Evaluating the Role of Past Clinical Information on Risk Adjustment.

Abstract

The objective of this study was to evaluate whether incorporating historical clinical information beyond 1 year improves risk adjustment. Administrative data from the Department of Veterans Affairs and Medicare (for veterans concurrently enrolled in Medicare) for fiscal years (FYs) 2011-2015. We regressed total annual costs on Medicare hierarchical condition category indicators and risk scores for FY 2015 in both a concurrent and a prospective model using 5-fold cross-validation. Regressions were repeated incorporating clinical information from FY 2011 to 2015. Model fit was appraised using R and mean squared predictive error (MSPE). All veterans affairs users (n=3,254,783) with diagnostic information FY 2011-2015. In a concurrent model, adding additional years of historical clinical information (FY 2011-2014) did not result in substantive gains in fit (R from 0.671 to 0.673) or predictive capability (MSPE from 1956 to 1950). In a prospective model, adding additional years of historical clinical information also did not result in substantive gains in fit (R from 0.334 to 0.344) or predictive capability (MSPE from 3988 to 3940). Incorporating historical clinical information yielded no material gain in risk adjustment fit.