Association of the Mandatory Medicare Bundled Payment With Joint Replacement Outcomes in Hospitals With Disadvantaged Patients

This cohort study examines changes in health care spending, use, and quality of care after the implementation of Medicare’s first mandatory bundled payment system at participating hospitals and among dual-eligible patients.


eFigure 3. Study Sample Selection Process
Notes: CJR is Comprehensive Care for Joint Replacement Model. MSA is metropolitan statistical area. HMO is health maintenance organization. LEJR is lower extremity joint replacements (i.e., hip and knee joint replacements). BCPI is Bundled Payments for Care Improvement Initiative. eAppendix 1. MSA Randomization for CJR Participation and Calculation of Sampling Weights CMS used a stratified clustered random sampling, with metropolitan statistical areas (MSAs) serving as the clusters, to assign MSAs for CJR participation (see eFigure 1). CJR participation was mandatory for all hospitals within a selected MSA.
1. From the 388 MSAs in the United States, CMS excluded 192 MSAs from CJR model participation due to their low hip and knee replacement volume or high rates of participation in the Bundled Payments for Care Improvement (BPCI). 2. The remaining 196 eligible MSAs were grouped into 8 strata based on their population size and historical care episode spending. CMS identified cut-points for strata using Kmeans factor analysis. 3. CMS then randomly selected MSAs for CJR participation within each stratum, with a higher sampling probability (oversampling) in the strata with the higher historical spending. They implemented random sampling using the "PROC SURVEYSELECT" statement in SAS. As a result of the random selection, CMS announced 75 treatment MSAs for CJR participation and 121 control MSAs. Additional information about the randomization process is available in the CJR Final Rule (Section III A). 1 CJR participation was mandatory for hospitals located in the 75 selected MSAs unless they were already participating in the BPCI model one, or phase two of model two or model four. 4. Under this sampling approach (described in the step 1-3), it would be possible to calculate sampling weights using the selection probability applied to each of 8 strata. 5. However, CMS retroactively changed the eligibility criteria for MSAs after the initial randomization described in the step 3. This led to the exclusion of 8 treatment MSAs that had been selected for the CJR participation and 17 control MSAs that were not selected for CJR but would have been excluded under this new criteria. Additionally, we excluded the San Juan, Puerto Rico MSA where the healthcare system was struck by Hurricane Maria in 2017. As a result of further exclusion of those 26 (=8+17+1) MSAs, we updated the selection probability for each MSA strata conditional on the new eligibility criteria: We then calculated updated raw selection and non-selection weights from these probabilities using equation (2).
Finally, we normalized the selection weights for each strata by dividing them by the sum of the raw strata-level selection weights, and normalize the non-selection weights by dividing them by the sum of the raw strata-level non-selection weights.

eAppendix 2. Measure of Hospitals Serving a High Percentage of Disadvantaged Joint Replacement Patient Population
We considered four measures to identify hospitals serving a high percentage of disadvantaged patient populations: (1) hospitals with a high proportion of hip and knee joint replacement patients who were dually eligible for both Medicaid and Medicare, (2) hospitals with a high disproportionate patient percentage (DPP), (3) hospitals with a high proportion of inpatient days attributable to patients eligible for Medicaid but not Medicare (Medicaid ratio) and (4) hospitals with a high proportion of hip and knee joint replacement patients living in areas of high poverty. Please note that DPP-based definition and Medicaid-ratio-based definition are measures of lowincome patients served for all inpatient services while the other two definitions are specific to hip and knee joint replacement patients. Measure Measure 4 -Hospitals with a high proportion of high and knee joint replacement patients living in areas of high poverty Patients were considered living in high poverty areas if they lived in a ZIP code where at least 20% of the residents over the age of 64 reported living under the federal poverty level. ZIP code level poverty rates were obtained from American Community Survey (ACS) 5-year estimates from 2013-2017 (ACS Table S1701).
Comparing Measures 1 to 4 Our goal was to identify which measure was best able to differentiate hospitals serving a high proportion of disadvantaged patients from other hospitals on outcomes common among disadvantaged joint replacement patients (e.g., high spending, high rates of institutional postacute care discharge, and high readmission rates). 6,7 For each measure, we chose a threshold, and then compared hospitals above that threshold with hospitals below that threshold on three outcomes (spending, discharge to institutional post-acute care, and readmission). We then repeated this process over a range of possible thresholds.
For example, for the DPP measure, we compared hospitals in the top 10% of DPP to those in the bottom 90%, then compared those in the top 15% to those in the bottom 85%, then those in the top 20% to those in the bottom 80%, and so on. By plotting outcomes across the continuum of possible thresholds, we were able to visually assess which measure and threshold maximized differentiation between two groups of hospitals while avoiding small group sizes. Once we chose a measure and threshold, we defined hospitals above the threshold as hospitals serving a high proportion of disadvantaged patients. We used the pre-CJR data for these calculations because the implementation of CJR may influence patient composition of each hospital.
As seen in Figure A below, the dual-eligible based definitionone using the proportion of dualeligible hip and knee joint replacement patients -provided better differentiation in outcomes than the three other definitions. Within this measure, differentiation was relatively stable across thresholds, and therefore we chose to define hospitals serving a high percentage of disadvantaged patient populations as those in the top 25% for proportion dual-eligible hip and knee joint replacement patients for both simplicity and to ensure there was a sufficient number of hospitals in each group.
CMS has also used the proportion of dual-eligibles as part of their value-based payment systems to identify hospitals serving disadvantaged patient populations after the dual-eligibility status was found to be the most powerful predictor of poor outcomes under federal payment systems. 3 One example is the Hospital Readmission Reduction Program, where target readmission rates for each hospital are set within strata based on the proportion of dual-eligible discharges. 1 We also preferred the dual-eligible-based measure because it was specific to hip and knee joint replacement patients. In contrast, DPP-based and Medicaid-ratio-based measures are calculated based on all patients with inpatient stays (regardless of their receipt of joint replacement surgeries). Notes: LEJR is lower extremity joint replacement (i.e., hip and knee joint replacement); DPP is disproportionate patient percentage. The "above threshold" line represents the mean outcome among hospitals above the quantile threshold while the "below threshold" line is the mean outcome among hospitals below that threshold. We defined hospitals above the selected threshold of the selected measure as those serving a high proportion of disadvantaged patient populations.

eAppendix 3. Standardization of Medicare-Allowed Payments
Based on Gottlieb et al. 4 with some modifications, we standardized all Medicare-allowed payments to remove payment differences driven by wage index, indirect costs of medical education, and other special payments by taking the following three steps: (1) Calculate provider wage index We calculated provider wage index using MSA wage index. 5 CMS annually update MSA wage index based on the cost of living across MSAs. The calculation of provider wage index for outpatient claims was different because the portion of the claim payment amount affected by MSA wage index was lower for outpatient claims (0.60) than it was for all other claim types (0.75).
= 0.40 + (0.60 × ) ℎ = 0.25 + (0.75 × ) (2) Standardize Medicare-allowed payments We standardized payments in index hospitalization claims and 90-day readmission claims differently from payments in carrier, outpatient, and post-acute care claims (i.e., skilled nursing facility, inpatient rehab, long-term care hospital, swing bed, and home health). Medicare-allowed payments for hospitals differed based on not only wage index but also teaching hospital status, safety-net hospital status, and others. The triple-difference approach measures the difference between two estimates derived through the difference-in-differences (DD) approach. In our case, we considered one DD estimate for high-dual hospitals and the other for low-dual hospitals. The DD estimate for high-dual hospitals is the average difference (occurring with the CJR implementation) in their outcomes in treatment MSAs subtracted by the average differences in the control MSAs. Likewise, the DD estimate for low-dual hospitals measures changes in outcomes under CJR model among low-dual hospitals. The difference between the DD estimates of high-and low-dual measures the differential changes between two groups of hospitals under CJR. More specifically, we estimated the following care-episode-level regression: where ℎ is an outcome variable for joint replacement that occurred in hospital ℎ located in MSA in year . takes a value of 1 if a joint replacement occurred in the treatment MSAs and 0 otherwise. takes a value of 1 if an observation occurred in 2016-2017 and 0 otherwise. ℎ ℎ takes a value of 1 for high-dual hospitals and 0 for low-dual hospitals. includes types of joint replacement, occurrence of major complications or comorbidities during the hospital stay, and patient age and gender. ℎ , , and represent hospital, year, and quarter fixed effects. 1 is the estimate of interest and measures the differential changes in outcomes between high-and low-dual hospitals under CJR. 2 is another estimate of interest and measures changes in outcomes in low-dual hospitals under CJR. We obtained changes in outcomes in high-dual hospitals under CJR by adding 1 and 2 . We clustered standard errors on MSAs to account for correlation in error terms within MSAs. Our analysis also included sample weights in regressions to correct for any bias caused by stratified sampling (eAppendix 1).

eAppendix 5. Comparison of Adjusted Outcomes Between Treatment and Control MSAs Across High-and Low-Dual Hospitals Each Year in the Pre-CJR Period
We compared adjusted outcomes between treatment and control MSAs across high-and lowdual hospitals each year prior to the CJR implementation. If we found no differences in outcomes between treatment and control MSAs each year in the pre-CJR period, that would suggest that random selection of MSAs for CJR participation worked well to select comparable treatment and control MSAs. This would also mean that we meet a parallel pre-trend assumption.
Our analysis included joint replacements that occurred prior to the public announcement of CJR model (between 2012 and 2014) in treatment and control MSAs. The unit of regression was each care episode that includes the index hospitalization and 90-day post-discharge care. We stratified the sample to two groups (joint replacements that occurred in high-and low-dual hospitals) and ran a linear ordinary least-squares regression within each of stratified group. We regressed outcomes on an indicator of treatment MSA (vs. control), indicators of year (with year 2012 as a reference group), and interactions between an indicator of treatment MSA and indicators of year. We also adjusted for patient characteristics (type of surgery, presence of major complications or comorbidities during hospital stay, age, and female), hospital characteristics (volume of hip/knee replacements, ownership, major teaching hospital) and MSA characteristics (post-acute care supply per 100,000 people, Medicare Advantage penetration rates, Herfindahl-Hirshman index of market concentration for hip/knee joint replacements in the MSA, indicator of 8 MSA groups based on historical spending and population size) and clustered standard errors on MSAs. We applied Bonferroni correction to outcomes to adjust for multiple testing and to keep familywise error rate at 0.05.
We found no significant differences in adjusted-outcomes when comparing treatment and control MSAs within high-dual hospitals and low-dual hospitals each year in the pre-CJR period (see eTable 2a and 2b). Our findings ensure comparability between treatment and control MSAs, and therefore we met a parallel pre-trend assumption.

eAppendix 6. Required Changes in Episode Spending Each Hospital Had/Has to Receive a Bonus (or Avoid a Penalty) in CJR Years 1-5
We assessed required savings that each hospital would have to make in order to receive a CJR bonus (or avoid a penalty) in CJR years 1-5. We calculated them as the difference between the hospital's CJR spending benchmark and historical episode spending.
CJR spending benchmark CMS sets CJR spending benchmark for each hospital as a weighted average of hospital historical spending and regional historical spending as seen in the following equation: where ℎ ℎ is the hospital's CJR spending benchmark.
ℎ is the hospital's historical episode spending and is the regional historical episode spending. is a weighting factor that determines the impact of hospital historical spending and regional historical spending on CJR spending benchmark. The values of were 2/3 in CJR years 1-2, 1/3 in CJR year 3, and 0 in CJR years 4-5. is a deflationary factor and was set to = (1 − 0.03).
CMS adjusted spending benchmark for hospital wage index, changes in pricing for services, and hospital performance on a composite quality score. We removed the effect of these adjustments from our calculations to focus on the impact of shift toward regional spending in setting CJR spending benchmark. We reported all dollar amounts in 2016 dollars.
We obtained four CJR spending benchmarks for each hospital because CMS sets separate CJR spending benchmark for four groups: (1) elective surgeries without the presence of major complications or comorbidities during the hospital stay (MCC), (2) elective surgeries with MCC, (3) fracture surgeries without MCC, and (4) fracture surgeries with MCC.
Each hospital's spending benchmarks for CJR years 1-2 are available online. 6 Hospital historical spending We obtained hospital historical spending for CJR years 1-2 by plugging spending benchmarks and regional historical spending into Equation (1). Regional historical spending for CJR years 1-2 is also available online. 6 We then estimated CJR spending benchmark for years 3-5 by plugging hospital and regional historical spending and weights (that change over time) into Equation (1).
Differences between CJR spending benchmark and hospital historical spending We calculated the difference between each hospital's historical spending in CJR year 1 with the estimated spending benchmarks for each year. The difference estimated per-episode savings that each hospital would need to achieve to receive a bonus (or avoid a penalty) in each year. Figure  3 in the main manuscript illustrates the mean required savings for high-and low-dual hospitals in CJR years 1-5. We calculated separate required savings for each of four groups (elective vs fracture surgery; with vs without MCC).