Outcomes by Race and Ethnicity Following a Medicare Bundled Payment Program for Joint Replacement

Key Points Question What outcomes were associated with race and ethnicity following implementation of the Comprehensive Care for Joint Replacement (CJR) program, a traditional Medicare bundled payment model? Findings In this cohort study of 309 834 hospitalizations for lower-extremity joint replacement in California from 2014 to 2017, the CJR model was associated with a larger increase in home discharge rates for Hispanic patients with Medicare Advantage vs non-Medicare coverage compared with their non-Hispanic White counterparts. Meaning These findings suggest that the CJR model may increase home discharge rates among Hispanic patients outside of traditional Medicare and thereby reduce discrepancy in postacute care.


eAppendix. Equations for the Event Study (A), Difference-in-Differences (B) and Triple Differences Analysis (C)
Below we show the main regression equations used in this study.In (A), we show the equation for the event study: Y ihmq = a hm + y q + q * Treated m + X' ihmq {3 + E ihmq (A) where Y ihmq is the outcome of interest (log adjusted length of stay or discharge home) for patient i treated in hospital h in MSA m and quarter-year q.The regression includes indicators for hospitals, a hm , to control for fixed differences in outcomes across hospitals and indicators for quarter-year (e.g, Q1 2014; Q2 2014, …) y q , to flexibly control for general time trends as well as a set of patient characteristics included in the discharge dataset, age and its square, sex (1=female; 0 otherwise), race/ethnicity, and whether the patient had a major complication or comorbidity (MS-DRG 470 or 469).Our interest is in q the coefficients on the quarter-year fixed-effects interacted with the treatment indicator for whether the MSA where the hospital was located was randomized to the CJR treatment.We omitted the interaction term for the first quarter of 2016 such that estimates are normalized to the quarter before CJR took effect.Consequently, the coefficients show the temporal difference in outcomes between treated and control hospitals relative to the reference period and allow us to assess whether our difference-in-differences estimates capture a change in outcomes that is credibly related to CJR.
In (B), we show the equation for the difference-in-differences analysis.
Y ihmq = a hm + y q + oPost mq + X' ihmq {3 + E ihmq (B) where Y ihmq , a hm , y q and X' ihmq are as defined above and post mq is an indicator that captures the interaction between the post CJR period (Q2 of 2016 and later) interacted with an indicator for the treated hospitals (i.e., hospitals located in MSAs randomized to CJR).Note that the main post-period effect is subsumed in the quarter indicators and the main MSA effects are subsumed in the hospital indicators.Our key parameter of interest is the coefficient o on post mq , which captures the differential change in outcomes in hospitals randomized to CJR after the program was implemented compared to the change for hospitals not randomized to participate in CJR.
In (C), we show the equation for the triple differences analysis: where Y ihmq , a hm , y q , X' ihmq and post mq are as defined above.Post qnw is an indicator that captures the interaction between post CJR period (Q2 of 2016 and later) interacted with an indicator for the racial minority patient.Treated mnw is an indicator that captures the interaction between an indicator for the treated hospitals (i.e., hospitals located in MSAs randomized to CJR) interacted with an indicator for the non-White patient.Post mqnw is an indicator that captures the triple interaction between post CJR period and an indicator for the treated hospitals interacted with an indicator for the non-White patient.Our key parameter of interest was is the coefficient ). on post mqnw , which captures the difference between the differential change in outcomes among racial minority patients in hospitals randomized to CJR after the program was implemented compared to the change for hospitals not randomized to participate in CJR and the differential change in outcomes among non-Hispanic White patients in hospitals randomized to CJR after the program was implemented compared to the change for hospitals not randomized to participate in CJR.

eFigure 2 .
Unadjusted Changes in Health Care Services Utilization Among Traditional Medicare Hispanic Patients vs Non-Hispanic White Patients in Treatment and Control MSAs in California From 2014 to 2017 (N = 99,132) eFigure 3. Changes in Health Care Services Utilization Among Non-Hispanic White and Hispanic Medicare Advantage Patients in Treatment Relative to Control MSAs in California From 2014 to 2017 (N = 72,895) A) Non-Hispanic White Patients (N=59,078) B) Hispanic Patients (N=13,817) NOTE: 0 quarter-year represents the second quarter of 2016.-year represents the second quarter of 2016.

eFigure 5 .
Changes in Proportion of Hispanic Patients in Treatment Relative to Control MSAs in California From 2014 to 2017 (N = 270,245) NOTE: 0 quarter-year represents the second quarter of 2016; Analysis was weighted based on the number of admissions per hospital.eFigure 6. Changes in Proportion of Racial Minority Patients in Treatment Relative to Control MSAs in California From 2014 to 2017 (N = 309,834) NOTE: 0 quarter-year represents the second quarter of 2016; Analysis was weighted based on the number of admissions per hospital.

eTable1. Characteristics of Patients in Treatment and Control MSAs in California by Race and Ethnicity From 2014 to 2017 (N=309,834)
NOTE: MCC = Major Complication or Comorbidity; Others include Native American, Eskimo, Aleut and others.