Total Hip Arthroplasty in the Bundled Payments for Care Improvement Advanced Model: Who Will Bust the Bundle?

Background Lower extremity joint replacement is the second largest inpatient expense to U.S. Centers for Medicaid & Medicare Services (CMS), prompting value-based care experimentation most recently seen in the Bundled Payment for Care Improvement Advanced Model (BPCI-A). BPCI-A tested a new target pricing methodology and welcomed acute care hospital, physician group practice (PGP), and convener participants. Objective To identify patient and facility factors predictive of primary total hip arthroplasty (THA) episode costs exceeding target payment for PGP clients of a value-based convener in the BPCI-A. Methods From 29 PGPs across BPCI-A Model Years 3-5 (2020-2022), 4,178 THAs were analyzed. There were 901 bundle busters (22%), defined as patients whose 90-day episode costs exceeded the final target price set by CMS during reconciliation. A predictive multivariable logistic regression model was built from demographics, prior post-acute care use, comorbidities, total comorbidity burden, major teaching hospital status, urban/rural designation, safety net status, census division and bed size. Model performance was calculated using the C-statistic, and the relative contribution of variables were assessed by Akaike information criterion. Results Significant predictors included model year (P = 0.001), MS-DRG (P < 0.001), prior post-acute care (P = 0.004), myasthenia gravis/myoneural disorders/inflammatory and toxic neuropathies (P = 0.049), coagulation defects and other hematologic disease (P = 0.011), metastatic cancer or acute leukemia (P = 0.029), and total HCC (P = 0.034). Compared to New England, six census divisions had lower odds of exceeding target prices (P ≤ 0.003). When compared to small bed size, large and extra-large bed size had higher odds of exceeding target price (P = 0.014). The model’s discriminative power was moderate (c-statistic = 0.678). Conclusions This is the first study of PGP-convener partnerships in BPCI-A and spans three model years across all census divisions utilizing real target price and reconciliation data, offering the most comprehensive view of PGP THA performance in BPCI-A. While several patient and facility factors were significantly associated with exceeding the target payment, they had limited discriminatory ability in predicting excess costs after THA for PGP-convener partners. As bundled payment models evolve, refined predictive models are needed to empower PGPs and conveners to strategically redesign care, especially as CMS signaled a future for PGP-specific models.