Data‐Led Policy Design Using Medicare Shared Savings Program (MSSP) Health Care Cost Trajectories

Abstract

Research ObjectiveThe Medicare Shared Savings Program (MSSP) encourages participating Accountable Care Organizations (ACOs) to deliver quality care at lower cost through shared savings models. While the MSSP model offers ACOs and the Centers for Medicare and Medicaid Services (CMS) an opportunity to reduce spending growth while promoting quality goals, ACOs need explicit guidance about those activities over which they have control (e.g. program offerings, care pathways) to improve upon cost and quality goals. Variation amongst geographically‐adjusted and risk‐adjusted total per beneficiary spending amongst ACOs suggests room for ACOs to consider further cost saving efforts. Existing studies do not clearly point to how to achieve those ends. This study presents a methodology for understanding MSSP cost trajectories and identifying those factors that may impact them.Study DesignThis study seeks to characterize and identify trajectories of health care spending for individual patients with type 2 diabetes who experience hospitalization for a major acute cardiovascular event (MACE) (e.g. myocardial infarction, stroke). This analysis uses a retrospective cohort design with the cohort created by an index event (MACE). We examine spending variability for at least six months following the acute event to estimate variability pre and post the event and to cluster patients into post‐event cost trajectories. We use a nonparametric random forest model to identify characteristics from prior to the index event that predict changes in cost trajectories post‐event. We test finding robustness using a RE‐EM model that allows for autocorrelation and combines the structure of mixed effects models for longitudinal data with the flexibility of tree‐based estimation methods. These models allow for testing a broad set of characteristics. These include patient characteristics (e.g. gender, age, percent in zipcode below federal poverty line), comorbid conditions (e.g. dementia, Charlson co‐morbidity index), service use (e.g. prescription count, eye exam, skilled nursing facility use), preventive activities (e.g. flu shot, mammogram) and clinical indicators (e.g. blood pressure, BMI).Population StudiedThis study uses claims and electronic medical record (EMR) data from 2015–2017 for an MSSP population from the largest ACO in South Carolina. In 2017, there were 58,472 attributed beneficiaries. 2015–2017 was one contract period for the ACO and an upside risk‐only arrangement (Track 1).Principal Findings4064 individuals met exclusionary criteria prior to the event (e.g. continuously enrolled from 2015–2017, no prior MACE, index event is non‐fatal). Fisher's F‐test finds strong evidence of variation (P value <2.2e‐16 [F = 19.516]) in costs for the sample before and after the index event. Further work to be presented will cluster patients by post‐event cost trajectory and examine the impact of characteristics (e.g eye exam, falls risk assessment) prior to the index event on changes in post‐event health care spending.ConclusionsVariability exists in our sample pre‐ and post‐index event suggesting reason to investigate drivers of differences in patient characteristics prior to their MACE across cost trajectory clusters.Implications for Policy or PracticeThis work contributes a methodology for examining patient cost trajectories that allows health systems and policy makers to point to specific services and programs for interventions that suggests ways for that cost trajectory to be intervened upon.Primary Funding SourceUniversity of South Carolina internal funding.