Estimating an intervention's effects on health care disparities within and between physician organizations: decomposing the effects of health system affiliation.

Vertically integrated health systems are expanding across the U.S., but there is no evidence that system-affiliated providers are better able to reduce socioeconomic disparities than unaffiliated providers. This paper develops and interprets statistical models for estimating the longitudinal effects of an exposure, health system affiliation, on within- and between-group disparities. Mixed effect difference-in-differences regression models were used to estimate the effect of physician organization (PO) health system affiliation on socioeconomic disparities in diabetes medication adherence for fee-for-service Medicare beneficiaries from 2013 to 2019, decomposing the effect of affiliation on disparities into within- and between-PO socioeconomic disparities. We find that after becoming affiliated, adherence for dually-eligible beneficiaries worsened relative to non dually-eligible beneficiaries within POs, but between-PO disparities were reduced. Simulation studies were used to evaluate the methods, and showed the importance of centering dual-eligibility status and its interaction with affiliation status at the PO-year level and including the PO-year level means as predictors to decompose effects of affiliation into within and between-PO effects.