Skip to main content
HHS Author Manuscripts logoLink to HHS Author Manuscripts
. Author manuscript; available in PMC: 2020 Sep 1.
Published in final edited form as: Med Care. 2019 Sep;57(9):695–701. doi: 10.1097/MLR.0000000000001169

Changes in hospital referral patterns to skilled nursing facilities under the Hospital Readmissions Reduction Program

Kunhee Lucy Kim 1, Li Li 2, Meng Kuang 2, Leora I Horwitz 1,3, Sunita M Desai 1
PMCID: PMC6944285  NIHMSID: NIHMS1065528  PMID: 31335756

Abstract

Background:

The Hospital Readmissions Reduction Program (HRRP) penalizes hospitals for higher-than-expected readmission rates. Almost 20% of Medicare fee-for-service (FFS) patients receive post-acute care in skilled nursing facilities (SNFs) after hospitalization. SNF patients have high readmission rates.

Objective:

To investigate the association between changes in hospital referral patterns to SNFs and HRRP penalty pressure.

Design:

We examined changes in the relationship between penalty pressure and outcomes before versus after HRRP Program announcement among 2,698 hospitals serving 6,936,393 Medicare FFS patients admitted for target conditions: acute myocardial infarction, heart failure, or pneumonia. Hospital-level penalty pressure was the expected penalty rate in the first year of the HRRP multiplied by Medicare discharge share.

Outcomes:

Informal integration measured by the percentage of referrals to hospitals’ most referred SNF; formal integration measured by SNF acquisition; readmission-based quality index of the SNFs to which a hospital referred discharged patients; referral rate to any SNF.

Results:

Hospitals facing the median level of penalty pressure had modest differential increases of 0.3 percentage points (pp) in the proportion of referrals to the most referred SNF and a 0.006 standard deviation increase in the average quality index of SNFs referred to. There was no statistically significant differential increases in formal acquisition of SNFs or referral rate to SNF.

Conclusion:

HRRP did not prompt substantial changes in hospital referral patterns to SNFs, although readmissions for patients referred to SNF differentially decreased more than for other patients, warranting investigation of other mechanisms underlying readmissions reduction.

Keywords: health policy, economics, quality, readmissions

Introduction

Pay-for-performance programs implemented under the 2010 Affordable Care Act (ACA) link provider payments and outcomes to incentivize health care quality improvement. The Hospital Readmissions Reduction Program (HRRP) is one such program which, in its first year, penalized hospitals up to 1% of their Medicare fee-for-service (FFS) revenue if risk-adjusted 30-day readmission rates for acute myocardial infarction (AMI), heart failure (HF), and pneumonia were higher than expected. The HRRP reduced readmission rates, even after accounting for changes in patient case mix or coding.16 However, little is known about how hospitals may have achieved readmission reductions.

About 40% of Medicare FFS patients are discharged to post-acute care (PAC) providers, half of which are to skilled nursing facilities (SNFs).7 Patients discharged to SNF, who are typically sicker and have higher readmission rates than those discharged home, are likely to be a high-priority population for hospitals aiming to reduce readmission rates. Changing if and where patients receive post-discharge skilled nursing care could be one strategy by which hospitals improved readmission rates in response to the HRRP. In particular, hospitals may have integrated with SNFs by concentrating referrals to fewer SNFs or acquiring SNFs, since hospital-SNF integration is associated with reduced readmission rates.811 Hospitals may also have referred patients to higher quality SNFs.

We investigate whether hospitals facing higher penalty pressure under HRRP integrated with SNFs—either formally through acquisition or informally through concentrating referrals to preferred providers—or increased referrals to higher quality SNFs. In secondary analyses, we also examined the association between HRRP penalty pressure and changes in the proportion of discharges referred to SNFs, severity of patients referred to SNFs, and 30-day readmission rates among patients referred to SNFs.

Methods

To study whether the HRRP Program was associated with changes in hospital referral patterns to SNFs, we empirically tested for evidence that hospitals expecting higher penalties at the start of the HRRP Program differentially changed SNF referral patterns compared to hospitals expecting lower penalties following announcement of the HRRP Program. Changes in the relationship between penalty pressure and the outcomes following HRRP Program announcement would serve as evidence that hospitals changed their referral behavior in response to the HRRP. Our empirical strategy constitutes a differences-in-differences design with a continuous treatment variable, where the treatment is hospitals’ expected penalty pressure at the start of the HRRP Program.

Study cohort and data sources

We examined AMI, HF, and pneumonia Medicare FFS admissions during July 1, 2008-June 30, 2016 in short-term acute care hospitals subject to the HRRP, using the Centers for Medicare and Medicaid Services’ (CMS) administrative claims database (Inpatient Standard Analytic File and Medicare Enrollment Database). We applied all of the HRRP’s exclusion criteria, except qualifying readmissions which were included as index admissions.12 A patient with a nonswing-bed claim in a SNF within 2 days of hospital discharge was defined as being referred to that SNF. Hospital characteristics and participation in other CMS value-based programs were drawn from the CMS Cost Reports, Hospital Compare, and relevant program websites. See Appendix.

Treatment variable: Hospitals’ penalty pressure in the first year of HRRP

The share of a hospital’s overall revenue at risk under the HRRP, referred to as penalty pressure, was driven by two factors: the penalty rate and the share of inpatient revenue from Medicare. We consider both to estimate each hospital’s penalty pressure.

The HRRP penalty rate, which was based on a hospital’s 30-day readmission rates for target conditions during the prior 3 years, is the percentage reduction in the hospital’s Medicare FFS base diagnosis-related group (DRG) payments. Readmission rates fell beginning in July 2010 after the HRRP was announced, suggesting hospitals began reducing readmissions even before details of the penalty were announced.13 Therefore, we constructed a hospital’s expected penalty rate as of July 2010 assuming that hospitals formed expectations of their penalty rates using information CMS had provided on the hospital’s risk-adjusted 30-day readmission rate relative to the national hospitals for the three target conditions between July 2007-June 2010.2 The expected penalty rate was estimated using a two-part model that predicted for each hospital in the first year of the HRRP the probability of being penalized (logit model) and the penalty rate conditional on being penalized (generalized linear model with gamma family and log link). Consistent with HRRP penalty rate calculation, the excess readmission ratios (ERRs) for the 3 target conditions in fiscal years 2008–2010 were used to predict penalty rate. No other covariates were used for prediction. To increase power, the prediction used all hospitals that could be penalized under the HRRP Program (i.e. we did not impose further sample restrictions at this stage).

This expected penalty rate was multiplied by the share of each hospital’s discharges attributable to Medicare in 2010 to construct the HRRP penalty pressure for each hospital.

Outcome measures

Our outcomes were constructed at the hospital-year level. Informal integration was measured as the share of a hospital’s referrals sent to its most-referred SNF in that year. As alternate measures of informal integration, in sensitivity analyses, we examined the Herfindahl-Hirschman index (HHI) of referrals (the sum of squared shares of a hospital’s referrals to each SNF) and the log count of SNFs accounting for 80% of referrals in descending order starting with the hospital’s most-referred SNF.11,13 For all informal integration outcomes except the last, higher levels indicate more concentration of referrals. On the last outcome, a decrease in the log count of SNFs accounting for 80% of hospital referrals indicates more concentration. Formal integration was given by whether a hospital reported owning a SNF, conditional on not owning one in 2008, the baseline year. To examine the quality of SNFs to which hospitals referred patients, we constructed a composite measure reflecting average quality for the set of SNFs to which each hospital referred patients.14 We calculated a quality index for each SNF, measured as its 30-day readmission rate in 2008, standardized across SNFs in its HRR. For each hospital, we computed the weighted average of this quality index across all SNFs to which a hospital referred patients in a given year, where weights reflect the proportion of the hospital’s referrals to the SNF. We multiplied this index by −1 so that a positive coefficient estimate can be interpreted as an increase in referrals to higher-quality SNFs. Since the index was calculated using SNF’s baseline readmission-based quality, any improvement in aggregate performance over time would reflect an increase in hospital referrals to higher quality SNFs rather than subsequent changes in SNF quality post-HRRP.

To test whether any estimated changes in referral patterns could be explained by changes in underlying characteristics of referred patients, as secondary outcomes, we examined the proportion of discharged patients referred to any SNF and the average comorbidity count per SNF-referred patient.12 Finally, we examined 30-day all-cause unplanned readmission rates among patients referred to a SNF and patients not referred to a SNF, applying the CMS Planned Readmission Algorithm version 4.0.15,16

Statistical Analysis

To investigate how the association between hospitals’ penalty pressure and each outcome changed following announcement of the HRRP, we estimated the following hospital-year level linear regression model:

Yht=β0+β1×PenaltyPressureh×Postt+αh+δt+γXht+ϵht

where Yℎt denotes an outcome for hospital h in year t; PenaltyPressure denotes hospital h’s penalty pressure in the first year of HRRP and therefore, does not vary with time; Postt is an indicator for years beginning 2011; α denotes hospital fixed effects to control for time-invariant hospital-level factors; δt denotes year fixed effects accounting for outcome trends during the study period; and Xℎt is a vector of time-varying hospital characteristics including demographic and clinical characteristics of a hospital’s patient populations served. We controlled for hospital characteristics including number of beds, ownership status, urban/rural classification, teaching status, and whether a hospital owned a SNF in the previous year. Because other value-based reforms may be associated with our outcomes,17 we controlled for participation in Stages 1 and 2 of meaningful use of the electronic health records (EHR) Incentive Program, participation in the Bundled Payments for Care Improvement Initiative (BPCI), penalization under the Hospital Acquired Condition (HAC) Reduction Program, and hospital payment adjustment factor under the Hospital Value Based Purchasing Program. We also controlled for the log-transformed number of SNFs operating in the hospital’s hospital referral region (HRR) in each year. Finally, we controlled for hospital-year level measures of patient characteristics including age (in 10-year bins), race/ethnicity, comorbidities used in the CMS measures, and an indicator for whether the beneficiary was dually eligible in Medicare and Medicaid.12 We estimated heteroscedasticity-robust standard errors clustered at the hospital level.18 We weighted each hospital-year observation by the number of discharges or referrals in the first year of the sample period.19

The coefficient β1 estimates the change in the outcome in the post-HRRP period relative to pre-HRRP period in response to a 1-percentage-point (pp) increase in the penalty pressure. Since the first-year’s HRRP penalty rate was capped at 1% and Medicare enrollees account for 40% of discharges in the average hospital, the coefficients estimate the response for hospitals facing a penalty that was well above the actual penalty faced by most hospitals under the Program. Thus, we also reported scaled estimates that can be interpreted as the absolute estimated differential change associated with an increase of the penalty pressure to the sample median (0.07 pp).

In estimating our models, we restricted the analysis to hospitals that had data in every year and that had at least 30 beds in every year and 50 discharges in AMI, HF, or PN per pre-HRRP year on average. Furthermore, for hospitals with few SNF referrals, changes in where a small number of patients are referred could result in large differences in our SNF referral pattern outcomes, leading to highly sensitive and more noisy estimates. As such, for all SNF referral pattern outcomes except the share of discharges referred to a SNF and readmission rates among patients not referred to a SNF, we excluded hospitals with fewer than 15 SNF referrals on average during the pre-HRRP period or with fewer than 2 referrals in any year during the study period. Analyses on SNF acquisition were limited to hospitals that did not own a SNF at baseline. See Appendix for further details on exclusion criteria.

Our empirical strategy assumes that hospitals with higher penalty pressure would have had similar changes in outcomes as those with lower penalty pressure in the absence of the HRRP. To examine the validity of this assumption, we compared pre-HRRP announcement trends in outcomes between hospitals with high versus low penalty pressure, by comparing mean outcomes in pre-HRRP announcement years among hospitals in the top versus bottom quartile of penalty pressure. We also tested for statistically significant differences in the association between penalty pressure and our outcomes in the years prior to HRRP announcement.

We conducted several robustness checks including unweighted estimation and restricting the post-HRRP period to 2013 to examine short-term responses. We also re-estimated analyses after replacing the expected penalty rates with the actual rates as well as using actual or expected penalty rates alone without multiplying them by Medicare share. We examined the association between penalty pressure and SNF referral pattern outcomes for each condition separately. We also tested for heterogeneity in effects by conducting analyses after stratifying for hospital size (as measured as beds), urban/rural classification, teaching status, whether a hospital participated in BPCI during the study period, and whether a hospital was located in a market with higher or lower than median number of SNFs in its market.

Results

Baseline hospital characteristics

Our sample included 2,698 hospitals with 6,936,393 hospital discharges for 3 target conditions which comprised 15% of total hospital discharges. Hospitals on average referred 22% of discharged patients to a SNF (AMI: 22%, HF: 21%, PN: 24%) (Table 1). Participation rates in CMS value-based programs were 75% for Meaningful Use Stage 2 of the EHR Incentive program and 7% for BPCI. Twenty-three percent of hospitals were penalized under the HAC reduction program.

Table 1.

Descriptive statistics of key variables

No. hospital-year observations 21,584 Mean SD
Basic hospital characteristics
Number of discharges, all conditions (share referred to SNF, %) 321 (21.7) 270 (8.3)
 AMI 63 (21.7) 76 (17.3)
 Heart Failure 149 (21.1) 132 (8.9)
 Pneumonia 110 (23.7) 82 (9.6)
Number of beds
 <= 100 29.7 45.7
 101–300 47 49.9
 301+ 23.3 42.3
Ownership
 Not-for-profit 63.9 48
 For-profit 21 40.7
 Government owned 15.1 35.8
Teaching 36.5 48.2
Urban 72.8 44.5
Participation in other value-based reforms since the start of each reform
Meaningful use of EHR stage 1 since 2011 (stage 2 since 2014) 38.4 (74.5) 48.6 (43.6)
Bundled Payments for Care Improvement Initiative 7.1 25.6
Penalized by hospital-acquired condition reduction program since 2015 23.3 42.3
Payment adjustment factor under the hospital value-based purchasing program since 2013 1.0 0.004
Demographics: Share of patients referred to SNF / not referred to SNF, % Black 9.2 / 9.8 14.9 / 14.9
White 86.2 / 84.3 18.2 / 18.4
Dual-eligible 28.5 / 21.3 18.9 / 15.1
Age 65–74a 17.1 / 32.5 9.7 / 8.5
Age 75–84 35.7 / 37.9 10.4 / 5.5
Age 85–94 41.1 / 26.4 12.0 / 7.5
Age 95+ 6.1 / 3.1 5.3 / 2.3
SNF market characteristics
Number of SNFs in the same HRR 84 82
HRRP penalty in the first year of HRRP, % (Median)
Penalty pressure based on expected penalty rate 0.11 (0.07) 0.11
 Expected penalty rate 0.29 (0.21) 0.26
 Share of Medicare discharges in 2010, % 36.8 (37.2) 11.2
Penalty pressure based on actual penalty rate 0.12 (0.06) 0.15
Actual penalty rate 0.31 (0.18) 0.34

Abbreviations: EHR, Electronic Health Records; HRR, Hospital Referral Region; HHI, Herfindahl-Hirschman index.

a

Reference level for categorical variables in the regression models.

Hospitals’ penalty pressure under the HRRP

The mean hospital penalty pressure was 0.11% (median: 0.07%). The mean expected penalty rate was 0.29%, and the mean percentage of discharges for Medicare beneficiaries was 37%. The expected and actual penalty rates had a correlation coefficient of 0.78, and their average difference was 0.02pp (eFigure 1). Hospitals in higher quartiles of expected penalty rate in the first year of the HRRP had higher actual penalty rates on average in subsequent years, suggesting a stable composition of hospitals in the high-penalty group over time (eFigure 2).

Baseline SNF referral patterns

Before the announcement of the HRRP, hospitals in the top and bottom quartiles by penalty pressure (subsequently labeled hospitals with “high” versus “low” penalty pressure, respectively) had similar trends across referral outcomes, supporting the parallel trends assumption. Coefficient estimates reflecting pre-intervention trend differences were small and not statistically significant for all outcomes except SNF quality (eFigure 3). Since SNF quality was the only outcome with a trend difference and the difference diminished in the following year, we interpreted this result as idiosyncratic.

High penalty pressure hospitals consistently concentrated more referrals to fewer SNFs, referred fewer patients to high-quality SNFs compared to low penalty pressure hospitals, and had readmission rates among their SNF-referred patients (Figure). Furthermore, high penalty pressure hospitals acquired SNFs at a higher rate than low penalty pressure hospitals immediately after the announcement of HRRP, though differences were smaller in later years.

Figure. Trends in outcomes among hospitals with penalty pressure in top versus bottom quartiles.

Figure.

Sample mean outcomes at the hospital-year level among hospitals in top versus bottom quartiles of penalty pressure are plotted over the study period. Penalty pressure is the expected penalty rate in the first year of HRRP multiplied by the share of Medicare discharges in 2010. The vertical lines dividing each panel distinguish between the pre-and post-HRRP periods.

Impact on SNF referral patterns and readmission rates

Penalty pressure was associated with modest but statistically significant differential changes in the concentration of SNF referrals after the introduction of the HRRP, and this result was largely robust to alternate claims-based integration measures (Table 2). In the post-versus pre-HRRP period, a 0.07% (median) increase in the penalty pressure was associated with a 0.3pp (unscaled coefficient estimate: 4.04%, P-value: 0.002) differential increase in the share of referrals to the most referred SNF. We saw an early increase in SNF acquisition among high penalty pressure hospitals (Figure, eTable 3), but on average over the full study period, the difference was not statistically significant (coefficient: 5pp, P-value: 0.14). Penalty pressure was associated with small but statistically significant differential increases in referrals to high-quality SNFs following introduction of HRRP. An increase in penalty pressure to the median was associated with a differential increase of 0.006 (coefficient: 0.08, P-value<0.001) in the SNF quality index, relative to a mean of quality index of −0.04 (standard deviation: 0.38).

Table 2.

The impact of HRRP penalty pressure on SNF referral patterns and 30-day readmission rate among SNF referred patients

Sample mean (SD) Estimated effect of HRRPa
Effect 95% CI P-value
Primary outcomesb
 Share of referrals to hospital’s most referred SNF, % 35 (20) 4.04 [1.53, 6.56] 0.002
 Probability of acquiring SNF given not having one 0.01 (0.4) 0.05 [‒0.02, 0.12] 0.142
 Average SNF performance at baseline, weighted by referral share ‒0.04 (0.38) 0.08 [0.04, 0.12] <0.001
Alternate claims-based measures of integration
 Referral HHI 0.23 (0.19) 0.02 [0.01, 0.04] 0.01
 Log Number of SNFs accounting for 80% of hospital’s referrals 1.7 (0.7) ‒0.06 [‒0.13, 0.01] 0.09
Secondary outcomes
 Discharges referred to SNF, % 21.7 (8.3) 0.05 [‒1.06, 1.16] 0.93
 Average comorbidity count per SNF referred patientc 0.13 (0.41) 0.01 [‒0.01, 0.02] 0.39
 30-day hospital readmission rate, %
  Patients referred to SNFs 24.7 (8.5) ‒12.68 [‒14.29, ‒11.07] <0.001
  Patients not referred to SNFs 18.9 (5) ‒8.31 [‒9.21, ‒7.41] <0.001
a

Coefficient estimate β1 on PenaltyPressure X Post from estimating a hospital-year level linear regression model adjusted for hospital fixed effects, year fixed effects, hospital characteristics, patient case-mix, and weighted using the number of discharges or referrals in the first year of the sample period to correct for heteroskedasticity in group-average dependent variables. Confidence intervals were constructed using standard errors clustered at the hospital level. Absolute differential changes associated with an increase in penalty pressure from zero to median (Δ=0.07pp) are reported in the text.

b

For all outcomes measured among SNF referred patients (readmission rate, comorbidity level, and referral HHI among SNF referred patients, and baseline readmission performance of SNFs referred to), we used the sample of hospitals with on average at least 15 referrals in each year during the pre-HRRP period (N=19,608); for outcomes measured among all discharged patients (readmission rate among all other patients and share of discharges referred to SNF), we used the sample of all hospitals (N=21,584); for the outcome of probability of acquiring SNF, we used the sample of hospitals without SNFs in the baseline period, 2008 (N = 14,360).

c

For average comorbidity count, we used 31 comorbidity categories used for risk adjustment in developing the CMS’s readmission publicly reported readmission measures.12

In secondary outcomes analyses, after the introduction of the HRRP, penalty pressure was not associated with differential changes in either the proportion of discharges referred to SNF (coefficient: 0.05pp, P-value: 0.93) or severity of patients (as measured by the number of comorbidities) referred to SNFs (coefficient: 0.01, P-value: 0.39).

Penalty pressure was associated with differential decreases in readmission rates for patients referred to a SNF, and the effect was greater than that among patients not referred to a SNF. Hospitals at the median penalty pressure had differential decreases of 0.89pp (coefficient: −12.7, P-value: <0.001) and 0.58pp (coefficient: −8.3, P-value: <0.001) in readmission rates for patients referred to a SNF and for patients not referred to a SNF, respectively (eFigure 3). In aggregate, the difference in mean unadjusted 30-day readmission rate among patients referred to SNFs between top and bottom quartiles of penalty pressure hospitals narrowed from 10pp in 2009 to 5pp in 2016.

Our robustness analyses without weighting, with a shorter post-HRRP period, or using alternate measures of penalty largely yielded similar conclusions though the estimates differed in magnitude and were less precisely estimated (eTables 34). Analyses stratified by condition suggest that changes in referral patterns that were observed were mostly driven by changes in patterns for pneumonia patients (eTable 5). Moreover, in stratified analyses, we found evidence suggestive of heterogeneous responses by hospital type (eTables 610).

Discussion

Our model estimates are largely statistically significant but small, leading us to conclude that overall, penalized hospitals did not substantially alter SNF referral patterns for target condition discharges in response to the HRRP. We found evidence of modest informal integration via concentration of referrals and of increased referrals to higher quality SNFs among hospitals facing high penalty pressure. High penalty pressure hospitals acquired SNFs at a higher rate immediately after introduction of the HRRP, but not differentially so over the full post HRRP period studied. We did not find that penalty pressure was associated with changes in the likelihood of a discharged patient being referred to SNF or in the severity of patients referred to SNF, suggesting the modest changes we did find were not driven by changes in patient mix.

By linking Medicare payments to an outcome without mandating specific processes, hospitals had the flexibility to adopt efficient practices that suited their systems. Contrary to our hypothesis, we did not find that altering SNF referral patterns was a major strategy undertaken by hospitals facing high penalty pressure.

Several factors may explain our findings. First, the revenue at stake in the first year may have been too small to motivate large-scale change. Second, despite existing evidence that patients referred to integrated SNFs are less likely to be re-hospitalized,8,10,11 integrating with PAC providers and investing in post-discharge care management may require substantial hospital investments, and patient preferences for specific SNFs may be hard to influence. High costs of managing referrals, SNF capacity constraints, and other barriers may limit hospitals’ ability to steer patients to specific SNFs. Differential increases in referral concentration observed several years after announcement of the HRRP suggest that it may take more time for hospitals to develop SNF preferred networks, although we found that hospitals accelerated formal SNF acquisition shortly after Program implementation. Third, reduced hospital DRG payments for patients discharged to PAC after short stays under Medicare’s PAC transfer policy may disincentivize increase of referrals to PAC, as most DRGs in HRRP target conditions were subject to this policy.20 However, we would expect the transfer payment policy to affect all hospitals.

Despite the lack of evidence supporting changes in hospitals’ SNF referral patterns, we found that patients referred to SNFs experienced differentially greater readmissions reduction when discharged from hospitals facing high penalty pressure. This differential readmission reduction exceeded that among other patients and cannot be explained by changes in the underlying characteristics of patients being discharged to SNFs that we observed. Though HRRP penalty pressure contributed to modest changes in referral patterns including referral concentration, the relatively small effects are unlikely to fully explain the readmissions reductions, highlighting the need for future research on other potential mechanisms.

Our study had several limitations. First, we examined hospital responses to the first year of the HRRP but hospitals were subject to different penalties in each subsequent year. Estimating the effects of penalty pressure in each year would be subject to bias, and hospitals’ relative revenue at risk was largely stable over the course of the study period suggesting effects accounting for yearly penalty rates would be similar. Second, formal hospital-SNF integration was defined by legal ownership and may have excluded other forms of integration such as common ownership or joint ventures which have become prevalent.21 However, we constructed claims-based measures to indirectly capture these and other forms of integration. Third, we could not control for hospitals’ participation in Accountable Care Organizations. However, previous research found no effect on PAC use outside the population of ACO-attributed Medicare beneficiaries which was limited during our study period, suggesting the impacts in our overall population would likely be minimal.22 We also could not control for the SNF Value-Based Purchasing Program which financially incentivized SNFs for reduced readmission rates, though it was implemented in 2015 near the end of our study period. Fourth, changes in referral patterns or other outcomes could have been driven by unobserved changes in the patient population treated by hospitals or other unobservable time-varying factors as a result of the HRRP. However, we controlled for a rich set of observable clinical and demographic characteristics in analyses, and we did not find differences in the likelihood of referral to SNF or severity of patients referred to SNF in response to the HRRP. Moreover, by controlling for hospital-fixed effects, our analysis controls for time-invariant factors. Finally, our measure of Medicare share used to calculate penalty pressure included both Medicare Advantage and FFS discharges, though only Medicare FFS revenue was subject to the penalty. Our results were similar when examining penalty rate alone.

Conclusion

We do not find evidence that hospitals responded to the HRRP by substantially altering SNF referral practices. However, readmissions for patients referred to a SNF declined substantially with penalty pressure and the differential reduction was greater than that for patients not referred to a SNF. Future research should explore other potential mechanisms underlying the readmissions reduction under the HRRP. In particular, our findings of differentially greater readmission reductions among SNF-referred patients (relative to non-SNF referred patients) and differential increases in the average SNF quality index among high penalty pressure hospitals suggest that improvements in quality of SNF care merit further research as a potentially prominent driver of readmission reductions.

Supplementary Material

Appendix

Grant Support:

This work was supported by a grant from the Agency for Healthcare Research and Quality (R01HS022882). The content is solely the responsibility of the authors and does not necessarily represent the official views of the Agency for Healthcare Research and Quality. The AHRQ did not play a role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; or preparation, review, and approval of the manuscript.

Footnotes

Disclosures:

Drs. Li and Horwitz and Ms. Kuang have performed work under contract to CMS to develop quality measures for readmissions. Drs. Kim and Desai have no conflicts of interest to disclose.

References

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Appendix

RESOURCES