Exploration of using constrained regression in Germany's morbidity-based risk adjustment.

Risk adjustment schemes are meant to compensate health plans adequately for their enrollees' expected health care costs so as to prevent incentives for risk selection in competitive health insurance systems with restrictions on risk-rating of premiums. However, important under-/overcompensation for specific groups of enrollees persists in many of the current risk adjustment schemes in place. While for some groups, a direct inclusion of the group in the risk adjustment scheme can solve this issue, others cannot be included in the scheme, e.g. because the status is not observed for every enrollee. Van Kleef et al. (Eur. J. Health Econ. 18, 1137-1156 (2017)) suggest constrained regression as a remedy. In this paper, we explore constrained regression in the context of the German morbidity-based risk adjustment scheme. We find that constrained regression is technically feasible in the German context and has the potential to improve upon the current base model in terms of overall under-/overcompensation and even individual model fit, particularly if the constraint is not set to fully eliminate under-/overcompensation in the respective group but to only partial elimination. Before implementation of constrained regression, a policy-discussion on which groups should be included in constructing an overall measure of under-/overcompensation is needed.