Skilled Nursing Facility Participation in a Voluntary Medicare Bundled Payment Program: Association with Facility Financial Performance

Abstract

Importance:

Model 3 of the Bundled Payments for Care Improvement (BPCI) is an alternative payment model in which an entity takes accountability for the episode costs. It is unclear how BPCI affected the overall skilled nursing facility (SNF) financial performance and the differences between facilities with differing racial/ethnic and socioeconomic status (SES) composition of the residents.

Objective:

To determine associations between BPCI participation and SNF finances and across-facility differences in SNF financial performance.

Design, Setting, and Participants:

Longitudinal study spanning 2010-2017, based on difference-in-differences analyses for 575 persistent-participation SNFs, 496 drop-out SNFs, and 13,630 eligible nonparticipating SNFs.

Main Outcome Measures:

Inflation-adjusted operating expenses, revenues, profit, and profit margin.

Results:

BPCI was associated with reductions of $0.63 million (M) in operating expenses, and $0.57M in operating revenues for the persistent-participation group, but had no impact on the drop-out group, compared to nonparticipating SNFs. Among persistent-participation SNFs, the BPCI-related declines were $0.74M in operating expenses, and $0.52M in operating revenues for majority-serving SNFs; and $1.33M and $0.82M in operating expenses and revenues, respectively, for non-Medicaid-dependent SNFs. The between-facility SES gaps in operating expenses were reduced (differential DID-estimate=$1.09M). Among drop-out SNFs, BPCI showed mixed effects on across-facility SES and racial/ethnic differences in operating expenses and revenues. The BPCI program showed no effect on operating profit measures.

Conclusions:

BPCI led to reduced operating expenses and revenues for SNFs that participated and remained in the program, but had no effect on operating profit indicators and mixed effects on SES and racial/ethnic differences across SNFs.

INTRODUCTION

To improve the value of healthcare and the efficiency of healthcare delivery, the Centers for Medicare and Medicaid Services (CMS) have proposed and tested several innovative payment models with a goal of providing better care, improving health outcomes, and lowering costs.¹ The transition from volume-based to value-based payments under current Medicare reforms has important implications for the delivery of post-acute care (PAC), because PAC is the fastest growing major spending category of overall Medicare expenditures,² and accounts for the largest share of geographic variation in total Medicare spending.³ Consequently, PAC is the target of several Medicare alternative payment models (APMs).

One of these APMs is the Medicare Bundled Payments for Care Improvement (BPCI), which is a voluntary program that started in 2013 and includes 4 separate bundled payment models differing in eligibility criteria for health care providers, types of services covered, reimbursement methodologies and conditions. These 4 models use the Medicare severity diagnosis-related group (MS-DRG) definition to identify episodes of care. Particularly, Model 1 focuses on all MS-DRGs for eligible beneficiaries, but other models allow participants to select up to 48 clinical episodes.⁴ For instance, major joint replacements, sepsis, and simple pneumonia and respiratory infections are the top 3 episodes of care (that started in PAC admission and included all care delivered in 30/60/90 days after the initiation) commonly selected by Model 3 SNFs.⁵ Additionally, Model 1 only covers all inpatient services for the participating acute-care hospitals.⁵ Model 2 allows physician group practices and acute-care hospitals to join, under which CMS bundles payments for care during an inpatient stay, PAC, and all other types of services (e.g., physician visits and laboratory tests) incurred up to 30/60/90 days after hospital discharge.⁵ Model 3, which is the focus of this study, targets any PAC (e.g., skilled nursing facility [SNF] care and inpatient rehabilitation) within 30/60/90 days following an SNF admission; PAC providers (e.g., SNFs, inpatient rehab facilities, and home health agencies) and physician group practices are eligible for participation in the model.⁵ Model 4 bundles payments for hospital and physician services during an inpatient stay and related readmissions within 30 days after discharge. ⁵ Under Models 1, 2, and 3, participating healthcare providers will either keep a portion of the savings if the aggregated fee-for-service payments (AFPs) are lower than a target price or repay some of the difference to CMS if the AFPs are higher than the target amount.^4,6 The target price is calculated based on historical payments, minus a discount for all BPCI-eligible episodes, which can vary depending on the episode length. For example, under Model 3, a discount of 3% is applied to the historical amount, regardless of the length of the episode. Model 4 participants are paid a single, prospectively determined amount.⁵ BPCI includes two phases. Phase 1 is a preparation stage for prospective participants to decide if to step into the risk-bearing phase. Phase 2 is a risk-bearing period with financial and performance responsibility for episodes of care.

SNFs make up approximately 89% of all Model 3 participants.⁷ Financially, SNFs are heavily dependent on Medicare, and their finances are likely to be affected by changes in Medicare reimbursement policy.^8,9 Specifically, the overall expenses, revenues, and profits of participating SNFs may be affected if care practices are changed in response to the BPCI financial incentives. For example, Barnett and colleagues revealed that BPCI Model 3 was associated with reduced institutional spending due to decreases in length of stay (LOS), and had no significant effect on mortality or readmissions among Medicare beneficiaries who underwent joint replacement procedures (the most commonly selected bundle) from 2013 through 2017.¹⁰ The changes in care practice might benefit other patients (including both in and not in the BPCI episode of care) as they are cared for in an SNF and share common facility features, staffing, and culture of practices. Therefore, the overall institutional payments, like operating expenses, might be reduced. However, prior studies mainly focused on associations between hospital participation in bundled payment models and inpatient costs and payments for selected episodes,^5,11–15 and results were mixed. Although some research^5,10 evaluated the effectiveness of BPCI Model 3, these analyses only focused on selected one/several bundles. No research has examined the relationship between BPCI participation and changes in SNF financial performance.

Nursing home care has been considered as a two-tiered system, with the lower tier consisting of facilities caring predominantly for Medicaid and/or racial and ethnic minority residents and characterized by more financial strain, lower staffing levels, and poorer quality of care than other facilities.^16,17 Bundled payment reforms may affect the existing racial/ethnic and socioeconomic status (SES) differences or disparities across SNFs.^18–20 SNFs offer important recuperation and rehabilitation services to Medicare beneficiaries recovering from a recent acute hospital stay. However, the quality and efficiency of PAC in SNFs is not optimal. Patients discharged to SNFs often were readmitted, failed to return to the community, and suffered from poor health outcomes.^19,21–29 Under the bundled payments, BPCI-participating SNFs are assumed to promote care efficiency and quality (e.g., shortening LOS, and reducing preventable rehospitalizations that are directly associated with costs and payments) to reduce costs and achieve savings. Improving the efficiency and quality of SNF care requires participating facilities to make substantial investments in organizational structure and process of care (e.g., electronic medical records implementation, improved discharge planning process, and additional training of staff). Since SNFs predominated by racial/ethnic minority and/or Medicaid residents have been consistently shown to have fewer resources available to invest due to a higher percentage of patients with Medicaid coverage and higher costs of caring for vulnerable patients,^30,31 their financial performance, such as operating expenses, may improve to a lesser degree compared to other participating SNFs. On the other hand, under bundled payments, facilities disproportionately serving vulnerable beneficiaries may have greater opportunities to achieve cost savings given their worse financial performance before participation in APMs.²⁰ For instance, accountable care organizations (ACOs) that served a high proportion of Medicare-Medicaid dually enrolled beneficiaries were more likely to realize cost reductions and share in savings under the Medicare Shared Savings Program (MSSP), even though they showed slightly higher preventable readmission rates and lower quality scores.^20,32 The overall effect of bundled payments on racial/ethnic and SES differences in financial performance across SNFs is unclear given the two different possibilities, and has not been empirically tested.

Therefore, this study aimed to evaluate whether SNF participation in BPCI was associated with changes in SNF financial performance, and whether the BPCI-associated effects differ by SNF groups thereby reducing or widening racial/ethnic and SES differences in financial performance across facilities.

METHODS

Data, Sample, and Study Period

Publicly available Medicare BPCI episode analytic files document detailed information on enrolled model and phase, targeted clinical episodes, and episode start and end dates for each BPCI-participating SNF. We linked the files to the corresponding “Nursing Home Compare” (NHC) files to identify 15,877 Medicare-certified BPCI-participating SNFs nationally. Outcomes were retrieved from the Medicare Cost Reports (MCR) that contain annual financial statement information³³ on free-standing SNFs. Therefore, we excluded 1,176 hospital-based SNFs from the study sample. Finally, of the 14, 701 free-standing SNFs, 575 participants were enrolled in the BPCI program persistently (at the risk-bearing phase) and did not drop out (i.e., persistently participating SNFs) during the period of October 1, 2013, to Jun 30, 2017, 496 participants joined the initiative at the risk-bearing phase initially then withdrew completely (i.e., drop-out SNFs) during the same period, and 13,630 facilities never participated in the initiative (at the risk-bearing phase) throughout the study period (i.e., eligible nonparticipants). See Figure 1A in supplemental online appendix for more details about the selection of the analytic sample.

Data for the years 2010-2017 from the NHC, LTCfocus (LTCFocus.org), CMS wage index, Area Health Resource Files, Master Beneficiary Summary (MBSF) base segment, Long-Term Care Minimum Data Set (MDS) 3.0, and ACO Public-Use files were exploited to define additional SNF and market characteristics. Our study period spanned from January 2010 to December 2017. See supplemental online appendix for more details about the definition of the study period.

Outcomes

The primary outcomes were (1) operating expenses; (2) operating revenues; (3) operating profit; and (4) operating profit margin. All financial accounting elements were derived from the MCR’s worksheets G and S.³³ As the fiscal year start dates and reporting periods vary by SNF, we constructed calendar-year financial measures and standardized them to 365 days. All outcomes were inflation adjusted to 2017 dollars using the medical care consumer price index.³⁴ See Tables 1.1A and 1.2A in supplemental online appendix for more details.

Independent variables

Three binary indicators that represented SNF BPCI participation status were defined. The first indicator defined if the SNFs participated in BPCI and did not drop out (persistent BPCI-participating SNFs, yes/no), the second indicator identified SNFs that joined BPCI initially then dropped out completely (BPCI-drop-out SNFs [that accounted for 46% of all participating SNFs], yes/no), and the last indicator identified eligible SNFs that never enrolled in BPCI nationally. The nonparticipants served as the reference group.

Each BPCI-participating SNF had its specific pre-BPCI and post-BPCI periods defined according to the SNF’s actual participation initiation date. An exposure measure was calculated as the proportion of months the SNF was exposed to the BPCI intervention in a given year. For instance, if a persistent-participating SNF’s participation date was October 1, 2013, the exposure variable was set to zero for 2010, 2011, and 2012, 3 of 12 = 0.25 in 2013, and 1 for each year from 2014-17. The value of BPCI exposure variable ranged from 0 to 1 in each year.

We defined two other dichotomous indicators to denote if the SNF disproportionately served racial/ethnic minority and Medicaid patients, respectively, using the MBSF base segment and MDS 3.0 files (see supplemental online appendix for more details). Following previous studies,^35,36 SNFs serving 35% (or higher) of minorities were categorized as minority-serving SNFs, and others were defined as majority-serving SNFs. We used 60% Medicaid enrollment as a cut-off point to categorize whether the SNF was predominated by Medicaid patients.

Covariates

Relevant facility-level and market-level covariates were adjusted for in regression analyses.^37–41 SNF variables included occupancy rate, chain affiliation (yes/no), a case mix index, number of beds, ownership (for-profit, nonprofit, or government-owned), overall five-star ratings, and number of selected BPCI-episodes. County-level characteristics included annual average per capita income, unemployment rate, a measure of SNF market concentration determined by the Herfindahl-Hirschman index, urban location (yes/no), CMS SNF wage index, Medicare Advantage (MA) penetration, and ACO penetration. A set of state dummies were used to account for unobserved geographic factors. Trends in financial outcomes were captured via year dummies.

Statistical Analysis

We performed bivariate analyses to compare differences in SNF and market characteristics among persistent-participating SNFs, drop-outs, and non-participating SNFs, and between the pre-BPCI and post-BPCI period, using ${\chi }^2$ tests for categorical variables and t-tests for continuous variables.

To evaluate the associations between BPCI participation and changes in SNF finances, a difference-in-differences (DID) framework was specified. Generalized linear models (GLMs) with log link function were fitted for the skewed and heavy-tailed financial outcomes,⁴² including operating expenses, revenues, and profit; and a GLM model with identity link function was fitted for operating profit margin with a value between 0 and 1. All adjusted analyses included two indicators for SNF participation (i.e., persistent-participating and drop-out SNFs), a measure for SNF BPCI exposure, their interactions, and covariates described above. To test the parallel trends assumption, we conducted GLM regressions of SNF financial indicators on BPCI participation status, year dummies, and their interactions, during the pre-intervention period (i.e., 2010-2012) (Table 2A in supplemental online appendix). We also plotted unadjusted outcome means by SNF participation status (Figure 1). Both statistical and graphic analyses demonstrated that this assumption was not violated.

Figure 1:

operating expenses, operating revenues, operating profit, and operating profit margin, by participation status.

Source/Notes: “Author’ analysis of data from the Medicare Cost Reports and Medicare BPCI episode analytic files, 2010-2017.

We fitted another set of DID models with a GLM approach to assess the relationships between SNF participation in bundled payments and potential between-facility racial/ethnic and SES differences in SNF finances by adding to the DID model triple interactions among the indicators for BPCI participation status, the SNF exposure measure, and two separate indicators for minority-serving and Medicaid-dependent SNFs. Additionally, we checked the robustness of results in the preferred main findings by examining alternative cut-off points including the highest quartile and highest quintile for the categorization of minority-serving versus (vs.) majority-serving SNFs, and Medicaid-dependent SNFs vs. non-Medicaid-dependent SNFs. All models used heteroscedasticity-robust standard errors to account for potential invalid assumption of homoscedasticity.⁴³

We reported marginal effects of GLM coefficients for easing interpretations of estimates. Statistical significance was set at the 2-tailed P<0.05. All analyses were performed using Stata (version13.1, Stata Corp) and SAS (version9.4. SAS Institute Inc). See supplemental online appendix for more details about the model specifications.

RESULTS

Our sample included 13,630 eligible nonparticipants, 575 persistent-participating SNFs, and 496 drop-outs between 2010 and 2017(Figure 1A). Compared with controls, BPCI-participating SNFs were more likely to have higher occupancy rate and case mix, to be part of chain, for-profit, and have more beds. BPCI-participating SNFs were located in more competitive urban areas with higher per capita income, and higher indexes of MA penetration and SNF wage (Table 1). Persistent-participating SNFs had lower proportions of Medicaid and racial/ethnic minority residents and higher five-star ratings, whereas drop-outs showed a higher percentage of racial/ethnic minority residents and lower five-star ratings than non-participating SNFs (Table 1).Within each group of SNFs, facilities showed similar characteristics between the pre- and post-BPCI period and among groups defined based on BPCI participation status (Tables 3A, 4A, and 5A in supplemental online appendix).

Table 1:

Characteristics of skilled nursing facilities (SNFs) in Model 3 of Bundled Payments for Care Improvement (BPCI) Initiative, 2017.

	Control (n=13,096)	Persistent (n=575)	Drop-out (n=488)	Persistent versus Control SNFs: P	Drop-out versus Control SNFs: P
	Mean ± SD or Prevalence (%)
Facility-level characteristics
Percent of Medicaid patients in nursing home	54.01(22.81)	46.97(19.95)	55.82 (19.61)	<0.001	0.08
Percent of Minority patients in nursing home	19.86(22.17)	16.34(18.07)	25.92(21.80)	<0.001	<0.001
Five-star ratings	3.29(1.37)	3.61(1.33)	2.89(1.38)	<0.001	<0.001
Occupancy rate	79.75(15.88)	81.72(13.00)	81.90(13.98)	<0.001	0.003
Case mixa	1.17(0.15)	1.22(0.11)	1.21(0.14)	<0.001	<0.001
Chain affiliation, %	55.85	85.57	80.12	<0.001	<0.001
Ownership, %
For-profit	72.96	81.57	93.85
Nonprofit	21.63	17.22	3.48
Government	5.41	1.22	2.66	<0.001	<0.001
Number of beds	108.98(58.12)	110.72(49.93)	128.44(56.13)	0.48	<0.001
Number of selected BPCI-episodes	--	12.61(12.02)	0.38(1.91)	--	--
Market-level characteristics
SNF market concentrationb	0.79(0.25)	0.85(0.19)	0.85(0.22)	<0.001	<0.001
Urban,%	72.96	83.83	84.63	<0.001	<0.001
Per capita income	48607.87(13983.31)	51894.11(15402.87)	50822.50(13848.16)	<0.001	<0.001
Unemployment rate (16+)	4.42(1.23)	4.44(1.01)	4.44(1.09)	0.72	0.76
MA penetrationc	0.30(0.13)	0.33(0.12)	0.31(0.12)	<0.001	0.06
ACO penetrationd	0.15 (0.12)	0.16 (0.11)	0.16 (0.11)	0.18	0.03
SNF wage index	0.95(0.19)	1.01 (0.21)	0.97 (0.16)	<0.001	0.05

^aThe average Resource Utilization Group Nursing Case Mix Index, measuring the intensity of care needed by different SNF residents.

^bAccording to Herfindahl-Hirschman Index, calculated as 1- the sum of squared shares of SNF beds for all SNFs in the county. Market competitiveness scores ranged from 0 (least competition) to 1(highest competition).

^cIt was calculated as the proportion of total Medicare Advantage beneficiary person-years to total Medicare beneficiaries in the county. Medicare Advantage penetration ranged from 0 to 1.

^dIt was calculated as the proportion of total assigned beneficiary person-years to total Medicare beneficiaries in the county. Accountable Care Organization penetration ranged from 0 to 1.

BPCI=Bundled Payments for Care Improvement; SNF=skilled nursing facility; MA=Medicare Advantage; ACO=Accountable Care Organization.

Figure 1 presents trends in the key financial outcomes for the three SNF groups. SNFs in each of the BPCI-participating groups showed decreasing operating expenses and operating revenues, and increasing then slightly declining operating profit and operating profit margin throughout the study period. For example, the average operating expenses were $10.77 million(M), $11.77M, and $13.33M for the control, persistent-participating, and drop-out SNFs in 2010, and decreased to $10.47M, $11.04M, and $11.83M, respectively, in 2017. The corresponding unadjusted reductions were $0.29M, $0.73M, and $1.50M.

Table 2 presents unadjusted outcomes in the pre-BPCI and post-BPCI periods and adjusted DID estimates after the launch of bundled payments for SNFs in the persistent-participating group relative to the control group. The BPCI policy was associated with reductions of $0.63M (95% confidence interval [CI], $−1.13M to $−0.13M, P=0.01) in operating expenses, and $0.57M (95% CI, $−1.00M to $−0.14M, P=0.009) in operating revenues. Additionally, the BPCI-associated decreases in operating expenses were $0.74M (95% CI, $−1.28M to $−0.19M, P=0.008) for majority-serving SNFs, and $1.33M (95% CI, $−1.95M to $ −0.71M, P<0.001) for non-Medicaid -dependent SNFs; while such associations were not found for both minority-serving SNFs and Medicaid-dependent SNFs. We further found that differences in the BPCI-related declines in expenses between minority- and majority-serving SNFs were insignificant, while such differences were significantly reduced between Medicaid and non-Medicaid-dependent SNFs (differential DID-estimate=$1.09M, 95% CI, $0.22M to $1.95M, P=0.01). BPCI was associated with declines in operating revenues by $1.02M (95% CI, $−2.01M to $−0.02M, P=0.05) for minority-serving SNFs, by $0.52M (95% CI, $−0.98M to $−0.06M, P=0.03) for majority-serving SNFs, by $0.68M (95% CI, $−1.28M to $−0.08M, P=0.03) for Medicaid-dependent SNFs, and by $0.82M (95% CI, $−1.36M to $−0.29M, P=0.003) for non-Medicaid-dependent SNFs. However, the reductions did not differ significantly between SNF groups defined by resident socio-demographics. Meanwhile, BPCI had no effect on operating profit or profit margin for persistent-participating SNFs, or SNFs in any subgroups.

Table 2:

Changes in operating expenses, operating revenues, operating profit, and operating profit margin, by participation status and skilled nursing facility (SNF) (BPCI persistent SNFs versus control SNFs).

	BPCI persistent SNFs			Control SNFs			DID, unadjusted	DID, adjusted		Differential DID, adjusted
	Pre-BPCI (1)	Post-BPCI (2)	Diff (3)=(2)-(1)	Pre-BPCI (4)	Post-BPCI (5)	Diff (6)=(5)-(4)	(7)=(3)-(6)	Estimate (95% CI) (8)	P	Estimate (95% CI) (9)=(8)minority/Medicaid-(8)majority/non-Medicaid	P
a Operating expenses, million $
All	11.75	11.32	−0.44	10.72	10.50	−0.22	−0.22	−0.63 (−1.13, −0.13)	0.01
Minority-serving	13.03	13.31	0.27	11.80	11.25	−0.55	0.82	−0.87 (−1.87, 0.13)	0.09
Majority-serving	11.59	11.03	−0.56	10.49	10.32	−0.17	−0.39	−0.74 (−1.28,−0.19)	0.008	−0.13 (−1.20, 0.95)	0.81
Medicaid-dependent	10.03	9.78	−0.25	11.51	11.50	−0.01	−0.24	−0.24 (−0.95,0.47)	0.51
Non-Medicaid-dependent	12.19	11.79	−0.40	9.44	9.11	−0.33	−0.08	−1.33 (−1.95, −0.71)	<0.001	1.09 (0.22, 1.95)	0.01
a Operating revenues, million $
All	14.28	14.48	0.19	12.25	12.50	0.25	−0.06	−0.57 (−1.00, −0.14)	0.009
Minority-serving	15.62	16.53	0.91	13.70	14.19	0.49	0.42	−1.02 (−2.01, −0.02)	0.05
Majority-serving	14.12	14.18	0.06	11.95	12.10	0.15	−0.09	−0.52 (−0.98, −0.06)	0.03	−0.49 (−1.54, 0.56)	0.36
Medicaid-dependent	12.02	12.19	0.16	10.99	11.34	0.35	−0.19	−0.68 (−1.28, −0.08)	0.03
Non-Medicaid-dependent	14.86	15.18	0.32	13.02	13.33	0.31	0.01	−0.82 (−1.36, −0.29)	0.003	0.14 (−0.58, 0.86)	0.70
a Operating profit, million $
All	2.49	3.12	0.63	1.69	1.97	0.28	0.35	−0.04 (−0.34, 0.27)	0.81
Minority-serving	2.58	3.22	0.64	2.20	2.63	0.43	0.21	−0.10 (−0.80, 0.60)	0.78
Majority-serving	2.48	3.11	0.62	1.59	1.82	0.23	0.40	−0.00 (−0.32, 0.32)	1.00	−0.10 (−0.84, 0.64)	0.80
Medicaid-dependent	2.00	2.40	0.41	1.66	2.02	0.36	0.05	−0.23 (−0.59, 0.14)	0.22
Non-Medicaid-dependent	2.62	3.34	0.72	1.71	1.94	0.22	0.50	0.22 (−0.14, 0.59)	0.22	−0.45 (−0.91, 0.00)	0.05
b Operating profit margin
All	15.72	19.16	3.44	11.19	12.74	1.55	1.90	1.17 (−4.62, 6.95)	0.69
Minority-serving	14.37	17.95	3.58	12.62	14.78	2.16	1.43	3.49 (−0.98, − 7.96)	0.13
Majority-serving	15.88	19.33	3.45	10.89	12.25	1.36	2.09	0.99 (−5.68, 7.66)	0.77	2.50 (−4.01, 9.02)	0.45
Medicaid-dependent	13.13	16.34	3.21	11.56	13.29	1.72	1.49	2.68 (−9.95, 15.30)	0.68
Non-Medicaid-dependent	16.38	20.02	3.65	10.96	12.34	1.38	2.27	3.48 (0.28, 6.68)	0.03	−0.81 (−13.13, 11.52)	0.90

^aGeneralized linear models with log link function adjusted for skilled nursing facility and geographic covariates listed in Table 1, as well as State and year dummies. The number scale is 1,000, 000.

^bGeneralized linear model adjusted for skilled nursing facility and geographic covariates listed in Table 1, as well as State and year dummies.

BPCI=Bundled Payments for Care Improvement; SNF=skilled nursing facility; DID= difference-in-differences; 95% CI=95% confidence interval.

Table 3 shows changes in the financial outcomes before and after the BPCI initiative for drop-out SNFs relative to non-participating (control) SNFs. BPCI was not associated with changes in financial performance for drop-out SNFs overall, but was significantly associated with changes for SNFs in several subgroups. Specifically, operating expenses were increased by $1.18M (95% CI, $0.02M to $2.33M, P=0.05) for minority-serving SNFs, and declined by $0.85M (95% CI, $ −1.31M to $−0.39M, P=0.001) for majority-serving SNFs, leading to exacerbated across-facility differences (differential adjusted DID =$2.03M, 95%CI, $0.81M to $3.24M, P=0.001) associated with bundled payments. Under the BPCI policy, Medicaid-dependent SNFs suggested increased operating expenses(adjusted DID= $1.05M, 95% CI, $0.33M to $2.07M, P=0.04), but non-Medicaid-dependent SNFs showed reduced operating expenses (adjusted DID=− $1.23M, 95% CI, $−1.72M to $−0.74M, P<0.001), resulting in reduced differences between the groups (differential adjusted DID=$2.28M, 95% CI, $1.17M to $3.39M, P<0.001). In terms of operating revenues, the BPCI-associated reductions were $0.82M (95% CI, $−1.28M to $−0.37M, P<0.001) for majority-serving SNFs, and $1.10M (95% CI, $−1.70M to $−0.51M, P<0.001) for non-Medicaid-dependent SNFs, but there were no BPCI-associated changes for minority-serving and Medicaid-dependent SNFs. Consequently, the across-SNF racial /ethnic gaps were significantly increased (differential adjusted DID=$1.56M, 95% CI, $0.37M to $2.74M, P=0.01), whereas the between-facility SES differences were significantly reduced (differential adjusted DID=$1.42M, 95% CI, $0.46M to $2.38M, P=0.004). The impact of BPCI on operating profit-related outcomes and across-SNF differences was statistically insignificant.

Table 3:

Changes in operating expenses, operating revenues, operating profit, and operating profit margin, by participation status and skilled nursing facility (SNF) (BPCI drop-out SNFs versus control SNFs).

	BPCI drop-out SNFs			Control SNFs			DID, unadjusted	DID, adjusted		Differential DID, adjusted
	Pre-BPCI (1)	Post-BPCI (2)	Diff (3)=(2)-(1)	Pre-BPCI (4)	Post-BPCI (5)	Diff (6)=(5)-(4)	(7)=(3)-(6)	Estimate (95% CI) (8)	P	Estimate (95% CI) (9)=(8)minority/Medicaid-(8)majority/non-Medicaid	P
a Operating expenses, million $
All	13.10	12.26	−0.85	10.72	10.50	−0.22	−0.63	−0.16 (−0.71, 0.38)	0.56
Minority-serving	15.27	14.29	−0.98	11.80	11.25	−0.55	−0.43	1.18 (0.02, 2.33)	0.05
Majority-serving	12.50	11.55	−0.94	10.49	10.32	−0.17	−0.77	−0.85 (−1.31, −0.39)	<0.001	2.03 (0.81, 3.24)	0.001
Medicaid-dependent	12.69	12.08	−0.61	11.51	11.50	−0.01	−0.60	1.05 (0.03. 2.07)	0.04
Non-Medicaid-dependent	13.30	12.38	−0.92	9.44	9.11	−0.33	−0.59	−1.23 (−1.72,−0.74)	<0.001	2.28 (1.17, 3.39)	<0.001
a Operating revenues, million $
All	17.21	16.89	−0.32	12.25	12.50	0.25	−0.57	−0.45 (−0.91, 0.01)	0.05
Minority-serving	19.45	19.23	−0.22	13.70	14.19	0.49	−0.71	0.73 (−0.39, 1.85)	0.20
Majority-serving	16.59	16.09	−0.50	11.95	12.10	0.15	−0.65	−0.82 (−1.28, −0.37)	<0.001	1.56 (0.37, 2.74)	0.01
Medicaid-dependent	15.70	15.88	0.18	10.99	11.34	0.35	−0.16	0.31 (−0.46, 1.09)	0.43
Non-Medicaid-dependent	17.93	17.60	−0.33	13.02	13.33	0.31	−0.64	−1.10 (−1.70, −0.51)	<0.001	1.42 (0.46, 2.38)	0.004
a Operating profit, million $
All	4.11	4.64	0.53	1.69	1.97	0.28	0.25	0.14 (−0.23, 0.52)	0.46
Minority-serving	4.18	4.94	0.76	2.20	2.63	0.43	0.33	0.67 (−0.13, 1.46)	0.10
Majority-serving	4.09	4.53	0.44	1.59	1.82	0.23	0.22	−0.01 (−0.43, 0.40)	0.95	0.68 (−0.21, 1.57)	0.13
Medicaid-dependent	3.01	3.80	0.79	1.66	2.02	0.36	0.44	0.64 (0.13, 1.16)	0.01
Non-Medicaid-dependent	4.63	5.22	0.60	1.71	1.94	0.22	0.37	0.05 (−0.47, 0.57)	0.86	0.59 (−0.13, 1.31)	0.11
b Operating profit margin
All	19.02	22.83	3.81	11.19	12.74	1.55	2.26	3.98 (−1.52, 9.48)	0.16
Minority-serving	18.30	22.17	3.87	12.62	14.78	2.16	1.71	4.56 (0.42, 8.70)	0.03
Majority-serving	19.22	23.05	3.84	10.89	12.25	1.36	2.48	4.50 (−2.31, 11.31)	0.20	0.06 (−6.97, 7.08)	0.99
Medicaid-dependent	15.36	20.40	5.04	11.56	13.29	1.72	3.32	5.68 (−6.42, 17.78)	0.36
Non-Medicaid-dependent	20.73	24.52	3.79	10.96	12.34	1.38	2.41	5.23 (1.55, 8.90)	0.005	0.45 (−11.80, 12.72)	0.94

^aGeneralized linear models with log link function adjusted for skilled nursing facility and geographic covariates listed in Table 1, as well as State and year dummies. The number scale is 1,000, 000.

^bGeneralized linear model adjusted for skilled nursing facility and geographic covariates listed in Table 1, as well as State and year dummies.

BPCI=Bundled Payments for Care Improvement; SNF=skilled nursing facility; DID=difference-in-differences; 95% CI=95% confidence interval.

Table 6A in supplemental online appendix showed that sensitivity analysis results were qualitatively similar to those in the main specifications, with the exception of the reduced SES differences in operating profit among drop-outs (4th highest quartile defined as the cut point: differential adjusted DID=$1.25M, 95% CI, $0.36M to $1.14M, P=0.006; 5th highest quintile defined as the cut point: differential adjusted DID=$1.61M, 95% CI, $0.60M to $2.63M, P=0.002).

DISCUSSION

In this analysis of the Medicare BPCI Model 3 from 2010 to 2017, SNFs that persistently enrolled in the program were able to reduce operating expenses and revenues without reducing their total operating profits or profit margins. SNFs that initially participated in BPCI but then dropped out of the program did not show these changes in financial performance measures. In addition, the BPCI effects on expenses and revenues were sometimes different for racial/ethnic minority-serving SNFs vs. other SNFs, and for Medicaid-dependent SNFs vs. other SNFs, leading to increased or decreased gaps between SNF groups. However, no evidence was found that BPCI had an impact on the existing between-SNF differences in operating profit or profit margin.

There are several possible reasons to explain the differential overall BPCI effects on operating expenses and operating revenues for persistent-participating SNFs and for drop-out SNFs. First, in our study, drop-outs were more likely to care for patients who were clinically and socio-demographically disadvantaged and had worse quality of care (measured by the overall five-star ratings) than persistent-participating SNFs. SNFs participating in BPCI need to invest substantial resources in structure and process of care in response to the financial incentives of bundled payments and avoid penalties. However, the essential resources for changes in care delivery were probably less available to participants^30,31 that disproportionately served medically and socially complex patients, i.e. the drop-out SNFs. Future research is needed to determine if resource strains or other factors may underlie the drop out of this subgroup of SNFs and the lack of BPCI effect on the financial status of these facilities. Another possible reason is that, as our data show, persistent-participating SNFs chose more episodes than drop-outs for bundled payments. Therefore, compared to the drop-out group, persistent-participating SNFs tended to have more of their patients subject to the BPCI and were more motivated for fundamental practice changes, perhaps leading to changes in expenses and revenues related to patient care. Specifically, persistent-participating SNFs may be more financially incentivized to reduce their average LOS, payments and costs (that are related to expenses and revenues directly) for the selected episodes of care, as documented in previous studies.^10–12,14 These changes in practice may spillover to SNF patients not in the targeted episodes of care as short-term post-acute patients of targeted and non-targeted episodes (and to a lesser extent, long-stayers as well) cared for in a SNF share common facility characteristics, staff, and organizational policies.

Financial profitability plays a vital role in the delivery of adequate nursing home care and the stability of facility operation. We found no evidence that BPCI Model 3 was associated with changes in operating profits or profit margins for participating SNFs, implying that SNF participation in BPCI did not harm the operation of the facility financially. This is likely due to the simultaneous declines in operating expenses and revenues over time that were associated with bundled payments. For example, for the persistent-participating group, the BPCI-associated reduction in operating expenses was $0.63M, and the corresponding decrease in operating revenues was $0.57M. The two relative changes offset each other, resulting in no considerable changes in operating profit.

These effects of BPCI Model 3 on the financial performance of persistent-participating SNFs, reduced operating expenses and revenues but no change in profits, suggest that bundled payment initiatives may serve as an effective APM to enhance the efficiency of PAC without financially hurting the operation of the facility, particularly given the current context that the Covid-19 crisis may have worsened the financial status of many SNFs and other long-term care facilities. Meanwhile, Holmgren and colleagues pointed out that only 3.7% of eligible PAC providers ever participated in BPCI Model 3, and 43.2% of those that enrolled dropped out.⁶ These statistics as well as our findings of the lack of effect on the financial performance of SNFs in the drop-out group indicate that policy makers need to improve strategies to attract more eligible facilities to participate and remain in this voluntary program in a more persistent way. For example, the program in the future could require an at least three-year participation period for those SNFs that enrolled in the program voluntarily and include extra bonuses for facilities that both remain in the program and demonstrate maintained or improved efficiency.

Our analyses further revealed that the differences in operating profit and profit margin between SNFs serving disproportionately more Medicaid or racial/ethnic minority residents and other SNFs remained unchanged under BPCI Model 3, and that the BPCI-associated reductions in operating expenses and revenues were lesser for minority-serving and Medicaid-dependent SNFs. These results were consistent with recent evidence^44,45 that under bundled payment models, healthcare institutions disproportionately serving socioeconomically disadvantaged patients were more likely to be financially penalized and less likely to receive bonuses, due to the fact that they were less able to achieve reduced costs of care or improved efficiency compared to their counterparts. Our results also suggest that current bundled payment programs do not help reduce the across-facility differences, or potential disparities, in profitability of care, which underscores the importance of future redesign of the bundled payment model to address the issue of healthcare inequalities. For example, under current BPCI Model 3, the target price for an episode of care is calculated solely based on a facility’s historic cost based on claims data without adjustment for other factors, such as the extent to which the facility is caring for medically or socioeconomically vulnerable patients. Going forward, CMS should adjust for patient sociodemographic factors in its target price algorithm to give facilities credit for caring for socially and clinically vulnerable patients.⁴⁵

This study has several limitations. First, we focused on SNF participants, the largest participating group of BPCI Model 3,⁷ so our findings may not generalize to other PAC participants, such as inpatient rehabilitation facilities, and home health agencies. Second, since financial data from SNF MCR files do not undergo systematic audits, financial information may not be completely accurate. However, the validity and reliability of the MCR data has been confirmed before, and the MCR is the only national data available for assessing the effect of payment policy change on SNF financial performance.³⁷ Third, results might be biased due to unmeasured confounders, such as local socio-economic and facility leadership factors that might be associated with the choice of BPCI participation, although our multivariable analyses controlled for important SNF, county, and state covariates that may be associated with SNF financial performance. Fourth, the overall trends in decreased expenses and revenues for SNFs in all groups might be due to other recent Medicare APMs. During our study period of 2010-2017, CMS initiated several value-based programs. For example, the MSSP started in 2012 and allowed groups of healthcare providers to assume collective accountability for the quality and cost of care furnished.⁴⁶ In addition, Model 2 of the BPCI covers beneficiaries’ inpatient stay plus all PAC use.⁵ Recent studies^41,47,48 showed that these APMs led to decreases in SNF use after hospital discharge, and among those discharged to SNF for post-acute care, reduced length of stay. We were not able to completely control for these confounding with current data although our use of the DID framework, assuming other APMs were affecting different SNF groups in roughly the same way, helped minimize bias due to uncontrolled confounding. Finally, this study primarily focused on the association of SNF BPCI enrollment with changes in financial outcomes. Other important SNF-specific outcomes (e.g., case mix, LOS, and unplanned readmissions), which were related to operating expenses and payments, were not captured. Further research needs to evaluate those outcomes to understand how the BPCI-enrolled SNFs cut their operating expenses and whether reducing care costs affect quality of SNF care.

In conclusion, BPCI was associated with reduced expenses and revenues for SNF in the persistent-participating group, but these associations were not found for SNFs that dropped out of BPCI Model 3. The associations between BPCI participation and changes in gaps in financial performance between minority-serving or Medicaid-dependent SNFs and their counterparts were mixed. Current bundled payment models may be improved by incorporating additional incentives to reduce program disenrollment and to address existing healthcare inequalities between facilities.