Performance of Statistical Models To Predict Mental Health and Substance Abuse Cost

• Activity Number: 333
• Type: Contributed
• Date/Time: Tuesday, August 8, 2006 : 2:00 PM to 3:50 PM
• Sponsor: Section on Health Policy Statistics
• Author(s): Maria Montez*+ and Cindy Christiansen and Susan L. Ettner and Susan Loveland and Amy K. Rosen
• Companies: Boston University and Boston University and University of California, Los Angeles and Boston University and Boston University
• Keywords: cost analysis ; GLM ; log-normal ; retransformation ; risk adjustment

Abstract

Our sample consisted of 525,620 VA patients with mental health (MH) or substance abuse (SA) diagnoses who incurred costs during FY99. We tested a log-normal (LN) model and generalized linear models on untransformed cost, defined by distributional assumption and link function: Gaussian Identity (GI); Gamma Log (GL); and Gamma Square Root (GSR). Risk-adjusters included age, sex, and 12 MH/SA categories. To determine the best model, predictive ability was evaluated using root mean square error (RMSE), mean absolute prediction error (MAPE), and predictive ratios among deciles of predicted cost. To study the effect of analyzing a random sample of the population on model choice, we recomputed these statistics using small size random samples. The GRS and the GI models had similar RMSEs; the GSR and the LN models had similar MAPEs. The GSR had some convergence problems with small samples.