Modeling the health care utilization of children in Medicaid

Managers of healthcare payment systems have two competing goals, containing costs while providing reimbursement adequate to insure service access to all enrollees. Reimbursements set too low may result in inadequate coverage, while reimbursement increases may expose the payer to adverse selection. Risk adjustment models are used to balance the goals of cost containment and adequate reimbursement, both prospectively to adjust future payments and retrospectively to reward providers who care for challenging patients. If ignored by payers, new risk adjustment technology may be used by profit-seeking firms to facilitate adverse selection.; This dissertation presents new risk adjustment technology which forecasts costs better than traditional methodologies. It does so by using finite mixture models to identify groups of patients with different health status characteristics and using these groups to stratify the traditional two-part risk adjustment model. Using claims data from all children under age 21 in Georgia Medicaid in 1999, a simple finite mixture model was estimated to find the best number of health status groups and assign individual-level probabilities of membership to each group. These group probabilities were then used to stratify the logistic and the continuous portions of a traditional two-part risk adjustment model. Expected cost estimates were derived for outpatient, prescription, and inpatient services, compared to those generated by the traditional two-part model, and then verified using unfitted data.; The mixture model identified four distinct health status groups with increasingly high costs and service probabilities. Using these groups to stratify the traditional two-part risk adjustment model substantially improved the predictive power for all outpatient, prescription, and inpatient services. If providers used this information for adverse selection in idealized circumstances, they could translate as much as 24% of total Medicaid spending into excess profit. These methods should have substantial power in any insurance setting where individuals from different risk categories are pooled.