Improving risk adjustment of self-reported mental health outcomes

Risk adjustment for mental health care is important for making meaningful comparisons of provider, program, and system performance. The purpose of this study was to compare the predictive value of three diagnosis-based risk-adjustment models for predicting self-reported mental health outcomes. Baseline and 3-month follow-up mental health assessments were obtained on 1,023 veterans in Veterans Health Administration mental health programs between 2004 and 2006. Least-squares regression models predicting mental health outcomes used the Behavior and Symptom Identification Scale-24, Veterans RAND-36, and Brief Symptom Inventory. Sequential models began with sociodemographics, added baseline self-reported mental health, and compared three psychiatric case mix schemes: two using six diagnostic categories and the other (psychiatric case mix system [PsyCMS]) using 46 categories. R (2) were lowest for sociodemographic models (0.010-0.074) and highest for models with the PsyCMS (0.187-0.425). The best predictive ability was obtained when baseline mental health and 1 year of psychiatric diagnoses were added to sociodemographic models; however, the "best" risk-adjustment model differed between inpatients and outpatients.