Potential of enhanced demographic and social data in prediction health care costs related to substance abuse medicaid privately insured populations

States often use capitated payment systems when purchasing health and behavioral health care for Medicaid clients. In capitated (prospective) payment programs, the payer sets payment rates for a future time period. A central problem in the development of Medicaid payment systems is that health plans may attempt to avoid enrolling people who are expected to be costly. In particular, because people with substance abuse may be high utilizers of services, health plans paid through capitation may try to avoid such enrollees - either by not enrolling such persons or by not providing access to high-quality substance abuse care. Risk-adjusted payments may improve this situation by reimbursing health plans based on the expected medical need of actual enrolled populations, rather than on the average community risk. However, risk-adjustment systems that only account for demographics and diagnoses are limited in their ability to reduce health plans' incentives to avoid covering people with substance abuse problems under capitation. Another challenge for risk-adjustment is the transient nature of Medicaid enrollment. Most of the literature on risk adjustment for persons with substance abuse problems includes only people who maintain continuous enrollment over the study period. This chapter reports on several risk-adjustment models focused on Oregon Medicaid clients involved with substance abuse treatment in a large health maintenance organization. Models were designed to predict substance abuse treatment expenditures or total health plan costs. Risk-adjustment models that incorporated new variables such as social, clinical, and Medicaid eligibility indicators predicted substance abuse and total health plan expenditures somewhat better than simpler models typical of current risk-adjustment systems. However, models predicting substance abuse expenditures still performed very poorly and explained no more than 11% of the variation in expenditures. On the other hand, models for total health plan expenditures performed better, explaining as much as 35% of variation in expenditures. Although addition of new variables improved prediction modestly, payments would likely be quite inaccurate in some cases. Risk adjustment for capitated Medicaid substance abuse care remains challenging.