Longer-Term Effects of Bundled Payments for Medical Conditions on Spending and Utilization: A Difference-In-Difference Analysis of the Bundled Payments for Care Improvement Initiative

Monday, June 24, 2019: 2:15 PM
Madison A (Marriott Wardman Park Hotel)

Presenter: Joshua Rolnick

Co-Authors: Joshua Liao; Xinshuo Ma; Eric Shan; Jingsan Zhu; Erkuan Wang; Qian Huang; Amol Navathe

Discussant: Michael Barnett

Background: Bundled payment, in which payment is grouped together for an episode of care, has shown savings for surgical conditions. Less is known about the effects of bundled payments for medical conditions. An initial study of Medicare’s voluntary program, Bundled Payments for Care Improvement (BPCI), examined BPCI participation for five medical conditions and found no association with changes in spending or utilization after 1 year. We extend this work by using longer-term data and different model specification to examine BPCI participation for four high-volume medical bundles: congestive heart failure (CHF), pneumonia, acute myocardial infarction (AMI), and chronic obstructive pulmonary disorder (COPD).

Methods: We used Medicare claims data from 2011–2016 to identify index hospitalizations for the four conditions. BPCI participants were compared to non-participants outside BPCI markets in adjusted episode-level models. Episodes attributed to BPCI physician groups were removed to avoid bias. The primary outcome was standardized total episode spending. Secondary outcomes included spending by type, readmission rates, and mortality. Models were adjusted for patient characteristics, time-varying market characteristics (hospital referral region), and hospital and quarterly time fixed effects. Ordinary least squares models were used for utilization and generalized linear models with a log-link function and gamma distribution for spending, with robust standard errors.

Results: 226 BPCI hospitals (265,351 episodes baseline period, 86,170 treatment period) were matched to 704 non-BPCI hospitals (207,228 episodes baseline period, 64,832 treatment period), with small differences in hospital and market characteristics after matching. In adjusted analysis, BPCI participation was not associated with a change in total spending (+0.1%, 95% CI -1.1% to 1.2%, p=0.93) or 90-day mortality (+0.3 percentage points (pp), 95% CI -0.1 to 0.6 pp, p=0.15). Skilled nursing facility spending decreased (-5.6%, 95% CI -9.4% to -1.6%, p=0.01), while spending on home health (+5.2%, 95% CI 2.3% to 8.3%, p<0.001) and readmissions (+3.6% ,95% CI 0.9% to 6.3%, p=0.01) increased. The 90-day readmission rate also increased (+1.4 pp, 95% CI 0.9 pp to 2.0 pp, p<0.001).

Conclusions: Hospital participation in BPCI was not associated with changes in total episode spending or mortality, but was associated with a shift toward spending on lower intensity post-acute care and higher readmissions. These findings suggest that policymakers should closely monitor patients hospitalized for medical conditions under BPCI-Advanced.