A Decision-Theoretic Approach to Identifying Future High-Cost Patients

This research examines how diagnosis-based risk adjustment systems might be used to help allocate funds to cover very-high-cost (VHC) patients. VHC patients are relatively rare in health care systems and therefore predictive models have a difficult time allocating resources to them. The authors studied the Veterans' Health Administration to compare methods of allocating funds to VHC patients. Data were drawn from three consecutive years of costs (2002-2004) as calculated by the Veterans Affairs (VA) Decision Support System, and included approximately 250,000 patients per year. The authors considered a method successful if it accurately predicted the proportion of VHC patients at each VA medical center under consideration.

The authors' analysis shows that Diagnostic Cost Groups, used in conjunction with the Bayesian decision-theoretic algorithm outlined in their paper, can be effective tools to allocate funds to VHC patients in a hospital network. Simply using current data to predict the future make-up of patient pools was also found to be very accurate for predicting the proportion of VHC patients in the population. The authors note that certain aspects of their study might affect its applicability to other health care networks. For example, VA patient pools have more men than women; also, the VA utilizes global budgeting, unlike most health care providers in the U.S.