Assessing Risk-Adjustment Approaches under Non-Random Selection

Abstract

Various approaches have been proposed to adjust for differences in enrollee risk in health plans. Because risk-selection strategies may have different effects on enrollment, we simulated three types of selection—dumping, skimming, and stinting. Concurrent diagnosis-based risk adjustment, and a hybrid using concurrent adjustment for about 8% of the cases and prospective adjustment for the rest, perform markedly better than prospective or demographic adjustments, both in terms of R 2 and the extent to which plans experience unwarranted gains or losses. The simulation approach offers a valuable tool for analysts in assessing various risk-adjustment strategies under different selection situations.