The Medicare Shared Savings Program: Savings, Selection, and Spillovers

We previously estimated spending reductions associated with provider participation in the Medicare Shared Savings Program (MSSP), using a difference-in-differences approach. By 2015, participation by providers entering the MSSP in 2012 was associated with a $302 differential reduction in annual total Medicare spending (P<0.001), participation by 2013 entrants was associated with a $139 differential reduction (P=0.009), and participation by 2014 entrants with a $36 differential reduction (P=0.41). Spending reductions were larger for physician groups than for hospital-integrated organizations and consistently grew over time within entry cohorts. We implemented several measures to minimize selection bias at patient and provider levels and found no evidence of residual selection in a series of falsification tests and sensitivity analyses. Differential changes in patient characteristics between ACO and non-ACO control groups were minimal; unadjusted estimates were very similar to adjusted estimates; holding constant the set of clinicians composing ACOs did not appreciably change estimates; and falsification tests in the pre-period and among large non-ACO groups did not reveal differential changes in spending in the absence of MSSP participation. We have implemented additional approaches to addressing risk selection, some of which can exacerbate other forms of selection bias. Our results are nevertheless robust and our conclusion of modest savings from the MSSP remains unaltered. We also examined spillover effects in a national commercially insured population unexposed to Medicare or commercial ACO contracts. We find evidence of spillover effects on spending, again limited to physician group ACOs, which have stronger incentives than health systems to engage in strategies that reduce spending for all the patients they serve. The evidence of commercial spillovers makes favorable risk selection at the patient level a less likely explanation for spending reductions in the MSSP; since favorable risk selection would presumably erode fee-for-service revenue, ACOs have incentives to limit selection to populations under risk contracts.