Identifying high-cost episodes in lower extremity joint replacement.

Understanding how to accurately identify high-cost episodes in joint replacement surgery remains a challenge for policymakers. Current models rely heavily on claims-based data, but these often fail to capture the full complexity of patient health and surgical outcomes. Prior research has shown that claims data alone may not fully represent the clinical variability among patients. This gap motivated the need to assess whether adding clinical data could improve the accuracy of cost prediction models. The Centers for Medicare & Medicaid Services (CMS) has implemented the Comprehensive Care for Joint Replacement (CJR) program, which uses claims data to adjust payments. However, the effectiveness of this approach in identifying high-cost episodes is uncertain. No prior work had resolved how much clinical data could enhance model performance. Existing models may misclassify patients due to limited data sources. Incorporating additional clinical variables could potentially improve the precision of cost group assignments. This uncertainty drove the current investigation into the role of clinical data in cost prediction.

The study aimed to evaluate whether claims-based risk adjustment models could be improved by incorporating clinical data to identify high-cost episodes in lower extremity joint replacement (LEJR) patients. The researchers focused on the Comprehensive Care for Joint Replacement (CJR) program provisions to assess the effectiveness of different data sources. They sought to determine if adding clinical variables could enhance the ability to assign patients to accurate cost groups. The study compared model performance across three populations: the full CMS Medicare population and two subsets with additional clinical data. The goal was to evaluate how much clinical data could improve the accuracy of cost prediction models. The researchers also wanted to assess whether functional status and other clinical variables could better identify patients in the top 20% of costs. The motivation stemmed from the limitations of claims-based models in capturing patient-specific factors. This approach could inform future policy decisions on bundled payment models.

The study used Medicare fee-for-service data linked with clinical data from two health systems: the High Value Healthcare Collaborative (HVHC) and Mayo Clinic. The researchers focused on LEJR episodes in 2013 that met CJR program criteria. They divided the data into three populations: the full CMS Medicare population and two subsets with additional clinical information. Multivariable logistic models were used to predict high-cost episodes. The models were evaluated using C-Statistic scores to measure performance. The researchers tested the impact of adding CMS-HCC categories to claims-based models. They also assessed the effect of including clinical variables, particularly functional status. The study compared model performance across the three populations to determine the value of clinical data in cost prediction.

The study found that claims-based models alone had low to moderate performance in identifying high-cost episodes. The CMS population had a C-Statistic of 0.714, while the HVHC and Mayo populations scored 0.628 and 0.587, respectively. Adding CMS-HCC categories improved model performance, with CMS reaching 0.758, HVHC at 0.692, and Mayo at 0.677. Clinical variables further enhanced performance, particularly in the Mayo population where functional status improved the C-Statistic to 0.783. The top 20% of episode costs were better identified with the inclusion of clinical data. The models showed that claims-based approaches alone were insufficient for accurate cost prediction. The addition of clinical data, especially functional status, significantly improved model accuracy. These findings suggest that clinical data can enhance the identification of high-cost episodes in joint replacement surgery.

The authors concluded that claims-based models alone may not be sufficient for accurately identifying high-cost episodes in joint replacement surgery. The addition of clinical data, particularly functional status, improved model performance in predicting high-cost episodes. These findings suggest that incorporating clinical variables could enhance the accuracy of cost prediction models. The results support the potential for using clinical data to improve payment adjustments in bundled LEJR procedures. The study highlights the limitations of claims-based models in capturing patient-specific factors. The authors propose that including clinical data could lead to more precise cost group assignments. The findings may inform future policy decisions regarding the CJR program and other bundled payment models. The study emphasizes the importance of evaluating new data elements for reimbursement decisions.