Evidenced Based- Analyzing articles
boopier
Article3EB.pdf
https://doi.org/10.1177/1077558717744611
Medical Care Research and Review 2019, Vol. 76(5) 643 –660
© The Author(s) 2017 Article reuse guidelines:
sagepub.com/journals-permissions DOI: 10.1177/1077558717744611
journals.sagepub.com/home/mcr
Empirical Research
Hospital Readmissions Reduction Program: Intended and Unintended Effects
Min Chen1 and David C. Grabowski2
Abstract This study examines whether the Hospital Readmissions Reduction Program (HRRP), which penalizes hospitals with excess readmissions for certain conditions, has reduced hospital readmissions and led to unintended consequences. Our analyses of Florida hospital administrative data between 2008 and 2014 find that the HRRP resulted in a reduction in the likelihood of readmissions by 1% to 2% for traditional Medicare (TM) beneficiaries with heart failure, pneumonia, or chronic obstructive pulmonary disease. Readmission rates for Medicare Advantage (MA) beneficiaries and privately insured patients with heart attack and heart failure decreased even more than TM patients with the same target condition (e.g., for heart attack, the likelihood for TM beneficiaries to be remitted is 2.2% higher than MA beneficiaries and 2.3% higher than privately insured patients). We do not find any evidence of cost-shifting, delayed readmission, or selection on discharge disposition or patient income. However, the HRRP reduced the likelihood of Hispanic patients with target conditions being admitted by 2% to 4%.
Keywords Medicare, readmissions, hospital, discharge
This article, submitted to Medical Care Research and Review on 30 June 2017, was revised and accepted for publication on November 6, 2017.
1Florida International University, Miami, FL, USA 2Harvard Medical School, Boston, MA, USA
Corresponding Author: Min Chen, College of Business, Florida International University, 11200 SW 8th Street, Miami, FL 33199, USA. Email: [email protected]
744611MCRXXX10.1177/1077558717744611Medical Care Research and ReviewChen and Grabowski research-article2017
644 Medical Care Research and Review 76(5)
Introduction
Hospital readmissions are common and costly. In 2011, the U.S. Medicare program paid for 1.8 million 30-day readmissions with a total cost of $24 billion (Hines, Barrett, Marguerit, Jiang, Joanna, & Steiner, 2014). Some readmissions could be prevented with better quality of care (Axon & Williams, 2011), and the Medicare Payment Advisory Commission (MedPAC) estimates that a 10% reduction in avoidable read- missions would save the Medicare program at least $1 billion (MedPAC, 2013). To achieve both better outcomes for patients and greater savings for Medicare, the Affordable Care Act (ACA) created the Hospital Readmissions Reduction Program (HRRP), which applies financial penalties to acute care hospitals with higher-than- expected readmission rates among Medicare fee-for-service (FFS) beneficiaries in the 30-days following discharge for certain target conditions.
Since October 2012, the HRRP has targeted three conditions: acute myocardial infarction (AMI), congestive heart failure, and pneumonia. Beginning in October 2014, total hip or knee replacement and chronic obstructive pulmonary disease (COPD) were also included in the program. The Centers for Medicare and Medicaid Services (CMS) calculates the average risk-adjusted, 30-day hospital-readmission rates for patients with each targeted condition and penalizes hospitals that perform worse than the national average. For Fiscal Year (FY) 2013, the maximum penalty for a hospital with excess readmissions was 1% of its total Medicare base payment. The penalty went up to 2% of the Medicare base payment for FY 2014, and 3% for FY 2015 forward (CMS, 2016).
New Contribution
Prior studies have examined the initial three target conditions (i.e., AMI, heart failure, and pneumonia) and suggested that the HRRP has lowered 30-day readmissions among Medicare FFS beneficiaries (Carey & Lin, 2015; Gerhardt et al., 2013; Zuckerman, Sheingold, Orav, Ruhter, & Epstein, 2016). Using Medicare FFS claims data, two recent articles compared the changes in readmission rates by hospital penalty status and confirmed that hospitals with the lowest pre-HRRP performance had the greatest improvement (Desai et al., 2016; Wasfy et al., 2017). How readmissions change among Medicare Advantage beneficiaries and privately insured patients, how- ever, is still somewhat unclear and vitally important. Because the HRRP penalties only apply to traditional Medicare patients, one way that a hospital could recoup lost Medicare reimbursements as a result of excess readmissions would be to readmit more privately insured or Medicare Advantage patients. In this study, we exploit a state- based all-payer dataset (through 2014) to examine the overall impact of the HRRP on readmissions among traditional Medicare, Medicare Advantage, and privately insured patients, respectively. We examine not only the aforementioned three originally tar- geted conditions but also the two new penalty conditions (i.e., COPD and total hip or knee replacement).
Furthermore, we explore several other potential consequences of the HRRP across targeted and nontargeted conditions. First, we examine the impact of the HRRP on
Chen and Grabowski 645
readmissions post–30 days to detect if the HRRP has simply delayed readmissions. Next, we examine whether the HRRP led to any “cherry picking” of low-risk patients at admission. Finally, we examine whether the HRRP led to increased skilled nursing facility (SNF) or home health agency (HHA) discharges.
Conceptual Framework
The HRRP is a very direct policy instrument. Hospitals are financially penalized for excess 30-day readmissions for the target conditions. Medicare’s goal in implementing the HRRP was to encourage hospitals to reduce 30-day readmissions through better hospital care. In response to the HRRP, we hypothesize that hospitals will lower read- missions for these target conditions assuming the cost of reducing readmissions is below the amount of the readmission penalty. We also assume that hospitals want to avoid any negative reputation effects associated with being penalized (Winborn, Alencherril, & Pagán, 2014), which might lead them to lower readmissions even if the cost of doing so exceeds the readmission penalty.
Because the HRRP is a relatively blunt policy, we expect it to incent hospitals to change their behaviors in both intended and unintended ways. In terms of unintended consequences, strong potential exists for what economists term the multitasking prob- lem in which providers direct their efforts toward those metrics for which they might be penalized while shirking on those metrics for which they are not penalized. Under the HRRP, hospitals would have the incentive to push any readmissions out past day 30 when they are no longer penalized for the readmission. Critics have suggested that hospitals might dodge the HRRP penalties by increasingly placing returning patients within 30 days of discharge on observation status (Himmelstein & Woolhandler, 2015; Noel-Miller & Lind, 2015). Observation stays are billed as outpatient services rather than readmissions to acute care and would not be counted in the HRRP penalty calcu- lation. Between 2006 and 2013, observation stays increased by 96% for Medicare patients (MedPAC, 2015). One recent study, however, did not find a statistically significant increase in observation stays for targeted versus nontargeted conditions (Zuckerman et al., 2016).
Another unintended consequence would be to discharge patients with a low risk of readmission to costlier postacute care settings because the hospitals are only at risk for readmissions under the HRRP and not postdischarge spending. Thus, at the margin, hospitals have the incentive to increase discharges to home health and skilled nursing facilities for the HRRP target conditions if such discharges would help hospitals reduce readmission rates. From Medicare’s perspective, spending on these postacute services would likely more than offset any potential savings from decreased 30-day readmissions.
Finally, the HRRP’s readmission measures adjust for demographic characteristics associated with higher rates of hospital readmissions (such as age) and severity. However, they do not allow risk adjustment based on patients’ race, ethnicity, or socio- economic status. Because patients with low socioeconomic status are found to have higher readmission rates than the overall population (Hu, Gonsahn, & Nerenz, 2014),
646 Medical Care Research and Review 76(5)
hospitals may respond to the omission of these risk factors by selecting patients on race and socioeconomic status associated with lower rates of hospital readmissions.
Method
Data and Outcome Variables
We construct our hospital admissions and readmissions measures using the State Inpatient Discharge data, collected and maintained by the Florida Agency for Health Care Administration. The data contain detailed information on all inpatient stays in Florida from Quarter 1 of 2008 to Quarter 4 of 2014 and a unique patient identifier that allows us to track a patient’s historical visits across hospitals over time. In addition, we used Medicare Hospital Compare data released in July 2009 (for the period July 2005– June 2008) to examine baseline risk-adjusted readmission rates at the inpatient pro- spective payment system (IPPS) hospitals in the United States.
We adapt methods used in the prior studies to construct index hospitalization and 30-day all-cause readmission at the patient level. Specifically, we code index hospital- izations as stays in which no inpatient discharge had occurred within the previous 30 days. Hence, a hospitalization is either an index stay or a readmission. We then iden- tify target conditions by the principal diagnosis or procedure of the index hospitaliza- tion, using Healthcare Cost and Utilization Project’s (HCUP’s) Clinical Classifications Software (CCS). CCS is a tool that collapses diagnosis and procedure codes from the International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM).1 We used the single level CCS diagnosis code 100 for AMI, 108 for heart failure, 122 for pneumonia, and 127 for COPD. The CCS procedure code used for total hip and knee replacement is 152-153. The ICD-9 codes used to identify total hip and knee replacement are 81.51 (primary hip replacement) and 81.54 (primary knee replacement). In addition, we follow the prior literature (Carey & Lin, 2015; Mellor, Daly, & Smith, 2016) and select gastrointestinal conditions with Medicare Severity Diagnosis Related Group (MS-DRG) codes 329-331, 377-379, and 391-392 to be our control group of Medicare index hospitalizations.2
Similarly, we define two additional indicator variables when readmission occurred within 45 days or 60 days, respectively, and compared them to the 30-day readmission to identify if readmission occurred within 31 to 45 days or 31 to 60 days. Finally, we use the disposition of the patient at discharge to code dummy variables indicating whether the patient was discharged to an SNF or HHA.
Control Variables
To control for heterogeneity associated with changes in readmission and other out- comes over time, our models include a rich set of patient-level covariates. The covari- ates include demographics such as sex, age group, race, primary payer, income category, and rural/urban location. We also constructed time-varying clinical measures for severity adjustment, including (1) indicators of high severity with major
Chen and Grabowski 647
complications/comorbidities based on the MS-DRG codes and (2) the number of comorbid conditions compiled from a set of 29 binary variables identifying coexisting medical conditions that are not directly related to the main reason for index admission (refer to HCUP’s Elixhauser Comorbidity Software for details).3
We identify and exclude certain index hospitalizations following the rules specified in the technical reports of constructing 30-day all cause readmission rates prepared for CMS: (1) hospitalizations during which patients died, (2) discharged against medical advice, and (3) discharged or transferred to another acute care facility. For AMI admis- sions, we also excluded cases with same-day discharges. The analysis sample contains 951,215 index admissions from 156 hospitals.
Statistical Analysis
We use a difference-in-differences (DD) method to compare changes in outcomes of patients in the treatment group before and after the HRRP relative to changes in out- comes of the control group. The treatment group consists of Medicare FFS beneficia- ries aged at least 65 years old and with one of the five HRRP target conditions as the primary diagnosis for their index admission. For each condition we use three different comparison groups for a total of 15 models. The first comparison group consists of hospital admissions among Medicare FFS patients aged 65 years and older and with gastrointestinal conditions as their primary diagnosis. The second comparison group includes hospitalizations of each of the five target conditions among Medicare Advantage patients aged 65 years and older. The third comparison group comprises privately insured patients with those five target conditions.
We estimate the following model:
Y Post Treatment Post
Treatment X Hospi iht t i t
i it
= +
+ + +
∗ +α µ µ
µ β 1 2
3 � ttalh iht+ε (1)
where Yiht is an indicator for a study outcome for patient i at hospital h in time period t. More specifically, we first examine if the patient was readmitted within 30 days of discharge and if there is any delayed readmission after 30 days but within 45 or 60 days of discharge. We also examine if the patient was discharged to a costlier postacute care setting (i.e., a SNF or a HHA). Finally, we examine whether the HRRP reduced the likelihood of admitting minority patients or lower income patients. Minority patients are indicated by whether the patient is Black or of Hispanic ethnicity. We iden- tify a patient to be in a lower income region if the patient resides in a ZIP code wherein the estimated annual median household income is in the bottom two quartiles. Each of these outcome measures represents a separate regression.
Postt is a dummy variable set to 1 if the observation is from the posttreatment period in either the treatment or a comparison group. We use 2008-2009 as the pre- HRRP period and 2012-2014 as the post-HRRP period for AMI, heart failure, and pneumonia. For the two newly added conditions (i.e., COPD and total hip or knee replacement), we use 2014 as the post-HRRP period. Treatmenti indicates whether the
648 Medical Care Research and Review 76(5)
index admission was a hospitalization targeted by the HRRP, and equals zero if the index admission was part of a comparison group. The interaction effect of Postt * Targeti represents our key variable of interest, the DD estimate of the impact of the HRRP. Xit is a vector that captures the time-varying patient characteristics (listed in Table 1). The hospital fixed effects (Hospitalh) are used to control for the unobserved, time-invariant differences across hospitals.
Thus, we use pre-HRRP levels for the target admissions and concurrent changes from the precontract to postcontract period in the nontarget admissions to establish counterfactuals that would be expected in the absence of HRRP program, and we esti- mate changes that differed from this expectation (i.e., the differential change or the change attributable to the HRRP). For all the regression analyses, the standard errors are clustered at the level of the hospital to allow for an arbitrary covariance matrix within the clusters.
Because penalties are based on whether a hospital’s readmission rate exceeds the national average, hospitals with a baseline readmission rate above the threshold are at greater risk of the penalty and thus have stronger incentives to improve. In July 2009, the CMS Hospital Compare website began to publicly report IPPS hospitals’ perfor- mance in 30-day readmission rates for AMI, heart failure, and pneumonia, respectively. For each IPPS hospital with more than 25 cases, its performance is classified into three categories: “better than U.S. national rate,” “no different than U.S. National Rate,” or “worse than U.S. national rate.” We use the national rate for the period July 2005 to June 2008 obtained from CMS’s Hospital Compare data as the baseline threshold rate and compare the hospital specific average 30-day readmission rates to the national average to define if a hospital is “at risk” for any penalty.4 Given that penalties are based on a hospital’s past 3-year average readmission performance, partial responses might be observed immediately after ACA passage but before penalties go into effect. Using historic readmission rates prior to ACA passage allows us to test the full effects of the HRRP. To examine how the impact of HRRP varies across hospitals with differ- ent risks of facing the penalty, we divide the sample into two groups based on whether patients were admitted into a hospital with its baseline readmission rate above the threshold rate, and then we re-estimate the DD model on both of the subsamples.
We further compare this DD estimate of patients treated at hospitals at risk for HRRP penalties versus those patients treated at hospitals not at risk for penalties. More formally, we estimate the triple difference model (DDD) specified below:
Y Post Target Risk Post Target Post
Risk iht t i h t i t
h
= + + +
+
∗ ∗ ∗
∗
α µ µθ 1 2 µµ γ
γ γ β 3 1
2 3
Target Risk Post
Target Risk X Year i h t
i h it t iht
∗ +
+ + + + +ε
(2)
Compared with Equation (1), the added variable Riskh is an indicator variable that specifies whether a hospital is at risk for HRRP penalties, which equals to 1 if hospital h’s baseline readmission rate is above the national average and 0 otherwise. The inter- action effect of Postt * Targeti * Riskh represents our key variable of interest, the triple difference estimate of the impact of the HRRP.
649
T a b
le 1
. D
es cr
ip ti ve
S ta
ti st
ic s.
A ll
Sa m
pl e
A M
I H
F PN
C O
PD H
IP /K
ne e
G I
(1
) (2
) (3
) (4
) (5
) (6
) (7
)
O ut
co m
e m
ea su
re s
R ea
dm it te
d w
it hi
n 30
d ay
s 0.
16 [
0. 37
] 0.
18 [
0. 38
] 0.
23 [
0. 42
] 0.
16 [
0. 37
] 0.
19 [
0. 39
] 0.
05 [
0. 25
] 0.
13 [
0. 34
]
R ea
dm it te
d w
it hi
n 31
-4 5
da ys
0. 03
[ 0.
17 ]
0. 03
[ 0.
17 ]
0. 05
[ 0.
21 ]
0. 03
[ 0.
17 ]
0. 04
[ 0.
20 ]
0. 01
[ 0.
12 ]
0. 03
[ 0.
16 ]
R
ea dm
it te
d w
it hi
n 31
-6 0
da ys
0. 05
[ 0.
22 ]
0. 05
[ 0.
21 ]
0. 08
[ 0.
27 ]
0. 05
[ 0.
22 ]
0. 07
[ 0.
25 ]
0. 02
[ 0.
15 ]
0. 05
[ 0.
21 ]
D
is ch
ar ge
d to
S N
F 0.
17 [
0. 37
] 0.
13 [
0. 33
] 0.
17 [
0. 37
] 0.
17 [
0. 38
] 0.
12 [
0. 32
] 0.
39 [
0. 49
] 0.
09 [
0. 29
]
D is
ch ar
ge d
to H
H A
0. 22
[ 0.
41 ]
0. 16
[ 0.
37 ]
0. 25
[ 0.
43 ]
0. 18
[ 0.
38 ]
0. 20
[ 0.
40 ]
0. 45
[ 0.
50 ]
0. 13
[ 0.
33 ]
Pa ti en
t ch
ar ac
te ri
st ic
s
A
ge (
65 -7
4 ye
ar s)
0. 29
[ 0.
45 ]
0. 29
[ 0.
46 ]
0. 24
[ 0.
43 ]
0. 25
[ 0.
43 ]
0. 38
[ 0.
49 ]
0. 39
[ 0.
49 ]
0. 26
[ 0.
44 ]
A
ge (
75 -8
5 ye
ar s)
0. 30
[ 0.
46 ]
0. 28
[ 0.
45 ]
0. 36
[ 0.
48 ]
0. 31
[ 0.
46 ]
0. 35
[ 0.
48 ]
0. 28
[ 0.
45 ]
0. 27
[ 0.
44 ]
A
ge ( ≥8
5 ye
ar s)
0. 20
[ 0.
40 ]
0. 19
[ 0.
39 ]
0. 33
[ 0.
47 ]
0. 25
[ 0.
43 ]
0. 17
[ 0.
38 ]
0. 09
[ 0.
29 ]
0. 16
[ 0.
37 ]
Fe
m al
e 0.
55 [
0. 50
] 0.
41 [
0. 49
] 0.
49 [
0. 50
] 0.
53 [
0. 50
] 0.
57 [
0. 49
] 0.
61 [
0. 49
] 0.
60 [
0. 49
]
R ac
e
W hi
te 0.
78 [
0. 41
] 0.
78 [
0. 42
] 0.
74 [
0. 44
] 0.
77 [
0. 42
] 0.
81 [
0. 39
] 0.
86 [
0. 35
] 0.
76 [
0. 43
]
B
la ck
0. 09
[ 0.
28 ]
0. 07
[ 0.
26 ]
0. 12
[ 0.
33 ]
0. 09
[ 0.
29 ]
0. 07
[ 0.
25 ]
0. 06
[ 0.
23 ]
0. 10
[ 0.
30 ]
H is
pa ni
c 0.
10 [
0. 30
] 0.
10 [
0. 30
] 0.
10 [
0. 30
] 0.
11 [
0. 31
] 0.
11 [
0. 31
] 0.
06 [
0. 24
] 0.
11 [
0. 31
]
O
th er
r ac
e 0.
03 [
0. 18
] 0.
04 [
0. 21
] 0.
03 [
0. 17
] 0.
03 [
0. 18
] 0.
01 [
0. 12
] 0.
03 [
0. 16
] 0.
04 [
0. 19
]
M ed
ia n
ho us
eh o ld
in co
m e
Q
ua rt
ile 1
0. 34
[ 0.
47 ]
0. 34
[ 0.
47 ]
0. 37
[ 0.
48 ]
0. 35
[ 0.
48 ]
0. 40
[ 0.
49 ]
0. 28
[ 0.
45 ]
0. 33
[ 0.
47 ]
Q ua
rt ile
2 0.
33 [
0. 47
] 0.
33 [
0. 47
] 0.
33 [
0. 47
] 0.
33 [
0. 47
] 0.
33 [
0. 47
] 0.
34 [
0. 47
] 0.
33 [
0. 47
]
Q
ua rt
ile 3
0. 24
[ 0.
42 ]
0. 23
[ 0.
42 ]
0. 22
[ 0.
41 ]
0. 23
[ 0.
42 ]
0. 21
[ 0.
41 ]
0. 27
[ 0.
44 ]
0. 24
[ 0.
43 ]
Q ua
rt ile
4 0.
09 [
0. 29
] 0.
09 [
0. 29
] 0.
08 [
0. 26
] 0.
09 [
0. 28
] 0.
06 [
0. 24
] 0.
11 [
0. 32
] 0.
09 [
0. 29
]
Li vi
ng lo
ca ti o n
R
ur al
a re
a 0.
06 [
0. 23
] 0.
06 [
0. 24
] 0.
06 [
0. 23
] 0.
06 [
0. 23
] 0.
08 [
0. 28
] 0.
06 [
0. 24
] 0.
05 [
0. 22
]
U
rb an
a re
a 0.
94 [
0. 23
] 0.
94 [
0. 24
] 0.
94 [
0. 23
] 0.
94 [
0. 23
] 0.
92 [
0. 28
] 0.
94 [
0. 24
] 0.
95 [
0. 22
]
Pr im
ar y
pa ye
r
Fe e
fo r
se rv
ic e
0. 54
[ 0.
50 ]
0. 49
[ 0.
50 ]
0. 66
[ 0.
47 ]
0. 59
[ 0.
49 ]
0. 62
[ 0.
49 ]
0. 52
[ 0.
50 ]
0. 46
[ 0.
50 ]
M ed
ic ar
e A
dv an
ta ge
0. 22
[ 0.
41 ]
0. 24
[ 0.
43 ]
0. 25
[ 0.
43 ]
0. 19
[ 0.
39 ]
0. 26
[ 0.
44 ]
0. 20
[ 0.
40 ]
0. 21
[ 0.
41 ]
Pr iv
at e
in su
ra nc
e 0.
24 [
0. 43
] 0.
27 [
0. 44
] 0.
09 [
0. 29
] 0.
22 [
0. 41
] 0.
12 [
0. 32
] 0.
28 [
0. 45
] 0.
33 [
0. 47
]
T o ta
l n um
be r
o f co
m o rb
id it ie
s 2.
92 [
1. 89
] 2.
77 [
1. 78
] 3.
56 [
1. 79
] 3.
55 [
2. 04
] 2.
83 [
1. 77
] 2.
22 [
1. 58
] 2.
72 [
1. 90
]
H ig
h se
ve ri
ty (
w it h
m aj
o r
co m
pl ic
at io
ns o
r co
m o rb
id it ie
s) 0.
19 [
0. 39
] 0.
17 [
0. 38
] 0.
30 [
0. 46
] 0.
26 [
0. 44
] 0.
32 [
0. 47
] 0.
05 [
0. 22
] 0.
15 [
0. 36
]
N 95
1, 21
5 10
6, 84
4 16
3, 37
8 13
6, 61
9 81
,6 22
15 0,
33 5
31 2,
41 7
N ot
e. A
M I =
a cu
te m
yo ca
rd ia
l i nf
ar ct
io n;
H F
= h
ea rt
f ai
lu re
; P N
= p
ne um
o ni
a; C
O PD
= c
hr o ni
c o bs
tr uc
ti ve
p ul
m o na
ry d
is ea
se ; H
IP =
t o ta
l h ip
o r
kn ee
a ng
io pl
as ty
/r ep
la ce
m en
t;
G I =
g as
tr o in
te st
in al
c o nd
it io
ns . M
ed ic
ar e
FF S
re fe
rs t
o M
ed ic
ar e
fe e-
fo r-
se rv
ic e
be ne
fic ia
ri es
. S ta
nd ar
d de
vi at
io ns
a re
in b
ra ck
et s.
650 Medical Care Research and Review 76(5)
As noted above, the DDD approach implicitly assumes that hospitals at-risk and not at-risk for the HRRP share the same readmission shocks in a given hospital and year that are unrelated to the HRRP policy. The DD approach, which instead used as controls the within-hospital readmission shocks among patients not included in the HRRP program, may actually be preferable. Because little basis exists for distin- guishing these approaches ex ante, these models are probably best viewed as com- plementary approaches for exploring the validity of this study’s key results.
We conduct additional analyses to explore potential sources of bias. We compare trends in each outcome between the targeted and nontargeted admissions during the pre-HRRP period. Similar pre-HRRP trends would support our assumption that changes from the pre-HRRP to post-HRRP periods would have been similar for the target and nontarget conditions in the absence of the HRRP program. Considering that CMS began publicly reporting hospital performance in July 2009 and hospitals might start to respond by changing their behavior since then, we restrict the pre-HRRP period to be the first two quarters of 2009 and reestimated all the specifications using the alternative sample and the results stay robust.
Results
We observe several notable trends when examining the 30-day all cause readmis- sions by condition from 2008 to 2014 (see Figure 1). First, the 30-day readmission rates of FFS patients followed similar trends from 2008 to 2009 across the five target conditions and gastrointestinal condition. Second, the 30-day readmission rates of FFS patients with each of the five target conditions decreased or stayed relatively stable from 2012 to 2014, while the FFS patients with gastrointestinal conditions
Figure 1. Thirty-day all-cause readmission trend by condition. Note. HF = heart failure; PN = pneumonia; HIP = total hip or knee replacement; GI = gastrointestinal conditions.
Chen and Grabowski 651
experienced an increase in their 30-day readmission rate during the same time period. Finally, within the same condition, the 30-day readmission rates of FFS and Medicare Advantage patients followed similar trends from 2008 to 2009. We there- fore use 2008-2009 as the pre-HRRP comparison period. When comparing across payers for a given target condition, we also observe that Medicare readmissions rates were consistently higher than the rates for privately insured patients. Table 1 reports the descriptive statistics of the whole sample and by each of the five target conditions as well as the gastrointestinal condition. Compared with the national average, our sample has slightly higher 30-day all-cause readmission rates in AMI and heart failure and comparable readmission rates in pneumonia, COPD, and total hip or knee replacement.
We next examine the DD estimates on HRRP targeted admissions using three dif- ferent comparison groups (see Table 2). Compared with Medicare FFS patients with gastrointestinal conditions as the primary diagnosis, there was a 1% to 2% decrease in 30-day readmissions for comparable heart failure, pneumonia, and COPD patients. However, when compared with Medicare Advantage patients with the same target condition, we observe a statistically significant increase in 30-day Medicare FFS read- mission for AMI, heart failure, and pneumonia. Similarly, when compared with privately insured patients, 30-day readmissions for Medicare FFS patients admitted with AMI and heart failure increased. The results reveal that although the HRRP tar- geted Medicare FFS patients only, hospital readmission rates declined substantially in the MA and privately insured population after the HRRP, especially among cardiac related admissions. This may suggest that there are spillover effects from the HRRP extending to MA and privately insured patients. We then restrict our attention to MA and privately insured patients admitted with one of the five HRRP target conditions and compare changes in their readmissions to those of MA and privately insured
Table 2. Difference-in-Differences (DID) Estimates of the Effect of the Hospital Readmissions Reduction Program on Medicare FFS 30-Day Readmissions.
Medicare FFS patients with GI conditions as
control
Medicare FFS with Medicare Advantage,
same condition as control
Medicare FFS with private insurance, same
condition as control
DID impact DID impact DID impact
(1) (2) (3)
(1) Heart attack −0.004 (0.005) 0.022*** (0.008) 0.023*** (0.007) (2) Heart failure −0.007** (0.004) 0.012** (0.005) 0.020*** (0.008) (3) Pneumonia −0.006* (0.003) 0.010* (0.006) 0.007 (0.005) (4) Chronic obstructive
pulmonary disease −0.018** (0.005) 0.005 (0.007) −0.006 (0.008)
(5) Total hip or knee angioplasty
0.013 (0.008) 0.01 (0.011) 0.002 (0.007)
Note. FFS = fee-for-service; GI = gastrointestinal. All models include control variables listed in Table 1 as well as hospital fixed effects. The standard errors are clustered at hospital level. Robust standard errors in parentheses. *p < .1. **p < .05. ***p < .01.
652 Medical Care Research and Review 76(5)
patients with gastrointestinal conditions, respectively. The DD estimates reported in Appendix Table A1 confirmed that after passage of the HRRP, hospitals reduced car- diac-related readmissions not only for Medicare FFS patients, but also for MA and privately insured patients.
Next, we reran the DD estimation conditional on hospitals’ baseline readmission performance (see Table 3). Compared with admissions with gastrointestinal condi- tions, index hospitalizations with target conditions at a hospital “at risk” for penalties had statistically significant lower 30-day readmissions.5 On the contrary, none of the five target conditions show significant reductions in 30-day readmissions at hospitals with baseline performance better than the national average and thus at less risk for the penalties. These findings suggest that the HRRP has been effective in improving the regulated quality dimensions of the low performers, but high performers at baseline lacked incentives to further reduce their readmissions.
Having examined the main effect on readmission rates using various comparison groups, we turn to other potential intended and unintended consequences of the program. We examine the HRRP on three different dimensions: post–30-day readmis- sions, discharge status, and potential patient selection on income and race (see Table 4). Because privately insured patients and Medicare beneficiaries differ in age and other characteristics that may confound the results, we focus on using Medicare FFS benefi- ciaries with gastrointestinal conditions and MA beneficiaries with the same target con- dition as controls. We found no strategic responses from hospitals in terms of postponing readmissions past 30 days. This is consistent with early findings using data up to year 2012 (Carey & Lin, 2015) and extends results from another study that
Table 3. Difference-in-Differences (DID) Estimates of the Effect of the Hospital Readmissions Reduction Program on 30-Day Readmissions Conditional on Initial Performance at Baseline.
Discharged from hospitals at risk for penalties
Discharged from hospitals not at risk for penalties
DID impact DID impact
(2) (4)
(1) Heart attack −0.016** (0.008) −0.006 (0.005) (2) Heart failure −0.009** (0.004) −0.004 (0.004) (3) Pneumonia −0.012** (0.005) −0.0004 (0.004) (4) Chronic obstructive
pulmonary disease −0.010* (0.005) −0.005 (0.006)
(5) Total hip or knee angioplasty
0.003 (0.009) 0.0006 (0.006)
Note. The comparison group consists of Medicare fee-for-service patients with gastrointestinal conditions as their primary diagnosis during index admissions. All models include control variables listed in Table 1 as well as hospital fixed effects. The standard errors are clustered at hospital level. Robust standard errors in parentheses. *p < .1. **p < .05. ***p < .01.
653
T a b
le 4
. H
o sp
it al
R ea
dm is
si o ns
R ed
uc ti o n
Pr o gr
am E
ffe ct
o n
D el
ay ed
R ea
dm is
si o n,
D is
ch ar
ge D
is po
si ti o n
an d
Pa ti en
t Se
le ct
io n.
M ed
ic ar
e FF
S pa
ti en
ts w
it h
G I co
nd it io
ns a
s co
nt ro
l M
ed ic
ar e
FF S
w it h
M ed
ic ar
e A
dv an
ta ge
, s am
e co
nd it io
n as
co
nt ro
l
A
M I
H F
PN C
O PD
H ip
/K ne
e A
M I
H F
PN C
O PD
H ip
/K ne
e
(1
) (2
) (3
) (4
) (5
) (6
) (7
) (8
) (9
) (1
0)
D el
ay ed
r ea
dm is
si o n
(1
) R
ea dm
it te
d w
it hi
n 31
-4 5
da ys
− 0.
00 1
0. 00
2 −
0. 00
2 −
0. 00
2 −
0. 00
3* −
0. 00
1 0.
00 5
− 0.
00 7*
* 0.
00 03
− 0.
00 2
(0 .0
03 )
(0 .0
02 )
(0 .0
02 )
(0 .0
03 )
(0 .0
02 )
(0 .0
04 )
(0 .0
03 )
(0 .0
03 )
(0 .0
03 )
(0 .0
02 )
(2
) R
ea dm
it te
d w
it hi
n 31
-6 0
da ys
− 0.
00 3
0. 00
3 −
0. 00
5* −
0. 00
5 −
0. 00
6* *
− 0.
00 5
0. 00
8 −
0. 00
4 −
0. 00
2 −
0. 00
00 7
(0 .0
04 )
(0 .0
03 )
(0 .0
03 )
(0 .0
04 )
(0 .0
02 )
(0 .0
05 )
(0 .0
05 )
(0 .0
04 )
(0 .0
04 )
(0 .0
02 )
(1 94
0. 6)
(1 32
8. 6)
(1 20
9. 4)
(8 28
.9 )
(2 27
8. 5)
(1 82
1. 9)
− 12
02 .8
(8 22
.4 )
(5 57
.4 )
(1 20
7. 3)
D is
ch ar
ge d
is po
si ti o n
(3
) D
is ch
ar ge
d to
a S
ki lle
d N
ur si
ng F
ac ili
ty −
0. 01
4* *
0. 00
1 −
0. 01
1* *
− 0.
01 3*
* −
0. 01
5 −
0. 01
5* *
− 0.
00 6
− 0.
00 1
0. 00
09 0.
07 3
(0 .0
06 )
(0 .0
05 )
(0 .0
05 )
(0 .0
06 )
(0 .0
31 )
(0 .0
06 )
(0 .0
06 )
(0 .0
06 )
(0 .0
05 )
(0 .1
49 )
(4
) D
is ch
ar ge
d to
a H
o m
e H
ea lt h
A ge
nc y
0. 00
7 0.
02 1*
0. 00
6 0.
00 5
0. 02
6 −
0. 00
5 −
0. 02
1* *
− 0.
01 6*
* −
0. 01
6* −
0. 02
7 (0
.0 07
) (0
.0 11
) (0
.0 05
) (0
.0 08
) (0
.0 30
) (0
.0 08
) (0
.0 09
) (0
.0 07
) (0
.0 09
) (0
.0 25
) In
co m
e
(5 )
B el
o w
2 5%
− 0.
01 2
0. 00
02 0.
00 3
0. 01
7 −
0. 00
2 0.
00 01
− 0.
00 2
− 0.
01 48
− 0.
02 0*
− 0.
00 9
(0 .0
09 )
(0 .0
06 )
(0 .0
07 )
(0 .0
10 )
(0 .0
17 )
(0 .0
09 )
(0 .0
10 )
(0 .0
11 )
(0 .0
11 )
(0 .0
12 )
(6
) B
el o w
5 0%
− 0.
00 4
− 0.
00 3
0. 00
06 −
0. 00
7 0.
02 0
0. 01
1 0.
01 0
0. 01
5 0.
01 5
0. 14
7 (0
.0 09
) (0
.0 06
) (0
.0 07
) (0
.0 10
) (0
.0 22
) (0
.0 10
) (0
.0 09
) (0
.0 11
) (0
.0 11
) (0
.1 82
) R
ac e
(7
) B
la ck
0. 00
4 0.
00 3
− 0.
00 5
0. 00
06 0.
01 1
0. 00
3 0.
00 1
0. 00
7 0.
00 4
0. 00
3 (0
.0 06
) (0
.0 04
) (0
.0 04
) (0
.0 04
) (0
.0 15
) (0
.0 06
) (0
.0 08
) (0
.0 10
) (0
.0 05
) (0
.0 04
)
(8 )
H is
pa ni
c −
0. 01
4 −
0. 01
8* *
− 0.
01 7
− 0.
01 6*
− 0.
04 0*
** −
0. 07
2* **
− 0.
06 1*
** 0.
00 4
− 0.
03 6*
** −
0. 06
6* (0
.0 10
) (0
.0 09
) (0
.0 11
) (0
.0 09
) (0
.0 15
) (0
.0 24
) (0
.0 20
) (0
.0 05
) (0
.0 12
) (0
.0 37
)
N ot
e. A
M I =
a cu
te m
yo ca
rd ia
l i nf
ar ct
io n;
H F
= h
ea rt
f ai
lu re
; P N
= p
ne um
o ni
a; C
O PD
= c
hr o ni
c o bs
tr uc
ti ve
p ul
m o na
ry d
is ea
se ; H
ip /K
ne e
= t
o ta
l h ip
o r
kn ee
a ng
io pl
as ty
/ re
pl ac
em en
t; G
I =
g as
tr o in
te st
in al
c o nd
it io
ns ; F
FS =
f ee
-f o r-
se rv
ic e.
A ll
m o de
ls in
cl ud
e co
nt ro
l v ar
ia bl
es li
st ed
in T
ab le
1 a
s w
el l a
s ye
ar a
nd h
o sp
it al
f ix
ed e
ffe ct
s. T
he r
o bu
st
st an
da rd
e rr
o rs
r ep
o rt
ed in
p ar
en th
es es
a re
c lu
st er
ed a
t ho
sp it al
le ve
l. *p
< .1
. * *p
< .0
5. *
** p
< .0
1.
654 Medical Care Research and Review 76(5)
focuses on AMI patients only (Mellor et al., 2016). Interestingly, for hip or knee replacement, readmissions outside the 30-day window also decreased. This might be partly due to the effect of the Bundled Payment for Care Improvement. The cost incen- tives of bundled payment for lower extremity joint replacement programs took effect from October 1, 2013, slightly overlapping with the period of this study; and are based on an episode of care that ends 90 days postdischarge. In terms of discharge status, there is a consistent trend of AMI patients being decreasingly discharged to SNF. Although we do not observe any selection against low-income patients, the percentage of Hispanic patients admitted with heart failure and COPD conditions as well as for total hip or knee replacement surgeries were on the decline. Taken together, the results show heterogeneous effects of the HRRP on various outcomes across different conditions.
We conduct additional analyses to explore the validity of this study’s key results and report the findings in the Appendix. The triple difference estimates presented in the Appendix Table A2 yield similar results as the DD estimates reported in Table 2, column (1), which serves as a sensitivity analysis. In Appendix Table A3, we report additional validation tests of the robustness of our main DD and triple differences results. We construct an indicator for 2009, which is prior to the HRRP implementa- tion, and estimate specifications (1) and (2) using the pre-HRRP years only. None of the 15 models show any evidence of significant differences in 30-day readmission in periods prior to the regulations. This partially validates the identifying assumption that, in the absence of the HRRP, the treatment and comparison groups would have had similar trends in readmission.
Discussion
We examine patients discharged by Florida hospitals using both DD and triple dif- ference models to identify the HRRP’s effects. Although our results are comparable to those in the prior studies when using similar treatment and control groups, the readmission reduction disappeared or reversed when compared with Medicare Advantage enrollees and privately insured patients, especially for the three initially targeted conditions. On the one hand, this may suggest that health care quality has some commonality across payer groups and hospitals’ efforts in reducing readmis- sions likely spilled over to other Medicare beneficiaries and privately insured patients not directly targeted by the HRRP. On the other hand, we must also acknowl- edge that the larger effect among Medicare Advantage and privately insured patients may suggest some other contemporaneous downward trend in readmissions for these particular conditions that is unrelated to the HRRP. Regardless, our results allay concerns that hospitals may engage in cost-shifting behaviors to offset Medicare HRRP penalties. Given that hospitals’ incentives to reduce readmissions under the HRRP only apply to selected Medicare patients, one way that a hospital could recoup its lost Medicare reimbursements due to excess readmissions would be to readmit more patients covered by other types of insurance. Our results, however, do not sup- port such hypothesized cost-shifting behavior.
Chen and Grabowski 655
There is also no evidence of selection on income or admission of Black patients due to the HRRP. However, the likelihood of Hispanic patients being admitted with heart failure, COPD, or total hip or knee replacement decreased by 2 to 4 percent- age points. Studies have found that Black and Hispanic patients experienced higher readmission rates than Whites for many diagnoses including the target conditions (heart failure, AMI, pneumonia) covered by the CMS readmissions policies (Alexander, Grumbach, Remy, Rowell, & Massie, 1999; Jiang, Andrews, Stryer, & Friedman, 2005; McHugh, Carthon, & Kang, 2010; Rathore et al., 2003). In addi- tion, many providers believe that minority patients tend to be less educated and less likely to comply with treatment and thus have higher risks for readmissions (Balsa & McGuire, 2003; Schulman et al., 1999; van Ryn & Burke, 2000). Because the current penalty formula does not adjust for hospitals that serve large shares of indi- gent or minority patients, they may avoid such patients who they believe will reduce their performance and result in financial penalties (Ryan, 2010). This raises a major issue of concern as the HRRP may divert resources away from the small percentage of U.S. hospitals caring for the large majority of elderly Black and Hispanic patients and exacerbate health care related disparities in access and outcomes (Bhalla & Kalkut, 2010; McHugh et al., 2010). The fact that there is no evidence of selection against low-income or Black patients under the HRRP is reassuring, but the decreased admission of Hispanic patients is concerning, especially given a rela- tively large Hispanic population in Florida. It is worth noting that the inpatient discharge data contains limited information on socioeconomic status and a richer set of sociodemographic variables are needed to keep monitoring disparities across different minority groups and better understand the underlying factors that may cause disparities.
In terms of discharge status, our study finds AMI patients were less likely to be discharged to an SNF. This is consistent with prior studies that report no correlation or positive relationship between SNF rates and readmission rates for AMI or heart failure patients (Allen et al., 2011; Chen et al., 2012; Manemann et al., 2017), which challenges perceptions that SNF care will necessarily reduce readmissions. A recent study has also found that deliberate reduction in intensive post-acute care discharges because of incentives created by the Bundled Payment for Care Improvement model is not associated with increase in readmission rates (Jubelt, Goldfeld, Chung, Blecker, & Horwitz, 2016). Nonclinical factors such as hospital ownership of a SNF facility or distance to SNF may affect SNF referral and obscure the relationship between SNF care and readmission risk. Hospitals with low use of SNFs, on the other hand, may employ other mechanisms such as home health care nurses to ensure that patients received sufficient follow-up after discharge and thus achieve low readmission rates.
This study has limitations. First, we rely on the secondary inpatient administra- tive data, which are limited in clinical details in different stages of the care delivery process especially after the patient being discharged. We also do not directly observe providers’ actual medical decision-making processes at point of care. Therefore, the specific mechanism that drives the reduced readmissions and lead
656 Medical Care Research and Review 76(5)
to the heterogeneous effects of the HRRP on different target conditions remains a black box yet to be investigated. Similarly, due to data limitations, we are not able to examine changes in the observation stays following the HRRP. Finally, although we use various comparison groups and the triple difference approach as a sensitiv- ity analysis, our results still rely on the common trend assumption underlying the DD study design. If hospitals undertook other unobserved quality-improving ini- tiatives contemporaneous with the HRRP on the target admissions, we would over- estimate the impact of HRRP.
In conclusion, this study found that the HRRP led to intended readmission reduc- tions among Florida traditional Medicare beneficiaries. Meanwhile, MA and pri- vately insured patients with heart attack and heart failure had even lower readmission rates. There is no evidence of cost shifting, delayed readmission, or selection on income; however, the HRRP reduced the likelihood of Hispanic patients with target conditions being admitted by 2 to 4 percentage points. Future research is needed to understand why readmissions have fallen and how current health care reforms may affect other outcome dimensions including racial disparity and longer term health outcomes.
Appendix
Table A1. Difference-in-Differences (DID) Estimates of the Effect of the Hospital Readmissions Reduction Program on Medicare Advantage and Privately Insured 30-day Readmissions.
Medicare advantage patients with GI
conditions as control
Privately insured patients with GI
conditions as control
DID Impact DID Impact
(1) (2)
(1) Heart attack (acute myocardial infarction [AMI])
−0.021*** (0.007) −0.015*** (0.005)
(2) Heart failure (HF) −0.009* (0.005) −0.016* (0.008) (3) Pneumonia (PNE) −0.004 (0.006) −0.004 (0.004) (4) Chronic obstructive
pulmonary disease (COPD) 0.004 (0.007) 0.005 (0.008)
(5) Total hip or knee angioplasty 0.005 (0.007) −0.023*** (0.006)
Note. GI = gastrointestinal. All models include control variables listed in Table 1 as well as hospital fixed effects. The standard errors are clustered at hospital level. Robust standard errors in parentheses. *p < .1. **p < .05. ***p < .01.
657
T a b
le A
2 .
D iff
er en
ce -i n-
D iff
er en
ce -i n-
D iff
er en
ce s
Es ti m
at es
o f H
o sp
it al
R ea
dm is
si o ns
R ed
uc ti o n
Pr o gr
am E
ffe ct
o n
30 -d
ay R
ea dm
is si
o ns
.
A M
I H
F PN
C O
PD H
ip /K
ne e
(1
) (2
) (3
) (4
) (5
)
Po st
0. 01
0* **
( 0.
00 3)
0. 01
3* **
( 0.
00 3)
0. 01
1* **
( 0.
00 3)
0. 02
0* **
( 0.
00 4)
0. 02
2* **
( 0.
00 4)
T re
at m
en t
0. 06
** *
(0 .0
04 )
0. 07
** *
(0 .0
03 )
0. 00
5 (0
.0 03
) 0.
04 3*
** (
0. 00
4) −
0. 03
7* **
( 0.
00 5)
A t-
ri sk
0. 00
05 (
0. 00
4) −
0. 02
8 (0
.0 19
) 0.
00 2
(0 .0
04 )
0. 01
4* *
(0 .0
06 )
0. 02
4* **
( 0.
00 6)
Po st
* T
re at
− 0.
00 2
(0 .0
05 )
− 0.
00 4
(0 .0
04 )
− 0.
00 2
(0 .0
04 )
− 0.
01 7*
* (0
.0 07
) 0.
00 8
(0 .0
10 )
Po st
* A
t- ri
sk 0.
00 1
(0 .0
06 )
− 0.
01 0*
( 0.
00 6)
− 0.
00 4
(0 .0
05 )
− 0.
00 9
(0 .0
06 )
− 0.
00 6
(0 .0
06 )
T re
at *
A t-
ri sk
0. 01
8* (
0. 01
0) 0.
01 8*
** (
0. 00
6) 0.
02 1*
** (
0. 00
6) −
0. 01
3* *
(0 .0
07 )
− 0.
01 1*
( 0.
00 6)
Po st
* T
re at
* A
t- ri
sk −
0. 00
9* (
0. 00
5) −
0. 00
6* (
0. 00
3) (0
.0 06
) −
0. 01
0* (
0. 00
9) 0.
01 1
(0 .0
14 )
N 19
0, 63
0 24
9, 70
8 22
2, 74
0 13
3, 91
9 13
5, 56
7
N ot
e. A
M I =
a cu
te m
yo ca
rd ia
l i nf
ar ct
io n;
H F
= h
ea rt
f ai
lu re
; P N
= p
ne um
o ni
a; C
O PD
= c
hr o ni
c o bs
tr uc
ti ve
p ul
m o na
ry d
is ea
se ; a
nd H
ip /K
ne e
= t
o ta
l h ip
o r
kn ee
a ng
io pl
as ty
/r ep
la ce
m en
t. T
he c
o m
pa ri
so n
gr o up
c o ns
is ts
o f M
ed ic
ar e
fe e-
fo r-
se rv
ic e
pa ti en
ts w
it h
ga st
ro in
te st
in al
c o nd
it io
ns a
s th
ei r
pr im
ar y
di ag
no si
s du
ri ng
in de
x ad
m is
si o ns
. A ll
m o de
ls in
cl ud
e co
nt ro
l v ar
ia bl
es li
st ed
in T
ab le
1 a
s w
el l a
s ye
ar a
nd h
o sp
it al
f ix
ed e
ffe ct
s. T
he s
ta nd
ar d
er ro
rs a
re c
lu st
er ed
a t
ho sp
it al
le ve
l. R
o bu
st s
ta nd
ar d
er ro
rs in
p ar
en th
es es
. *p
< .1
. * *p
< .0
5. *
** p
< .0
1.
658 Medical Care Research and Review 76(5)
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
Notes
1. More details about the Clinical Classifications Software (CCS) for ICD-9-CM can be found here: https://www.hcup-us.ahrq.gov/toolssoftware/ccs/ccs.jsp
2. Alternatively, we also identify controls using CCS codes 138-140 and 153-155, and our results remain robust.
3. https://www.hcup-us.ahrq.gov/toolssoftware/comorbidity/comorbidity.jsp 4. For the five conditions we examined, the national average rate is 19.9 for AMI, 24.5 for
heart failure, 18.2 for pneumonia, 20.7 for COPD, and 5.2 for total hip or knee replacement. 5. The only exception is total hip or knee replacement, of which the DD estimate is insignificant.
References
Alexander, M., Grumbach, K., Remy, L., Rowell, R., & Massie, B. M. (1999). Congestive heart failure hospitalizations and survival in California: Patterns according to race/ethnicity. American Heart Journal, 137, 919-927.
Table A3. Specification Checks: Difference-in-Differences (DD) and Difference-in- Difference-in-Differences (DDD) Models With Alternate Control Groups.
Medicare FFS patients with GI
conditions as control
Medicare Advantage patients
as control
Medicare FFS patients with GI conditions as
control
DD estimates on pre-HRRP years
DD estimates on pre-HRRP years
DDD estimates on pre-HRRP years
(1) (2) (3)
(1) Heart attack (AMI) −0.001 (0.006) 0.001 (0.008) 0.014 (0.014) (2) Heart failure (HF) −0.001 (0.005) 0.011 (0.008) −0.004 (0.010) (3) Pneumonia (PNE) −0.007 (0.005) −0.006 (0.008) −0.007 (0.010) (4) Chronic obstructive
pulmonary disease (COPD)
0.006 (0.005) 0.0086 (0.008) 0.014 (0.011)
(5) Total hip or knee angioplasty
−0.001 (0.004) 0.004 (0.007) −0.009 (0.009)
Note. AMI = acute myocardial infarction; GI = gastrointestinal conditions; FFS = fee-for-service; HRRP = Hospital Readmissions Reduction Program. All models include control variables listed in Table 1 as well as hospital fixed effects. The standard errors are clustered at hospital level. Robust standard errors in parentheses. *p < .1. **p < .05. ***p < .01.
Chen and Grabowski 659
Allen, L. A., Hernandez, A. F., Peterson, E. D., Curtis, L. H., Dai, D., Masoudi, F. A., Fonarow, G. C. (2011). Discharge to a skilled nursing facility and subsequent clinical outcomes among older patients hospitalized for heart failure. Circulation. Heart Failure, 4, 293-300. https://doi.org/10.1161/CIRCHEARTFAILURE.110.959171
Axon, R. N., & Williams, M. V. (2011). Hospital readmission as an accountability measure. JAMA Journal of the American Medical Association, 305, 504-505. https://doi.org/10.1001/ jama.2011.72
Balsa, A. I., & McGuire, T. G. (2003). Prejudice, clinical uncertainty and stereotyping as sources of health disparities. Journal of Health Economics, 22, 89-116.
Bhalla, R., & Kalkut, G. (2010). Could Medicare readmission policy exacerbate health care system inequity? Annals of Internal Medicine, 152, 114-117. doi:10.7326/0003-4819-152- 2-201001190-00185
Carey, K., & Lin, M.-Y. (2015). Readmissions to New York hospitals fell for three target con- ditions from 2008 To 2012, consistent with Medicare goals. Health Affairs, 34, 978-985. doi:10.1377/hlthaff.2014.1408
Centers for Medicare & Medicaid Services. (2016). Readmissions Reduction Program (HRRP). Retrieved from https://www.cms.gov/Medicare/Medicare-Fee-for-Service-Payment/ AcuteInpatientPPS/Readmissions-Reduction-Program.html
Chen, J., Ross, J. S., Carlson, M. D. A., Lin, Z., Normand, S.-L. T., Bernheim, S. M., . . . Krumholz, H. M. (2012). Hospital referral to skilled nursing facilities and readmission rates after heart failure or myocardial infarction. American Journal of Medicine, 125, 100. e1-100.e9. doi:10.1016/j.amjmed.2011.06.011
Desai, N. R., Ross, J. S., Kwon, J. Y., Herrin, J., Dharmarajan, K., Bernheim, S. M., . . . Horwitz, L. I. (2016). Association between hospital penalty status under the hospital readmission reduction program and readmission rates for target and nontarget condi- tions. JAMA Journal of the American Medical Association, 316, 2647-2656. doi:10.1001/ jama.2016.18533
Gerhardt, G., Yemane, A., Hickman, P., Oelschlaeger, A., Rollins, E., & Brennan, N. (2013). Medicare readmission rates showed meaningful decline in 2012. Medicare & Medicaid Research Review, 3(2), E1-E12. doi:10.5600/mmrr.003.02.b01
Himmelstein, D., & Woolhandler, S. (2015). Quality improvement: “Become good at cheating and you never need to become good at anything else.” Retrieved from http://healthaffairs. org/blog/2015/08/27/quality-improvement-become-good-at-cheating-and-you-never- need-to-become-good-at-anything-else/
Hines, A. L., Barrett, Marguerit, L., Jiang, H., Joanna, & Steiner, C. A. (2014). Conditions with the largest number of adult hospital readmissions by payer, 2011 (Healthcare Cost and Utilization Project). Rockville, MD: Agency for Healthcare Research and Quality.
Hu, J., Gonsahn, M. D., & Nerenz, D. R. (2014). Socioeconomic status and readmissions: Evidence from an urban teaching hospital. Health Affairs, 33, 778-785. doi:10.1377/ hlthaff.2013.0816
Jiang, H. J., Andrews, R., Stryer, D., & Friedman, B. (2005). Racial/ethnic disparities in poten- tially preventable readmissions: The case of diabetes. American Journal of Public Health, 95, 1561-1567. doi:10.2105/AJPH.2004.044222
Jubelt, L. E., Goldfeld, K. S., Chung, W., Blecker, S. B., & Horwitz, L. I. (2016). Changes in discharge location and readmission rates under Medicare bundled payment. JAMA Internal Medicine, 176, 115-117. doi:10.1001/jamainternmed.2015.6265
660 Medical Care Research and Review 76(5)
Manemann, S. M., Chamberlain, A. M., Boyd, C. M., Weston, S. A., Killian, J., Leibson, C. L., . . . Roger, V. L. (2017). Skilled nursing facility use and hospitalizations in heart fail- ure: A community linkage study. Mayo Clinic Proceedings, 92, 490-499. doi:10.1016/j. mayocp.2017.01.014
McHugh, M. D., Carthon, J. M. B., & Kang, X. L. (2010). Medicare readmissions policies and racial and ethnic health disparities: A cautionary tale. Policy, Politics & Nursing Practice, 11, 309-316. doi:10.1177/1527154411398490
MedPAC. (2013, June). Refining the hospital readmissions reduction program. Retrieved from http://www.medpac.gov/docs/default-source/reports/jun13_ch04.pdf
MedPAC. (2015). Hospital short-stay policy issues. Retrieved from http://www.medpac.gov/ docs/default-source/reports/chapter-7-hospital-short-stay-policy-issues-june-2015-report-. pdf?sfvrsn=0
Mellor, J., Daly, M., & Smith, M. (2016). Does it pay to penalize hospitals for excess readmis- sions? Intended and unintended consequences of Medicare’s hospital readmissions reduc- tions program. Health Economics, 26, 1037-1051. doi:10.1002/hec.3382
Noel-Miller, C., & Lind, K. (2015, October 28). Is observation status substituting for hospital readmission? Retrieved from http://healthaffairs.org/blog/2015/10/28/is-observation- status-substituting-for-hospital-readmission/
Rathore, S. S., Foody, J. M., Wang, Y., Smith, G. L., Herrin, J., Masoudi, F. A., . . . Krumholz, H. M. (2003). Race, quality of care, and outcomes of elderly patients hospitalized with heart failure. JAMA Journal of the American Medical Association, 289, 2517-2524. doi:10.1001/jama.289.19.2517
Ryan, A. M. (2010). Has pay-for-performance decreased access for minority patients? Health Services Research, 45, 6-23. doi:10.1111/j.1475-6773.2009.01050.x
Schulman, K. A., Berlin, J. A., Harless, W., Kerner, J. F., Sistrunk, S., Gersh, B. J., . . . Escarce, J. J. (1999). The effect of race and sex on physicians’ recommendations for cardiac catheterization. New England Journal of Medicine, 340, 618-626. doi:10.1056/ NEJM199902253400806
van Ryn, M., & Burke, J. (2000). The effect of patient race and socio-economic status on physi- cians’ perceptions of patients. Social Science & Medicine, 50, 813-828.
Wasfy, J. H., Zigler, C. M., Choirat, C., Wang, Y., Dominici, F., & Yeh, R. W. (2017). Readmission rates after passage of the hospital readmissions reduction program: A pre– post analysis. Annals of Internal Medicine, 166, 324-331. doi:10.7326/M16-0185
Winborn, M. S., Alencherril, J., & Pagán, J. A. (2014). A news media analysis of the economic and reputational penalties of the hospital readmissions reduction program. INQUIRY: Journal of Health Care Organization, Provision, and Financing, 51, 0046958014561635. doi:10.1177/0046958014561635
Zuckerman, R. B., Sheingold, S. H., Orav, E. J., Ruhter, J., & Epstein, A. M. (2016). Readmissions, observation, and the hospital readmissions reduction program. New England Journal of Medicine, 374, 1543-1551. doi:10.1056/NEJMsa1513024
