Keywords: Risk adjustment, Medicare, Plan payment
In the conventional framework for designing health plan payment models, the regulator chooses risk adjustor variables, estimates risk adjustment weights, and adds other policy parameters, but the data from which estimates are derived are taken as given. In this paper we explore an entirely novel approach: using the data itself as a policy tool. In a mirror image of current practice, we take the risk adjustors, the estimation method, and other plan payment features as given, and change the data used for estimation to achieve a policy objective. Using Medicare data we apply these ideas to two areas of misallocation in health care: undercompensation for individuals with mental health diagnoses and disparities in health care spending between racial/ethnic groups. We transfer spending to the group of interest, re-fit the risk adjustment model on the modified outcome, and illustrate the relationship between the transfer amount and targeted measure. We show spending can be transferred between disease groups to eliminate undercompensation with a minimal impact to overall fit of the risk adjustment model, while correcting disparities requires shifting much larger amounts of spending.