Evaluating the Impact of Payer and Provider Integration on Medicare Advantage Enrollee Hospitalization and Enrollment Outcomes
Discussant: Dr. Steven D. Pizer
We identified 28 plans integrated with 299 hospitals operating nationally in 2015 using publically available MA and hospital characteristic data. We selected enrollees who are members of both integrated and non-integrated MA contracts using the Medicare Beneficiary Summary File linked to HEDIS data to assign contract IDs. We use the MedPAR hospitalization files to identify MA enrollees who were admitted to a hospital in 2015, and if the hospital was integrated. While administrative claims data were only recently made available for MA enrollees, the MedPAR file includes complete MA data for all hospitals which are eligible for DSH and several other payment modifiers, accounting for over 90% of all MA hospitalizations. We limit to counties where both integrated contracts and integrated hospitals operate.
To estimate the effect of integration on outcomes, we fit linear probability models including a flag if an enrollee was a member of an integrated plan, a flag if they were admitted to an integrated hospital and their interaction. Control variables include patient and hospitalization characteristics and HSA fixed effects. Our primary outcomes of interest are 90-day unplanned readmissions and 90-day mortality post hospitalization, and plan disenrollment. We also assess if integration results in different care practices within the hospital.
Our analysis includes 1,417,396 MA enrollees who were hospitalized in markets where integrated plans and providers operated. We find that enrollees who receive care in a fully integrated setting are less likely to disenroll from Medicare Advantage or switch plans in the following year, and have marginally lower readmission and mortality rates than did enrollees not hospitalized in systems with their own MA plan. These enrollees also have more diagnosis codes but less ICU use than Traditional Medicare or other MA enrollees admitted to the same hospitals which may be indicative of upcoding. We find little difference in readmission rates.
As the popularity of these models continues to grow, the potential advantages of coordination may need to be balanced against the increased cost to Medicare associated with more diagnostic complexity.