased between 2003 and 2004 for low users, however they did not change during this time period for high users. Conclusions: This change in benefit design had a small effect on total payments by the health plan. Using finite mixture models, we were able to conclude that most of the effect was in the low users, and did not affect utilization in high users
