AS Young, VISN 22 MIRECC and UCLA; K Kapur, RAND

Objectives: The VA and other publicly funded mental health systems are implementing managed care mechanisms, such as capitation, to control costs. Capitated reimbursement can increase the flexibility of services and create incentives to make care less expensive. It also creates incentives for providers to under-treat or avoid patients who have particularly high treatment costs. One approach to decreasing this undesirable incentive has been to risk adjust payments to providers. However, most research in this area has focused on Medicare and private populations, and risk adjustment for high utilizing, severely mentally ill patients has received too little attention. The objectives of this study are to examine patterns of utilization over time in high cost psychiatric patients, and to develop and test risk adjustment models in this population. We evaluate the extent to which risk adjustment models predict future costs, including costs during the first year of a capitation program for high-cost patients that was not risk adjusted.

Methods: We use longitudinal administrative data from the Los Angeles County Department of Mental Health for the years 1988 to 1994. The sample consists of 1,956 clients with severe mental illness who had high treatment costs. We examine patterns of utilization over time, and the extent to which costs changed from year to year. We estimate several modified two-part models of 1993 cost that use client-based variables such as demographics, living conditions, diagnoses, and prior mental health costs to explain the variation in mental health costs.

Results: No consistent patterns of treatment use were discernable. In patients with an average annual treatment cost of greater than $20,000, 90% of costs were due to acute and chronic hospitalization. The average annual change in treatment costs per patient was $40,863 (SD=$21,654). A risk adjustment model that incorporated demographic characteristics, diagnoses, and cost data from two previous years explained about 16 percent of the in-sample variation and 10 percent of the out-of-sample variation in costs. A model that excluded prior cost covariates explained only 5 percent of the variation in costs.

Conclusions: There were very large fluctuations in annual treatment costs in this population, and risk adjustment models had only modest success. Current risk adjustment methods may have some use in reducing risk selection in capitation programs. However, there is a need for improved risk adjustment techniques in public mental health, and these are likely to require additional data beyond that currently available. In the meantime, blended payment schemes that combine risk adjustment with risk corridors or partial fee-for-service payments should be explored.

Impact: This study suggests that current risk adjustment methods do not have the requisite predictive power to be used as the sole approach to adjusting capitation rates in mental health. A blended contract design may better reduce incentives for risk selection and under-treatment by using a partly risk adjusted capitation payment while not relying completely on the accuracy of risk adjustment models. Risk adjustment models estimated using data sets containing better predictors of rehospitalization are likely to have higher predictive power.