Does Admission to Hospital Affect Trends in Survival and Dependency After Stroke Using the South London Stroke Register?
Background and Purpose—Despite guidelines for specialist assessment in hospital for stroke, it is important to identify patient characteristics, trends, and outcome in patients not admitted to hospital compared with patients admitted to hospital.
Methods—Population-based stroke register of first in a life time strokes between 1995 and 2012 were examined. Baseline data included admission or nonadmission, case mix, stroke subtype, and risk factors before stroke. Survival curves were estimated with Kaplan–Meier methods. Logistic regression was used to determine factors associated with poor outcome (dead and dependency: Barthel index, <15) at 3 months and 1 year.
Results—Three thousand four hundred sixty-four patients were admitted to hospital for stroke. Patients admitted were more likely have more severe impairments (P<0.001). There was a significant trend for increasing admission over time; 1995 to 2000 (82%), 2001 to 2006 (90%), and 2007 to 2012 (94%); P<0.001. When survival analysis was stratified according to Barthel index ≥15 at day 7, there were no significant differences in survival curves between admission and nonadmission groups in 1995 to 2000 (P=0.15) or 2001 to 2006 (P=0.06), but there was a significant trend for higher survival rates for nonadmission in the 2007 to 2012 cohort (P=0.025). Admission to hospital (stroke unit) compared with nonadmission was also associated with poor outcome in the 2001 to 2006 time period (odds ratio, 2.66; confidence interval, 1.17–6.04) and the 2007 to 2012 time period (odds ratio, 5.26; confidence interval, 1.27–21.81).
Conclusion—There is a survival advantage from 2007 onward and lower levels of dependency from 2001 onward after adjusting for case mix for those patients who are not admitted to hospital, which requires further explanation.
Admission to hospital and access to specialist stroke services is a cornerstone of high-quality stroke care.1 National and international guidelines in stroke that we are aware of have supported the approach that patients with suspected stroke should be admitted to hospital to receive a range of evidence-based interventions.1–3 Despite this consensus, nonadmission rates for patients with stroke vary between 15% and 22%.4,5
Previous data from the South London Stroke Register (SLSR) suggested that death and disability were more likely to occur in patients who were admitted to hospital for stroke compared with patients who were not despite adjustment for case mix.4 It was hypothesized that aspects of hospital care may have been detrimental, which required further exploration. However, a randomized controlled trial comparing stroke unit care, inpatient stroke team care, and avoidance of hospital admission demonstrated that mortality at 3 months was significantly lower for patients assigned to stroke unit.6
Over the past 2 decades, there has been a policy drive for increasing admission for stroke predominately fueled by guidelines, highlighting access to acute specialist care, although many directives are now focusing efforts in reducing emergency admissions for many other conditions despite equivocal evidence.1,7 It is therefore important to identify patient characteristics of those who do not seek hospital admission for stroke and to ascertain whether they receive evidence-based interventions. The aims of this study are to explore the trends, the processes of care, and differences in outcome between patients admitted to hospital versus nonadmission after first-ever stroke using a population stroke register in South London.
Identification of Patients
Data for this analysis were derived from the SLSR, a population-based stroke register that has prospectively recorded first-ever strokes in patients within a geographically defined area of South London since 1995. Hospital surveillance for stroke included 2 teaching hospitals within and 3 outside the study area. Methods of patient notification and data collection have been described previously.8,9 Multiple, overlapping sources were used to register nonadmitted patients with stroke and admitted patients with stroke by trained study nurses/fieldworkers. All general practitioners within and on the borders of the study were contacted regularly and asked to notify the SLSR of patients. Regular communication with general practitioners was achieved by telephone contact and quarterly newsletters. Referrals of nonadmitted stroke to a neurovascular clinic or domiciliary visit to patients by the study team were also available to general practitioners. Community therapists were also contacted regularly.8,9 Data collected between 1995 and 2012 were used in this analysis. At the 2001 census, the population of the SLSR area was 271 817 (63% whites, 9% black Caribbean, 15% black African, and 13% other ethnic groups). Stroke diagnosis, using the World Health Organization clinical definition, was verified by study clinician, and patients were examined within 48 hours of referral. Patients who died before admission and incurred a stroke while in hospital were excluded from the analysis. All patients and their relatives gave written informed consent to participate in the study. Few patients had declined to be registered (1%).
Sociodemography and Case Mix
Sociodemographic data collected included age; sex; ethnic origin (1991 census question) stratified into white, black (black Caribbean, black African, and black other), and other ethnic group; and socioeconomic status (manual and nonmanual occupation). Clinical details at the time of maximal impairment within 72 hours of onset were obtained (swallowing: using the 3-oz water swallow test and urinary incontinence). The level of consciousness was assessed using the Glasgow Coma Scale score dichotomized into Glasgow Coma Scale scores <13 (impaired consciousness) and ≥ 13.10 Activities of daily living previous prior to stroke were assessed using the Barthel index (BI) and were classified as 0 to 14 (moderate/severe disability) and 15 to 20 (mild disability/independent).11 Stroke subtype was categorized using the Oxford Community Stroke Project classification: total anterior circulatory infarction (TACI), partial anterior circulatory infarction (PACI), posterior circulatory infarction (POCI), lacunar infarction (LACI), primary intracerebral hemorrhage (PICH), subarachnoid hemorrhage (SAH), or unclassified stroke (no pathological confirmation of stroke subtype).12
Previous Risk Factors
Previous history of hypertension (>140 mm Hg systolic or >90 mm Hg diastolic), diabetes mellitus, atrial fibrillation, previous transient ischemic attack, alcohol drinking status (yes/no) and smoking history (current, ex-smoker, and never smoked), and previous ischemic heart disease was recorded from general practice or hospital records.
Effective Interventions After Stroke
Patients were classified as (1) not admitted to hospital; (2) admitted/transferred to stroke unit at any time during hospital admission; and (3) admitted to nonstroke unit (managed in a general medical/geriatric ward). We examined a range of indicators of the process of care after an acute stroke suggested to be useful proxy measures for quality of stroke care.13 These included access to brain imaging (computed tomography, magnetic resonance imaging, or both), swallow assessment, the use of antihypertensive agents during the first 3 months of stroke, and the use of antiplatelet, anticoagulant, and cholesterol-lowering agents in ischemic stroke during the same time period.
Outcome as measured by the BI was categorized into good (BI≥15) and poor (death or dependency [moderate/severe] BI, 0–14).11 These were assessed 7 days, 3 months, and at 1 and 5 and 10 years after stroke. Survival time was calculated from the date of stroke to date of death.
Data were available from January 1, 1995, and we were able to obtain complete records up to December 31, 2012. We included all index cases up to December 31, 2012, and incorporated follow-up until May 31, 2013. Survival time was confirmed by the Office for National Statistics. Patients with no record of death were censored at May 31, 2013. Continuous variables were summarized as mean (SD) and categorical data as percentage. One-way ANOVA was used to test differences in continuous variables where appropriate, and the χ2 test was used for proportions. Survival curves were made among patients with stroke by consecutive time periods (per 6 years), ethnic groups, and for those with BI scores ≥15 at day 7 using the Kaplan–Meier method (unadjusted) and Log-rank tests. Multivariate survival analyses were undertaken using Cox proportional hazards models to determine the prognostic value of sociodemographic factors, case mix, stroke subtype, effective intervention, and risk factors before stroke. The event studied was all-cause mortality. The prognostic value of sociodemographic factors, case mix, effective intervention, and risk factors before stroke for 3-month and 1-year outcome was also examined by using multivariate logistic regression. Sensitivity analyses were carried out to assess possible effects of missing data by comparing the observed and complete case analyses with missing data analyses using various imputation methods, where missing data for survivors were imputed at all time points using a best- and then worst-case scenario for binary outcomes. Loss to follow-up rates varied by time point (after accounting for deaths): 3 months (24%) and 1 year (17.9%). All tests were 2 tailed, and P value <0.05 was considered statistically significant. Hazard ratios with 95% confidence interval (CI) for prognostic factors were calculated in Cox models, whereas odds ratios (ORs) with 95% CI were calculated in multivariate logistic models. All statistical analyses were performed with statistical software R, version 2.15.2.
Between January 1995 and December 2012, 3459 of a total of 3917 patients with first-ever stroke were admitted to hospital, and 458 (12%) patients were managed in the community. Table 1 describes the sociodemographic characteristics, case mix, stroke subtype, risk factors before stroke, and process of stroke care between patients admitted to hospital and nonadmitted patients. Patients admitted to hospital (stroke unit and nonstroke unit) were younger compared with those not admitted (69.9 versus 68.6 versus 71 years; P=0.002). Patients of black African Caribbean origin (P<0.0001) were more likely to be admitted. Patients not admitted were more likely to have a previous history of hypertension (P=0.03) and be a current smoker (P<0.0001) and current drinker of alcohol (P<0.0001) compared with patients admitted to hospital; however, there was a higher frequency of atrial fibrillation in patients admitted to hospital (P=0.0005). Patients admitted had more severe clinical impairments for stroke, such as incontinence, Glasgow Coma Scale score <13, failed swallow, and being disabled at day 7 (BI<15) compared with nonadmitted patients (P<0.0001). There were significant differences in stroke subtype between both groups with higher rates of TACI and intracerebral hemorrhage in admitted patients but lower rates of lacunar infarction in this group (P<0.0001). The swallow test was less likely to be performed in nonadmitted patients (23.1%) compared with admitted patients to stroke units (7.2%) and admitted patients to nonstroke units (11.7%; P<0.0001). Brain imaging consisting of both computed tomography and magnetic resonance imaging was also more likely to occur in patients to admitted stroke units (37.5%) compared with patients admitted nonstroke units (12.7%) and nonadmitted patients (12%; P<0.0001).
Table I in the online-only Data Supplement illustrates an increasing trend for admission to hospital over the 18-year period: 1995 to 2000 (82%), 2001 to 2006 (90%), and 2007 to 2012 (94%); P<0.001. When comparing characteristics across the 6-year time periods, nonadmitted patients became younger. Increasing trends of black African Caribbean patients not admitted were observed across all time periods. In this study, 23% (1995–2000), 67% (2001–2006), and 82% (2007–2012) of patients were managed in stroke units. The distributions of the day 7 BI <15 for nonadmitted patients were 13.5% (1995–2000), 11% (2001–2006), and 6.3% (2007–2012). The distribution of combined brain imaging using computed tomography and magnetic resonance imaging for admitted patients (stroke units) was 7.9% (1995–2000), 15.1% (2001–2006), and 74.1% (2007–2012).
Among the 3917 patients with first-ever stroke between January 1995 and December 2012, the median survival was 40.5 months (stroke units) compared with 40.4 months (nonstroke units) and 80.4 months (nonadmission; P<0.0001). The 7-day case fatality rate was 13.1% (stroke units) compared with 22.7% (nonstroke units) and 0.4% (nonadmission; P<0.0001). Ninety-day case fatality rate was 23.9% (stroke units) compared with 37.8% (nonstroke units) and 2.6% (nonadmission; P<0.0001).
When we compared patients registering in each consecutive 6-year period from 1995 to 2012, we found that survival was overall greater for nonadmission compared with admitted patients in the Kaplan–Meier analysis (Log-rank test, P<0.0001) with an overall improvement in survival over time for each period (P<0.0001; Figure 1A). When further analysis was stratified by ethnicity, there was a survival advantage for nonadmitted patients for both white (P<0.0001) and black Caribbean patients (P=0.01) but not black African patients (P=0.44). However for the whole cohort, when the analyses were stratified by day 7 BI ≥15 as a measure of case mix, there was no survival advantage between the groups (P=0.09; Figure 1B). When this analysis was stratified by each 6-year period, the survival advantage for nonadmission was shown in the most recent cohort of 2007 to 2012 (Log-rank test, P=0.025).
Factors affecting all-cause mortality are described in Table 2 across 6-year time periods. Multivariate survival analysis showed that increasing age, severe clinical impairments for stroke (Glasgow Coma Scale score, <13; urinary incontinence and failed swallow), and atrial fibrillation across all cohorts was associated with mortality. Being African Caribbean in the 1995 to 2000 period (hazard ratio, 0.7; 95% CI, 0.58–0.85) and the 2001 to 2006 (hazard ratio, 0.73; 95% CI, 0.59–0.91) period was associated with improved survival. Nonstroke unit management compared with nonadmission was associated with increased mortality across all time periods after adjusting for potential confounding factors. Stroke unit management compared with nonadmission was associated with increased mortality in all time periods but was significant in the 2007 to 2012 time period (hazard ratio, 7.56; 95% CI, 2.47–23.13) after adjusting for confounding factors.
Table II in the online-only Data Supplement shows the factors associated with poor outcome for each 6-year period at 1 year after adjusting for case mix. Increasing age and severe clinical impairments for stroke were associated with poor outcome across all time periods. Management in nonstroke units compared with nonadmission was more likely to be associated with poor outcome in the 2001 to 2006 time period (OR, 4.04; 95% CI, 1.69–9.67) and the 2007 to 2012 time-period cohort (OR, 12.61; 95% CI, 2.67–59.67). Management in stroke units compared with nonadmission was also associated with poor outcome in the 2001 to 2006 time period (OR, 2.66; 95% CI, 1.17–6.04) and the 2007–12 time period (OR, 5.26; 95% CI, 1.27–21.81). After 3 months (full data not shown), although the direction of the effects was similar across all time periods, there were no significant effects of nonadmission on poor outcome.
Figure 2 shows the distribution of poor outcomes (death and dependency: BI<15) and good outcome (BI≥15) across all time points of 7 days, 3 months, 1 year, 2 years, 5 years, and 10 years. Poor outcomes were evident across all time points for admitted patients compared with nonadmitted patients.
At 3 months, of the 2094 admitted patients with ischemic stroke, 368 (17.6%) were not prescribed antiplatelet agents compared with 37 (9.4%) of the 392 patients who were not admitted (P=0.02). Similarly, 760 (36.3%) of admitted patients with stroke were not prescribed cholesterol-lowering agents compared with 164 (41.8%) who were not admitted (P<0.001). Of the 351 patients who were admitted to hospital for ischemic stroke and were in atrial fibrillation, only 97 (27.6%) were prescribed warfarin compared with 4 (8.5%) of the 47 patients who were not admitted (P=0.008). There were no differences in antihypertensive prescription between the 333 patients admitted with known hypertension (73.6%) versus the 44 patients not admitted (81.8%); P=0.32).
This is the largest detailed study to date analyzing the differences in patients with stroke admitted to hospital compared with nonadmitted patients in an unbiased population in South London. The main finding from this study is that admitted patients had worse survival and poorer outcomes than patients with stroke who did not seek hospital admission after adjusting for confounding variables. This is at odds with the randomized controlled trial by Kalra et al6 where stroke unit admission was shown to be more effective than home support and mobile stroke team support for patients in a highly selected population of moderately severe strokes recruited within 72 hours. In addition to this, the study was conducted during 1995 to 1999 before widespread uptake of evidence-based interventions.
As the prognosis of the groups is not comparable, it is necessary to adjust for case mix variables in this nonrandomized comparison because of confounding by treatment indication. We have however adjusted for many of the case mix variables suggested by Davenport et al14 and validated case mix variables that have been shown to be predictive of poor outcome and are sufficient in quality and precision. These included age, sex, previous stroke BI, living conditions and socioeconomic status, stroke subtype, clinical assessments of maximal impairment (Glasgow Coma Scale, failed swallow, and incontinence), and previous stroke risk factors. It is however possible that some confounding factors remain unadjusted for such as markers of frailty, physiological variables, and more detailed measures of case mix, such as Charlson and Acute Physiology and Chronic Health Evaluation scores.15 Even when careful case mix adjustment is made using clinical data on stroke severity, no allowance can be made for a bias that exits between different types of patients using different services. It is also possible that since the clustering of patients within hospitals was not taken into account, unmeasured differences between hospitals may also account for some of the observed differences in admission and outcomes. There are also other factors that may differ between both groups, which may influence complex decision making for hospital admission, including patient choice, general practitioner advice, policy for local stroke services, and availability of resources.16
There are many possible interpretations to explain poorer outcomes for hospital admission. It has been acknowledged that hospital admission may be associated with many hazards, which includes malnutrition, sleep deprivation, risk of infection, pain, falls, and prescription of new drugs, which can lead to deconditioning as a result of loss of physiological homeostasis, particularly in older patients.17 Krumholz et al17,18 warn that many patients who are hospitalized are not only recovering from their acute illness within a complex pathway but are susceptible and exposed to a period of heightened risk for a wide range of adverse health events. It is also possible that patients are being discharged from hospital with higher levels of disability into the community, and this may account for their worse outcomes.
Although there is a strong evidence base that stroke units improves outcome, there was low uptake of such units in the earlier cohorts but a clear trend of improvement in the provision of these services. It is therefore possible that a significant proportion of patents were not in receipt of the important processes of care that are beneficial for patients with stroke.19 Interestingly, antiplatelet therapy was used less at 3 months in admitted patients compared with nonadmitted patients.
There was evidence of increasing hospital admission over time with a decreasing trend of the proportion of patients who were disabled at day 7 (BI<15) managed in hospital. This can be explained by increasing public awareness of stroke through campaigning and the reorganization of stroke care in London, promoting emergency access to hospital.20 In the latter cohort (2007–2012), there was an increasing trend of higher rates of black Caribbean patients being managed in the community. Previous data from the SLSR have demonstrated that black African Caribbean patients have a survival advantage over white groups, particularly in the patients aged >65 years old, and this may explain in part the survival advantage of nonadmission in the latter cohorts.21
In the latter cohort (2007–2012), patients who were not admitted were fewer in number, milder in severity, and had better outcome compared with earlier cohorts. In patients with nondisabling stroke, prompt specialist assessment is recommended, but guidelines do not specifically detail in which setting, this should occur.1 In a study comparing outpatient management of minor ischemic stroke with admission, there was no difference in 30-day admission rates for nonadmitted patients and 30-day readmission rates post discharge from hospital with lower costs with avoiding admission.22 In addition to this, higher rates of secondary prevention measures were used in nonadmitted patients. It therefore could be argued that for certain types of patients with stroke, there is no evidence that avoiding admission leads to poorer outcome.
Any general conclusion on poorer outcomes for admitted patients needs to be tempered by lower rates of statins and anticoagulation therapy use for patients not admitted to hospital, and it has been argued that admission to hospital may facilitate and reinforce secondary prevention compliance.22 The lower rates of swallow assessment in this group may reflect the timing and delay of specialist assessment. There were also lower rates of combined brain imaging overall in nonadmitted patients although there were similar rates of brain imaging for the latter cohort. This is in keeping with the need for magnetic resonance imaging brain imaging required for greater sensitivity for mild stroke in particular lacunar stroke.
There are strengths and limitations to this observational cohort study. The data were derived from a multiethnic population-based register whose outcomes had previously been described using the 1995 to 1998 cohort but now has the advantage of studying a large sample size of almost 4000 patients for a 18-year period with long-term follow up data, allowing statistical power to determine differences in survival and functional outcome in both groups. The strength of these analyses is the collection of processes of care variables over time with inclusion of new processes as the evidence for their use becomes established across different time periods. The loss to follow-up rates, once deaths are accounted for, in this study are around 20% at each time point. This loss to follow-up may introduce bias, yet estimates from analyses of the patients with complete data did not differ significantly from those presented here. In many cohort and stroke register studies, loss to follow-up rates are not often presented. We also acknowledge that inner city populations, such as in South London, are mobile with large numbers of migrant families, which can make follow-up challenging. Other reasons include the inability to complete follow-up because of cognitive impairment and refusal of patients to be assessed repeatedly. Efforts were also made for all patients’ changes of address to be recorded from hospital, general practice, or family sources. If patients had moved to another country, postal questionnaires were often sent and returned.9
The use of the BI at day 7 as a categorical measure of case severity can be argued, but results from analyses using dichotomized BI did not differ significantly from that using continuous BI. The ceiling and floor effects of the BI are acknowledged although this index is recognized in clinical studies and trials as an appropriate measure. Any patient experiencing stroke should be in receipt of specialist stroke care from the outset as soon as possible with the best evidence continuing to support stroke unit.1 The results of our analysis do not support that hospitalization for stroke should be avoided, but it does generate hypotheses, suggesting that potential hazards may exist with hospital admission and that for some patients with mild stroke, nonadmission with appropriate specialist intervention may be feasible and beneficial. Conversely, if nonadmission for stroke is considered particularly for patients with mild stroke, it is vital that access to specialist’s assessments, diagnostics, and secondary prevention are received. Randomized trials comparing nonadmission with urgent outpatient management versus hospital admission may be warranted in the future but may be difficult to conduct in the current arena of hyperacute stroke care.
We thank patients, their families, and the fieldworkers who have collected data for the South London Stroke Register since 1995.
Sources of Funding
The study was supported by Northern & Yorkshire NHS Research and Development Programme in Cardiovascular Disease and Stroke, Guy’s and St Thomas’ Hospital Charity, Stanley Thomas Johnson Foundation, The Stroke Association, Department of Health Healthcare Quality Improvement Partnership Grant, National Institute for Health Research (NIHR) Programme Grant (RP-PG-0407-10184), and the NIHR Biomedical Research Centre award to Guy’s and St Thomas’ Hospital in partnership with King’s College London.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.116.014136/-/DC1.
- Received May 20, 2016.
- Revision received June 20, 2016.
- Accepted June 30, 2016.
- © 2016 American Heart Association, Inc.
- 1.↵Royal College of Physicians Intercollegiate Stroke Working Party. National Clinical Guidelines for Stroke. 4th ed. London, UK: RCGP; 2012.
- Jaunch E,
- Saver J,
- Adams HP Jr,
- Bruno A,
- Connors J,
- Demaerschalk B,
- et al
- Bhalla A,
- Dundas R,
- Rudd AG,
- Wolfe CD
- Paul N,
- Koton S,
- Simoni M,
- Geraghty O,
- Luengo-Fernandez R,
- Rothwell P
- Kalra L,
- Evans A,
- Perez I,
- Knapp M,
- Donaldson N,
- Swift CG
- Purdy S
- Stewart JA,
- Dundas R,
- Howard RS,
- Rudd AG,
- Wolfe CD
- Addo J,
- Bhalla A,
- Crichton S,
- Rudd A,
- Mckevitt C,
- Wolfe CD
- Davenport RJ,
- Dennis MS,
- Warlow CP
- Phan TG,
- Clissold B,
- Ly J,
- Ma H,
- Moran C,
- Srikanth V
- Young J
- Langhorne P,
- Stott DJ,
- Robertson L,
- MacDonald J,
- Jones L,
- McAlpine C,
- et al
- Langhorne P,
- Pollock A
- Ramsay AI,
- Morris S,
- Hoffman A,
- Hunter R,
- Boaden R,
- McKevitt C,
- et al
- Wang Y,
- Rudd AG,
- Wolfe CDA
- Csiba L