You are on page 1of 8

ORIGINAL CONTRIBUTION

AssociationBetweenStrokeCenterHospitalization
for Acute Ischemic Stroke and Mortality
Ying Xian, MD, PhD Context Although stroke centers are widely accepted and supported, little is known
Robert G. Holloway, MD, MPH about their effect on patient outcomes.
Paul S. Chan, MD, MSc Objective To examine the association between admission to stroke centers for acute
ischemic stroke and mortality.
Katia Noyes, PhD, MPH
Design, Setting, and Participants Observational study using data from the
Manish N. Shah, MD, MPH
New York Statewide Planning and Research Cooperative System. We compared
Henry H. Ting, MD, MBA mortality for patients admitted with acute ischemic stroke (n = 30 947) between
Andre R. Chappel, BA 2005 and 2006 at designated stroke centers and nondesignated hospitals using dif-
ferential distance to hospitals as an instrumental variable to adjust for potential pre-
Eric D. Peterson, MD, MPH hospital selection bias. Patients were followed up for mortality for 1 year after the
Bruce Friedman, PhD, MPH index hospitalization through 2007. To assess whether our findings were specific to
stroke, we also compared mortality for patients admitted with gastrointestinal hem-

S
TROKE IS THE LEADING CAUSE OF orrhage (n=39 409) or acute myocardial infarction (n=40 024) at designated stroke
serious long-term disability and centers and nondesignated hospitals.
the third leading cause of mor- Main Outcome Measure Thirty-day all-cause mortality.
tality in the United States.1 Re-
sponding to the need for improve- Results Among 30 947 patients with acute ischemic stroke, 15 297 (49.4%) were
admitted to designated stroke centers. Using the instrumental variable analysis, ad-
ments in acute stroke care, the Brain mission to designated stroke centers was associated with lower 30-day all-cause mor-
Attack Coalition (BAC) published rec- tality (10.1% vs 12.5%; adjusted mortality difference, 2.5%; 95% confidence in-
ommendations for the establishment of terval [CI], 3.6% to 1.4%; P.001) and greater use of thrombolytic therapy (4.8%
primary stroke centers in 2000.2 In De- vs 1.7%; adjusted difference, 2.2%; 95% CI, 1.6% to 2.8%; P.001). Differences in
cember 2003, the Joint Commission be- mortality also were observed at 1-day, 7-day, and 1-year follow-up. The outcome dif-
gan certifying stroke centers based on ferences were specific for stroke, as stroke centers and nondesignated hospitals had
BAC criteria.3 Now, nearly 700 of the similar 30-day all-cause mortality rates among those with gastrointestinal hemor-
5000 acute care hospitals in the United rhage (5.0% vs 5.8%; adjusted mortality difference, 0.3%; 95% CI, 0.5% to 1.0%;
P=.50) or acute myocardial infarction (10.5% vs 12.7%; adjusted mortality differ-
States are Joint Commissioncertified
ence, 0.1%; 95% CI, 0.9% to 1.1%; P=.83).
stroke centers. 4 Some states, such
as New York, Massachusetts, and Conclusion Among patients with acute ischemic stroke, admission to a designated
Florida, have established their own stroke center was associated with modestly lower mortality and more frequent use of
thrombolytic therapy.
designation programs using the BAC
JAMA. 2011;305(4):373-380 www.jama.com
core criteria.
Despite widespread support for the
term mortality.9 Therefore, our goal was versity of Rochesters institutional re-
stroke center concept, there is limited
to evaluate the association between ad- view board, with waiver for informed
empirical evidence demonstrating that
mission to stroke centers for acute is- consent.
admission to a stroke center is associ-
chemic stroke and mortality. We identified 33 090 hospitalized pa-
ated with lower mortality. Prior stud-
tients, 18 years of age or older, with a
ies have largely focused on stroke pro- METHODS
cesses of care, such as treatment Author Affiliations: Duke Clinical Research Institute,
The primary data source was the New
timeline and use of thrombolytic Durham, North Carolina (Drs Xian and Peterson); Uni-
York Statewide Planning and Re- versity of Rochester, Rochester, New York (Drs Hol-
therapy.5-8 There is comparably less in- loway, Noyes, Shah, and Friedman and Mr Chap-
search Cooperative System (SPARCS),
formation on whether better care at pel); Saint Lukes Mid America Heart Institute, Kansas
a comprehensive reporting system that City, Missouri (Dr Chan); and Mayo Clinic, Roches-
stroke centers improves acute or long-
collects patient-level data from every ter, Minnesota (Dr Ting).
Corresponding Author: Ying Xian, MD, PhD, Duke
hospital admission in New York State. Clinical Research Institute, 2400 Pratt St, Durham, NC
For editorial comment see p 408.
This study was approved by the Uni- 27705 (ying.xian@duke.edu).

2011 American Medical Association. All rights reserved. (Reprinted) JAMA, January 26, 2011Vol 305, No. 4 373

Downloaded From: on 12/08/2017


STROKE CENTER HOSPITALIZATION AND MORTALITY

principal diagnosis of acute ischemic Of 244 New York hospitals, 104 cular disease, diabetes mellitus with or
stroke between January 1, 2005, and (42.6%) became state-designated stroke without complications, renal insuffi-
December 31, 2006. An ischemic stroke centers by the end of 2006 (eFigure, ciency, cancer, metastatic carcinoma,
diagnosis was verified through the In- available at http://www.jama.com). Be- liver disease, chronic obstructive pul-
ternational Classification of Diseases, cause some hospitals became stroke monary disease, dementia, connective
Ninth Revision, Clinical Modification centers during the study period, we as- tissue disease disorder, and peptic ul-
(ICD-9-CM) codes 433.x1, 434.x1, and signed stroke center status for each pa- cer disease. These comorbidities were
436. We limited our study sample to tient based on the hospitals designa- used to construct a modified version of
only patients presenting with an ini- tion at the time of admission. the Charlson comorbidity index tai-
tial stroke admission during the study lored for ischemic stroke.12 Hospital
period. We excluded 548 patients Outcome Measures characteristics, such as size and aca-
(1.7%) who lived outside of New York Evaluation of in-hospital mortality may demic affiliation, were obtained from
State and 123 patients (0.4%) with miss- be confounded by different lengths of the NYSDOH and the American Hos-
ing data. To avoid a bias against non- stay between stroke centers and non- pital Association Annual Survey. We
designated hospitals, we also ex- designated hospitals. Moreover, the also determined whether a patient lived
cluded 1472 patients (4.4%) for whom CMS is considering including 30-day in a rural or urban area by applying the
the distance from their home resi- ischemic stroke mortality as one of Rural-Urban Commuting Area Codes
dence to the admitting hospital was its publicly reported measures of hos- classification system to the patients resi-
greater than 20 miles, since these pa- pital quality of care.11 Therefore, as our dential zip code.13
tients would be less likely to receive primary outcome, we examined 30-
thrombolytic therapy. Consistent with day all-cause mortality among those Instrumental Variable
the Centers for Medicare & Medicaid who were and were not admitted to Because it would be impractical to ran-
Services (CMS), all transfer patients a stroke center. As secondary out- domize patients with acute ischemic
were assigned to the transferring hos- comes, we evaluated 1-day, 7-day, and stroke to designated stroke centers or
pital. The final sample included 30 947 1-year all-cause mortality for a sensi- nondesignated hospitals, researchers
patients. tivity analysis. Follow-up ended on the must rely on observational data to as-
date of death or 1 year after the index sess the association of stroke centers
Stroke Center Designation hospitalization through 2007, which- with mortality. However, both mea-
The New York State Stroke Center ever came first. Mortality after dis- sured and unmeasured confounding in-
Designation program is a collabora- charge was determined through the So- herent in observational studies may lead
tion between the New York State De- cial Security Administration Death to selection bias for treatment. For ex-
partment of Health (NYSDOH), the Master File. In addition, we explored ample, EMS personnel may systemati-
American Heart Association (AHA), and how the use of thrombolytic therapy cally transport more severely ill pa-
the New York State Quality Improve- (ICD-9-CM procedure code 99.10 tients to stroke centers. Standard
ment Organization.7,10 Beginning in and/or diagnosis-related group 559), statistical approaches, such as multi-
2004, all New York hospitals were in- discharge to skilled nursing facilities, variate logistic regression or propen-
vited to apply to the NYSDOH for stroke and all-cause readmission within 30 sity score analysis, cannot account for
center designation if they met the BAC days of the index hospital discharge dif- unmeasured confounding because they
criteria. These criteria are organized fered by whether a patient was admit- can only adjust for measured covari-
around 11 aspects of stroke care: acute ted to a designated stroke center. Pa- ates.14,15 One approach is to use instru-
stroke teams, written care protocols, tients who died during the index mental variable analysis (an economet-
emergency medical services (EMS), hospitalization were excluded from the ric method) to help minimize
emergency departments, stroke units, readmission analyses. unmeasured confounding.16,17
neurosurgical services, commitment The key notion behind instrumen-
and support of the medical organiza- Other Study Variables tal variable analysis is that the instru-
tion, neuroimaging services, labora- The SPARCS reporting system pro- ment is highly correlated with the treat-
tory services, outcome and quality im- vided data on patient characteristics, in- ment (stroke center vs nondesignated
provement activities, and continuing cluding sociodemographic informa- hospital) but is otherwise unrelated to
medical education.2 Hospitals were tion (age, sex, race/ethnicity, and observed or unobserved prognostic risk
evaluated for stroke center designa- insurance status) and comorbidities factors so that it does not directly or in-
tion with an initial hospital survey, fol- (differentiated from complications directly affect patient outcomes ex-
lowed by an on-site review and inspec- using a present-on-admission indica- cept through treatment.16,17 This is simi-
tion, to ensure hospital compliance with tor). Comorbidities included prior myo- lar to a randomized controlled trial in
the BAC criteria and preparedness to cardial infarction, congestive heart which the randomization process as-
operate as a stroke center. failure, atrial fibrillation, peripheral vas- signs patients to treatment groups, but
374 JAMA, January 26, 2011Vol 305, No. 4 (Reprinted) 2011 American Medical Association. All rights reserved.

Downloaded From: on 12/08/2017


STROKE CENTER HOSPITALIZATION AND MORTALITY

_ _
the randomization itself is not directly where xsc and x nsc denote the mean we performed subgroup analyses for pa-
associated with outcomes. of a covariate in stroke center and non- tients living in the New York metropoli-
In the case of stroke center admis- designated hospital patients while ssc2 tan area and for those in upstate New
2
sion, we used differential distance, and snsc denote the variance. Standard- York. Third, white individuals and mem-
which is an instrumental variable that ized difference for binary variables was bers of minority race/ethnic groups of-
has been used in prior studies of acute calculated as follows: ten live in different neighborhoods and
myocardial infarction and trauma.16-20 may have systematically used different
Differential distance was calculated as 100 (Psc Pnsc) hospitals. We stratified the analysis by
d=
the difference between the straight- Psc(1 Psc) + Pnsc(1 Pnsc) race and checked whether the effect of
line distance from a patients resi- 2 stroke centers varied by race/ethnicity
dence to the nearest stroke center mi- group. Fourth, to determine if the mor-
nus the straight-line distance from this where Psc and Pnsc denote the preva- tality findings were specific to stroke, we
patients residence to the nearest hos- lence of the binary variable. An abso- compared mortality among patients ad-
pital of any type. The differential dis- lute standardized difference greater than mitted at designated and nondesig-
tance is the additional distance, if any, 10 (approximately equivalent to nated hospitals for 2 other acute life-
beyond the nearest hospital to reach a P .05) indicates significant imbal- threatening conditionsgastrointestinal
stroke center. ance of a baseline covariate, whereas a (GI) hemorrhage and acute myocardial
The choice of differential distance as smaller value supports the balance as- infarction (AMI). Both conditions are
an effective instrumental variable is sumption between groups. quality indicators recommended by the
based on 2 assumptions: first, it is logi- We then assessed whether admis- Agency for Healthcare Research and
cal to assume that a patient trans- sion to a designated stroke center was Quality to assess a hospitals quality of
ported by private vehicle will go to the associated with lower mortality using care.25 If adjusted mortality was lower for
nearest hospital. Importantly, the New an instrumental variable analysis esti- either of these 2 conditions in desig-
York State Stroke Protocol requires EMS mated by a simultaneous 2-equation bi- nated stroke centers, this would sug-
personnel to transport stroke patients variate probit model.23 The first equa- gest that lower stroke mortality would
to the nearest stoke center if the pre- tion estimated the probability of stroke be due to these hospitals overall com-
hospital time is less than 2 hours.21 Pa- center admission as a function of dif- mitment to quality improvement, rather
tients who live close to a stroke center ferential distance and other covari- than to these hospitals implementation
are more likely to be transported to the ates. The second equation assessed the of actions specific to stroke.
stroke center. The second assumption association of stroke center admission All tests were evaluated at a 2-sided sig-
is that patients cannot predict if and with mortality, adjusted for other pa- nificance level of P.05. The analyses
when they will have a stroke, and there- tient and hospital factors. Estimating 2 were performed using SAS 9.2 (SAS In-
fore, they do not choose their resi- equations jointly using a bivariate pro- stitute, Cary, North Carolina) and Stata
dence based on proximity to a given bit approach provides consistent esti- 11 (StataCorp, College Station, Texas).
hospital. Thus, distance to each type mates of the treatment effect.23,24 The
of hospital is highly predictive of instrumental variableadjusted mor- RESULTS
whether the patient was admitted to a tality estimate (technically, the aver- Among 30 947 patients with acute is-
stroke center but is not associated with age marginal effect) can be inter- chemic stroke, 15 297 (49.4%) were ad-
disease characteristics such as stroke se- preted as the mean predicted difference mitted to designated stroke centers
verity. in the probability of death for stroke pa- (n=104) and 15 650 (50.6%) to non-
tients who received treatment at des- designated hospitals. TABLE 1 com-
Statistical Analyses
ignated stroke centers because they pares baseline characteristics of the
Baseline characteristics were com- lived relatively closer to stroke cen- study cohort. Patients admitted to
pared between patients admitted to des- ters vs patients who received treat- stroke centers were more frequently
ignated and nondesignated hospitals ment at nondesignated hospitals be- younger, non-Hispanic black, less likely
using the standardized difference. This cause they lived farther away. to live in a rural area, and more likely
method has been previously used to as- We examined the robustness of our to be admitted at a hospital with more
sess the comparability of study partici- findings in several ways. First, we sought beds and an academic affiliation. Pa-
pants.22 We calculated standardized dif- to determine whether admission to a des- tients admitted to stroke centers were
ference for continuous variables as ignated stroke center was associated with relatively healthier with respect to the
follows: lower 1-day, 7-day, and 1-year all- prevalence of comorbidities, although
cause mortality by repeating the analy- none of the differences were statisti-
_ _
100 (x sc x nsc) ses for these time points. Second, be- cally significant.
d=
s2sc + s2nsc cause the majority of stroke centers are In the assessment of the assump-
2 located in the city of New York (eFigure), tion that the instrumental variable is
2011 American Medical Association. All rights reserved. (Reprinted) JAMA, January 26, 2011Vol 305, No. 4 375

Downloaded From: on 12/08/2017


STROKE CENTER HOSPITALIZATION AND MORTALITY

highly correlated with the variable of ment of the assumption that the instru- tus according to the differential distance
interest, we found from a logistic re- mental variable does not independently to a stroke center.
gression model that the differential dis- affect patient outcomes so that it is not TABLE 2 shows baseline characteris-
tance was highly predictive of whether associated with other potential con- tics relative to whether a patient lived
a patient was admitted to a stroke cen- founders of the outcome, we exam- closer to a stroke center (differential dis-
ter (C statistic = 0.88). In the assess- ined the balance of observed health sta- tance=0 miles or 0 miles). Although
there were small differences in certain
Table 1. Baseline Characteristics of Study Cohort
measures, age and the prevalence of most
comorbidities were more similar than
Designated Nondesignated Absolute
Stroke Center Hospital Standardized were the groups in Table 1, as reflected
(n = 15 297) (n = 15 650) Difference a by diminishing standardized differences.
Age, y 12.5 Despite the similarity in observed health
Mean (SD) 72.3 (14.3) 74.0 (13.8) status, the differential distance groups
Median (IQR) 75 (63-83) 77 (65-84) differed substantially in their probabil-
Male sex, No. (%) 6957 (45.5) 6837 (43.7) 3.6 ity of being admitted to a stroke center.
Race/ethnicity, No. (%) The high correlation between differen-
Non-Hispanic white 8865 (58.0) 11 649 (74.4) 35.4
tial distance and stroke center, as well
Non-Hispanic black 3337 (21.8) 2303 (14.7) 18.5
Hispanic 1507 (9.9) 784 (5.0) 18.5
as the balance in observable health sta-
Other 1588 (10.4) 914 (5.8) 16.7
tus, provide validation of the key instru-
Insurance, No. (%)
mental variable assumptions. However,
Medicare 10 386 (67.9) 11 557 (73.9) 13.1 we did not have data to examine unmea-
Medicaid 1475 (9.6) 953 (6.1) 13.2 sured stroke severity.
Private insurance 2840 (18.6) 2562 (16.4) 5.8
Other insurance 185 (1.2) 220 (1.4) 1.7 Mortality and Other Outcomes
Self-pay 411 (2.7) 358 (2.3) 2.6 Mortality rates among patients admitted
Rural location 219 (1.4) 2556 (16.3) 54.3 to stroke centers and nondesignated hos-
Charlson comorbidity index score 6.6 pitals are summarized in TABLE 3. The
Mean (SD) 1.1 (1.4) 1.2 (1.5) overall30-dayall-causemortalityratewas
Median (IQR) 1 (0-2) 1 (0-2) 10.1%forpatientsadmittedtostrokecen-
Charlson comorbid condition, No. (%) ters and 12.5% for patients admitted to
MI 1051 (6.9) 1185 (7.6) 2.7
nondesignatedhospitals(unadjustedmor-
CHF 2146 (14.0) 2386 (15.3) 3.4
tality difference, 2.4%; P.001). Using
PVD 882 (5.8) 1084 (6.9) 4.7
instrumental variable analysis, we found
Dementia 947 (6.2) 1083 (6.9) 2.9
admission to a designated stroke center
COPD 1936 (12.7) 2503 (16.0) 9.5
was associated with a 2.5% absolute re-
Connective tissue disease 275 (1.8) 372 (2.4) 4.1
duction in 30-day all-cause mortality (ad-
Peptic ulcer disease 126 (0.8) 144 (0.9) 1.0
DM without complications 4557 (29.8) 4322 (27.6) 4.8
justed mortality difference, 2.5%; 95%
DM with complications 442 (2.9) 586 (3.7) 4.8
confidence interval [CI], 3.6% to 1.4%;
Renal disease 1150 (7.5) 1063 (6.8) 2.8
P.001).
Cancer 485 (3.2) 584 (3.7) 3.1
Use of thrombolytic therapy was 4.8%
Metastatic carcinoma 189 (1.2) 258 (1.7) 3.5
(739/15 297) for patients admitted at
Liver disease 116 (0.8) 133 (0.9) 1.0 stroke centers and 1.7% (266/15 650) for
Atrial fibrillation, No. (%) 3046 (19.9) 3485 (22.3) 5.8 patients admitted at nondesignated hos-
Hospital pitals (P.001). Admission to a stroke
Teaching hospital, No. (%) 11 264 (73.6) 7998 (51.1) 47.8 center was associated with increased use
Total beds, No. (SD) 458 (221) 308 (181) 74.4 of thrombolytic therapy (adjusted dif-
Distance to the hospital, median (IQR), 2.7 (1.2-5.6) 3.1 (0.8-6.7) 10.8 ference in thrombolysis use, 2.2%; 95%
mile b
CI, 1.6% to 2.8%; P.001). However,
Differential distance to stroke center, 0 (0-0.2) 11.5 (2.0-46.7) 86.2
median (IQR), mile c further adjustment for use of thrombo-
Abbreviations: CHF, congestive heart failure; COPD, chronic obstructive pulmonary disease; DM, diabetes mellitus; lytic therapy in the instrumental vari-
IQR, interquartile range; MI, myocardial infarction; PVD, peripheral vascular disease.
a Refer to the Methods section for equations used to calculate standardized differences. An absolute standardized able models did not substantially alter
difference 10 (approximately equivalent to P.05) indicates significant imbalance of a baseline covariate. the association of stroke center admis-
b Distance from the patients residence to the admitting hospital (5-digit zip code centroid to 5-digit zip code centroid).
c Distance from the patients residence to the nearest New York State designated stroke center (5-digit zip code cen- sion with lower 30-day mortality (ad-
troid to 5-digit zip code centroid) minus the distance from the patients residence to the nearest hospital of any type. justed mortality difference, 2.7%; 95%
Differential distance equals 0 if the nearest hospital is a designated stroke center.
CI, 3.8% to 1.6%; P.001). Among
376 JAMA, January 26, 2011Vol 305, No. 4 (Reprinted) 2011 American Medical Association. All rights reserved.

Downloaded From: on 12/08/2017


STROKE CENTER HOSPITALIZATION AND MORTALITY

Table 2. Patient Characteristics by Differential Distance to a Designated Stroke Center a


Non-Hispanic White Individuals Non-Hispanic Black Individuals

Differential Differential Absolute Differential Differential Absolute


Distance = 0 Miles Distance 0 Miles Standardized Distance = 0 Miles Distance 0 Miles Standardized
(n = 7516) b (n = 12 998) b Difference c (n = 2829) b (n = 2811) b Difference c
Age, y 0.4 4.2
Mean (SD) 75.8 (13.1) 75.9 (13.1) 66.7 (14.4) 67.3 (14.1)
Median (IQR) 79 (68-85) 79 (69-85) 67 (57-78) 68 (57-78)
Charlson comorbidity index score 5.2 0.9
Mean (SD) 1.1 (1.4) 1.2 (1.5) 1.2 (1.4) 1.2 (1.4)
Median (IQR) 1 (0-2) 1 (0-2) 1 (0-2) 1 (0-2)
Charlson comorbid condition, No. (%)
MI 560 (7.5) 1107 (8.5) 3.9 134 (4.7) 147 (5.2) 2.2
CHF 1178 (15.7) 2023 (15.6) 0.3 352 (12.4) 422 (15.0) 7.5
PVD 508 (6.8) 980 (7.5) 3.0 142 (5.0) 106 (3.8) 6.1
Dementia 519 (6.9) 895 (6.9) 0.1 158 (5.6) 183 (6.5) 3.9
COPD 1060 (14.1) 2211 (17.0) 8.0 281 (9.9) 322 (11.4) 4.9
Connective tissue disease 148 (2.0) 325 (2.5) 3.6 49 (1.7) 53 (1.9) 1.2
Peptic ulcer disease 60 (0.8) 116 (0.9) 1.0 20 (0.7) 22 (0.8) 0.9
DM without complications 1765 (23.5) 3259 (25.1) 3.7 1074 (38.0) 1057 (37.6) 0.7
DM with complications 204 (2.7) 422 (3.3) 3.1 115 (4.1) 119 (4.2) 0.8
Renal disease 504 (6.7) 846 (6.5) 0.8 261 (9.2) 261 (9.3) 0.2
Cancer 299 (4.0) 507 (3.9) 0.4 73 (2.6) 60 (2.1) 2.9
Metastatic carcinoma 121 (1.6) 214 (1.7) 0.3 29 (1.0) 27 (1.0) 0.7
Liver disease 53 (0.7) 93 (0.7) 0.1 18 (0.6) 25 (0.9) 2.9
Atrial fibrillation, No. (%) 2033 (27.1) 3563 (27.4) 0.8 326 (11.5) 341 (12.1) 1.9
Admit to stroke center, No. (%) 6286 (83.6) 2579 (19.8) 165.8 2279 (80.1) 1058 (37.6) 97.0
Abbreviations: CHF, congestive heart failure; COPD, chronic obstructive pulmonary disease; DM, diabetes mellitus; IQR, interquartile range; MI, myocardial infarction; PVD, peripheral
vascular disease.
a Similar results for Hispanic and other race/ethnicity group (not shown in the table).
b Distance from the patients residence to the nearest New York State designated stroke center (5-digit zip code centroid to 5-digit zip code centroid) minus the distance from the patients
residence to the nearest hospital of any type. Differential distance equals 0 if the nearest hospital is a designated stroke center.
c Refer to the Methods section for equations used to calculate standardized differences. An absolute standardized difference 10 (approximately equivalent to P.05) indicates sig-
nificant imbalance of a baseline covariate.

those surviving to hospital discharge, Table 3. Mortality at Designated Stroke Centers and Nondesignated Hospitals
there was no difference in rates of 30- No. (%)
day all-cause readmission (14.8% vs
14.2%; adjusted difference, 1.1%; 95% Designated Stroke Nondesignated Adjusted Mortality P
Center (n = 15 297) Hospital (n = 15 650) Difference (95% CI) a Value
CI, 0.3% to 2.6%; P = .12) and dis-
1d 90 (0.6) 134 (0.9) 0.3 (0.6 to 0.0) .04
charge to a skilled nursing facility (24.6%
7d 665 (4.3) 842 (5.4) 1.3 (2.1 to 0.6) .001
vs 28.5%; adjusted difference, 0.5%; 30 d 1543 (10.1) 1951 (12.5) 2.5 (3.6 to 1.4) .001
95% CI, 2.1% to 1.2%; P=.56). 1y 3412 (22.3) 4067 (26.0) 3.0 (4.4 to 1.5) .001
Abbreviation: CI, confidence interval.
Sensitivity Analyses a Negative values indicate lower mortality at designated stroke center vs nondesignated hospital. Adjusted for age,
sex, race, health insurance status, rural status, 13 Charlson comorbid conditions, atrial fibrillation, hospital teaching
In sensitivity analyses examining status, and total number of hospital beds by using the instrumental variable analysis.
whether the association between stroke
center admission and mortality varied
by race, location, or time points, lower Specificity Analyses pitals (5.0% vs 5.8%; adjusted mortal-
all-cause mortality was observed within In analyses to assess whether the lower ity difference, 0.3%; 95% CI, 0.5%
the first hospital day and at 7 days, and mortality at designated stroke centers to 1.0%; P =.50). Similarly, 30-day all-
the difference was sustained at 1 year was specific to stroke, we examined cause mortality for AMI did not signifi-
after the index hospitalization (Table 3). mortality rates for patients admitted cantly differ between the 2 groups
Subgroup analyses of the New York with GI hemorrhage (n = 39 409) and (10.5% vs 12.7%; adjusted mortality dif-
metropolitan area and upstate New AMI (n =40 024) at stroke centers and ference, 0.1%; 95% CI, 0.9% to 1.1%;
York and stratified analyses by race/ nondesignated hospitals. Thirty-day all- P=.83). For these 2 conditions, there
ethnicity found similar results of lower cause mortality for GI hemorrhage was also were no differences in 1-day or
all-cause mortality at designated stroke comparable for patients admitted to 7-day mortality (TABLE 5). Based on
centers (TABLE 4). stroke centers and nondesignated hos- sample size and observed mortality
2011 American Medical Association. All rights reserved. (Reprinted) JAMA, January 26, 2011Vol 305, No. 4 377

Downloaded From: on 12/08/2017


STROKE CENTER HOSPITALIZATION AND MORTALITY

ited information on patient out-


Table 4. Sensitivity Analysis: 30-Day Mortality at Designated Stroke Centers and
Nondesignated Hospitals comes.5-8 To date, only 1 study in Finland
No./Total No. (%) has reported lower 1-year stroke case fa-
tality associated with stroke centers.26
Designated Nondesignated Adjusted Mortality P Our study extends the findings from this
Stroke Center Hospital Difference (95% CI) a Value
Location b
prior study, as systems of stroke care in
Metropolitan New York 1034/11 120 (9.3) 715/6881 (10.4) 2.0 (3.4 to 0.5) .01 the United States may differ substan-
Upstate New York 509/4177 (12.2) 1236/8769 (14.1) 2.0 (3.8 to 0.3) .02 tially from other national health care sys-
Race/ethnicity c tems (especially those with universal
Non-Hispanic white 1091/8865 (12.3) 1635/11 649 (14.0) 2.5 (3.9 to 1.1) .001 health coverage). We were able to re-
Non-Hispanic black 204/3337 (6.1) 148/2303 (6.4) 2.4 (4.8 to 0.0) .05 port both short-term and 1-year mortal-
Hispanic 117/1507 (7.8) 73/784 (9.3) 5.2 (9.4 to 0.9) .02 ity outcomes and to demonstrate that
Other 131/1588 (8.3) 95/914 (10.4) 3.2 (7.3 to 0.9) .12
lower mortality was specific to stroke at
Abbreviation: CI, confidence interval.
a Negative values indicate lower mortality at designated stroke center vs nondesignated hospital. designated stroke centers.
b Stratified by location.
c Stratified by race/ethnicity group. Geographic patterns of stroke triage
are likely to be nonrandom. Designated
stroke centers and nondesignated hos-
Table 5. Specificity Analysis: Mortality for Gastrointestinal Hemorrhage and Acute pitals may treat different groups of pa-
Myocardial Infarction at Designated Stroke Centers and Nondesignated Hospitals tients in terms of demographics and dis-
Designated Nondesignated Adjusted Mortality P ease severity. For instance, it is possible
Stroke Center Hospital Difference (95% CI) a Value that EMS personnel may systematically
GI hemorrhage (n = 17 481) (n = 21 928)
transport more severely ill patients to
1d 119 (0.7) 178 (0.8) 0.2 (0.1 to 0.5) .24
stroke centers,4 which is consistent with
7d 352 (2.0) 523 (2.4) 0.2 (0.3 to 0.7) .46
our finding of a greater instrument vari-
30 d 871 (5.0) 1261 (5.8) 0.3 (0.5 to 1.0) .50
ableadjusted mortality difference (in ab-
AMI (n = 16 833) (n = 23 191)
solute value) compared with the unad-
1d 363 (2.2) 688 (3.0) 0.3 (0.7 to 0.2) .30
justed difference (eg, 2.5% vs 2.4% for
7d 983 (5.8) 1689 (7.3) 0.5 (1.3 to 0.2) .17
30-day all-cause mortality). Moreover,
30 d 1775 (10.5) 2950 (12.7) 0.1 (0.9 to 1.1) .83
prior studies have reported that stroke
Abbreviations: AMI, acute myocardial infarction; CI, confidence interval; GI, gastrointestinal.
a Negative values indicate lower mortality at designated stroke center vs nondesignated hospital. Adjusted for age, centers are more likely to admit pa-
sex, race, health insurance status, rural status, Charlson comorbidity conditions, hospital teaching status, and total
number of hospital beds by using the instrumental variable analysis. tients with hemorrhagic stroke, which is
associated with higher mortality com-
pared with ischemic stroke.7,8 Indeed, we
rates, a retrospective power analysis in- 1 year after stroke occurrence and was found a similar pattern, in which nearly
dicated that our study had more than independent of patient and hospital 60% (4193/7243) of patients with hem-
90% statistical power to detect a 0.1% characteristics. Importantly, the lower orrhagic stroke in New York were ad-
mortality difference for AMI and 70% mortality at designated stroke centers mitted to a stroke center during our study
power to detect a 0.3% mortality dif- was specific to stroke and was not found period.
ference for GI hemorrhage. for other acute life-threatening condi- In the absence of randomized con-
tions, suggesting that the mortality ben- trolled trials, controlling for treat-
COMMENT efit was related to stroke center desig- ment patterns is often difficult and as-
Reduced mortality and increased use of nation, rather than to overall quality sessments of mortality outcomes may
acute stroke therapies are 2 expected improvement efforts at designated be biased given the presence of treat-
benefits of primary stroke centers.2 Nev- stroke centers. Even though the differ- ment selection. Our analysis sought to
ertheless, limited empirical evidence ences in outcomes between stroke cen- address these concerns by using an in-
supports the benefits of stroke cen- ters and nondesignated hospitals were strumental variable analysis to con-
tersin particular, outcome-based modest, our study suggests that the trol for the selection bias (both mea-
quality measures.9 In this large obser- implementation and establishment of sured and unmeasured) inherent in
vational study, we found that patients a BAC-recommended stroke system of observational studies. After adjusting
admitted to stroke centers were more care was associated with improve- for patient and hospital characteris-
likely to receive thrombolytic therapy ment in some outcomes for patients tics and the potential for unmeasured
and had lower 30-day mortality rates with acute ischemic stroke. selection bias with the instrumental
when compared with patients admit- Previous evaluations of stroke center variable analysis, we found that admis-
ted to nondesignated hospitals. This quality performance have primarily fo- sion to a stroke center for an acute is-
survival benefit was sustained for up to cused on process measures with lim- chemic stroke was associated with a
378 JAMA, January 26, 2011Vol 305, No. 4 (Reprinted) 2011 American Medical Association. All rights reserved.

Downloaded From: on 12/08/2017


STROKE CENTER HOSPITALIZATION AND MORTALITY

2.5% absolute reduction in 30-day all- vival benefit at the end of 1-year follow- neurological and functional status at
cause mortality. up. Collectively, it is likely that the com- discharge, since these measures were
The BAC recommendations serve as binations of these efforts improve the not collected in the SPARCS database,
the cornerstone for the establishment structure and process of stroke care and nor were we able to assess cause-
of primary stroke centers. Previous subsequently contribute to improved specific mortality. Nonetheless, our
studies have shown reduced mortality patient outcomes. study was able to report on the rela-
among patients who were treated by Since stroke center certification is tionship between stroke center admis-
neurologists or who received orga- voluntary, it is possible that hospitals sion and all-cause mortalityan out-
nized care in a stroke care unit.27-29 Al- were already committed to quality im- come that has not been routinely
though we cannot determine which in- provement and would have achieved reported.
dividual components of the BAC criteria these results regardless of designa- Third, while our sensitivity and
for stroke center designation were most tion. A recent evaluation of Joint Com- specificity analyses suggest that lower
important for the lower mortality ob- missioncertified primary stroke cen- mortality associated with stroke cen-
served in this study, it is likely that the ters found that certified hospitals had ter admission may be due to the imple-
BAC criteria cannot be examined as in- better outcomes than noncertified hos- mentation of the BAC criteria as part
dividually isolated units. Rather, the 11 pitals even before the certification pro- of stroke center designation, other qual-
core criteria combined establish the in- gram began.32 Based on our data, we ity-improvement initiatives (eg, the
frastructure and define an approach for cannot definitively establish if the des- AHA Get With The Guidelines
optimizing care for acute ischemic ignation program resulted in reduced Stroke program), economic incen-
stroke. By emphasizing an integrated mortality or if higher-quality hospi- tives from pay for performance, and
and organized system of care with EMS, tals participated in designation. How- public reporting could also influence
hospital emergency departments, acute ever, this concern is mitigated by our stroke care and outcomes.
stroke teams, stroke units, and neuro- specificity analysis, in which we exam- Fourth, our study only included data
imaging services, the BAC criteria ined mortality rates at designated and from New York. The generalizability of
facilitate rapid transportation, evalua- nondesignated hospitals for 2 other life- our findings to other states and agen-
tion, and treatment. Moreover, avail- threatening conditionsGI hemor- cies certifying stroke centers remains
ability of stroke protocols standard- rhage and AMI. Hospitals committed to to be established. Fifth, we were un-
izes acute stroke care and minimizes quality improvement prior to stroke able to assess acute treatments other
protocol violation. These efforts are fur- center designation would be expected than thrombolytic therapy, including
ther enhanced by the BAC criterias em- to demonstrate lower mortality for other the use of life-sustaining interven-
phasis on surveillance of outcomes, medical conditions, as well as for stroke. tions and end-of-life care, which may
quality initiatives, and continuing edu- Nevertheless, the lower mortality ob- affect short-term or intermediate sur-
cational programs. served in this study was specific to vival.33 Sixth, many hospitals were tran-
Of equal importance are the im- stroke, thus suggesting that the lower sitioning to stroke center during our
proved performance measures, which stroke mortality could not be ex- study period. Defining a stroke center
may also affect downstream care and plained simply by the fact that hospi- based on designation status on the ad-
outcomes. Previous studies have found tals who received stroke designation mission date may have underesti-
association between process of care per- were more likely to implement hospital- mated the mortality differences with
formance measures and mortality out- wide quality improvement. stroke center admission. Nonetheless,
comes for patients with cardiovascu- Our study should be interpreted in our conservative approach still dem-
lar disease and stroke.30,31 It is possible the context of the following limita- onstrates a lower risk of death associ-
that improved guideline-based treat- tions. First, the SPARCS database did ated with stroke centers. Finally, the in-
ment, more frequent use of thrombo- not include information on stroke se- strumental variable approach assumes
lytic therapy, enhanced secondary pre- verity. The differences in mortality may that differential distance has no inde-
vention and risk factor management, be due to patient case mix, as opposed pendent effect on patient outcomes ex-
early rehabilitation, and patient edu- to variation in the quality of acute stroke cept through its impact on the likeli-
cation programs also may contribute to care. However, selection bias is more hood of receiving treatment at the
the lower mortality rates among patients likely to be against admission to a stroke designated stroke center. The assump-
treated at stroke centers. However, these center rather than favor stroke cen- tion by its very nature is unproven.
efforts may not have any appreciable ters.4 However, this assumption would gen-
short-term or immediate life-saving Second, we were unable to assess erally be satisfied if a patients resi-
effect, which is consistent with our find- other performance and outcomes, such dence is not associated with stroke se-
ings of minimal mortality difference as eligibility and contradiction of verity, which appears reasonable.
at day 1 and similar readmission rates thrombolytic therapy, thrombolysis- Moreover, differential distance has been
at 30 days compared with greater sur- related hemorrhage, quality of life, and widely and successfully used as an in-
2011 American Medical Association. All rights reserved. (Reprinted) JAMA, January 26, 2011Vol 305, No. 4 379

Downloaded From: on 12/08/2017


STROKE CENTER HOSPITALIZATION AND MORTALITY

strument to control for selection bias Administrative, technical, or material support: Xian, in the preparation, review, or approval of the manu-
Holloway, Friedman. script.
in a variety of clinical settings.16-20 Study supervision: Xian, Holloway, Noyes, Shah, Disclaimer: The content is solely the responsibility of
In conclusion, we found that admis- Friedman. the authors and does not necessarily represent the of-
Conflict of Interest Disclosures: All authors have com- ficial views of the AHA and AHRQ. This study used a
sion to designated stroke centers in New pleted and submitted the ICMJE Form for Disclosure linked SPARCS-SSADMF database. The interpreta-
York State was associated with a mod- of Potential Conflicts of Interest. Dr Holloway re- tion and reporting of these data are the sole respon-
estly lower risk of death for patients ported being a consultant for Milliman Guidelines, re- sibility of the authors. Dr Peterson, a contributing edi-
viewing neurology guidelines. Dr Peterson reported tor for JAMA, was not involved in the editorial review
with an acute ischemic stroke, and the receiving research grants from Schering Plough, Bristol- of or the decision to publish this article.
lower mortality in designated stroke Myers Squibb, Merck/Schering Plough, Sanofi Aventis, Previous Presentation: This study was presented in part
and Saint Judes and being a consultant/advisory board at the 2010 American Heart Association Quality of Care
centers appeared specific to stroke. member to Pfizer and Bayer Corp. No other disclo- and Outcomes Research Conference; Washington, DC;
sures were reported. May 20, 2010.
Author Contributions: Dr Xian had full access to all Funding/Support: This study was funded in part by a Online-Only Material: The eFigure is available at http:
of the data in the study and takes responsibility for predoctoral fellowship 0815772D from the Ameri- //www.jama.com.
the integrity of the data and the accuracy of the data can Heart Association (AHA) Founders Affiliate (Dr Additional Contributions: We acknowledge the in-
analysis. Xian). This study received infrastructure support from sightful contributions of Laine Thomas, PhD; Wen-
Study concept and design: Xian, Holloway, Friedman. the Agency for Healthcare Research and Quality qin Pan, PhD; and Margueritte Cox, MS (Duke Clini-
Acquisition of data: Xian, Friedman. (AHRQ) (U18HS016964). Dr Chan is supported by a cal Research Institute, Durham, North Carolina), for
Analysis and interpretation of data: Xian, Holloway, Career Development Grant Award (K23HL102224) assistance with statistical analyses and geomapping.
Chan, Noyes, Shah, Ting, Chappel, Peterson, Friedman. from the National Heart, Lung, and Blood Institute. We thank Zainab Magdon-Ismail, EdM, MPH (Ameri-
Drafting of the manuscript: Xian, Chan. Dr Shah is supported by the Paul B. Beeson Career De- can Heart Association, Founders Affiliate); Toby I.
Critical revision of the manuscript for important in- velopment Award (National Institute on Aging Gropen, MD (Long Island College Hospital, New York);
tellectual content: Xian, Holloway, Chan, Noyes, Shah, 1K23AG028942). Nancy R. Barhydt, DrPH, RN; and Anna Colello, Esq
Ting, Chappel, Peterson, Friedman. Role of the Sponsors: The funding organizations had (New York State Department of Health), for assis-
Statistical analysis: Xian. no role in the design and conduct of the study; in the tance with data. None were compensated for their con-
Obtained funding: Xian, Holloway, Friedman. collection, analysis, and interpretation of the data; or tributions.

REFERENCES
1. Lloyd-Jones D, Adams RJ, Brown TM, et al; Ameri- man Services. http://www.hospitalcompare.hhs http://www.nyhealth.gov/nysdoh/ems/protocolsnew
can Heart Association Statistics Committee and Stroke .gov/. Accessed May 21, 2010. .htm. Accessed April 11, 2010.
Statistics Subcommittee. Heart disease and stroke sta- 12. Goldstein LB, Samsa GP, Matchar DB, Horner RD. 22. Austin PC. Using the standardized difference to
tistics: 2010 update: a report from the American Heart Charlson Index comorbidity adjustment for ischemic compare the prevalence of a binary variable between
Association [published correction appears in Circula- stroke outcome studies. Stroke. 2004;35(8):1941- two groups in observational research. Comm Statist
tion. 2010;121(12):e260]. Circulation. 2010;121 1945. Simulation Comput. 2009;38(6):1228-1234.
(7):e46-e215. 13. Measuring rurality: rural-urban commuting area 23. Greene WH. Econometric Analysis. 5th ed. Up-
2. Alberts MJ, Hademenos G, Latchaw RE, et al. Rec- codes. US Dept of Agriculture. http://www.ers.usda per Saddle River, NJ: Prentice Hall; 2003.
ommendations for the establishment of primary stroke .gov/briefing/Rurality/RuralUrbanCommutingAreas/. 24. Bhattacharya J, Goldman D, McCaffrey D. Esti-
centers: Brain Attack Coalition. JAMA. 2000;283 Accessed June 1, 2010. mating probit models with self-selected treatments.
(23):3102-3109. 14. Rosenbaum PR, Rubin DB. Reducing bias in ob- Stat Med. 2006;25(3):389-413.
3. Joint Commission primary stroke center certification. servational studies using subclassification on the pro- 25. Inpatient quality indicators overview. Agency for
http://www.jointcommission.org/certification pensity score. J Am Stat Assoc. 1984;79(387): Healthcare Research and Quality quality indicators:
/primary_stroke_centers.aspx. Accessed March 7, 516-524. February 2006. http://www.qualityindicators.ahrq
2010. 15. Stukel TA, Fisher ES, Wennberg DE, Alter DA, .gov/iqi_overview.htm. Accessed April 11, 2010.
4. Alberts MJ. Stroke centers: proof of concept and Gottlieb DJ, Vermeulen MJ. Analysis of obser- 26. Meretoja A, Roine RO, Kaste M, et al. Effective-
the concept of proof. Stroke. 2010;41(6):1100- vational studies in the presence of treatment selec- ness of primary and comprehensive stroke centers:
1101. tion bias: effects of invasive cardiac management on PERFECT stroke: a nationwide observational study from
5. Lattimore SU, Chalela J, Davis L, et al; NINDS Sub- AMI survival using propensity score and instrumen- Finland. Stroke. 2010;41(6):1102-1107.
urban Hospital Stroke Center. Impact of establishing tal variable methods. JAMA. 2007;297(3):278- 27. Stroke Unit Trialists Collaboration. Organised in-
a primary stroke center at a community hospital on 285. patient (stroke unit) care for stroke. Cochrane Data-
the use of thrombolytic therapy: the NINDS Subur- 16. McClellan M, McNeil BJ, Newhouse JP. Does base Syst Rev. 2007;(4):CD000197.
ban Hospital Stroke Center experience. Stroke. 2003; more intensive treatment of acute myocardial infarc- 28. Mitchell JB, Ballard DJ, Whisnant JP, Ammering
34(6):e55-e57. tion in the elderly reduce mortality? analysis using in- CJ, Samsa GP, Matchar DB. What role do neurolo-
6. Douglas VC, Tong DC, Gillum LA, et al. Do the Brain strumental variables. JAMA. 1994;272(11):859- gists play in determining the costs and outcomes of
Attack Coalitions criteria for stroke centers improve 866. stroke patients? Stroke. 1996;27(11):1937-1943.
care for ischemic stroke? Neurology. 2005;64(3): 17. Newhouse JP, McClellan M. Econometrics in out- 29. Goldstein LB, Matchar DB, Hoff-Lindquist J, Samsa
422-427. comes research: the use of instrumental variables. Annu GP, Horner RD. VA Stroke Study: neurologist care is
7. Gropen TI, Gagliano PJ, Blake CA, et al; NYSDOH Rev Public Health. 1998;19(1):17-34. associated with increased testing but improved
Stroke Center Designation Project Workgroup. Qual- 18. Beck CA, Penrod J, Gyorkos TW, Shapiro S, Pilote outcomes. Neurology. 2003;61(6):792-796.
ity improvement in acute stroke: the New York State L. Does aggressive care following acute myo- 30. Peterson ED, Roe MT, Mulgund J, et al. Associa-
Stroke Center Designation Project. Neurology. 2006; cardial infarction reduce mortality? analysis with in- tion between hospital process performance and out-
67(1):88-93. strumental variables to compare effectiveness in Ca- comes among patients with acute coronary syndromes.
8. Stradling D, Yu W, Langdorf ML, et al. Stroke care nadian and United States patient populations. Health JAMA. 2006;295(16):1912-1920.
delivery before vs after JCAHO stroke center Serv Res. 2003;38(6 pt 1):1423-1440. 31. Bravata DM, Wells CK, Lo AC, et al. Processes of
certification. Neurology. 2007;68(6):469-470. 19. McConnell KJ, Newgard CD, Mullins RJ, Arthur care associated with acute stroke outcomes. Arch In-
9. Fonarow GC, Gregory T, Driskill M, et al. Hospital M, Hedges JR. Mortality benefit of transfer to level I tern Med. 2010;170(9):804-810.
certification for optimizing cardiovascular disease and versus level II trauma centers for head-injured patients. 32. Lichtman JH, Allen NB, Wang Y, Watanabe E, Jones
stroke quality of care and outcomes. Circulation. 2010; Health Serv Res. 2005;40(2):435-457. SB, Goldstein LB. Stroke patient outcomes in US hos-
122(23):2459-2469. 20. Pracht EE, Tepas JJ III, Celso BG, Langland-Orban pitals before the start of the Joint Commission Pri-
10. New York State Department of Health desig- B, Flint L. Survival advantage associated with treat- mary Stroke Center certification program. Stroke. 2009;
nated stroke centers. http://www.health.state.ny.us ment of injury at designated trauma centers: a bivar- 40(11):3574-3579.
/nysdoh/ems/stroke/stroke.htm. Accessed March 7, iate probit model with instrumental variables. Med Care 33. Holloway RG, Quill TE. Mortality as a measure
2010. Res Rev. 2007;64(1):83-97. of quality: implications for palliative and end-of-life
11. Hospital Compare. US Dept of Health and Hu- 21. New York State EMT/AEMT BLS stroke protocols. care. JAMA. 2007;298(7):802-804.

380 JAMA, January 26, 2011Vol 305, No. 4 (Reprinted) 2011 American Medical Association. All rights reserved.

Downloaded From: on 12/08/2017