The effects of Accountable Care Organizations forming preferred Skilled Nursing Facility networks on market share, patient composition, and outcomes

Abstract

Background

Through participation in payment reforms such as bundled payment and accountable care organizations (ACOs), hospitals are increasingly financially responsible for healthcare use and adverse health events occurring after hospital discharge. To improve management and coordination of post-discharge care, ACO hospitals are establishing a closer relationship with skilled nursing facilities (SNFs) through the formation of preferred SNF networks.

Research Design

We evaluated the effects of preferred SNF network formation on care patterns and outcomes. We included 10 ACOs that established preferred SNF networks between 2014 and 2015 in the sample. We first investigated whether hospitals “steer” patients to preferred SNFs by examining the percentage of patients sent to preferred SNFs within each hospital before and after network formation. We then used a difference-in-difference model with SNF fixed effects to evaluate the changes in patient composition and outcomes of preferred SNF patients from ACO hospitals after network formation relative to patients from other hospitals.

Results

We found that preferred network formation was not associated with higher market share or better outcomes for preferred SNF patients from ACO hospitals. However, we found a small increase in the average number of Elixhauser comorbidities for patients from ACO hospitals after network formation, relative to patients from non-ACO hospitals.

Conclusion

After preferred SNF network formation, there is some evidence that ACO hospitals sent more complex patients to preferred SNFs, but there was no change in the volume of patients received by these SNFs. Furthermore, preferred network formation was not associated with improvement in patient outcomes.

Introduction

Medicare expenditure on post-acute care rose rapidly in the 1990s and early 2000s (1). This growth has been often attributed to Medicare’s separate payment to hospitals and post-acute care providers, which provided little incentive for efficient use of post-acute care or coordination of care across settings (2). In response to these concerns, the Affordable Care Act included several new payment approaches, such as bundled payment and accountable care organizations (ACOs), where a hospital or a physician group was responsible for post-discharge costs and patient outcomes (3).

A key challenge facing hospitals participating in ACOs was how to coordinate and use post-acute care more efficiently. One strategy hospitals used to influence post-discharge care was to establish closer relationships with skilled nursing facilities (SNFs) through informal integration, whereby hospitals selectively formed strong ties with SNFs while remaining under separate ownership. An example of such informal integration is “preferred SNF networks”. Preferred SNFs are selected based on historical costs, quality of care, or historical volume (4–8). ACOs established preferred SNF networks with several expectations. First, while hospitals are not allowed to compel patients to choose a particular SNF, they can provide patients with a suggested list of preferred providers (“soft steering”) (5, 9). To the extent that these facilities provide lower cost and better quality of care, channeling patients to these SNFs might improve ACO performance on cost and quality. Second, preferred networks might be able to promote care coordination between hospitals and preferred SNFs through electronic health record exchange, improving continuity of care by having ACO-affiliated clinicians and staff remain actively involved over the course of the entire episode (8, 10, 11). Finally, in each Medicare ACO program, ACOs were allowed to designate a set of partner SNFs where attributed patients could be admitted without the typically required preceding 3-day hospital stay (known as the 3-day waiver program), which might facilitate greater coordination and efficiency in post-acute care (12, 13). However, a few qualitative studies showed that it might be too early to tell whether those expectations are met and there exist some challenges and complexities. For example, ACO hospitals are not able to choose SNFs for patients and many patients chose SNFs based on location instead of quality of care (29). Therefore, it remains an open question whether establishing preferred SNF networks will significantly improve outcomes or lower costs.

Several studies have examined the association between hospital-SNF collaboration and patient outcomes and found a strong negative correlation between the proportion of discharges received by a SNF from a hospital and the rate of hospital readmission (14–16, 18), but such positive effect may not be present in all cases (17). Very few studies have investigated the effects of preferred SNF networks established by ACO hospitals. McHugh et al. found that hospitals developing SNF networks had reductions in readmission for patients discharged to SNFs vs. hospitals without networks, but failed to infer a causal relationship due to a lack of information on when the preferred networks were formed (7). Huckfeldt et al. found preferred SNFs exhibited better performance prior to being selected to participate in preferred SNF networks, indicating that health systems selected SNFs with lower resource use and better quality to participate in their networks (9). However, whether preferred network formation affects patient referral patterns and the effects of preferred network formation on patient outcomes and cost remains unknown.

In this study, we investigated whether preferred SNF network formation improved outcomes and lowered costs for patients discharged from hospitals to preferred SNFs, and whether it led to change in market share and patient composition of preferred SNFs. We investigated these effects in 10 health systems that participated in Medicare ACO programs and established preferred SNF networks between 2014 and 2015. We identified SNFs that were included in an ACO’s preferred network and focused on patients discharged to these preferred SNFs. Within each SNF, we compared outcomes of patients admitted to this SNF from hospitals that participated in the ACO vs. patients from other hospitals. We compared outcomes both before and after preferred network formation to estimate how network formation changes outcomes.

Methods

Study Population

We selected 10 ACOs that formed preferred SNF networks between 2014 and early 2015. We determined which SNFs were in an ACO’s preferred network and which hospitals participated in the ACOs based on public information on ACOs’ websites. We reported information about the 10 ACOs including the ACO types, the start dates of the ACOs and the preferred networks, Hospital Referral Regions (HRRs) in which they were located (19), whether they participated in the 3-day waiver program, whether they are physician-led ACOs, and the dates when the preferred SNF list that we used in the study was published (Table 1). We searched archives of these webpages to find the version for the list that was as close to the initial network start date as possible.

Table 1.

Background information for ACOs and their preferred SNF networks in the sample

	ACO name	ACO type	ACO start date	Preferred SNF network start date	HRR	3-day waiver	Date list published	Physician-led?
1	Allina	Pioneer & Next Generation	2012	2014	MN	yes	1/1/16	no
2	Monarch	Pioneer & Next Generation	2012	2014	CA	yes	8/1/15	yes
3	Michigan Pioneer	Pioneer & Next Generation	2012	2014	MI	yes	2015	no
4	Banner	Pioneer & MSSP	2012	2014	AZ	yes	08/2014	no
5	OSF Healthcare	Pioneer & MSSP	2012	2015	IL, MI, MO	yes	3/17/16	no
6	Partners	Pioneer & Next Generation	2012	2014	MA	yes	01/2017	no
7	Atrius	Pioneer & Next Generation	2012	2014	MA	yes	3/31/16	no
8	Cleveland Clinic	MSSP	2015	2015	OH	no	5/18/17	no
9	BJC HealthCare ACO	MSSP	2012	2015	MO	no	1/8/16	no
10	Torrance Memorial Integrated Physicians	Next Generation	2012	2015	CA	no	8/15/16	yes

The study population included Medicare fee-for-service beneficiaries discharged from hospitals to freestanding SNFs that are included in the preferred networks of the 10 ACOs between January 2012 and December 2013 and between January 2015 and September 2016. We were unable to identify the exact date of network formation for all ACOs. However, we were able to identify that ACOs in our sample formed their networks in 2014 or early 2015, thus we designed 2012–2013 as the pre period, 2015–2016 as the post period and allowed 2014 to be a buffer period. The unit of analysis was an episode of care that included an initial hospital stay, a subsequent SNF stay and the 90 days following SNF discharge.

We only included freestanding SNFs in our sample and excluded hospital-based SNFs for two reasons. First, hospital-based SNFs account for only about 5% of SNFs in the market and are structurally different from free-standing SNFs. Second, the sample size of hospital-based SNFs was not large enough to estimate a separate analysis of hospital-based SNFs. However, in a supplemental analysis, we further compared whether the effects of network formation were different for patients from freestanding SNFs vs. in-hospital SNFs (see text, Supplemental Digital Content 1, DDD Model).

Data Sources

We obtained information on baseline hospital characteristics from the 2012 American Hospital Association hospital survey (20). We identified SNF stays, preceding hospitalizations, hospital readmissions and inpatient rehabilitation facility (IRF) stays using Medicare Provider Analysis and Review (MEDPAR) files (21). We identified home health episodes in Home Health Standard Analytic Files (22). We identified custodial nursing home stays using the Minimum Data Set (MDS) (23). We obtained patient characteristics, Medicare enrollment information, mortality and patient comorbidities from the Master Beneficiary Summary files (24).

Study Measures

SNF Episode Outcomes

The first set of outcomes included healthcare use during the index SNF stay and preceding hospitalization, including the length of the preceding hospitalization, the initial SNF stay, and Medicare spending on the initial SNF stay. Next, we examined patient outcomes during the 90 days following the initial SNF discharge, including hospital readmissions, community residence on the last day of the episode, mortality, and total Medicare spending (initial SNF stay and all hospital and post-acute care spending in the subsequent 90 days). Finally, we examined subsequent post-acute care utilization in the 90-day episode, including whether a patient received any SNF, home health, IRF, long-term care or custodial nursing home care, and total post-acute spending. The final outcome was the change in functional status, measured by the difference between the last and the first non-missing Activity of Daily Living (ADL) score obtained from MDS files.

Primary Explanatory Variables

Each patient in our sample first stayed in a hospital then was discharged to a SNF that was part of an ACO’s preferred network. We refer to these SNFs as “preferred SNFs”. Patients were admitted to preferred SNFs from two types of hospitals. First, patients were admitted from hospitals that participated in the same ACO as a preferred SNF. We designated the hospital as the SNF’s “ACO hospital”. Second, patients were admitted to preferred SNFs from hospitals that did not participate in any ACOs. We designated these hospitals as non-ACO hospitals. We categorized patients into the treatment group if a patient was from a SNF’s ACO hospital, and patients into the control group if a patient was admitted from a non-ACO hospital. We only included patients of preferred SNFs in our sample, as only these SNFs had variation in treatment status across patients. We excluded patients from hospitals that belong to some other ACOs besides the 10 ACOs in our sample, as we do not have information on their preferred SNF lists thus cannot assign them to treatment or control group.

Patient and Hospital Characteristics

We controlled for patient characteristics including demographic measures (gender, age, and race and ethnicity), socio-economic status (Medicaid coverage and eligibility for the Part D low-income subsidy), Medicare Severity-Diagnosis Related Group (MS-DRG) for the preceding hospital stay, and the comorbidities listed on the preceding hospital claim (25). Coding of comorbid conditions accounted for the transition from ICD9 to ICD10 that occurred during the analysis period. We also controlled for hospital characteristics including ownership (non-profit, for-profit, or government), teaching status, hospital size indicated by the number of beds, and urban location.

Empirical Approach

Market Share

First, we evaluated the effects of network formation on market share. We hypothesized that hospitals would be more likely to discharge patients to their preferred SNFs after network formation. We used a sample of all patients from ACO hospitals, both sent to preferred and non-preferred SNFs, to calculate the percentage of patients sent to preferred SNFs in all patients from each hospital in pre and post period, and examined within-hospital changes in the probability of discharging patients to preferred SNFs after network formation (see text, Supplemental Digital Content 1, Market Share).

Patient Composition

Next, we evaluated the effects of network formation on the composition of patients. We hypothesized that ACO hospitals would be more likely to discharge more complex patients to preferred SNFs after network formation as more complex patients might benefit more from care coordination. We first evaluated the changes in characteristics of patients admitted to preferred SNFs from the SNF’s ACO hospital after network formation. Next, we compared the pre-post changes in characteristics of patients from ACO hospitals (treatment group) vs. patients from non-ACO hospitals (control group), by running a difference-in-difference (DID) regression model with SNF fixed effects (see text, Supplemental Digital Content 1, Patient Composition).

Patient Outcomes

Finally, we evaluated the effects of network formation on patient outcomes. We hypothesized that network formation would improve outcomes of preferred SNF patients from the SNF’s ACO hospital vs. patients from non-ACO hospitals, due to improved coordination between partnering hospitals and SNFs. We first evaluated the pre-post changes in outcomes of preferred SNF patients from the SNF’s ACO hospitals (treatment group), then compared the pre-post changes in outcomes for the treatment group relative to the control group (patients from non-ACO hospitals) by estimating DID models with SNF fixed effects (see text, Supplemental Digital Content 1, Patient Outcomes).

We also conducted an event study to test the identifying assumption that time trends in outcomes for the treatment and control groups did not differ prior to network formation (see text, Supplemental Digital Content 1, Event Study).

It is important to note that the changes in outcomes after network formation for patients from ACO hospitals could come from two sources. First, better care coordination after network formation could lead to improved outcomes for patients from ACO hospitals relative to patients from non-ACO hospitals. Second, the changes in patient outcomes could be driven by unobservable changes in patient characteristics for treatment group relative to control group. To address this issue, we adopted an approach developed by Oster to explore the extent to which our results may be sensitive to selection based on unobservables (see text, Supplemental Digital Content 1, Oster Bounds) (26).

Results

Table 1 summarized basic information of the 10 ACOs included in our sample. Most of the ACOs started in 2012 and they formed preferred SNF networks between 2014 and 2015. The 10 ACOs covered various parts of the US. The majority of these ACOs participated in the 3-day waiver program and only 2 ACOs are physician-led ACOs.

Market Share

Table 2 displays the average percentage of patients who went to preferred SNFs in all SNF patients from each hospital before and after network formation and whether the within-hospital pre-post differences are statistically significant. Prior to network formation, hospitals affiliated with the 10 ACOs sent 30.9% patients to their future preferred SNFs; after network formation, they sent marginally higher 31.4% patients to preferred SNFs. The slight increase was not statistically significant. Hospitals that did not participate in the 10 ACOs sent 12.4% patients to preferred SNFs in the pre-period and the number remained the same after network formation. We also reported the fraction of all SNF patients within an HRR that went to preferred SNFs before and after network formation (see Appendix Table 1, Supplemental Digital Content 2). Consistent with Table 2 we did not find significant pre-post changes in the market shares of these preferred SNFs.

Table 2.

Percent of patients sent from hospitals to preferred SNFs in pre and post period (%)

	Pre	Post	Pre-post difference	Difference with SNF FE
ACO-hospitals
Overall (%)	30.9	31.4	0.5	0.4
Partners (%)	43.1	45.4	2.3	1.1
Cleveland (%)	27.1	25.8	−1.3	−0.9
Atrius (%)	46.1	45.2	−0.9	−0.4
Banner (%)	41.8	44.4	2.6	2.7*
Monarch (%)	21.9	23.3	1.4	1.1
Allina (%)	41.2	37.6	−3.6	−0.4
Michigan (%)	30.0	29.3	−0.7	−0.8
OSF (%)	20.7	28.0	7.3	5.5**
BJC (%)	5.7	5.1	−0.6	−1.5
Torrance (%)	83.8	79.1	−4.7	−4.6**
Non-ACO hospitals (%)	12.4	12.4	0.0	0.3

Patient Characteristics

Table 3 displays demographic, socio-economic, clinical and functional characteristics of preferred SNF patients from ACO hospitals before and after network formation, changes in characteristics after ACO formation, and the differential change relative to patients from non-ACO hospitals (DID estimates). We found that patients in the treatment group were less likely to be enrolled in Medicaid or the Part D low income subsidy and had more comorbidities in the post-period compared to the pre-period. The DID regressions showed that preferred network formation was generally not associated with significant changes in observed characteristics for patients from ACO hospitals relative to non-ACO hospitals. The exceptions were a small decrease in the fraction of Hispanic patients and a small increase in the average number of Elixhauser comorbidities for patients from ACO hospitals after network formation.

Table 3.

Change in patient composition after preferred SNF network was formed

	Patients from ACO hospitals		Pre-post diff for patients from ACO hospitals with SNF FE	Pre-post diff for patients from ACO hospitals relative to patients from non-ACO hospitals (DID) with SNF FE
	Pre	Post
Number of episodes	22267	17378
Demographic characteristics
Age	82.2	81.8	−0.3**	−0.4
Male (%)	35.0	37.2	2.0**	0.2
White (%)	89.8	89.6	−0.6	0.1
Black (%)	6.4	6.5	0.6	0.1
Asian (%)	1.6	1.6	0.0	0.2
Hispanic (%)	1.0	0.7	−0.4**	−0.3*
Other (%)	1.3	1.7	0.3**	−0.1
Socioeconomic characteristics
Dual eligible (%)	16.5	14.6	−1.9***	−0.7
Low income subsidy (%)	18.0	15.9	−2.0***	−1.0
Clinical status
Elixhauser comorbidities counts	2.5	3.1	0.6***	0.1*
Selected comorbidities (%)
Hypertension	56.7	64.6	7.8***	0.9
Diabetes	22.8	27.6	4.8***	1.1
Congestive heart failure	16.8	20.3	3.6***	0.5
Renal failure	15.6	22.2	6.7***	1.7
Metastatic cancer	2.5	2.6	0.0	−0.2
Top 5 MS-DRGs (%)
470: Major Joint Replacement or Reattachment of Lower Extremity w/o MCC	11.9	9.9	−2.2***	1.2
871: Septicemia or severe sepsis w/o MV 96+ hours w MCC	3.7	6.0	2.3***	0.5
481: Hip & femur procedures except major joint w CC	3.6	3.7	0.1	−0.2
690: Kidney & urinary tract infections w/o MCC	2.6	1.7	−0.8***	−0.3
291: Heart failure & shock w MCC	1.8	2.4	0.6***	0.1
Functional status	14.5	14.4	−0.1	−0.1

Patient Outcomes

Table 4 first shows regression results that evaluated changes in outcomes for patients from ACO hospitals after network formation. We found that within each SNF, patients from ACO hospitals had shorter initial SNF stay, lower initial SNF payment, higher return to community rate after network formation, compared to the pre-period. They also had higher mortality rate and higher use of post-acute care. However, when we compared the pre-post changes in outcomes for patients from ACO hospitals relative to the control group (patients from non-ACO hospitals), we did not observe significant relative differences in any the outcomes. In a supplemental analysis, we used a DDD model to compare the DID estimates for patients admitted to freestanding preferred SNFs to patients from in-hospital preferred SNFs (Appendix Table 5, Supplemental Digital Content 2). We found evidence that patients from in-hospital SNFs had larger reductions in PAC payments compared to patients from freestanding SNFs, which resulted from lower use of SNFs and higher use of HHA, suggesting that network formation might have bigger effect on in-hospital SNFs vs. freestanding SNFs.

Table 4.

Change in patient outcomes after preferred SNF network was formed

	Patients from ACO hospitals		Patients from non-ACO hospitals		Unadjusted models with SNF FE			Adjusted models with SNF FE
	Pre	Post	Pre	Post	Pre-post diff for patients from ACO hospitals	Pre-post diff for patients from non-ACO hospitals	DID	Pre-post diff for patients from ACO hospitals	Pre-post diff for patients from non-ACO hospitals	DID
Number of episodes	22267	17378	15958	12172
Initial SNF use and hospitalization
Hospital length of stay	6.1	6.2	5.8	6.0	0.0	0.2*	−0.1	−0.2***	0.1*	−0.2
Initial SNF length of stay	25.0	22.6	25.8	22.5	−2.6***	−2.7***	0.1	−2.5***	−2.8***	0.5
Initial SNF payment ($)	11096.3	10704.0	11271.1	10758.9	−518.5*	−439.5**	−82.8	−518.2**	−532.8**	75.2
Episode outcomes occurring within 90 days of initial SNF discharge
Readmission rate (%)	35.8	35.7	32.7	33.2	0.1	1.0	−0.7	−1.9**	−1.2	−0.6
Return to community rate (%)	68.0	71.4	70.4	73.6	3.5***	1.9	1.5	4.3***	3.4***	0.8
Mortality rate (%)	16.1	18.5	14.9	16.8	2.3***	2.3***	0.1	1.3**	0.9	0.5
Total Medicare payment ($)	21142.2	21516.5	20621.7	20510.4	224.5	24.2	226.6	−397.3	−581.1*	307.1
Post-acute care use within 90 days of initial SNF discharge
PAC payment ($)	5145.5	5054.2	4914.2	4673.5	−126.4	−255.5*	146.7	−253.1**	−409.4***	148.3
Any use of PAC (%)	71.4	74.1	68.1	71.1	2.4***	2.2**	0.3	1.9**	2.0*	0.0
Any use of SNF (%)	23.7	22.6	22.2	21.4	−1.0*	−0.2	−0.8	−2.0***	−1.3*	−0.6
Any use of HHA (%)	61.4	65.6	58.0	62.7	4.0***	3.2***	0.9	3.8***	3.3***	0.5
Any use of IRF (%)	0.9	1.1	1.3	1.6	0.2	0.2	−0.1	0.2	0.1	−0.1
Any use of custodial nursing home (%)	16.0	10.9	15.1	10.6	−5.0***	−3.4***	−1.7	−4.6***	−3.5***	−1.0
Any use of long-term care (%)	1.5	1.2	1.1	0.9	−0.3*	−0.2	−0.1	−0.5**	−0.4*	−0.1
Change in functional status
Difference in ADL score	−2.2	−1.9	−2.5	−2.2	0.3***	0.3**	0.0	0.2***	0.2*	0.0

We conducted an event study which tested the identifying assumption that time trends in outcomes for the treatment and control groups did not differ prior to network formation (see Appendix Table 2, Supplemental Digital Content 2). In general, the results supported the parallel time trends assumption as for most time points in the pre-period the treatment-control difference was not significantly different and there was no systematic trend.

We also constructed bounds for the estimated DID effects that considered bias due to selection on unobservables and found that estimates accounting for selection on unobservables were similar to estimates accounting for selection on observables only. In addition, both sets of estimates are consistent with the finding that network formation was not associated with outcomes among patients admitted from a SNF’s ACO hospital (see Appendix Table 3, Supplemental Digital Content 2).

One potential explanation for our finding is that preferred network formation did change outcomes for preferred SNFs’ patients but the effects were not isolated to patients from ACO hospitals. That is, practice pattern spillover effects might have led to similar changes in outcomes for both patients from ACO hospitals and non-ACO hospitals. To investigate this issue, we also compared the performance between preferred SNFs and non-preferred SNFs pre and post network formation (see Appendix Table 4, Supplemental Digital Content 2). In both pre and post period, patients of preferred SNFs had better outcomes and lower costs compared to patients sent to out-of-network SNFs, and we find no evidence that performance of preferred SNFs relative to out-of-network SNFs improved after network formation. Thus, it is unlikely that spillover effects within a SNF from ACO to non-ACO patients are a potential explanation for our main finding that network formation had little impact on patient outcomes.

Discussion

ACOs have an incentive to improve post-acute care outcomes and lower costs. Establishing preferred SNF networks is one approach ACOs and their affiliated hospitals are using to improve care coordination with SNFs. In our study, we hypothesized that after the formation of these preferred networks, hospitals would send more of their patients or more complex patients to their preferred SNFs and have improved patient outcomes. We used a DID analysis with SNF fixed effects to test the hypothesis. In particular, we compared within-SNF changes in outcomes after network formation for preferred SNF patients from ACO hospital relative to patients from non-ACO hospitals. We found little change in market share and outcomes and a modest increase in the number of comorbidities of patients from ACO hospitals after network formation.

Similar to previous studies, we found that preferred SNFs themselves were better than non-preferred SNFs, even before preferred networks were actually established (9, 27). This result suggests that hospitals were able to accurately identify higher quality and lower cost SNFs to be included in their networks. However, when we investigated the effects of network establishment, we did not find hospitals sent more patients to preferred SNFs or that preferred SNF patients from ACO hospitals had improved outcomes.

There are several possible explanations for our null results. First, even though hospitals may have accurate information on the quality of SNFs – and despite the fact that hospitals are financially responsible for post-discharge care and outcomes – they are not allowed to recommend specific SNFs under Medicare regulations guaranteeing freedom of choice of provider (28). Hospital discharge planners may not provide patients with quality information on potential SNFs for fear of violating these regulations (29). For example, according to a qualitative study, patients were not given quality-of-care information about SNFs when discharged from hospitals (29). Most patients choose SNFs based on location, which would not be altered by preferred network formation. Therefore, we believe that allowing a “harder steering” and giving ACOs greater controls over patient choice on SNFs could help preferred SNF networks to meet the expectations of lowering cost and improving quality. Second, prior research on the performance of ACO models shows mixed results. Some studies document that ACOs are associated with modest reduction or slower increase in spending and improved performance on certain quality measures, while other studies find no evidence of savings or improved performance (30–33). It remains an open question whether ACOs can achieve significant cost savings and improvements in quality of care.

Our results also provided some evidence that ACO hospitals send more complex patients to preferred SNFs after network formation. Patients with multiple comorbid conditions usually have complex medical needs and are responsible for the greatest proportion of spending. Sending those patients to preferred SNFs may be a way for ACO hospitals to better manage them, in order to achieve the largest effects on patient outcomes and financial rewards (31). Our finding suggests that ACOs developed preferred SNF networks and targeted its care coordination and management efforts toward patients with more complex needs.

Our study has several limitations. First, we examined 10 ACOs that publicly posted the list of their preferred SNFs, but we do not have information about other ACOs. Therefore, our results might not be representative of other preferred networks. Second, although we carefully controlled for hospital and patient characteristics, we acknowledge that there may be unobserved differences between patients from ACO hospitals vs. non-ACO hospitals, which may confound the observed results. Although we calculated bounds for our DID estimation that accounted for potential omitted variables bias and showed robust results, these bounds are subject to assumptions and thus not conclusive. Finally, we identified patients discharged from hospitals that were part of an ACO. However, it is possible that not all patients from these hospitals were attributed to the ACO because the attribution is through patients’ primary care physicians. Unfortunately, we do not have data on outpatient care and office visits to determine ACO attribution.

In conclusion, after the formation of preferred SNF networks, there is some evidence that ACO hospitals sent more complex patients to their preferred SNFs, but there was no change in the volume of patients received by these SNFs. Furthermore, preferred SNF network formation was not associated with improvement in patient outcomes. This situation might be ameliorated by giving ACOs greater control over patient choice of SNFs.