Association of Medicare Mandatory Bundled Payment Reform With Joint Replacement Surgery Use for Beneficiaries With Alzheimer Disease and Related Dementias

This cohort study evaluates the association of the Comprehensive Care for Joint Replacement model with utilization of joint replacements for Medicare beneficiaries with Alzheimer disease and related dementias.

Inpatient stays excluded: -Not admitted for MS-DRG 469/470 (n=85,578,540) -Did not meet MBSF inclusion criteria listed above (n=2,385,064) -Not admitted for short, inpatient stays (n=6,198) -Not admitted to hospitals reimbursed by the Inpatient Prospective Payment System (n=1,764) -Non-elective stays (n=182,375) -Medicare not primary payor for inpatient stay (n=67,677) -Expired during 90-day episode (n=451) -Admitted for fracture (n=7,073) -Not enrolled in Medicare Parts A and B at admission (n=17,750) -Admitted to hospital participating in BPCI Model 1 or in risk-bearing phase of Models 2 or 4 (n=128,457) -Not admitted for total hip or total knee replacement (n=5,750) -Duplicate observation (n=103) -Qualifying stay during the episode of previous stay (both stays/episodes dropped) (n=26,891) Inclusion criteria from merging the MBSF and MedPAR files (2013-2017) Note: The files were merged at the beneficiary-level. Two binary indicators (one for hip and one for knee replacements) were created to represent whether a beneficiary had one or more qualifying stay during the year.
-Beneficiaries with joint replacements that meet CJR criteria or beneficiaries who did not undergo any joint replacements -Beneficiaries with diagnosis of rheumatoid arthritis/osteoarthritis. Alternate hypothesis: In the pre-CJR period, the use of THRs/TKRs for beneficiaries with ADRD (compared to those without ADRD) is statistically significantly different between treatment and control MSAs.
Test for parallel trends assumption for triple differences models To assess whether the trends in the use of joint replacements for Medicare beneficiaries with or without ADRD were parallel in the period before the CJR was implemented (parallel trends assumption for the triple differences models), we estimated the following models (separate for hip and knee replacements). The data for these models was limited to 2013-2015 (pre-CJR period).  We used the Wald test to separately test for the statistical significance of , , and .
A p-value <0.05 on the test for any of the three estimates in a model represented a violation of the parallel trends assumption.
Results of the parallel trends tests for triple differences models The parallel trends tests were statistically significant only for hip replacements in the main analysis.
Hypotheses for triple differences model: Null hypothesis: With CJR implementation, the change in the use of THRs/TKRs for beneficiaries with ADRD (compared to those without ADRD) is not statistically significantly different between treatment and control MSAs.
Alternate hypothesis: With CJR implementation, the change in the use of THRs/TKRs for beneficiaries with ADRD (compared to those without ADRD) is statistically significantly different between treatment and control MSAs.

Model estimation for triple differences models
We estimated the following triple differences models (separate for hip and knee replacements) to assess the differential effect of the CJR. The triple differences approach includes estimating three differences: First, the difference between the pre-and post-CJR rates for each ADRD/non-ADRD group in CJR and non-CJR MSAs (difference 1). Second, the difference in difference 1 between CJR and non-CJR MSAs for each ADRD/non-ADRD group (difference 2). Third, the difference in difference 2 between beneficiaries with ADRD compared to those without ADRD (difference 3). Because of the violation of the parallel trends assumption, we included interactions of with and (for the hip replacement models) to account for the differential trends in the pre-CJR period. 1

eAppendix 3. Weighting strategy to account for MSA selection probability
We used the method by the Lewin Group to account for an MSA's probability of selection into the treatment (CJR MSAs) or control (non-CJR MSAs) group. 3 In this approach, the CJR MSAs were assigned a weight of 1 and the non-CJR MSAs were assigned weights to represent the CJR MSAs. These weights for the non-CJR MSAs were obtained by dividing the number of CJR MSAs in each of the 8 strata (constructed by the CMS using quartiles of pre-period episode spending and whether the MSA had above or below median population) by the number of non-CJR MSAs in that stratum.  Notes: Adjusted rates from patient-level multivariable linear regression models with robust/sandwich estimators of variance. The models assessed CJR's association with the use of surgeries for ADRD beneficiaries (versus beneficiaries without ADRD) in CJR MSAs versus non-CJR MSAs. The models controlled for age, sex, race/ethnicity, dual-eligibility, comorbidities, calendar year (and relevant interactions with CJR MSA and ADRD indicator), MSA fixed effects, and MSA weights. a Percentage point difference in the rates of surgeries for beneficiaries with ADRD (versus beneficiaries without ADRD) in CJR MSAs with CJR implementation versus non-CJR MSAs ("triple difference").