Objective. To compare the performance of various risk adjustment models in
behavioral health applications such as setting mental health and substance
abuse (MH/SA) capitation payments or overall capitation payments for popula
tions including MH/SA users.
Data Sources/Study Design. The 1991-93 administrative data from the Michiga
n Medicaid program were used. We compared mean absolute prediction error fo
r several risk adjustment models and simulated the profits and losses that
behavioral health care carve outs and integrated health plans would experie
nce under risk adjustment if they enrolled beneficiaries with a history of
MH/SA problems. Models included basic demographic adjustment, Adjusted Diag
nostic Groups, Hierarchical Condition Categories, and specifications design
ed for behavioral health.
Principal Findings. Differences in predictive ability among risk adjustment
models were small and generally insignificant. Specifications based on rel
atively few MH/SA diagnostic categories did as well as or better than model
s controlling for additional variables such as medical diagnoses at predict
ing MH/SA expenditures among adults. Simulation analyses revealed that amon
g both adults and minors considerable scope remained for behavioral health
care carve outs to make profits or losses after risk adjustment based on di
fferential enrollment of severely ill patients. Similarly, integrated healt
h plans have strong financial incentives to avoid MH/SA users even after ad
justment.
Conclusions. Current risk adjustment methodologies do not eliminate the fin
ancial incentives fur integrated health plans and behavioral health care ca
rve-out plans to avoid high-utilizing patients with psychiatric disorders.