Individual level health risk scores have myriad applications including risk-adjusted payments, cost projection, measuring savings and assessing quality of care for members. Health risk scores are generally a function of demographic and condition/diagnosis-based category variables. Usually, models do not explicitly account for the impact of the provider’s approach to resource utilization, so member risk for a given condition is the same regardless of whether the attributed provider practices aggressively or conservatively. The result is that variation in practice pattern between providers remains an unmeasured part of the risk score model error. We propose a new risk score model framework that explicitly models provider impact on risk as a random effect in a statistical linear mixed model. One critical application is to more accurately evaluate provider efficiency. If risk scores are used to measure patient outcomes with only member-based morbidity, and then used to normalize the risk of a physician’s panel, providers A and B with very different patient risk profiles but similar average scores would be evaluated the same, even though their patients’ care needs may be quite different.
Robert Jason Reed, FSA, MAAA
Senior Director, Advanced Analytics
Ben Guszkiewicz, ASA, MAAA