3101 — Predicting Mental Health Outcomes in the VA: What is the Best Model for Risk Adjustment?

Chatterjee S (Center for Health Quality, Outcomes and Economic Research (CHQOER), ENRM VA, Bedford, MA), Rosen AK (Center for Health Quality, Outcomes and Economic Research (CHQOER), ENRM VA, Bedford, MA), Seal P (Center for Health Quality, Outcomes and Economic Research (CHQOER), ENRM VA, Bedford, MA), Glickman M (Center for Health Quality, Outcomes and Economic Research (CHQOER), ENRM VA, Bedford, MA), Spiro, III A (Massachusetts Veterans Epidemiology and Research Information Center (MAVERIC), Boston VA Healthcare System, Boston, MA), Eisen SV (Center for Health Quality, Outcomes and Economic Research (CHQOER), ENRM VA, Bedford, MA)

Objectives:
Accurate risk adjustment is critical for comparing mental health outcomes across individuals, programs, or healthcare systems. The objective of this study was to examine whether the Psychiatric Case-Mix System (PsyCMS), a recently validated risk-adjustment measure designed to predict costs for VA psychiatric patients, also performs well in predicting mental health (MH) clinical outcomes.

Methods:
Our sample included 1,346 veterans receiving MH services from two VA health centers from 2004-2006. Participants completed self-report mental health assessments (Behavior and Symptom Identification Scale or BASIS-24 and Veterans SF-36) at enrollment in the study and three months later. Prior mental health utilization, ICD-9-CM psychiatric diagnoses, and Global Assessment of Functioning (GAF) ratings were obtained from administrative data files. Five regression models were tested to determine whether the PsyCMS increased prediction of MH clinical outcomes (change in BASIS-24, MCS (from the SF-36), and GAF) over and above demographic characteristics, previous utilization, and broad MH diagnostic categories. Model 1 included demographic characteristics. Model 2 added baseline and follow-up level of care (inpatient vs. outpatient) and enrollment site. Model 3 added pre-enrollment mental health utilization. Model 4 added six broad diagnostic categories (schizophrenia/schizoaffective, depressive disorder, bipolar, PTSD/anxiety, substance abuse and “other”). Model 5 added the 46 PsyCMS diagnostic categories. R2 and partial F tests were computed to determine the statistical significance of adding the PsyCMS versus the six diagnostic groups to the other predictors.

Results:
All models were significant at p < 0.05. The PsyCMS categories added 4-5 % to the R2 compared to the six diagnostic groups, and increased the R2 by 6-7% when added to model 3. The partial F test showed that the PsyCMS was statistically significantly for change in BASIS, MCS, and GAF (all p-values < .01), after controlling for all other variables.

Implications:
Models incorporating standard predictors and the PsyCMS performed significantly better in explaining variation in MH clinical outcomes than models using standard predictors and major diagnostic categories alone.

Impacts:
As the VA seeks to improve the quality of mental health care, measurement of mental health outcomes is increasingly important. The PsyCMS, previously validated to predict costs, may also be a useful risk-adjustment tool for examining mental health outcomes.