Does admission to hospital improve the outcome for stroke patients?

Transcription

1 Age and Ageing 2001; 30: 197±203 # 2001, British Geriatrics Society Does admission to hospital improve the outcome for stroke patients? AJAY BHALLA, RUTH DUNDAS, ANTHONY G. RUDD 1,CHARLES D. A. WOLFE Department of Public Health Sciences, Guy's, King's and St Thomas' Hospital School of Medicine, Capital House, 42 Weston Street, London SE1 3DQ, UK 1 Department of Elderly Care, Guy's and St Thomas' Hospital, London, UK Address correspondence to: A. Bhalla. Fax: q44) : bhalla@ajay1.freeserve.co.uk Abstract Objectives: to identify the factors associated with hospital admission and the differences in management and outcome of stroke patients between hospital and home. Design: a prospective community stroke register 1995±8) with multiple noti cation sources. Setting: an inner city multi-ethnic population of in South London, UK. Participants: 975 subjects with rst in a lifetime strokes, whether or not they were admitted to hospital. Patients dying suddenly and those already hospitalized at the time of stroke were excluded. Main outcome measures: factors associated with hospital admission; differences in management in the acute phase of stroke; mortality and dependency assessed by the Barthel index 3 months post-stroke. Results: 812 patients were admitted to hospital for stroke; 163 were managed in the community. Factors independently associated with hospital admission included stroke severity, pre-stroke independence, atrial brillation, having an intracranial haemorrhage and having a non-lacunar infarction. Computed tomography scan rates were higher in admitted 78%) than non-admitted patients 63%; P = 0.001). By 3 months, %) of the admitted patients had died compared with 13 8%) of non-admitted patients P-0.001). Of the admitted patients, %) had a Barthel index 018 compared with %) of those who were not admitted P-0.001). After adjusting for case-mix variables, the odds ratios for death and dependency Barthel index-18) in admitted and non-admitted patients were ±5.12) and ±4.22) respectively. Conclusion: patients with clinical indicators for a more severe stroke were more likely to be admitted to hospital. Hospitalized stroke patients may have poorer survival and disability rates than those who remain at home, even after adjustment for case mix. There may be some aspects of acute hospital care that may be detrimental to outcome in certain groups of stroke patients. This requires further investigation. Keywords: home, hospital, outcome, stroke Introduction In the United Kingdom, hospital admission rates for stroke vary, with rates upwards of 70% being reported [1]. This variation in admission rates may re ect local policy, differences in stroke severity or social support [2]. Stroke accounts for the use of one- fth of medical beds [3]. There is increasing demand and pressure for acute hospital beds, partly explained by the increasingly ageing population associated with its growing morbidity [4]. Increasing provision of services in the community could potentially reduce the high costs incurred by health services and reduce the pressure on acute hospitals. The Intercollegiate Stroke Working Party indicate that stroke patients should only be managed at home if an appropriate diagnostic assessment can be undertaken, if care services are able to provide exible support and if the services are part of a specialist stroke service [5]. Hospital-at-home schemes have been evaluated for elderly patients with stroke, but their effectiveness remains unproven [6]. There is a suggestion that home-based care allows some stroke patients to reach their full rehabilitation potential as well as addressing their psychosocial needs if community stroke services can be organized ef ciently [7]. However, in view of the undisputed evidence of organized co-ordinated multidisciplinary care in stroke units, hospital-based 197

2 A. Bhalla et al. care may be a more effective way of delivering services and improving outcome [8]. The development of potential therapies for acute stroke such as thrombolysis with recombinant tissue plasminogen activator, may also result in a shift towards secondary care [9]. Until this happens, it is unclear if admission for stroke is necessary. Here, we report the factors associated with hospital admission for stroke patients, the differences in management of hospitalized and community patients, and differences in outcome. Methods A community-based register was established in Southwark and North Lambeth, South London the South London Stroke Register), to estimate the incidence, natural history and resource use of stroke in a multi-ethnic population. This register provides a nearcomplete and unbiased list of patients who have had a rst in a lifetime stroke out of a population of about individuals [10]. Data were collected from 1 January 1995 to 31 December 1998 at the time of stroke and 3 months later, using face-to-face interviews with the patient and/or carer wherever possible. Medical information was also con rmed from hospital and general practitioner notes. All surviving patients were examined at the time of maximum impairment within 72 h of stroke onset by one of the South London Stroke Register doctors. We obtained ethical approval from the Guy's and St Thomas' ethical committee. Initial data included: 1. Socio-demography: age, sex, social class and ethnicity [1]. Ethnic origin was strati ed into three groups: black black Caribbean, black African and black other), white and other Asian, Pakistani, Indian, Bangladeshi, Chinese and other). Social class categories were grouped into non-manual I, II, III non-manual), manual III manual, IV, V ) and economically inactive student, unemployed, unable to work because of disability, being a carer and being retired). Activities of daily living before stroke according to Barthel index [12]), living conditions before the stroke living alone, living with carer, living in an institution including: sheltered accommodation, residential or nursing home) and social support before the stroke meals on wheels, home help, day centre, day hospital, district nurse, community psychiatric nurse, respite care) were also collected. 2. Case severity: measures of clinical state at time of maximal impairment within 72 h) included level of consciousness Glasgow coma scale [29]), transient loss of consciousness at stroke outset, dysphagia, dysphasia, motor impairments weakness and paralysis) and incontinence. 3. Co-morbidity: we identi ed a history of hypertension, myocardial infarction, diabetes mellitus and atrial brillation through hospital and general practitioner records. Stroke subtype was classi ed according to the Oxford Community Stroke Project classi cation [13] as total anterior circulation stroke, partial anterior circulation stroke, posterior circulation stroke, lacunar infarction, primary intracerebral haemorrhage, subarachnoid haemorrhage or unclassi ed strokeðno pathological con rmation of subtype. 4. Diagnostic tests: full blood count, blood glucose, erythrocyte sedimentation rate, serum cholesterol, chest radiograph, electrocardiogram, echocardiogram, brain imaging computed tomography, magnetic resonance imaging) and carotid Doppler. We also noted whether the patient was managed on a stroke unit. 5. Outcome: patients were assessed during face-to-face interviews at 3 months post-stroke with con rmation of medical data from general practice, hospital and therapy records. We recorded the place of residence alone, with carer, hospital or institution), use of general practice and hospital and diagnostic services. We also recorded co-morbidities after stroke and the management of hypertension and the use of aspirin self-reported). Outcome assessment included death and dependency at 3 months Barthel index 18, independent; Barthel index-18, dependent) [12]. Any recurrence of stroke within 3 months were also noted. Statistical analysis We examined the univariate associations between hospitalized and community patients and sociodemography, case severity, co-morbidity, stroke subtype and diagnostic tests using Fisher's exact test. We estimated the associations between patient characteristics and admission to hospital by multiple logistic regression, after adjusting for age, sex, pre-stroke Barthel index and variables that were signi cant in the univariate analyses between admitted and non-admitted patients. Patients having a primary intracerebral haemorrhage and subarachnoid haemorrhage were not included in the model as only three and no patients respectively were managed at home. Stroke subtype was categorized into lacunar and non-lacunar, as there was an exaggerated imbalance in the total anterior circulation stroke group, where no patient managed in the community had this subtype, compared with 140 patients who were admitted. We used multiple logistic regression to compare death rates and functional outcome between hospitalized patients and community patients. When modelling for outcome, we selected variables by a backward stepwise procedure with signi cance set at the 5% level. The variables that were included in the model are shown in Table 1. These variables were either signi cantly associated with mortality in the univariate analysis or considered to be clinically relevant. The goodness of t of 198

3 Hospital versus home for stroke the model was assessed using the Hosmer±Lemeshow test [30]. Results Between 1 January 1995 and 31 December 1998, 1139 patients with rst stroke were registered with the South London Stroke Register, of whom 17 died suddenly Table 1. Case-mix variables in multiple regression modelling Age years) Sex Social class Ethnicity Hypertension pre-stroke Atrial brillation pre-stroke Myocardial infarction pre-stroke Diabetes mellitus pre-stroke Pre-stroke independence Living conditions pre-stroke Stroke subtype non-lacunar vs lacunar infarction) Incontinence within 72 h of stroke onset Paralysis within 72 h of stroke onset Dysphasia within 72 h of stroke onset Dysphagia within 72 h of stroke onset Glasgow coma scale within 72 h of stroke onset Transient loss of consciousness at stroke onset at home. Of the 1122 remaining patients, %) were in hospital at the time of their stroke and were excluded from the analysis, %) were admitted to hospital and %) remained in the community. There were no differences in admission rates between those below and above 65 years 86% versus 82%, P = 0.26). Most socio-demographic measurements were comparable for hospitalized and community stroke patients, but there were signi cant differences in living conditions Table 2). Of 370 patients living alone, %) were admitted, compared with %) of 513 patients who were living with someone P = 0.016). Of 89 patients who were living in an institution sheltered home, residential home or nursing home), 78 88%) were admitted. Table 3 summarizes the differences in co-morbidity and treatment of risk factors before stroke. Table 4 shows the univariate relationship between admission to hospital and severity of stroke. Patients with transient loss of consciousness at stroke onset and a Glasgow coma scale of -13 within 72 h of stroke were more likely to be admitted to hospital. Patients with dysphasia, dysphagia, weakness and paralysis were also more likely to be admitted. Table 5 shows the odds ratios for the variables associated with hospital admission in the univariate analysis and multivariate analysis using multiple logistic regression. Initial transient loss of consciousness, Table 2. Socio-demographic details of hospitalized and community patients Variable Hospitalized n = 812) Community n = 163) P-value a Sex male) ) 88 54) Ethnic group White ) ) Black ) 18 11) Other 42 5) 9 6) Living conditions before stroke Alone ) 50 31) With other s) ) ) Institution 78 10) 11 7) Missing 2 0) 1 0) Social class Non-manual ) 39 24) Manual ) ) Inactive ) 13 8) Missing 1 0) 0 0) Support services before stroke Meals on wheels 44 5) 6 4) Home help 85 10) 17 10) Day centre 22 3) 7 4) Day hospital 5 1) 2 1) District nurse 57 7) 8 5) Community psychiatric nurse 8 1) 3 2) Respite care 4 0) 1 1) Other 50 6) 12 7)

4 A. Bhalla et al. incontinence, dysphagia, paralysis, increasing pre-stroke Barthel index, atrial brillation and a non-lacunar infarction were all independently associated with hospital admission. Table 3. Co-morbidity and treatment of risk factors before stroke Variable Hospitalized Community P-value Hypertension ) ) a Hypertensive therapy ) 61 49) a Missing 4 1) 0 0) Atrial brillation ) 20 12) a Missing 4 0) 1 1) Anticoagulation 20 11) 0 0) a Missing 2 1) 0 0) Myocardial infarction ) 21 13) a Missing 6 1) 1 1) Diabetes mellitus ) 22 14) a Barthel index before stroke Median b Inter-quartile range 19±20 19±20 Missing 4 0) 1 1) Stroke subtype a TACI ) 0 0) PACI ) 41 25) POCI 86 10) 26 16) LACI ) 62 38) PICH ) 3 1) SAH 53 7) 0 0) Unclassi ed 78 10) 31 19) TACI, total anterior circulation infarct; PACI, partial anterior circulation infarct; POCI, posterior circulation infarct; LACI, lacunar infarction; PICH, primary intracerebral haemorrhage; SAH, subarachnoid haemorrhage. b Mann±Whitney U test. The use of diagnostic services by hospital admissions and community patients are shown in Table 6. Hospitalized patients were more likely to have a computed tomography brain scan, a chest radiograph and a electrocardiogram, while carotid Doppler examination was more frequent in community patients. Out of 812 patients admitted to hospital, %) were managed on a stroke unit during their stay. Overall, %) patients were alive at 3 months. Data were missing for ve hospitalized and one community patient. These patients refused follow-up, migrated or were untraceable through noti cation sources. The following variables were signi cantly associated with death at 3 months: age, sex, ethnicity, atrial Table 4. Comparison of clinical indicators of stroke severity between admitted and community patients Initial symptom Hospitalized Community P-value a Loss of consciousness ) 3 2) at onset Missing 66 8) 5 3) GCS assessment within ) 3 2) h -13) Missing 3 0) 0 0) Dysphagia within 72 h ) 21 13) Missing 13 2) 2 1) Dysphasia within 72 h ) 23 14) Missing ) 4 2) Incontinent/catheterized ) 15 9) Missing 14 2) 2 1) Motor de cit within 72 h None 88 11) 37 23) Any weakness ) ) Any paralysis ) 4 2) Missing 13 1) 2 1) GCS, Glasgow coma scale. Table 5. Univariate and multivariate analysis for factors associated with hospital admission Univariate Multivariate Variable Odds ratio 95% CI Odds ratio 95% CI P-value Transient loss of consciousness at 24 h , , Incontinence at 72 h , , Motor de cit at 72 h None 1.00 ± 1.00 ± Weakness , , 1.86 ± Paralysis , , Dysphagia , , Atrial brillation , , Barthel index before stroke , , Stroke subtype Non-lacunar infarction 1.00 ± 1.00 ± Lacunar infarction , , 1.15 Unclassi ed , , 0.32 CI, con dence interval. 200

5 Hospital versus home for stroke Table 6. Use of diagnostic tests in the acute phase of stroke Test Hospitalized Community P-value a Full blood count ) ) Glucose ) ) Cholesterol ) 96 59) Electrocardiogram ) ) ESR ) 86 53) Echocardiogram ) 48 29) CT scan ) ) MRI scan 77 9) 20 12) Carotid Doppler ) 63 39) Chest radiograph ) ) ESR, erythrocyte sedimentation rate; CT, computed tomography; MRI, magnetic resonance imaging. brillation, myocardial infarction, pre-stroke Barthel index, living conditions pre-stroke, stroke subtype and clinical indicators for stroke severity P-0.001). By 3 months, 285 admitted patients 35%) had died, compared with 13 8%) of the patients managed in the community P-0.001). After adjusting for case-mix variables see Table 1), the odds ratio for death in admitted patients compared with community patients was 2.21 [95% con dence interval CI) 0.96±5.12]. The nal model consisted of age, sex, dysphagia, paralysis, atrial brillation, pre-stroke Barthel index, incontinence, Glasgow coma scale score and stroke subtype Hosmer± Lemeshow goodness of t, P = 0.249). When missing data were taken into account, the odds ratio for the worse-case scenario assuming all patients missing were dead) was % CI 0.85±4.24). Of the patients managed in the community, %) achieved a Barthel score of 018 compared with %) of those admitted to hospital P-0.001). After adjusting for case-mix variables, the odds ratio for dependency Barthel index -18) in admitted patients compared with patients managed in the community was % CI 1.35±4.22). The nal model consisted of age, sex, living conditions prestroke, hypertension pre-stroke, pre-stroke Barthel index, incontinence and paralysis Hosmer±Lemeshow goodness of t, P = 0.646). When missing data were taken into account, the odds ratio for dependency for the worse-case scenario assuming all missing were dependent) was % CI 1.16±3.65). One hundred and nineteen 37%) of the 320 patients who were admitted and lived alone were discharged to their original homes. Thirty-eight 76%) of the 50 patients who were managed in the community and lived alone remained in the same household. By 3 months, 90 patients remained in hospital, all of whom had been initially admitted. Eleven 2%) of the admitted patients had recurrent strokes, as did 10 7%) of those managed at home P = 0.003). General practitioners reviewed %) of those admitted compared with 75 50%) of those not admitted P = 0.25). Of those admitted, %) were reviewed by a hospital physician as outpatients, compared with 70 47%) of those who were not admitted P = 0.29). Of hospitalized patients, %) were diagnosed as hypertensive at 3 months, compared with %) of community patients P = 0.139). Among those diagnosed as hypertensive, %) of those who were admitted were prescribed antihypertensive treatment, compared with 72 64%) of those who were managed at home P = 0.147). One hundred and eight 72%) community patients were prescribed anti-platelet agents, compared with %) admitted patients P = 0.001). When the analysis was restricted to patients with cerebral infarction, %) admitted patients were prescribed anti-platelet agents compared with 92 80%) community patients P = 0.33). Signi cantly more patients admitted to hospital 11%) were prescribed anticoagulant therapy compared with those who were not admitted 3%; P = 0.004). This signi cance was still apparent after the analysis was restricted to patients with cerebral infarction P = 0.006). Twenty-one 58%) of 37 patients who were in atrial brillation and were receiving warfarin and not antiplatelet therapy were admitted compared with two 33%) of six patients who were managed at home P = 0.286). Discussion This is the rst detailed analysis of admitted and nonadmitted stroke patients in an unbiased sample of all stroke patients in a population. An unexpected nding of this study was that hospitalized stroke patients appeared to have higher mortality and disability rates than those who remained at home. This is at variance with a previous study by Wade and colleagues which demonstrated no such difference [6]. One possible explanation is that we have not adjusted for all possible case-mix differences between the groups; this is a limitation of non-randomized controlled studies. Davenport and colleagues demonstrated that in non-randomized studies, interpretation of outcome data may be in uenced by case-mix variables and that adjustments are essential to allow meaningful comparisons [14]. We have included most of the case-mix variables suggested by Davenport and colleagues see Table 1). However, although we adjusted for stroke subtype, we subsequently categorized strokes into lacunar and non-lacunar infarction. It is therefore possible that important subtype differences may have been masked. It is also possible that collection of case-mix variables may have been in uenced by whether patients were admitted or not. The CIs for the odds ratio for death do, however, cross 1 at 0.96 but also approach 5, which, although not statistically 201

6 A. Bhalla et al. signi cant, may be clinically relevant. The outcome differences may be real but we remain unclear as to why this should be so. There are several possible explanations. Perhaps some aspects of acute hospital care are detrimental to outcome in certain groups of stroke patients. Evidence from a survey of acute stroke treatment in UK hospitals suggests that some hospital-based treatments are delivered haphazardly in the UK with no evidence for their use [21]. There is also evidence that hospital-based stroke services are organized haphazardly [22]. A Stroke Association survey found that only 52% of stroke patients were managed by a system of organized stroke care [23]. Inadequate acute supportive measures to maintain hydration or nutrition may be harmful in particular subgroups of patients. Increased rates of infection during hospital stay, such as urinary tract and chest infections, may also have been an important factor contributing to increased mortality [24]. We know that stroke units save lives [7] in hospital, but in this study only 32% of patients were managed on a stroke unit. It was therefore possible that most of the patients admitted to hospital did not receive early supportive treatment, early identi cation of complications and focussed multidisciplinary specialist stroke care [25]. Although emotional outcome was not measured, there is evidence that psychological factors are prevalent in hospitalized stroke patients, and this may affect recovery [26]. Any general conclusion regarding poorer outcomes with hospitalized patients should be tempered by the ndings of higher recurrence rates in patients managed at home. One-quarter of patients who were managed alone at home did not remain in the same household after 3 months. Between 25 and 50% of all stroke patients are not admitted to hospital in the UK [1, 2, 15]. Thus, our hospital admission rate of 72% was higher than that quoted in several previous studies [2, 15]. There is debate as to whether all stroke patients should be admitted to hospital [16]. This debate is similar to the debate over the management of myocardial infarction 20 years ago, before the introduction of coronary care units [17]. It is important to try to understand what clinical and social factors are relevant in deciding to admit patients with stroke. Stroke patients may be admitted to hospital for diagnostic reasons, rehabilitation, management of complications or nursing care [18]. Previous research has suggested that stroke severity, social isolation and nursing care are important factors related to hospital admission [2, 4, 19]. In addition, we have demonstrated that, if a patient was previously independent before the stroke, was in atrial brillation and had a non-lacunar infarction, they were also more likely to be admitted. Patients with an intracranial haemorrhage were also more likely to be admitted. Most patients with severe stroke and who lived alone were admitted, possibly re ecting a lack of ability to provide intensive support when required. This may also have re ected a desire among hospital patients for support in the acute phase [20]. However, despite this, many patients not admitted did have medical investigations: this implies that general practitioners had access to pathology and radiology departments to investigate their stroke patients. Investigation of carotid disease was performed more often in non-admitted patients, which may be because more patients with non-disabling strokes remained at home. Higher rates of partial anterior and posterior circulation infarct stroke events were apparent in patients managed at home: this may have also explained the higher recurrence rates in this group [12]. A higher but not signi cant proportion of patients who were not admitted were more likely to be referred to hospital outpatients, which may explain the use of carotid investigation. Follow-up rates by general practitioners and physicians were similar between the two groups and were in concordance with previous studies [1]. Use of secondary preventive measures such as aspirin and antihypertensive therapy) after 3 months did not differ signi cantly between admitted and non-admitted patients, but these measures still remain underused, as also shown in another local study [27]. Anticoagulation was more likely to be used in patients admitted to hospital, possibly re ecting the higher prevalence of atrial brillation 3 months after stroke or the reluctance of general practitioners to commence anticoagulation. This nding may also partly explain the increase in recurrence rates in patients who were not admitted. To elucidate the true effects of care in hospital and at home on outcome, a randomized controlled trial may be required. This may be dif cult to achieve. Kalra and colleagues are attempting to address this question, and preliminary ndings suggest that stroke unit rehabilitation is superior to domiciliary specialist care at 3 months but not at 1 year [28]. The potential implementation of acute stroke therapies may change current practice. Further investigation of the effect of acute stroke support in hospitalized patients is required. Keypoints. Patients who had clinical indicators for a more severe stroke, atrial brillation and non-lacunar infarction were likely to be admitted to hospital.. There is inequity in the provision of diagnostic services to admitted and non-admitted stroke patients.. After adjusting for case-mix variables, admitted stroke patients appear to have poorer survival and disability rates than those who remain at home; unmeasured case-mix could explain these ndings.. Some aspects of acute hospital care may be detrimental in certain groups of stroke patients; this issue requires further investigation. 202

7 Hospital versus home for stroke Acknowledgements We would like to thanks all participants of the South London Community Stroke Register Project and all the general practitioners in the area who have supported the study. References 1. Wolfe CDA, Taub NA, Woodrow J et al. Patterns of acute care in three districts of southern England. J Epidemiol Community Health 1993; 47: 144±8. 2. Bamford J, Sandercock P, Warlow C, Gray M. Why are patients with acute stroke admitted to hospital. Br Med J 1986; 1: 1369± Wade DT. Stroke acute cerebrovascular disease). In Stevens A, Rafferty J eds. Health Care Needs Assessment, volume 1. Oxford: Radcliffe Medical Press, 1994; 111± Isard PA, Forbes JF. The cost of stroke to the National Health Service in Scotland. Cerebrovasc Dis 1992; 2: 47± Intercollegiate Working Party for Stroke. National Clinical Guidelines. London: The Royal College of Physicians, Wade DT, Hewer RL, Skilbeck CE et al. Controlled trial of home-care service for acute stroke patients. Lancet 1985; i: 323±6. 7. Andrews K, Stewart J. Stroke recovery: he can but does he? Rheumatol Rehabil 1979; 18: 43±8. 8. Stroke Unit Trialists Collaboration. Collaborative systematic review of the randomised trials of organised inpatient stroke unit) care after stroke. Br Med J 1997; 314: 1151±9. 9. Wardlaw JM, Yamaguchi T, del Zoppo G. Thrombolytic therapy versus control in acute ischaemic stroke Cochrane review). In Warlow CP, van Gijn J and Sandercock PAG eds. Cochrane Database of Systematic Reviews. Stroke Module. Oxford: Update Software/BMJ Publishing, Stewart JA, Dundas R, Howard RS et al. Ethnic differences in incidence of stroke: prospective study with stroke register. Br Med J 1999; 318: 967± Majeed F, Cook D, Poleniecki J, Martin D. Using data from the 1991 census. Br Med J 1995; 310: 1511± Mahoney FI, Barthel DW. Functional evaluation: the Barthel index. Md State Med J 1965; 14: 61± Bamford J, Sandercock P, Dennis M et al. Classi cation and natural history of clinically identi able subtypes of cerebral infarction. Lancet 1991; 337: 1521± Davenport RJ, Dennis MS, Warlow CP. Effect of correcting outcome data for case mix: an example from stroke medicine. Br Med J 1996; 312: 1503± Wade DT, Hewer RL. Hospital admission for acute stroke: who, for how long and to what effect? J Epidemiol Community Health 1985; 39: 347± World Health Organisation. Pan European Consensus Meeting on Stroke Management. Helsingborg, Sweden: World Health Organisation, Adgey AAJ, Crampton RS. Hospital or home for acute myocardial infarction: another look at whether or not we should bother to care. Am Heart J 1981; 102: 473± Wade DT, Langton Hewer R. Why admit stroke patients to hospital? Lancet 1983; 807± Brocklehurst JC, Andrews K, Morris P et al. Why admit stroke patients to hospital? Age Ageing 1978; 7: 100± Pound P, Bury M, Gompertz P, Ebrahim S. Stroke patients' views on their admission to hospital. Br Med J 1995; 311: 18± Lindley RI, Amayo EO, Marshall J et al. Acute stroke treatment in UK hospitals: the Stroke Association survey of consultant opinion. J Roy Coll Phys London 1995; 29: 525± Rudd AG, Irwin P, Rutledge Z et al. The national sentinel audit: an old tool for raising standards of care. J Roy Coll Phys London 1999; 33: 460± Ebrahim S, Redfern J. Stroke CareÐA Matter of Chance. A National Survey of Stroke Services Commissioned by the Stroke Association. London: Stroke Association, Davenport RJ, Dennis MS, Wellwood I, Warlow CP. Complications after acute stroke. Stroke 1996; 27: 415± Inredavik B, Bakke RPT, Slordahl SA et al. Treatment in a combined acute and rehabilitation stroke unit. Which aspects are most important? Stroke 1999; 30: 917± Sinyor D, Amato P, Kaloupek DG et al. Post-stroke depression: relationships to functional impairment, coping strategies and rehabilitation outcome. Stroke 1986; 17: 1102± Hillen T, Dundas R, Lawrence E et al. Antithrombotic and antihypertensive management 3 months after ischaemic stroke: a prospective study in an inner city population. Stroke 2000; 31: 469± Evans A, Perez I, Melbourn A, Steadman J, Kalra L. Alternative strategies in stroke: a randomised controlled trial of three strategies of stroke management and rehabilitation. Cerebrovasc Dis 2000; 10: Teasdale G, Jennett B. Assessment of coma and impaired consciousness: a practical scale. Lancet 1974; 2: 81± Hosmer DW, Lemeshow S. Goodness of t tests for the multiple logistic regression model. Communications Statistics 1980; A9: 1043±69. Received 4 April 2000; accepted in revised form15 January