Outcomes for Patients With In-Hospital Stroke: A Multicenter

Study From the Australian Stroke Clinical Registry (AuSCR)

Dominique A. Cadilhac, PhD,*,†,1 Monique F. Kilkenny, PhD,*,†,1

Natasha A. Lannin, PhD,‡,§ Helen M. Dewey, PhD,║ Christopher R. Levi, MBBS,{
Kelvin Hill, BAppSc,** Brenda Grabsch, BSW,† Rohan Grimley, MBBS,*,††
David Blacker, MBBS,‡‡ Amanda G. Thrift, PhD,* Sandy Middleton, PhD,§§
Craig S. Anderson, PhD,║║,{{ and Geoffrey A. Donnan, MD†,
On behalf of the Australian Stroke Clinical Registry Consortium

Background: The quality of care and outcomes for people who experience stroke whilst
in hospital for another condition has not been previously studied in Australia. Aims: To
explore differences in long-term outcomes among patients with in-hospital events
treated in stroke units (SUs) compared to those managed in other hospital wards.
Methods: Forty-five hospitals participating in the Australian Stroke Clinical Registry
between January 2010 and December 2014 contributed data. Survival of all patients
with in-hospital stroke to 180 days after stroke and health-related quality of life, using
EQ-5D-3L among 73% eligible, were compared using multilevel, multivariable regres-
sion models. Models were adjusted for age, sex, index of relative socioeconomic disad-
vantage, ability to walk, stroke type, transfer from another hospital, and history of
stroke. Results: Among 20,786 stroke events, 1182 (5.1%) occurred in-hospital (median
age 77 years, 49% male). Patients with in-hospital stroke treated in SUs died less often
within 30 days (Hazard Ratio 0.56; 95% CI 0.39-0.81) than those not admitted to SUs.
Survivors reported similar health-related quality of life between 90 and 180 days com-
pared to those treated in other wards (coefficient = 0.01, 95% CI 0.06-0.09, P = .78).
Patients managed in SUs more often received recommended management (e.g. swal-
lowing screening). Conclusion: The benefits of SU care may extend to patients
experiencing in-hospital stroke. Validation, including accounting for potential residual
confounding factors, is required.
Key Words: Stroke—stroke unit—stroke management—hospitals—outcome
© 2019 Elsevier Inc. All rights reserved.

From the *Stroke and Ageing Research, Department of Medicine, School of Clinical Sciences at Monash Health, Monash University, Clayton,
Australia; †Stroke Division, the Florey Institute of Neuroscience and Mental Health, University of Melbourne, Heidelberg, Victoria, Australia;
‡School of Allied Health (Occupational Therapy), La Trobe University, Heidelberg, Australia; §Occupational Therapy Department, Alfred Health,
Prahran, Australia; ║Eastern Health Clinical School, Faculty of Medicine, Nursing and Health Science, Monash University, Box Hill, Australia;
{Hunter Medical Research Institute, Newcastle, Australia; **Stroke Foundation, Melbourne, Australia; ††Sunshine Coast Clinical School, The Uni-
versity of Queensland, Birtinya, Australia; ‡‡Sir Charles Gairdner Hospital, Perth, Australia; §§Nursing Research Institute, St Vincent's Health
Australia (Sydney) and Australian Catholic University, Sydney, Australia; ║║George Institute for Global Health, University of Sydney, Sydney,
Australia; and {{Neurology Department, Royal Prince Alfred Hospital, Sydney, Australia.
Received May 5, 2018; revision received January 24, 2019; accepted January 25, 2019.
Grant support: The following authors received fellowship grants from the National Health and Medical Research Council (NHMRC): DAC
(1063761 co-funded Heart Foundation), CRL (1043913), MFK (1109426), AGT (1042600), and CSA (1081356). AuSCR was supported by grants from
the NHMRC (1034415), Monash University, Queensland Health, Stroke Foundation, Allergan, Ipsen, and Boehringer Ingelheim.
Address correspondence to Dominique Cadilhac, PhD, School of Clinical Sciences at Monash Health, Monash University, Level 3 Hudson Insti-
tute Building, 27-31 Wright Street Clayton VIC 3168, Clayton, Australia. E-mail: dominique.cadilhac@monash.edu.
1
Contributed equally to this work.
1052-3057/$ - see front matter
© 2019 Elsevier Inc. All rights reserved.
https://doi.org/10.1016/j.jstrokecerebrovasdis.2019.01.026

1302 Journal of Stroke and Cerebrovascular Diseases, Vol. 28, No. 5 (May), 2019: pp 13021310
JMCTD PATIENTS HAVE INCREASED IMT 1303

Introduction indicators, and outcomes (see below).9 Among individu-

als with multiple episodes of care registered within
Given the high prevalence of cardiovascular risk factors
90-180 days of an index event, only the first episode was
among hospitalized patients, stroke can occur in a broad
followed up. Annual linkage of AuSCR data to the
range of patients while they are in hospital receiving man-
National Death Index ensures survival status is known
agement for another condition including major surgery.1
for all registrants.
Stroke unit (SU) care, that includes co-located patients
within a hospital receiving specialized, interdisciplinary
Patient Characteristics
management can significantly reduce death and disabil-
ity.2 Therefore, all patients with stroke should be treated Patient characteristics included age, sex, country of
in SUs irrespective of age, stroke severity or type.3 birth, and language spoken at home. The variable “able to
Approximately, 4%-15% of patients experience a stroke walk unaided” is used for assessing stroke severity in
while in hospital for another condition (commonly termed AuSCR. It is a global measure of disability that is nor-
“in-hospital” stroke).4-6 Compared with patients present- mally assessed at the time of hospital admission. How-
ing from the community, patients with in-hospital ische- ever, for patients who experience their stroke when
mic strokes have worse outcomes and receive less already in hospital, this is assessed within the first
evidence-based care,7 with very few treated in SUs.8 There 24 hours of this event. This simple measure has been vali-
may be several justifiable reasons why patients with in- dated by Counsell and colleagues (relative risk for 30-day
hospital stroke are not managed in SUs. However, once survival 1.63 95% CI 1.15-2.31)11; and we have found this
their primary condition has stabilized, management in the variable to be a reliable predictor of independence at time
SU to support rehabilitation from the effects of stroke is of hospital discharge and for survival.12,13 Place of resi-
important for reducing disability,7 as well as ensuring dence was used to derive the “index of relative socio-eco-
commencement of secondary prevention therapies. nomic advantage and disadvantage” (IRSAD) score for
To our knowledge, evidence on the long-term outcome each patient. The scores were divided into quintiles, with
of patients with in-hospital stroke is lacking. We aimed to quintile 1 representing the most disadvantaged patients.
explore whether patients with in-hospital events had bet- Clinician classification of stroke type (ischemic, ICH,
ter survival and health-related quality of life (HR-QoL) undetermined, as well as whether the event occurred in
within 180 days if managed in an SU, compared to other hospital) and the ICD-10 discharge codes are both col-
wards, and to describe any potential explanatory factors lected in AuSCR.9 For our analyses, if the clinical stroke
for the associations observed. type was recorded as undetermined it was recoded to
ischemic or ICH if the discharge ICD-10 code was not I64.
Methods Discharge ICD-10 codes were also used to explore the
principle reason for hospital admission among patients
This was an observational study using data from the
experiencing in-hospital stroke, and were also used to
Australian Stroke Clinical Registry (AuSCR) obtained
develop the Charlson Comorbidity Index14 for patients
from 2010-2014. The AuSCR design has been detailed pre-
where we had information from administrative hospital
viously. In brief, the AuSCR is a voluntary, clinical quality
records.
registry used in Australian hospitals to capture data on
consecutive patients admitted with stroke or transient
ischemic attack or events that occur in hospital.9 The Process of Care Indicators
AuSCR commenced in 2009 and was initially funded via All patients with stroke should receive important pro-
research and educational grants, but is now mainly sup- cesses of care for which they are eligible. In AuSCR, four
ported by state government grants (since 2012). nationally agreed processes of care are collected: admitted
The majority of hospitals that contribute data to AuSCR to an SU; received intravenous thrombolysis (for ischemic
are public and located in the eastern (most densely popu- stroke); discharged on an antihypertensive agent; and
lated) states.10 Cases are entered prospectively in the received a discharge care plan.9 For a subset of patients
AuSCR based on clinical diagnosis of stroke during the from the state of Queensland, four additional processes
admission. Case ascertainment is checked annually using have been collected since 2012 and are also reported here:
International Classification of Diseases (ICD)-10 discharge time to first mobilization, dysphagia screen, aspirin
codes obtained from the hospital administrative system within 48 hours and being discharged on antiplatelets or
and compared to the cases entered in the registry at each antithrombotics if an ischemic event.
hospital (see Supplemental methods).
We included all episodes of ischemic, intracerebral
Outcome Assessment
hemorrhage (ICH), and undetermined stroke for the anal-
yses. Transient ischemic attack was excluded. The AuSCR We report in-hospital deaths within 7-days, 30-days,
includes a minimum dataset of personal information 90-days, and 180-days, discharge to home, and length of
(e.g. name, address), clinical data, process of care stay. For the subset of eligible patients who were
1304 D.A. CADILHAC ET AL.

followed-up (i.e. within 180 days of admission, first- undertaken for all our logistic regression models. Where
registered event),15 HR-QoL was measured using the relevant, we have reported pseudo R2 for each model to
EuroQol (EQ-5D-3L) questionnaire,16 and readmissions, provide an indication of the potential importance of varia-
recurrent strokes, and the modified Rankin Score (from bles not included in the models on the outcomes of inter-
July 2014) at follow-up were also obtained via self-report. est, as well as E-values to assess the potential contribution
Individual responses for the EQ5D were converted into a of unmeasured confounding.19 The analyses were
utility score using the method reported by Viney and col- performed using STATA 12.1 (Statcorp, College Station,
leagues.17 This approach was used since it is more respon- USA, 2014).
sive for survivors of stroke than using the EQ5D Visual
Analogue Scale score, and incorporates deaths and health Ethics and Patient Consent
states considered worse than death (e.g. utility score
Ethics approvals were obtained from all participating
below zero) as part of the summary measure.18
hospitals; Monash University (CF11/3537-2011001884);
and the Australian Institute of Health and Welfare. Con-
Statistical Analysis sistent with the approach recommended for clinical qual-
ity disease registries to reduce selection bias,20 AuSCR
For patient demographic variables, such as age, sex,
uses an “opt-out” process plus there is a waiver of consent
and country of birth we only used valid data. Consistent
for patients who die in hospital.9
with standard quality of care monitoring practice, when
data were missing for process of care indicators we
Results
assumed the response was negative to avoid overestima-
tion. Descriptive analyses were used to initially compare Between 2010 and 2014, 20745 episodes of care were
the characteristics of patients by the subgroups of interest. registered in AuSCR for 19,642 individuals from 45 hospi-
Pearson x2 tests were used for categorical variables, tals. Among these episodes, 1182 stroke events (overall
and the Kruskal Wallis test was used for the continuous 5.1% of total sample) occurred while the patient was in
variables. hospital for another condition. Compared with being
Models were adjusted for age, sex, IRSAD, ability to admitted from the community, patients experiencing in-
walk, type of stroke, transfer from another hospital and hospital events were older (median age 77 versus 76 years,
documented history of stroke. We included cases trans- P < .001); more likely to be female (49% versus 46%
ferred as an independent variable since this group P = .06); more often born in Australia or identify as having
differed from nontransferred patients: they were more an indigenous background and more likely to be able to
often male, younger, experienced more severe strokes and walk within 24 hours of stroke onset (38% versus 34%,
intracerebral hemorrhages, and were more common in P = .001). Type of stroke was similar between groups (see
the in-hospital stroke group (14% versus 4% if not trans- Supplementary Table e-1). Overall, fewer patients with in-
ferred). Regression models were used to investigate differ- hospital events compared with community-onset events
ences in outcomes based on whether or not patients with accessed an SU (63% versus community-onset 81%;
in-hospital stroke received management in an SU. P < .001) (see Supplementary Table e-2). Other differences
The primary analysis was run on patients with com- in receiving processes of care, outcomes and discharge
plete (nonimputed) data. Sensitivity analyses were under- diagnoses between patients with in-hospital events and
taken using a dataset whereby some variables with >1% community-onset events are available in supplementary
missing and unknown responses were recoded as “no” Tables e-2 to e-5. To our knowledge, none of the patients
for documented history of previous stroke, or “yes” for were treated with mechanical intubation or were treated
“able to walk.” In further sensitivity analyses we also with mechanical thrombectomy since this treatment was
included the Charlson Comorbidity Index in our models. still being assessed in clinical trials and this information
Cox proportional hazards regression analysis was used to was not captured.
calculate differences in the risk of mortality within and up Overall, 4389 patients died within 180-days of the
to 7, 30, 90, and 180 days. The analyses using the 30, 90, index event, and this was more common among those
and 180 day time points excluded those who had died at with a stroke that occurred while in hospital for another
the previous time point (for example, 30-day analysis condition (P < .001). Compared with being admitted
excluded deaths at 7 days, and 90 day analysis excluded from the community, patients experiencing in-hospital
deaths at 30 days). Differences between HR-QoL utility events were more likely to die within 180 days of
scores were assessed using median regression to account stroke (risk adjusted HR: 1.68: 95% CI: 1.49, 1.89; see
for the J-shape distribution. In each model, level or cluster Supplementary Table e-4).
was defined as hospital to account for potential residual Among the 1182 registered in-hospital events (median
confounding. All P values were two-sided with P < .05 age 77 years, 49% female) from 43 hospitals, demo-
considered significant for all analyses. Goodness-of-fit graphics, comorbidities and clinical characteristics were
tests including the Pearson Correlation x2 were similar between those treated and not treated in an SU
JMCTD PATIENTS HAVE INCREASED IMT 1305

(Table 1). Fewer patients with in-hospital stroke treated in greater access to evidence-based care as highlighted in the
an SU had an ICH. Patients with an in-hospital event who present study, and complementary evidence from studies
were treated in an SU more often received a range of pro- in the United States and Ireland.4,8 There is strong
cesses of care (e.g. mobilized during admission: 79% SU, evidence for a net benefit of thrombolysis for combined
52% other wards and intravenous thrombolysis: 14% SU, death and dependency, particularly for thrombolysis
6% other wards; Fig 1) and were more often discharged to administered within 3 hours.22 In the current study,
rehabilitation (Table 1) compared to those not treated in patients managed in an SU were more likely to receive
an SU. acute care interventions such as thrombolysis and aspirin,
Compared with patients not treated in an SU, there and were more often mobilized, received swallow screen-
were fewer deaths up to 180-days for patients with in-hos- ing, and prescribed prevention medications at discharge.
pital events treated in an SU. In multivariable analyses, These interventions are most often associated with better
treatment in an SU was associated with a reduced hazard outcomes.23-25
of death at 7 and 30 days after admission when compared Strengths of our study include prospective registration
to patients not treated in an SU (Table 2). Amongst those of a large number of consecutive patients admitted to a
who survived to 30 days after admission, there were no variety of metropolitan and regional areas of Australia.
further differences in deaths to 90 days after admission We also explicitly recorded management in SUs in all
between those treated in an SU and those treated in an individuals. In other studies of in-hospital stroke, there
alternate ward (HR 0.81, 95% CI 0.49-1.32). has been a reliance on optional reporting of these cases or
Among those who were eligible for follow-up via sur- a derived location of stroke from administrative data.4
vey (61% of patients with in-hospital events; median fol- Importantly, the proportion of missing data for in-hospi-
low-up time: 101 days, Interquartile range: 97, 107 days), tal stroke in our study was less than 2%. Furthermore,
self-reported readmission to hospital or recurrent stroke random auditing of AuSCR hospital data demonstrated
between patients with in-hospital events treated in an SU less than 1% discrepancy between auditors on the record-
and patients treated in other wards was similar (Table 3). ing of in-hospital strokes.10 We had >80% power for our
HR-QoL between 90 and 180 days among patients with main outcome analyses with 1182 patients with in-hospi-
in-hospital events treated and not treated in an SU was tal stroke (see Supplemental methods).
similar. In a small subanalysis, patients with in-hospital Limitations of our study include the potential for selec-
events were more likely to report being independent tion or referral bias. That is, less complex patients or those
(mRS 0-2) at 3-6 months follow-up when treated in an SU without competing conditions may have been more often
(21/39; 54%) than those not treated in an SU (1/8; 13%; transferred to an SU. We also acknowledge that we had
P = .03). Sensitivity analyses indicated that the results limited or incomplete data on comorbidities. In our study,
derived from our models were robust for example deaths patients treated in and not in an SU had similar comorbid-
up to 7 days: Original model 0.47 (95%CI 0.28, 0.77) with ity profiles, and in a subset of the patients both groups
Charlson Index as a covariate HR: 0.39 (95%CI 0.19, 0.79). had a median Charlson comorbidity index of 3. Our
results indicate a larger survival benefit than what was
observed in clinical trials of SU versus general wards2 and
Discussion
so requires validation. Important predictors of stroke
To our knowledge, this is the first comparison of long- mortality include age, stroke severity, and pre-existing
term outcomes, in terms of mortality (within 180 days) conditions such as atrial fibrillation and diabetes.26 In a
and HR-QoL (median 101 days), between patients with recent paper using the AuSCR data13 we highlighted the
in-hospital stroke events treated in an SU or in other importance of using appropriate risk adjustment variables
wards. In this large Australian sample, patients with in- and methods for comparing mortality outcomes for
hospital strokes treated in other wards experienced stroke, especially the need to account for stroke severity.
greater early mortality (within 30 days) than patients Since the AuSCR contains a pragmatic minimum dataset
treated in an SU. Amongst those who survived to 30 days of variables we were unable to adjust for additional pre-
after admission, there was no evidence that treatment in existing co-morbidities and we were unable to describe
an SU affected survival to 90 or 180 days, although the dif- the type of ward where the patient was managed for
ference in hazard ratio remained when deaths in the first patients not admitted to an SU. We acknowledge that
7 days were included (data not shown). In those who having a limited number of variables to include in our
were followed up between 90 and 180 days after admis- models could result in important predictors of outcome
sion, overall HR-QoL was similar between patients being omitted, while too many variables may lead to
treated in an SU and those treated in an alternate ward. overfitting (where false-positive predictors are errone-
Our study also provides evidence that SU care is associ- ously included in the model), under-fitting, or paradoxical
ated with increased survival for patients with stroke irre- fitting (where a variable with a positive association with
spective of whether the onset was in the community or in the outcome is found to have a negative association).11
hospital.21 This is most likely explained by evidence of Consistent with observational designs, unmeasured
1306 D.A. CADILHAC ET AL.

Table 1. Demographic and clinical characteristics of patients with in-hospital events according to treatment in a stroke unit

Characteristics SU Non-SU P value

N = 750 N = 432
n (%) n (%)
Female 352/749 (47) 220/428 (51) .15
Age (in years)
Less than 65 145/750 (19) 118/428 (28) .003
65-74 162/750 (22) 91/428 (21)
75-84 267/750 (36) 118/428 (28)
85+ 176/750 (24) 101/428 (24)
Median age in years (Q1, Q3) 78 (69, 85) 76 (64, 85) .024
Australian born 520/750 (69) 299/432 (69) .97
Indigenous background* 14/741 (2) 13/420 (3) .19
Previous stroke 197/703 (28) 87/393 (22) .03
Index of relative socio-economic advantage and disadvantage
Quintile 1 (most disadvantaged) 139/742 (19) 92/424 (22) .007
Quintile 2 178/742 (24) 84/424 (20)
Quintile 3 81/742 (11) 73/424 (17)
Quintile 4 141/742 (19) 81/424 (19)
Quintile 5 (most advantaged) 203/742 (27) 94/424 (22)
Type of stroke
Intracerebral hemorrhage 98/750 (13) 86/432 (20) <.001
Ischemic 626/750 (84) 292/432 (68)
Undetermined stroke 26/750 (4) 54/432 (13)
Stroke severity
Able to walky 253/697 (36) 161/383 (42) .06
Cases with additional administrative data from 2010-2013# N = 262 N = 192
Charlson Comorbidity Index (median) 3 (3, 4) 3 (2, 5) .02
Peripheral vascular disease 4 (2) 13 (5) .11
Congestive heart failure/myocardial infarct 51 (27) 59 (23) .32
Renal or Liver disease 29 (15) 36 (14) .68
Dementia 10 (5) 16 (6) .68
Chronic pulmonary disease 13 (7) 17 (6) .91
Cancer 24 (13) 22 (8) .15
Transfer from another hospital 284/747 (38) 135/428 (32) .03
Median length of stay (days) 11 (6,20) 13 (6,22) .26
Died In Hospital 96/730 (13) 109/417 (26) <.001
Discharge destination
Home 135/634 (21) 98/308 (32) <.001
Rehabilitation 292/634 (46) 106/308 (34) .001
Aged care 44/634 (7) 25/308 (8) .52
SU, Stroke unit.
*Identifies as Aboriginal and/or Torres Strait islander; Q1: 25th percentile; Q3: 75th percentile;
†
within first 24 hours of stroke onset;
#
patient-level data linked for AuSCR registrants with hospital administrative records for the index admission from New South
Wales, Victoria, Queensland and Western Australia.

confounding is a limitation. The observed estimates could likely to achieve independence following rehabilitation.
be explained away by an unmeasured confounder associ- However, the benefits of stroke unit care appear to be inde-
ated with both SU care and the outcome if the confounder pendent of the co-morbidity profile of patients, and in these
was of a magnitude equivalent to the E-value estimates circumstances the involvement of the stroke team is still
generated (from 1.0 to 11.3) above and beyond the mea- required.27 It may also be that those transferred to the SU
sured confounding factors, but a weaker confounding fac- were less often for palliative care where active intervention
tor could not do so.19 is minimized, but this remains unclear until our data can
It is plausible that patients with multiple conditions (e.g. be comprehensively linked with administrative records. In
renal failure) could justifiably not be transferred into the complementary cross-sectional data from the 2015 acute
SU as other acute conditions may take priority over stroke national audit (»40 cases per hospital in 112 hospitals),28
management, or because patients may be perceived as less 15% of patients with in-hospital events (23/151) were
JMCTD PATIENTS HAVE INCREASED IMT 1307

Discharged on anthromboc medicaon, if an ischemic

stroke*+

Aspirin within 48 hours, if an ischemic stroke*+

Swallow screen or assessment*+

Mobilized during admission*+

Discharged to the community with a care plan

Discharged on an anhypertensive medicaon*

Received thrombolysis, if an ischemic stroke*

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Other wards Stroke unit
Figure 1. Processes of care received by patients with in-hospital stroke managed, or not managed, in stroke units. Legend: *P value < .05; +variables only col-
lected in hospitals located in Queensland.

documented as receiving palliative care at some stage dur- other stroke populations as reliable for predicting stroke
ing their acute hospital stay. Only three of these 23 patients outcome.11,12 In recent validation work by Sim and col-
(13%) receiving palliative care for in hospital stroke were leagues the use of simple variables (including ability to
managed in the SU (unpublished data, Stroke Foundation walk) performed similarly well to a model that included
2015). These data provide some preliminary evidence that the NIHSS and age.29 We have also shown this variable to
palliative care may be more often undertaken outside the reliably account for differences in patient case-mix when
stroke unit among these types of patients. using the registry data to compare hospital mortality
“Ability to walk” may be considered a crude measure rates.13 Since 2015 we have also collected the NIHSS in
of stroke severity when compared with other popular and the registry, but missing data remains an issue. To
validated methods such as the National Institutes of account for potential differences in the severity of comor-
Health Stroke Scale, a direct measure of neurological bidities known to affect this type of stroke,7 in sensitivity
impairment. However, it is reliable to collect, does not analyses we included the Charlson Comorbidity Index in
require certification training, and has been validated in our models and the results were similar.

Table 2. Survival analysis of patients with in-hospital stroke managed and not managed in a stroke unit

SU Non-SU Model*
Time to death n (%) n (%) P value HR 95% CI P value E value
Up to 7 days 38/750 (5) 51/432 (12) <.001 0.47 0.28, 0.77 .003 2.753
8 to 30 days 115/750 (15) 121/432 (28) <.001 0.56 0.39, 0.81 .002 2.350
31 to 90 days 169/750 (23) 151/432 (35) <.001 0.81 0.49, 1.32 .396 1.584
91 to 180 days 206/750 (28) 163/432 (38) <.001 1.11 0.57, 2.16 .757 1.359
CI, confidence interval; HR, hazard ratio; SU, stroke unit (reference category).
*Model adjusted for age, sex, index of relative socioeconomic disadvantage, ability to walk within first 24 hours of stroke onset,
type of stroke, transfer from another hospital and documented history of stroke. 30-day regression excluded deaths up to
7 days, 90-day regression excluded deaths up to 30 days, 180-day regression excluded deaths up to 90 days. The lowest possi-
ble E-value is 1,19 but there are no specific guidelines on the range of E-values. If an E-value is deemed small (residual con-
founding is a threat) and where it is larger (residual confounding may not be a problem).30 Therefore, our E-values indicate
the minimum strength of association between SU care and the outcome that would be required by an unmeasured confounder
to fully explain away each of the SU care and outcome models listed.
 1308
Table 3. Outcomes at 90-180 day follow-up of patients with in-hospital stroke managed and not managed in a stroke unit

SU Non-SU Model*
Follow-up n (%) n (%) P value OR 95% CI P value E-value Pseudo-R2
Readmission 76/305 (25) 30/126 (24) .81 0.79 0.45, 1.37 .40 1.500 0.040
Recurrent stroke 13/305 (4) 3/126 (2) .35 5.92 0.74, 47.0 .09 11.317 0.077
EQ-5D-dimensions
Mobility 155/302 (51) 59/124 (48) .48 1.24 0.78, 1.99 .36 1.469 0.022
Self-care 156/304 (51) 57/125 (46) .28 1.16 0.71, 1.92 .55 1.365 0.048
Usual activities 226/304 (74) 94/125 (75) .85 0.75 0.42, 1.33 .32 1.577 0.057
Pain/discomfort 178/302 (59) 81/125 (65) .26 0.72 0.44, 1.17 .19 1.637 0.023
Anxiety/depression 165/302 (55) 71/123 (58) .56 0.99 0.61, 1.61 .98 1.076 0.020
Median VAS (Q1, Q3) 60 (44,80) 60 (50,77) .62 coefficient 95% CI P value
EQ-5D-3L DCE 0.68 (0.45, 0.80) 0.66 (0.45, 0.79) .99 0.01 -0.06, 0.09 .78 N/A 0.026
CI, confidence interval; OR, odds ratio; Q1, 25th percentile; Q3, 75th percentile; SU, stroke unit (reference category).
*All models are adjusted for age, sex, index of relative socioeconomic advantage and disadvantage, ability to walk within first 24 hours of stroke onset, type of stroke, transfer from
another hospital and documented history of stroke; For assessment of the goodness-of-fit, all P > .05, indicating that the models were a good fit; EQ-5D-3L, EuroQoL -5 dimen-
sion-3 level instrument,16 VAS, Visual Analogue Scale score of the EQ5D; DCE, discrete choice experiment utilities determined using the method by Viney et al. based on the
EQ-5D dimensions.17 The lowest possible E-Value is 1,19 but there are no specific guidelines on the range of E-values. If an E-value is deemed small (residual confounding is a
threat) and where it is larger (residual confounding may not be a problem).30 Therefore, our E-values indicate the minimum strength of association between SU care and the out-
come that would be required by an unmeasured confounder to fully explain away each of the SU care and outcome models listed. Pseudo R2 is a statistic that indicates the preci-
sion of a logistic regression model and can be used to assess the goodness-of-fit in comparison to another model. That is, where the Pseudo R2 value is larger it has a better ability
to predict the outcome.

D.A. CADILHAC ET AL.

JMCTD PATIENTS HAVE INCREASED IMT 1309

Not all cases of in-hospital strokes may have been cap- References
tured in our dataset. Hospital staffs are responsible for pro-
viding data to assess their case ascertainment each year. 1. Park HJ, Cho HJ, Kim YD, et al. Comparison of the char-
acteristics for in-hospital and out-of-hospital ischaemic
When we compared the hospitals with incomplete case
strokes. Eur J Neurol 2009;16(5):582-588.
ascertainment to those with 100% case-ascertainment, we 2. Stroke Unit Trialists Collaboration. Organised inpatient
found that the proportion of in-hospital strokes was simi- (stroke unit) care for stroke. Cochrane Database Syst Rev
lar, ranging from 5%-11% for individual hospitals. 2013;9:CD000197.
In future research we plan to address some of these lim- 3. Farooq MU, Reeves MJ, Gargano J, et al. In-hospital
stroke in a statewide stroke registry. Cerebrovasc Dis
itations by enriching the AuSCR data using data linkage
2008;25(1-2):12-20.
techniques to merge hospital admissions and emergency 4. Cumbler E, Wald H, Bhatt DL, et al. Quality of care and
datasets which contain additional information on comor- outcomes for in-hospital ischemic stroke: findings from
bidities, palliative care, procedure codes and hospital care the National Get With The Guidelines-Stroke. Stroke
in an intensive care unit. We will be able to expand the 2014;45(1):231-238.
5. Fonarow GC, Reeves MJ, Smith EE, et al. Characteristics,
variables available for case-mix adjustment or other fea-
performance measures, and in-hospital outcomes of the
tures of hospital care without requiring hospital staff to first one million stroke and transient ischemic attack
collect more data. admissions in get with the guidelines-stroke. Circ Cardio-
vasc Qual Outcomes 2010;3(3):291-302.
6. Blacker DJ. In-hospital stroke. Lancet Neurol 2003;2
Conclusion
(12):741-746.
From this large, multicenter cohort study we observed 7. Chen S, Singh RJ, Kamal N, et al. Improving care for
acute in-hospital ischemic strokes: a narrative review. Int
an association with treatment in an SU and better early
J Stroke 2018. 1747493018790029.
(within 30-days) survival for patients experiencing a 8. Briggs R, McDonagh R, Mahon O, et al. In-hospital
stroke in hospital compared to their counterparts man- stroke: characteristics and outcomes. Ir Med J 2015;108
aged in other types of wards. Although referral bias due (1):24-25.
to comorbidity or palliation decisions may contribute to 9. Cadilhac DA, Lannin NA, Anderson CS, et al. Protocol
and pilot data for establishing the Australian Stroke Clin-
the observed differences in outcomes; the benefits of SU
ical Registry. Int J Stroke 2010;5(3):217-226.
care may extend to patients experiencing in-hospital 10. Cadilhac DA, Lannin NA, Anderson CS, et al. The Aus-
stroke and where feasible, management in an SU should tralian Stroke Clinical Registry Annual Report 2014.
be considered. Future work is required to clarify whether Annual report. Melbourne: The Florey Institute of Neuro-
the established benefit of SU care is also applicable to the science and Mental Health; 2015 November 2014. Report
No.: 1.
subgroup of in-hospital stroke not managed in SUs and
11. Counsell C, Dennis M, McDowall M, et al. Predicting out-
whether engagement of the stroke service team in their come after acute and subacute stroke: development and
care is of value where transfer is not possible. validation of new prognostic models. Stroke 2002;33
(4):1041-1047.
Acknowledgments: We acknowledge Joyce Lim and Sab- 12. Cadilhac D, Kilkenny M, Churilov L, et al. Identification
of a reliable subset of process indicators for clinical audit
rina Small from The George Institute for Global Health (Syd- in stroke care: an example from Australia. Clin Audit
ney), and Francis Kung, Karen Moss, Steven Street, Renee 2010;2:67-77.
Stojanovic, Robin Armstrong, Enna Stroil-Salama, Kate Paice, 13. Cadilhac DA, Kilkenny MF, Levi CR, et al. Risk-adjusted
Kasey Wallis, Adele Gibbs and Alison Dias from AuSCR hospital mortality rates for stroke: evidence from the
Office who contributed to AuSCR operations during this study Australian Stroke Clinical Registry (AuSCR). Med J Aust
2017;206(8):345-350.
period. Staffs from the Stroke Foundation are acknowledged 14. Quan H, Li B, Couris CM, et al. Updating and validating
for their contributions to patient follow-up. We would also the Charlson comorbidity index and score for risk adjust-
like to thank Joosup Kim (Monash University) for assistance ment in hospital discharge abstracts using data from 6
with aspects of the data analysis, and the hospital staff for their countries. Am J Epidemiol 2011;173(6):676-682.
diligence regarding data collection for AuSCR. Hospital site 15. Cadilhac DA, Kim J, Lannin NA, et al. Better outcomes
for hospitalized patients with TIA when in stroke units:
investigators who provided data between 2010 and 2014 are an observational study. Neurology 2016;86(22):2042-
also acknowledged (see Supplementary material 1: co-investi- 2048.
gators). We are grateful to the patients and families who have 16. The EuroQol Group. EuroQol-a new facility for the mea-
contributed information. surement of health-related quality of life. Health Policy
1990;16(3):199-208.
17. Viney R, Norman R, Brazier J, et al. An Australian dis-
Supplementary Materials crete choice experiment to value EQ-5D health states.
Health Econ 2014;23(6):729-742.
Supplementary material associated with this article can
18. Golicki D, Niewada M, Karlinska A, et al. Comparing
be found in the online version at doi:10.1016/j.jstrokecere responsiveness of the EQ-5D-5L, EQ-5D-3L and EQ VAS
brovasdis.2019.01.026. in stroke patients. Qual Life Res 2014.
1310 D.A. CADILHAC ET AL.

19. VanderWeele TJ, Ding P. Sensitivity analysis in observa- hyperglycaemia, and swallowing dysfunction in acute
tional research: introducing the E-value. Ann Intern Med stroke (QASC): a cluster randomised controlled trial. Lan-
2017;167(4):268-274. cet 2011;378(9804):1699-1706.
20. Clark AM, Jamieson R, Findlay IN. Registries and 26. Feigin VL, Lawes CM, Bennett DA, et al. Worldwide
informed consent. N Engl J Med 2004;351(6):612-614. stroke incidence and early case fatality reported in 56
author reply -4. population-based studies: a systematic review. Lancet
21. Cadilhac DA, Andrew NE, Lannin NA, et al. Quality of acute Neurol 2009;8(4):355-369.
care and long-term quality of life and survival: The Austra- 27. Jorgensen HS, Kammersgaard LP, Houth J, et al. Who
lian Stroke Clinical Registry. Stroke 2017;48(4):1026-1032. benefits from treatment and rehabilitation in a stroke
22. Wardlaw JM, Murray V, Berge E, et al. Thrombolysis for Unit? A community-based study. Stroke 2000;31(2):434-
acute ischaemic stroke. Cochrane Database Syst Rev 439.
2014;(7):CD000213. 28. National Stroke Foundation. National Stroke Audit Clini-
23. Cadilhac DA, Carter RC, Thrift AG, Dewey HM. Why invest cal Report: Acute Services. Melbourne: National Stroke
in a national public health program for stroke? An example Foundation; 2015 December.
using Australian data to estimate the potential benefits and 29. Sim J, Teece L, Dennis MS, et al. Validation and recalibra-
cost implications. Health Policy 2007;83(2-3):287-294. tion of two multivariable prognostic models for survival
24. Lynch E, Hillier S, Cadilhac D. When should physical and independence in acute stroke. PLoS One 2016;11(5):
rehabilitation commence after stroke: a systematic e0153527.
review. Int J Stroke 2014;9(4):468-478. 30. Ioannidis JPA, Tan YJ, Blum MR. Limitations and misin-
25. Middleton S, McElduff P, Ward J, et al. Implementation terpretations of e-values for sensitivity analyses of obser-
of evidence-based treatment protocols to manage fever, vational studies. Ann Intern Med 2019;170(2):108-111.