Variations in Case Fatality and Dependency From Stroke in Western and Central Europe
Background and Purpose—There are significant variations in mortality rates from stroke in Europe. A European Union BIOMED Concerted Action was established to assess and determine the reasons for the variations in case fatality and disability after stroke.
Methods—Hospital-based stroke registers were established in 12 centers in 7 western and central European countries to collect demographic, clinical, and resource use details at the time of first-ever stroke during 1993–1994. At 3 months, details of survival, activity of daily living score, and use of health services were recorded. Multinomial logistic regression was used to estimate the relationship between centers and outcome (dead, functionally independent, functionally dependent), with adjustment for case mix and resource use variables, and to predict outcomes for the full cohort. This should minimize the bias due to loss to follow-up.
Results—A total of 4534 stroke events were registered. The mean age was 71.9 years (SD, 12.53). There were significant differences between centers for all case mix and resource use variables (P<0.001). Multinomial logistic regression modeling of outcome indicated that for those patients initially unconscious (588), center was not significantly related to outcome (P=0.427). For those initially conscious, there were wide variations in death and dependency between centers after adjustment for case mix, type of bed, and use of CT scan. The predicted proportion dead at 3 months ranged from 42% (95% CI, 35% to 49%) in one UK center to 19% (95% CI, 14% to 24%) in France.
Conclusions—Areas with high mortality rates within western and central Europe have been identified for stroke outcome, and there appears to be opportunity for considerable health gain in certain centers. Adjustment for case mix and health service resource use does not explain these differences in outcome. Although there are true differences in outcome, the aspects of care that need to be altered to improve outcome remain unclear despite detailed data collection. Comparisons of outcome of the same design used in the present study do not allow rational policy decisions to be made.
There are significant variations in mortality from stroke in those aged <65 years in Europe, a group in whom stroke has been considered avoidable.1 2 The Monitoring Trends and Determinants in Cardiovascular Disease (MONICA) study of stroke incidence in subjects aged <75 years demonstrated large differences in stroke incidence in Europe.3 Case fatality rates also varied from 15% to 49% at 28 days in men and from 18% to 57% in women. This 3-fold variation in case fatality may be influenced by differential rates of case ascertainment of stroke, by stroke subtype or severity, or may be a result of different methods of management of stroke. Variation in stroke management is evident, as reflected in the differences in the use of brain CT scans between the MONICA centers (0% to 76%).4
Comparison of hospital mortality rates, both crude and adjusted for case mix, for various conditions is politically attractive but fraught with difficulties of interpretation. Iezzoni et al5 caution against comparisons between hospital survival rates when the routine data may be poor. The same group also demonstrated that agreement in identifying the worst or best hospitals, using 11 non–stroke-specific scores to adjust for case mix, was only fair to good.6
A survey of consultant opinion in the United Kingdom demonstrated that well-organized stroke services occur irregularly and haphazardly, and an audit of stroke care in southern England identified poor adherence to stroke guidelines.7 8 These types of variation in incidence, management, and survival have to be borne in mind when one considers the guidelines the European Stroke Council has set for stroke services in the millenium and how they can be achieved in such diverse settings.9
Linking variation in the management of stroke to outcome is difficult, particularly when sites within 2 countries are compared or even when sites within a country are compared.10 A European BIOMED project involving 12 centers (22 hospitals) in 7 countries was established to address issues related to the resource use, cost, and outcome of stroke care.11 12
This study aims to assess the variations in mortality and disability 3 months after a first-in-a-lifetime stroke and to link these data with variations in the process of care. The study demonstrates the potential uses and abuses of such comparisons.
Subjects and Methods
A European Union BIOMED Concerted Action was initiated to establish the relationships between resource use, costs, and outcome of different packages of care for stroke. The specific objectives have been outlined previously.11 12 The project has involved 12 centers (22 hospitals) in 7 European countries, namely, England, France, Germany, Hungary, Italy, Portugal, and Spain. The hospitals are not necessarily typical of care in their country, but they all provide general acute care to the local population. The following centers were teaching hospitals: UK 1 to UK 4, Hungary, Germany 1, and Germany 2. UK 1 had a stroke rehabilitation ward, and the centers in France, Germany 1, and Hungary had acute stroke-monitoring facilities. The database and outcome scales were agreed on and discussed at 2 workshops before data collection commencement, and the study team visited each center to oversee data collection. A manual of definitions was distributed to all centers. The clinical data items were routinely collected in the centers and have been collected for population studies in Europe previously13 14 15 ; they consist of clinical assessments undertaken at bedside in each center. Issues regarding data collection, interpretation, and quality were discussed at the site visits and at 6 monthly meetings of the group during the 3 years of the project.
Hospital-based registers were established in September 1993 for 1 calendar year to collect patient-based data prospectively. Data collection related to first-ever stroke admissions according to the World Health Organization definition.13 Stroke-specific data questionnaires were developed, and well-validated case mix variables were collected by dedicated data collectors in each center10 ; these were developed from previous register questionnaires formulated by participants14 15 and were in agreement with those used by the MONICA Stroke Study.3
Baseline information used for these analyses included demographic factors (ie, age, sex, prestroke modified Rankin score,16 living condition before stroke); case mix, including risk factors at the time of maximum impairment within the first 7 days (level of consciousness subsequently dichotomized into 2 categories of coma or noncoma,17 site of plegia/paresis, speech or swallowing problems as a result of stroke, incontinence [including catheterization]); type of stroke (cerebral infarction or hemorrhage, subarachnoid hemorrhage); use of hospital resources (type of bed, eg, medical, neurological, intensive care, surgery, rehabilitation, private, other; length of stay in acute hospital wards); use of major diagnostic tests (brain imaging, angiography, carotid Doppler); and major therapeutic interventions (neurosurgery). The risk factors/comorbidities considered were hypertension, diabetes mellitus, previous stroke and transient ischemic attack, atrial fibrillation, and myocardial infarction. These risk factors were defined locally from history and examination and were in accord with the MONICA Project.3
At the 3-month follow-up, the patients were reassessed, usually by face-to-face interview, except at 1 UK center (UK 2) where follow-up was by mailed questionnaire. These data were supplemented with information from case notes and other routine hospital and general practice sources and from information from caregivers.
The living conditions at 3 months were noted (alone, with companion, institutionalized). The clinical state was reassessed in a manner identical to that of the initial assessment. The use of further major diagnostic tests (brain imaging, angiography, carotid Doppler, echocardiogram), rehabilitation, and visits to the physician/family doctor were documented. Outcome was assessed in terms of survival to 3 months (including whether stroke was the cause of death) and disability (Barthel and modified Rankin scales), which were validated by face-to-face interviews or mailed questionnaires.16 18 19
Analysis included univariate comparisons between centers on initial demography, case mix, type of bed, length of stay, use of brain imaging, and outcomes at 3 months. The univariate associations between centers and categorical variables were examined with the χ2 test. One-way ANOVA was used to investigate the univariate associations between centers and continuous variables.
Outcome at 3 months was defined as either dead, alive with a Barthel score of <20, alive with a Barthel score of 20 (independent), or unknown (including those known to be alive but with unknown Barthel score and those whose vital status was not known). Multinomial logistic regression was used to examine the variables associated with loss to follow-up in each center. Multinomial logistic regression was then used to examine the relationship between center and outcome (for those whose outcome was known), with adjustment for case mix variables, risk factors, which are also categorized as comorbidities (vide infra), length of stay (<3 or >3 weeks), and use of CT scan.
The case mix variables considered were age (years), sex, prestroke modified Rankin score, incontinence, and limb deficit at time of maximum impairment (none, deficit, paralysis). The risk factors/comorbidities considered were hypertension, diabetes mellitus, previous stroke or transient ischemic attack, atrial fibrillation, and myocardial infarction. These data on risk factors were collected by each center using local definitions from either history or examination.
Two separate modeling procedures were performed, one for those initially not in coma and one for those initially in coma. For those patients initially in coma, it is not possible to control for case mix variables such as swallowing and deficit because these cannot be assessed at examination. For those patients initially not in coma, the case mix variables included in all models were age, sex, prestroke Rankin score, incontinence at time of maximum impairment, and limb deficit at time of maximum impairment (none, hemiparesis, hemiplegia). An initial model was fitted, including the effect of center and interactions between center and all case mix/comorbidity variables. A separate model was fitted to the data from each center and used to predict outcome for all those in the entire initial data set (including all centers) who had been conscious initially. For those patients initially in coma, a separate multinomial logit model for each center was used to model the probabilities of each outcome. The model from each center was then used to predict outcome for those in the entire initial data set who had been in coma initially.
The bootstrap method (with 100 replications) was used to obtain the standard errors of the predicted probabilities of each outcome for each center and hence the confidence intervals.20 This procedure was then repeated, including length of stay and use of CT scan. It was again repeated with adjustment for prestroke risk factors/comorbidities.
For each center, the model relating case mix to outcome uses only those cases in which outcome is known. Using this model to predict outcome for the entire initial cohort assumes that the relationship between case mix and outcome is the same in those not followed up as in those followed up. Centers with a small number of cases followed up will tend to yield estimates with larger standard errors than those with a large number of cases followed up. Thus, follow-up rates should not bias the estimates but only affect their uncertainty. The prediction of outcome for each center is based on a model fitted to subjects with known outcome from that center only. Thus, the prediction for each center is not affected by any possible bias due to loss to follow-up in other centers.
A total of 4534 first-in-a-lifetime strokes were registered in the 12 centers. The number registered by each center at the time of stroke, the proportion who had died by 3 months, and the proportion followed up are detailed in Table 1⇓. The rates of loss to follow-up varied from 0% to 35%, with 6 centers having loss to follow-up rates of <10%.
Six of the centers had rates of loss to follow-up <10% (Table 1⇑). For the remaining 6 centers, multinomial logistic regression was used to examine the factors associated with whether the person was followed up alive, dead, or lost to follow-up. For all centers except Hungary and UK 5, some case mix variables were associated with the probability of being lost to follow-up. Increased chance of loss to follow-up increased with increasing age (Portugal), having a prestroke handicap (Portugal, Spain, Germany 1), being incontinent (Portugal, Germany 1), being continent (Spain), having no deficit after stroke (Spain, Germany 2), and being unconscious (Portugal).
The mean age of the entire cohort was 71.9 years (SD 12.53; range, 14 to 119 years), with significant variation between centers (P<0.001). Of the total of 4534 subjects, 2273 (50.13%) were female, with significant differences in the proportion of each sex between centers (P=0.014).
The differences in case mix variables between centers and use of key resources are described in Table 2⇓. There were significant differences between centers for all variables (P<0.001 in all cases). There were also obvious differences between centers in the use of medical and neurological beds. In UK 5, 40% of patients were in beds dedicated to care of the elderly. The living conditions and Rankin and Barthel scores at 3 months are shown in Table 1⇑. The proportion of patients living at home alone ranged from 7% to 57%.
Initial multinomial logit modeling of outcome (dead, Barthel <20, Barthel=20) considered case mix variables only. Initial unconsciousness was significantly associated with outcome in all centers. For those initially in coma (n=588), center was not significantly related to outcome (P=0.43). No case mix variables were significantly related to outcome in this group.
For those initially not in coma, the following variables were included in all models: age, prestroke modified Rankin score, sex, incontinence, and deficit. An initial model was fitted, including the effect of center and interactions between center and all case mix variables. Significant interactions existed between center and both incontinence and deficit. Because of this, a separate model was fitted to the data from each center for those initially not in coma. The models from each center (one for those in coma, one for those not in coma) were then used to predict outcome for the entire initial data set (Table 3⇓). This shows that the predicted death rate for the initial 4534 patients would range from 19% if they were all treated in Germany 2 to 39% if they were all treated in UK 5.
Further modeling, including length of stay and use of brain imaging, was undertaken to determine whether these aspects of service delivery affected outcome since they varied considerably between centers. For those unconscious at time of maximum impairment, no variables were significantly related to outcome. For those not in coma at the time of maximum impairment, the case mix variables used before were included in each model, together with brain imaging and length of stay (<10 days, 10 to 21 days, >21 days). Type of bed was considered a variable, but, because centers differed widely in the types of bed in which patients were treated, the variable could not be used uniformly across centers. Neither neurology bed (yes/no) nor stroke unit (yes/no) was significant for any center, and therefore this was not included in the model. The coefficients from the model for each center were estimated. The models from each center were then used to predict outcome for the entire initial data set (Table 4⇓). This shows that the predicted death rate for the entire 4534 patients would range from 19% if they were all treated in France to 42% if they were all treated in UK 5.
An additional model included the risk factors for stroke as well as the variables in the previous model. There were still significant variations in outcome, and the rank order for case fatality and disability did not change materially. The results are therefore not presented here.
This large European study has described significant differences in 3-month outcome in terms of case fatality or dependency for stroke that are unexplained by conventional case mix variable adjustment. The UK centers also appear to have consistently worse outcomes than the rest of Europe.
Data Collection and Case Mix
The study had the advantages of including >4500 patients in 22 units in 12 centers in 7 European states in 1993–1994. Data were prospectively collected by dedicated personnel in each center using a standardized questionnaire determined at the beginning of the project11 on the basis of previously published register questionnaires.14 15 There was regular 6-month feedback with centers at meetings to discuss problems with data collection and interpretation of questions, and a manual was written to aid standardization. The centers were self-selected and contained physicians or researchers with an interest in stroke. Because there is often only one center per country, the centers are not necessarily representative of their country, but we believe that they reflect the types of stroke care practiced in western and central Europe.
It was not possible to collect data on social class because the standardized collection and comparison across countries are problematic, particularly in the elderly. Any differences between centers and countries should ideally control for social class, which may be a confounding factor although not an excuse for increased mortality.
The data items collected on stroke at the time of maximum impairment have been used in many previous registers13 14 15 and are considered robust for stroke research, although between-center, interobserver studies have not been undertaken. The level of detail decided on was a balance between those factors known to predict outcome21 and the feasibility of data collection in 12 centers across Europe in a routine manner. Davenport et al22 discuss the confounding influence that variations in case mix exert on clinical outcome. They chose 19 different indicators of case mix that they considered of likely clinical importance, although the rationale for them is not validated. When their indicators are compared with those used in this study, there is broad agreement except that our study did not collect data on employment, poststroke hypertension, and subtype of stroke according to the Bamford classification. The study was able to adjust for case mix but unable to control for selection biases that may have varied between centers, particularly with respect to variations in the categorization of coma, incontinence, and paralysis.
The 3-month outcomes included case fatality after stroke, which was obtained at a local level, and there may have been differences in the postmortem validation of the cause of death. The 3-month time point necessitates the use of increased resources for follow-up, but it is biologically more logical and overcomes the in-hospital case fatality rate biases. Case fatality is only one outcome, and dependency should also be considered. The Barthel Index is the gold standard for stroke outcome research; it is easily administered by questionnaire either face to face, by telephone, or by mail.19
The follow-up rates were good in 6 centers, but several centers suffered from low follow-up rates, illustrating the practical difficulties of outcome assessment other than with the use of in-hospital case fatality. In 4 of the 6 centers with low follow-up rates, the probability of being followed up was associated with one or more case mix variables. The general pattern seemed to be that older people, people with a previous handicap, those with a deficit after stroke, or those with incontinence were more likely to be missed than followed up alive. However, only in Portugal was probability of follow-up associated with unconsciousness at baseline, which is the case mix variable strongly associated with outcome in all centers. Thus, there is evidence that loss to follow-up was not random in these 4 centers.
The important distinction when data are missing is between the case in which the probability of being missed depends on the actual outcome, even when case mix variables have been taken into account, and the case in which the probability of being missed is independent of the actual outcome once case mix variables have been taken into account. It is difficult to distinguish between these 2 cases unless further information on the missing data is available. However, it seems plausible that the relationship between outcome and case mix may be the same in those who are followed up as in those who are not followed up. If this is not the case for a given center, the only results to be biased will be the predicted outcome for that center only, since predictions for each center depend only on data from that center. Results from the 4 centers in question should be interpreted with this caution in mind.
The techniques for statistical analyses we used were not, however, dependent on 100% follow-up rates, and the only assumption made about the patients who were lost to follow-up was that the same relationship between case mix and outcome existed as in those who were followed up.
Comparisons Between Centers
In univariate analyses there were significant differences in initial stroke severity, indicating either different hospital admission policies, with some centers focusing on more severe strokes, or differences in the subtypes of stroke in each area. These variations in case mix make adjustment essential when outcomes are compared. There were significant differences in length of stay, type of bed used, and use of brain imaging. These variations have previously been reported, and after we controlled for case mix, there were unexplained increased lengths of stay in certain centers, particularly in the United Kingdom.11 The relationship between the type of bed and outcome is reasonably well established, with a stroke unit, however defined, reducing mortality rates.23 The use of brain imaging, although encouraged by stroke specialists, has not been shown to improve outcome in patients with stroke.
Multinomial modeling enables the outcomes to be adjusted for case mix and several resource use variables. In estimation of the model for each center, it is assumed that the relationship between case mix and outcome is the same in those followed up and those not followed up (see above). Those with missing data from the center of interest have been treated in the same manner as all cases from other centers, and the model from that center was used to predict their outcome. The model and predicted outcome for each center do not depend in any way on follow-up in any other center.
The assumption about the relationship between case mix and outcome in those not followed up obviously cannot be examined. However, even among those 6 centers with high follow-up rates, considerable variation in predicted outcome is apparent (Table 3⇑). This conclusion would be unaltered by any change in predicted outcomes in the other 6 centers if further information became available on those lost to follow-up. Since follow-up was unrelated to case mix variables in Hungary and UK 5, it seems likely that the predicted outcome for these centers will also be valid, given the assumptions about the relationship between case mix and outcome in those not followed up.
Thus, only the predicted outcomes for Portugal, Spain, Germany 1, and Germany 2 are potentially biased by unknown outcome for those not followed up. The relationship between case mix and outcome would have to be quite different in those not followed up to substantially change these results. However, even then, the order of the centers in terms of predicted outcome might change, but the fact that there is unexplained variation between the centers remains.
There appeared to be no differences in outcome between centers for those patients initially in coma, reflecting the generally poor outcome in this group of patients.21 For the majority of patients who did not lose consciousness, the study shows significant differences in death and dependency unexplained by case mix in these 12 centers, and the size of the difference is unlikely to be explained by chance, bias, and confounding, as previously explained. This is a real effect, but perhaps we have not controlled for all possible confounding variables. Obvious exclusions from the data collection include social class, subtype of stroke, more detailed case mix variables, and measures of comorbidities that may affect outcome. However, the study was large enough for us to be certain that these results are a true representation of care in Europe, with carefully collected data from each center and with well-validated case mix and comorbidity variables. The addition to the analyses of risk factors such as hypertension, atrial fibrillation, and previous transient ischemic attack did not alter the findings, and it could be argued that further control for confounders will not substantially alter these dramatic findings.
Variations in mortality and morbidity rates between centers and countries for certain conditions are well documented and used by healthcare planners and politicians either to set targets for their reduction or to conduct a confidential inquiry to explore the reasons for high rates and consequently to improve health and alter case fatality.24 Similarly, the World Health Organization has set targets for the reduction of case fatality for stroke by the year 2005 without clearly indicating what interventions will effectively enable all countries and hospitals to achieve these targets.25
In the United Kingdom, league tables for in-hospital death rates have been published in Scotland, but these are difficult to interpret because case mix is not considered adequately and in-hospital death rates are confounded by hospital discharge policies.22 Mant and Hicks26 have discussed the advantages and disadvantages of publishing hospital-specific death rates. They advocate using process measures based on the results of randomized controlled trials that are able to detect differences between hospitals that would not be identified by comparing hospital-specific mortality. The acute interventions for stroke are, however, very limited.
The United Kingdom has a worse outcome than the rest of Europe in terms of mortality rates, and these data are difficult to explain. One feature that recurred during discussions was the low intervention rates of English physicians for controlling abnormal physiology, for example, high blood pressure and blood glucose levels in the acute phase of stroke and the use of supposedly disease-modifying drugs. Another component of care not investigated in this study is the delay from stroke onset to hospital admission. However, there is no good evidence to suggest reduced delay to admission, and management of these factors in the acute phase alters outcome. The data from this study do show striking differences, and it is difficult to know how to use such data. It is also difficult to know which aspect of acute care needs to be altered to improve outcome or what other case mix variables need to be included in such data sets to reduce the chance of bias, particularly since the evidence base for acute interventions is not strong. Ongoing research in the BIOMED program is addressing links between resource input, case mix, acute management, and outcome in a more detailed manner.
In conclusion, this study has compiled detailed data on first-in-a-lifetime stokes in 12 centers in Europe and has shown, after adjustment for case mix, significant differences in outcome in patients initially conscious after their stroke. Areas with high mortality rates within Europe have been identified, but the course of action to improve outcome in these centers remains in doubt. There appears to be opportunity for considerable health gain in certain centers, particularly the United Kingdom. The study cautions against comparisons of this nature, which, although leading to further understanding of the relationship between case mix, resources, and outcome, do not allow rational policy decisions to be made.
D.H. Barer, Y. Ellul, Department of Medicine for the Elderly, Newcastle General Hospital, England; S. Ebrahim, M. Ayana, P. Gompertz, R. Harwood, P. Pound, Department of Primary Care and Population Sciences, Royal Free Hospital School of Medicine, London, England; H. Rogers, Center for Health Service Research, University of Newcastle, England; A. Rudd, Department of Care of the Elderly, Guy’s & St Thomas’ Hospital, London, England; M. Giroud, M. Menassa, M. Lemesle, Service de Neurologie, Center Hospitalier Regional et Universitaire de Dijon, France; K. Kunze, Neurologischen Universitatsklinik, Hamburg-Eppendorf, Germany; J. Berger, Institute of Mathematics and Computer Science in Medicine, University Hospital Eppendorf, Germany; B. Haussler, W. Mall, H. Nolting, Institut fur Gesundheits und Sozialforschung GmbH (IGES), Berlin, Germany; Z. Nagy, C. Ovary, Z. Vokoq, National Stroke Center, Budapest, Hungary; D. Inzitari, A. Di Carlo, M. Lamassa, P. Vanni, G. Pracucci, S. Spolveri, G. Landini, F. Cordopatri, L. Bagnoli, Dipartimento di Scienze Neurologiche & Psichiatriche, Ospedale Careggi, Italy; M. Carrageta, J. Namora, I. Remidios, A. Santos, Dr J. Coisinha, Hospital Garcia de Orta, Almada, Portugal; J. Dias, Divisao de Epidemiologia, Direccao Geral de Saude, Lisboa, Portugal; A. Arias, P. Casquero, S. Montserrat, M. Torrent, Direccion Provincial INSALUD, Gabinete de Estudios, Palma de Mallorca, Spain.
This study was supported by the European Union Biomed I program and Glaxo Wellcome (K.T.). We thank C. Price for manuscript preparation.
Reprint requests to Dr C. Wolfe, Department of Public Health Sciences, The Guy’s, Kings College, and St Thomas’ Hospital Medical and Dental School, Guy’s Campus, London SE1 9RT, UK.
A list of study participants is given in the Appendix.
- Received August 18, 1998.
- Revision received November 2, 1998.
- Accepted November 3, 1998.
- Copyright © 1999 by American Heart Association
Holland WW, project director. European Community Atlas of Avoidable Death. Vol 1. 2nd ed. Oxford, UK: Oxford University Press; 1991.
Annual World Health Statistics. Geneva, Switzerland: World Health Organization; 1990, 1991, 1992.
Thorvaldsen P, Asplund K, Kuulasmaa K, Rajakangas AM, Schvoll M. Stroke incidence, case fatality and mortality in the WHO MONICA Project. Stroke. 1995;26:361–367.
Asplund K, Bonita R, Kuulasmaa K, Rajakangas AM, Schaeduch M, Suzuki K, Thoraldsen P, Tuomilehto J. Multinational comparisons of stroke epidemiology. Stroke. 1995;26:355–360.
Iezzoni LI, Shwartz M, Ash AS, Mackiernan YD. Predicting in-hospital mortality for stroke patients: results differ across severity-measurement methods. Med Decis Making. 1996;16:348–356.
Iezzoni I, Schwartz M, Ash A, Hughes J, Daley J, Mackiernan Y. Using severity-adjusted stroke mortality rates to judge hospitals. Int J Qual Health Care. 1995;7:81–94.
Lindley RI, Amayo EO, Marshall J, Sandercock PAG, Dennis M, Warlow CP. Hospital services for patients with acute stroke in the United Kingdom: the Stroke Association survey of consultant opinion. Age Ageing. 1995;24:525–532.
Wolfe CDA, Stojcevic N, Rudd AG, Warburton F, Beech R. The uptake and costs of guidelines for stroke in a district of Southern England. J Epidemiol Community Health. 1997;51:520–525.
Gompertz P, Pound P, Brittan J, Ebrahim S. How useful are non-random comparisons of outcomes and quality of care in purchasing hospital stroke services? Age Ageing.. 1995;24:137–141.
Beech R, Ratcliffe M, Tilling K, Wolfe C. Hospital services for stroke care: a European perspective. Stroke. 1996;27:1958–1964.
Wolfe CDA, Taub NA, Woodrow J, Richardson E, Warburton FG, Burney PGJ. Does incidence, severity, or case fatality of stroke vary in southern England? J Epidemiol Community Health. 1993;47:139–143.
Giroud M, Milan C, Beuriat P, Gras P, Essayagh E, Arveux P, Durnas R. Incidence and survival rates during a two year period of intracerebral and subarachnoid haemorrhages, cortical infarcts, lacunes and transient ischaemic attacks: the stroke registry of Dijon: 1985–89. Int J Epidemiol. 1991;20:892–899.
Mooney CZ, Duval RD. Bootstrapping: A Non-Parametric Approach to Statistical Inference. Newbury Park, Calif: Sage Publications; 1993.
Taub NA, Wolfe CDA, Richardson E, Burney PGJ. Predicting the disability of first-time stroke sufferers at 1 year: 12 month follow up of a population-based cohort in southeast England. Stroke. 1994;25:352–357.
Davenport RJ, Dennis MS, Warlow CP. Effect of correcting outcome data for case-mix: an example from stroke medicine. BMJ. 1996;312:1503–1505.
Department of Health. Health of the Nation: a Strategy for Health in England (Cm 1986). London, UK: HMSO; 1992.
World Health Organization. Pan-European Consensus Meeting on Stroke Management. Helsingborg, Sweden: World Health Organization; 1995.
Mant J, Hicks N. Detecting differences in quality of care: the sensitivity of measures of process and outcome in treating acute myocardial infarction. BMJ. 1995;311:793–796.