Hybrid (classification and regression) approach for selection of members for disease management programs
*Ognian Konstantinov Asparouhov, MEDai, Inc.
Donghui Wu, MEDai, Inc
Keywords: prediction of future healthcare cost, disease management, sensitivity, classification, regression
Traditionally the most “impactable” members for disease management (DM) programs are selected between the top 2-5% predicted high risk/cost members. The health care cost prediction is a regression problem and does not necessary maximize sensitivity/actual cost concentration which are more important criteria for DM purposes. We propose an approach combining regression and classification to maximize the above criteria. Our comparative study (based on commercial members) aims to show some advantages of using classification (alone or in combination with the regression) for indirect maximization of the above mentioned criteria in the following: -improvement of the overall sensitivity; -identification of low cost movers; -involvement of classification outputs(probability for future complications/ hospitalizations/emergency room visits) in regression models (prediction of future cost)