Active Redesign of a Medicaid Care Management Strategy for Greater Return on Investment: Predicting Impactability

摘要

Care management of high-cost/high-needs patients is an increasingly common strategy to reduce health care costs. A variety of targeting methodologies have emerged to identify patients with high historical or predicted health care utilization, but the more pertinent question for program planners is how to identify those who are most likely to benefit from care management intervention. This paper describes the evolution of complex care management targeting strategies in Community Care of North Carolina's (CCNC) work with the statewide non-dual Medicaid population, culminating in the development of an Impactability Score that uses administrative data to predict achievable savings. It describes CCNC's pragmatic approach for estimating intervention effects in a historical cohort of 23,455 individuals, using a control population of 14,839 to determine expected spending at an individual level, against which actual spending could be compared. The actual-to-expected spending difference was then used as the dependent variable in a multivariate model to determine the predictive contribution of a multitude of demographic, clinical, and utilization characteristics. The coefficients from this model yielded the information required to build predictive models for prospective use. Model variables related to medication adherence and historical utilization unexplained by disease burden proved to be more important predictors of impactability than any given diagnosis or event, disease profile, or overall costs of care. Comparison of this approach to alternative targeting strategies (emergency department super-utilizers, inpatient super-utilizers, or patients with highest Hierarchical Condition Category risk scores) suggests a 2- to 3-fold higher return on investment using impactability-based targeting.