Predicting Survival for 1 Year Among Different Subtypes of Stroke. Results From the Perth Community Stroke Study

Transcription

1 935 Predicting Survival for Year Among Different Subtypes of Stroke Results From the Perth Community Stroke Study Craig S. Anderson, MBBS, BMedSc, FRACP; Konrad D. Jamrozik, MBBS, DPhil, FAFPHM; Robyn J. Broadhurst, BA, BSc; Edward G. Stewart-Wynne, MBChB, FCP(SA), FRACP Downloaded from by on August 2, 28 Background and Purpose Few studies have evaluated the factors influencing or predicting long-term survival after stroke in an unselected series of patients in whom the underlying cerebrovascular pathology is clearly defined. Moreover, the relative importance of risk factors for stroke, including sociodemographic and premorbid variables, has not been described in detail. Methods The study cohort consisted of 492 patients with stroke who were registered with a population-based study of acute cerebrovascular disease undertaken in Perth, Western Australia, during an 8-month period in 989 and 99. Objective evidence of the pathological basis of the stroke was obtained in 86% of cases, and all deaths among patients during a follow-up of year were reviewed. Results One hundred twenty patients (24%) died within 28 days of the onset of stroke. Among the different subtypes of stroke, the -year case fatality (mean, 38%) varied from 6% and 6% for boundary zone infarction and lacunar infarction, respectively, to 42% and 46% for subarachnoid hemorrhage and primary intracerebral hemorrhage, respectively. Using Cox proportional-hazards analysis, a predictive model was Renewed interest in case fatality and the predictors of death from stroke can be traced to several sources. Information on survival after stroke, as well as data on those factors that determine a high or low likelihood of recovery, helps to identify those patients who may benefit more than others from specific medical treatment and rehabilitation services and from the counseling of family members. Increased financial constraints on the healthcare system underscore the importance of such information in developing cost-effective ways of providing both short-term treatment and long-term care for patients with stroke. Finally, interest in the natural history of stroke has been rekindled by the advent of potentially effective treatments for acute ischemic stroke, which are now undergoing evaluation in clinical trials. Received March 28, 994; final revision received June 24, 994; accepted July 5, 994. From the Department of Medicine, Flinders University of South Australia (C.S.A.); the Department of Public Health, University of Western Australia, Nedlands (K.D.J., R.J.B.); and the Stroke Unit, Department of Neurology, Royal Perth Hospital (Western Australia) (E.G.S.-W.). Reprint requests to Dr Craig S. Anderson, Department of Medicine, Flinders University of South Australia, Bedford Park, South Australia American Heart Association, Inc. developed on 32 patients with acute stroke (test sample). The best model contained five baseline variables that were independent predictors of death within year: coma (relative risk [RR], 3.; 95% confidence interval [CI],. to 8.4), urinary incontinence (RR, 3.9; 95% CI,.4 to.6), cardiac failure (RR, 6.5; 95% CI, 2.8 to 5.), severe paresis (RR, 4.9; 95% CI,.6 to 5.5), and atrial fibrillation (RR, 2.; 95% CI,. to 3.5). The sensitivity, specificity, and negative predictive value of this model for predicting death were 9%, 83%, and 95%, respectively. When applied to a second randomly selected validation sample of 7 events, sensitivity was 94%, specificity 62%, and negative predictive value 92%, indicating stability of the model. Conclusions Although the case fatality, timing, and cause of death vary considerably among the different pathological subtypes of stroke, simple clinical measures that reflect the severity of the neurological deficit and associated cardiac disease at onset independently predict death by year and may help to direct management. (Stroke. 994;25: ) Key Words Australia epidemiology prognosis stroke outcome Although the literature on survival after stroke is extensive, particularly with regard to the acute period, the majority of studies are based on selected series of patients referred to the hospital or for rehabilitation. 29 Although these studies are able to describe the characteristics and outcome of pathologically homogeneous stroke entities, such patients are often not representative of all individuals who suffer a stroke. Conversely, few population-based studies have been able to define accurately the natural history of well-defined pathological subtypes of stroke 2 and thus provide the information needed to plan clinical trials of specific treatments for particular subtypes of stroke. While the apparent significance of prognostic factors identified by univariate analyses has been criticized, the practical utility of several multivariate computer models has also been questioned. Few models have been tested in an independent sample, 46 and even then they appear to be only slightly better than univariate analyses at predicting survival. Since complex systems that rely on "black-box" calculations are often too difficult and too time-consuming for routine clinical use, it has been argued that clinicians should only use simple clinical variables to help predict outcome when making decisions about individual patients. In the acute phase, the chance of surviving a stroke is generally a complex function of the extent of neurolog-

2 Downloaded from by on August 2, Stroke Vol 25, October 994 ical damage and the complications of stroke-related disability. After this time, systemic factors (in particular, the extent of atherosclerosis) assume greater importance. 7 However, other physiological, functional, and social factors might also influence survival after stroke and other diseases. 8 " 2 It is only recently that these factors have been considered, 2 even though they may have confounded other studies that have emphasized the adverse effect of increasing age on survival after stroke. 28 Drawing on a large population-based study of patients in whom the pathological subtype of stroke had been defined accurately, the present investigation examined the effects of sociodemographic variables, risk factors for cerebrovascular disease, various clinical features, and premorbid levels of disability and social activity on the frequency and causes of death in the year after stroke. Subjects and Methods Overall Study Design The data analyzed and reported here are from the Perth Community Stroke Study. The protocol of this study has been described in detail elsewhere 22 and was approved by the human rights and ethics committee at each participating institution. In brief, we registered all possible episodes of acute cerebrovascular disease that occurred between February 2, 989, and August 9, 99, inclusive, in a geographically defined segment of Perth, Western Australia. Based on the Australian Bureau of Statistics 986 Census, the estimated population of the study area (on June 3, 989) was persons (69 8 males and 69 7 females). Comparison with census figures for the remainder of Perth suggests that the study population contained proportions that were slightly higher than average for elderly persons and those born overseas, particularly from southern European countries. Otherwise, the socioeconomic characteristics and patterns of admission to the hospital and length of time in the hospital for stroke were generally representative of those of the whole city (total population,.2 million). Every effort was made to ensure complete ascertainment of cases whether patients were managed at home or admitted to hospitals either within Perth or elsewhere if temporarily resident outside the study area. Cases were detected from a variety of overlapping sources that included notifications from general practitioners within the study area; scrutiny of attendance at and admissions to all acute-care hospitals, rehabilitation centers, and nursing homes within and around the study area; and prospective surveillance of hospital morbidity data and of requests for computed tomographic (CT) scans and carotid duplex ultrasound scans at Royal Perth Hospital and Sir Charles Gairdner Hospital, two acute-care public teaching hospitals that lie within and adjacent to the study area, respectively. Additional surveillance activities included reviews of all reports by the Perth Coroner and the Department of Neuropathology at Royal Perth Hospital, the major referral center for neuropathology analysis in the state, for residents of the study area. By arrangement with the Registrar-General for Births, Deaths, and Marriages, the Perth office of the Australian Bureau of Statistics provided the University Department of Public Health with a computer disk containing all deaths registered in the state each month. Additional cases were sought by scanning each disk electronically for deaths that occurred during the study period and for key words such as "stroke," "cerebral," and "cerebrovascular accident" in the multiple causes of death entered by a doctor for a patient resident within the study area. Final attempts to ensure completeness of ascertainment were augmented by retrospective checking of computerized hospital discharge data. Western Australia is unique among the Australian states in having had, since 972, a computerized data system that covers all inpatient separations (discharges, transfers, and deaths) from every hospital in the state. Each entry includes the date of admission and separation, multiple diagnostic codes, codes identifying both the hospital and the doctor involved in each case, and the name and address of the patient. From a total of 883 possible acute cerebrovascular events registered in the Perth Community Stroke Study, a final diagnosis of stroke was made in 536. The remainder included 23 transient ischemic attacks and 224 events that were considered to be noncerebrovascular in etiology. This report describes the outcome of the patients with events registered with a final diagnosis of stroke. Definitions Stroke was defined according to World Health Organization (WHO) criteria as "rapidly developing symptoms and/or signs of focal, and at times global, loss of cerebral function, with symptoms lasting more than 24 hours or leading to death with no apparent cause other than that of vascular origin." 23 The term "global" refers mainly to subarachnoid hemorrhage. Each event was classified as being the patient's first-ever-in-alifetime (first-ever) stroke or a recurrent event. For patients with multiple events, the index event was defined as that event which occurred nearest to the time when the patient was first registered. A case "managed in hospital" is one in which admission occurred within 28 days of the onset of stroke. Twenty-eight-day and -year case fatalities were defined as the proportion of all events resulting in death within 28 days and 365 days of onset, respectively. In addition, we prospectively applied criteria that combined clinical data with the results of brain imaging or necropsy, so that strokes could be classified into several "pathologically distinct" subtypes. 24 "Appendix A" summarizes the criteria for these specific diagnoses, which include types of cerebral infarction, namely, large-artery occlusive, embolic, lacunar, and boundary zone infarcts; primary intracerebral hemorrhage; and subarachnoid hemorrhage. In cases in which there was neither brain imaging nor necropsy, the stroke was defined as being of undetermined type. Baseline Assessment All cases meeting the clinical criteria for inclusion were reviewed by a single investigator (C.S.A.), who conducted a standardized interview and physical examination. Those patients who were not personally assessed were classified as cases on the basis of the history and neurological examination obtained from medical records. After each patient (or closest relative) had given consent, information obtained at baseline included data on associated illnesses, risk factors for cardiovascular disease ("Appendix B"), and the patterns of disability and social activity in the immediate premorbid period. If the patient was unconscious or otherwise not assessable, information was obtained from the patient's spouse or another reliable proxy, hospital records, or the general practitioner or from the necropsy report when this was the only information available. The physical signs recorded for each patient at the onset of stroke included an assessment of the level of consciousness, the severity of limb paresis, and the presence or absence of urinary incontinence, cardiac failure, and atrial fibrillation. Level of consciousness at the time of presentation to an attending medical officer was estimated from examination of the patient or by extrapolation of data from the medical records and was measured by means of the Glasgow Coma Scale. 25 This scale examines motor function, verbal responses, and eye opening and results in a score from 3 to 5. We defined a score of 3 to 9 as "comatose," to 4 as "drowsy," and the top score of 5 as "normal." The presence or absence of paresis was evaluated in all patients. However, the severity of paresis was only measured in those patients assessed by the

3 Anderson et al Predictors of Survival After Stroke 937 Downloaded from by on August 2, 28 first author within 2 weeks of the onset of stroke using the Motricity Index. 26 This index results in a score from (total paralysis) to (normal); a score of to 5 was defined as "severe paresis," 5 to 94 as "moderate paresis," and a top score of 95 to as "normal or minimal paresis." Those patients not assessed personally soon after onset, either because they died rapidly or were entered into the study late, were assumed to be plegic if data indicated that they were comatose at onset. Information on urinary incontinence was obtained from medical records or by asking nursing staff or relatives. Patients were assessed as incontinent if they had accidents, needed help, or needed an indwelling catheter during admission to the hospital. The diagnosis of cardiac failure was made in the presence of cardiomegaly, gallop rhythm, raised jugular venous pressure, and pulmonary edema that was confirmed on chest radiography. Atrial fibrillation must have been confirmed on an electrocardiogram within month after the onset of stroke. Hypertension after onset was defined according to the levels of blood pressure outlined in "Appendix B." Premorbid level of disability was assessed with the modified scaling of the Barthel Index of activities of daily living (ADL) The Barthel Index is a quick, reliable, and wellvalidated measure of mobility and ADL that gives a score between and 2 in -point increments. The top score of 2 implies "independence" but not necessarily normal living. Patients were arbitrarily divided into three grades according to their usual level of function during the month before the stroke. Scores of to 9 and to 9 were defined as "severe" and "moderate" dependency, respectively. Information on each item of ADL was first sought by self-report, but allowance was made for those patients who were confused or otherwise unassessable by asking a nurse or relative. Premorbid social function was assessed in the same way by means of the Frenchay Activities Index, 29 which is based on the frequency with which a patient performs 5 separate activities (for example, shopping, reading books, cooking) and gives a total score between 5 (no activities) and 6 (full activities). Social function was graded according to scores of 5 to 29 as "inactive," 3 to 44 as "active," and 45 to 6 as "very active." Follow-up Procedures The same investigator who had made the baseline assessments prospectively followed up patients until 2 months after their index event or the time of death if it occurred before that date. All surviving patients in whom the final diagnosis was a stroke and who were not known to have died were contacted again at 4 and 2 months. Patients who agreed to be reinterviewed were generally seen in their own home or usual place of residence. All deaths were confirmed from official mortality data supplied by the Perth office of the Australian Bureau of Statistics. The causes of death are reported for year after the patients' index stroke registered in the study. After review of all available clinical information, investigations, and the findings at necropsy (if performed) and of the patient's death certificate, one of us (C.S.A.) classified the cause of each death using the diagnostic criteria outlined in "Appendix C." Statistical Analysis The principal aim of the statistical analysis was to develop a model for predicting survival during year after acute stroke. The x 2 test was used to evaluate the crude associations between death and categorical baseline variables. Variables associated with death in the univariate analyses (P<2) and present in at least % of events or judged to be clinically relevant in their own right, possible confounders, and interaction effects within the data were assessed in a multivariate model. The Kaplan-Meier method was used to establish survival curves for the different subtypes of stroke, and these were compared with the log-rank test. A split-sample technique 3 was used for internal validation of the model; two thirds of the cohort were randomly selected to form a test sample. Independent predictors of death at year were identified from these data by means of the Cox proportionalhazards technique and forward stepwise regression (EGRET software 3 ). The model was then applied to data from the remaining third of the cohort (validation sample) to examine its stability. The final model was chosen based on predictive ability and simplicity. Data are reported with 95% confidence intervals (CI), and probability values are two-tailed. Results Characteristics of the Study Population The study cohort consisted of 492 patients with acute stroke (52% men) aged 3 to 96 years (mean±sd age, 73 ±3 years; median, 76 years) who were followed up for year. Three hundred fifty-three patients (72%; 95% CI, 67% to 77%) had experienced their first-ever stroke during the 8-month study period. The majority of patients (83%) were assessed personally by C.S.A., with a median delay of 5 days after the onset of symptoms. Nineteen percent (95% CI, 6% to 22%) of patients were managed entirely outside the hospital during the acute phase. CT scans were performed after 375 events (76%), and magnetic resonance imaging and necropsy were used to document the underlying stroke pathology in 9 (4%) and 3 (6%) index events, respectively. Two-dimensional echocardiography and cervical Doppler studies were each performed on 7% of patients. Another 5% of patients underwent cerebral angiography, and the final diagnoses in these were subarachnoid hemorrhage (4), primary intracerebral hemorrhage (4), and cerebral infarction (7). On the basis of the clinical history and the results of these investigations, 86% of events had a defined "pathological" diagnosis for the stroke. Cerebral infarction accounted for 7% of all strokes (95% CI, 67% to 75%): large-artery (occlusive) stroke in 5%, embolic in %, lacunar in 6%, and boundary zone infarction in 4%. The remainder included primary intracerebral hemorrhage in % (95% CI, 8% to 4%) and subarachnoid hemorrhage in 4% (95% CI, 2% to 6%). In 4% the pathological basis of the stroke was undetermined. Patients in this category were more likely to be elderly (>85 years), female, dependent, inactive, and resident of a nursing home before the onset of stroke. Deaths patient was lost to follow-up. The overall casefatality rate at 28 days was 24% (95% CI, 2% to 28%) and at year was 38% (95% CI, % to 42%). The in-hospital case-fatality rate was 27% (95% CI, 23% to 3%). Table shows the case-fatality rates at 28 days and year for the different subtypes of stroke. Fig shows the survival curves for each subtype of stroke. The prognoses for lacunar infarction and boundary zone infarction were each significantly different from those for the other subtypes of stroke (log-rank statistic, P<.l). Table 2 shows the underlying causes of death in the 88 patients who died during the year of follow-up after their index event. Necropsy was performed in 64 (%) of patients. Fourteen patients died after recurrent strokes, and death was the direct result of the stroke in 2 of these cases. Overall, 5 (8%) and 4 (22%) deaths were due to coronary heart disease and pneumonia,

4 938 Stroke Vol 25, October 994 TABLE. Case-Fatality Ratios for the Subtypes of Stroke at 28 Days and Year for the Study Area, Perth, Western Australia, Dead at 28 Days Dead at Year Subtype and Sequence of Stroke. % 95% Cl. % 95% Cl First-ever stroke Cerebral infarction (n=247) Large-artery occlusion (n=67) Embolic (n=46) Lacunar (n=22) Boundary zone (n=2) Primary intracerebral hemorrhage (n=44) Subarachnoid hemorrhage (n=8) Undetermined (n=44) Total (n=353) Downloaded from by on August 2, 28 Recurrent stroke Cerebral infarction (n=2) Large-artery occlusion (n=78) Embolic (n=9) Lacunar (n=9) Boundary zone (n=6) Primary intracerebral hemorrhage (n=2) Subarachnoid hemorrhage (n=) Undetermined (n=24) Total (n=39) Cl indicates confidence interval respectively. Eleven deaths (6%) were secondary to cancer: bowel (2), lung (2), pharynx (), brain (), prostate (), and unknown primary (). There were only two confirmed cases of pulmonary embolism. Fig 2 shows the immediate cause of death during year among the major subtypes of stroke. The stroke directly caused the death in the great majority of patients who died within the first week of onset. The causes of death in each of the 5 cases of lacunar infarction were myocardial infarction (2), septicemia ?n - - Nv i. \ Boundary zone infarction : Lacunar infarction ^. Large artery infarction Subarachnoid haemorrhage Primary intracerebral haemorrhage Months FIG. Line graph shows Kaplan-Meier curves for death during year after onset according to the different subtypes of stroke (), acute gastrointestinal hemorrhage secondary to bowel cancer (), and lung cancer (). The single death after a boundary zone infarct was the result of pneumonia in a patient with premorbid dementia. Predictors of Survival Among the 492 patients in this study, 235 (48%) were either dead or had impaired communication due to aphasia or altered consciousness at the time of baseline assessment. In these cases, data concerning the onset of the stroke and other medical and social factors were obtained from a close relative or friend and from medical records. Data on one patient were obtained by monitoring his hospital separation record, but he refused to be interviewed. Medical records and/or death certificates were used as the sole source of information on the study variables in only 2 (4%) patients. Data on a history of claudication could not be obtained from % of patients, but information on other cardiovascular risk factors was missing in less than 2% of patients. Missing data from the physical examination at baseline included the severity of paresis (in 8%), hypertension after onset (8%), cardiac failure (7%), atrial fibrillation (5%), level of consciousness (6%), and incontinence (5%). The distribution of age, sex, and cardiovascular risk factors for these patients was similar to that in those patients with complete data. However, patients with missing data on physical signs were significantly more likely to have died within the first 4 days after onset, to have been inactive or dependent, and to

5 Anderson et al Predictors of Survival After Stroke 939 Downloaded from by on August 2, 28 TABLE 2. Causes of Death in 88 Patients Who Died Within Year of Acute Stroke Cause of Death Stroke. 9 % 58. Index event Recurrent event Cardiac Sudden death Chronic coronary artery disease Acute myocardial infarction Infective endocarditis Ruptured aorta* Infection Pneumonia Septicemia Peritonitis Pseudomembranous colitis Cancer Pulmonary embolism Renal failure Gastrointestinal hemorrhage (unknown cause) *This patient died after a spontaneous rupture of the thoracic aorta related to shrapnel embedded during World War II. have been managed outside of the hospital compared with other patients (all P<.5). Of the 492 events, 32 were randomly assigned to the test sample and the remainder to the validation sample. Table 3 shows that the characteristics of the two samples were similar. The factors found on univariate analysis to be associated with -year mortality were increasing age; subtype of stroke; coma, severe paresis, incontinence, cardiac failure, hypertension, and atrial fibrillation at onset; "unpartnered" marital status, living alone, and living in an institution at onset; previous stroke; history of claudication; premorbid disability; and premorbid social inactivity. In the stepwise Cox proportional-hazards analysis, the most parsimonious multivariate model associated with -year mortality in the test sample included the following factors: coma (relative risk [RR], 3.; 95% CI,. to 8.4), cardiac failure (RR, 6.5; 95% CI, 2.8 to 5.), urinary incontinence (RR, 3.9; 95% CI,.4 to.6), severe paresis (RR, 4.9; 95% CI,.6 to 5.5), and atrial fibrillation (RR, 2.; 95% CI,. to 3.5) at onset (Table 4). Whether a given stroke was a first-ever or recurrent event did not carry independent prognostic significance. We created a second model restricted to those patients who survived to 4 days after onset because conscious level was so powerful a predictor in the first model. These same variables were in the second model, with slight alterations in their hazard ratios. In view of the high case fatality associated with primary intracerebral hemorrhage and subarachnoid hemorrhage and the relatively large number of cases of undetermined stroke, a third model was constructed confined to cases of cerebral infarction (n=224) from the test sample. Coma (RR, 2.6; 95% CI,.8 to 8.3), cardiac failure (RR, 6.7; II ^^_ Cerebral infarction (n=97) Primary intracerebral haemorrhage (n=26) Subarachnoid haemorrhage (n=8) UIIUCL&III Undetermined stroke (n=s7) week(5) Index stroke I I Recurrent stroke ED Coronary heart disease E2 Pneumonia EZ3 Pulmonary embolism Other causes months FIG 2. Bar graph shows timing and causes of death according to the major subtypes of stroke. 95% CI, 2.4 to 8.8), incontinence (RR, 7.3; 95% CI, 2.5 to 2.3), and atrial fibrillation (RR, 2.3; 95% CI,. to 5.) were retained in the model. However, paresis was replaced by unpartnered marital status (RR, 2.42; 95% CI,. to 5.4) and a history of claudication (RR, 2.2; 95% CI,. to 4.7) to form the most parsimonious model. These six variables remained independent predictors of death during year when the analysis was further restricted to the subgroup of patients with cerebral infarction who survived to day 4 after onset. Validation Sample Table 5 demonstrates the stability of these models. Coma, cardiac failure, incontinence, severe paresis, and atrial fibrillation had suitable sensitivity and specificity as predictors of death among all strokes when any one was used as a dichotomized variable. Conversely, patients who presented as alert and continent and had mild to moderate paresis and no evidence of cardiac failure or atrial fibrillation had a very low likelihood of death during the first year after onset. Among patients with cerebral infarction, the specificity and positive predictive values of the variables were reduced because the overall case-fatality rate is lower in this subgroup. Conversely, the negative predictive value for the variables remained high. Discussion This analysis of the outcome of acute stroke differs from most others for several reasons. First, we included all cases of stroke occurring in a defined population. Selection bias is the most obvious explanation for differences in case-fatality ratios for stroke between hospital and population-based series, the former emphasizing more severe cases but also excluding many patients with hemorrhagic stroke who die rapidly before reaching the hospital The variable case mix of other studies often limits the degree to which the information

6 94 Stroke Vol 25, October 994 Downloaded from by on August 2, 28 TABLE 3. Selected Baseline Variables for Patients With Stroke* Variables, % Sociodemographic Age, yt Male sex Lives alone Institutional caret Living with a partner Premorbid dependence (ADL) Premorbid level of social activity! Very active Active Inactive Subtype of stroke Cerebral infarction Primary intracerebral hemorrhage Subarachnoid hemorrhage Undetermined Medical historyll Previous stroke History of hypertension History of angina History of myocardial infarction History of claudication History of diabetes mellitus Test Sample (n=32) 75 (6) Validation Sample (n=7) 77 (8) ADL indicates activities of daily living. *Data for missing or unknown values not included. tmedian (Q3-Q), where Q3-Q is a measure of the spread of the distribution representing the difference between the values of the 75th (quartile 3) and the 25th (quartile ) centiles. tdefined as residence with supervised care in a hostel or nursing home or hospital. Physical disability estimated by the Barthel Index of ADL. Social activity estimated by the Frenchay Activities Index. ^Explanations of definitions and categories are provided in "Subjects and Methods." # Level of consciousness was assessed by the Glasgow Coma Scale. "Severity of paresis was assessed by the Motricity Index on 355 patients (72%) seen by the first author within 4 days of a stroke. Another 5 patients (%) who died rapidly or were entered into the study late were assumed to be plegic because they were comatose at onset. Variables, % Smoking status Never Ex-smoker Current Clinical features at onset Level of consciousness* rmal Drowsy Comatose Severity of paresis** rmal/mild Moderate Severe Urinary incontinence Cardiac failure Atrial fibrillation Hypertension Test Sample (n=32) Validation Sample (n=7) obtained can be applied to the population from which it was derived. Second, whereas CT scanning, and other diagnostic investigations were: not routinely ; available in many early population-based studies, 2 we were able to classify "objectively" almost iill of our cases into one of several distinct pathological subtypes of stroke. Some previous studies used a clinical scoring system that appears more accurate than an unstructured clinical assessment in defining the type of stroke, 2 but many small hemorrhagic strokes may have been misdiagnosed as cerebral infarction, raising questions about the apparent prognosis of the hemorrhagic stroke in these populations. Selection bias, interobserver variation, and measurement error may influence the factors that ultimately enter a multivariate model and thus the predictive index. These problems were minimized in the present study because a single observer applied well-standardized and validated definitions and instruments to most patients early after the onset of stroke, and follow-up was both prospective and complete. However, patients were not assessed at uniform time intervals during the hyperacute phase of stroke, when the impact of any therapy for ischemic stroke may be greatest. Thus, the

7 Anderson et al Predictors of Survival After Stroke 94 TABLE 4. Characteristics Associated With -Year Mortality in the Test Sample Variable All strokes (n=32) ^-Coefficient SE RR 95% Cl Level of consciousness Drowsy Comatose Cardiac failure Urinary incontinence Severity of paresis Moderate Severe Atrial fibrillation Cerebral infarction (n=224) Level of consciousness Drowsy Comatose Cardiac failure Urinary incontinence Atrial fibrillation Unpartnered marital status Claudication RR indicates relative risk; Cl, confidence interval. Downloaded from by on August 2, 28 prognostic variables we identified may not be entirely suitable for making decisions regarding conventional and experimental therapies for the management of stroke in the hours immediately after its onset. In common with most other population-based studies, 25% of the patients in Perth died within the first month, almost 4% died within the first year of onset, and roughly 25% of patients managed in the hospital died during the initial admission. Although it has been known for a long time that primary intracerebral hemorrhage and subarachnoid hemorrhage are most likely to have lethal consequences, objectively proven lacunar infarction and boundary zone infarction had the most favorable prognosis in the Perth Community Stroke Study, with no deaths occurring within the first 28 days and -year case-fatality ratios of 6% and 6%, respectively. These findings are in agreement with others. 2 Data on the prognoses for different pathological subtypes of stroke depend to some extent on the criteria used and the process by which information related to diagnosis is obtained for each case of stroke. Although neuropathology is the "gold standard" for diagnosis in cerebrovascular disease, not all patients with stroke die from the condition. Therefore, epidemiological studies of stroke require a reasonable balance of subjective and objective criteria that preserves diagnostic sensitivity without sacrificing specificity. In this regard, the advent of CT has been a major advance because it is now relatively easy to distinguish hemorrhage from infarction accurately among those patients who undergo this investigation. However, characterization of ischemic stroke and identification of specific etiologies for stroke have proven more difficult, and problems arise because access of patients to CT and other modern neuroimaging is often delayed or not available in many countries. Even when accurate diagnostic information is collected in a standardized and prospective manner, a significant minority of cases will include cerebral infarcts of uncer- TABLE 5. Characteristics of the Model Using the Presence of Any One of the Variables as the Predictor of Death Within Year* Test Sample Validation Sample All Strokes (n=32) Cerebral Infarction (n=224) All Strokes (n=7) Cerebral Infarction (n=25) Sensitivity Specificity PPV NPV PPV indicates positive predictive value; NPV, negative predictive value. All values are percentages. *Using any of the variables outlined in Table 4.

8 Downloaded from by on August 2, Stroke Vol 25, October 994 tain etiology. 32 For a population-based study, we consider obtaining a proportional frequency of 4% for the undetermined subtype of stroke a very good result, although we recognize that the contribution of embolic stroke was probably underestimated because relatively few patients underwent carotid angiography or echocardiography. Among the other categories of cerebral infarction, the frequencies of boundary zone infarction and lacunar infarction are likely to be very accurate because the criteria for these diagnoses were highly specific. Our analytic approach was to identify the variables that best predict survival to assist clinicians in the initial management of patients with stroke in the general population. Despite the variable prognoses for survival among the different subtypes of stroke, it appears that in the first few days after onset, the risk of death is best gauged from three clinical variables coma, incontinence, and paresis that reflect the severity of the neurological deficit rather than the pathology, and two other cardiac variables heart failure and atrial fibrillation. In this study patients who were comatose, incontinent, or had severe paresis at presentation had approximately two to seven times the risk of dying within the first year of a stroke compared with those who were alert, continent, or had mild to moderate paresis, respectively. Loss of consciousness and incontinence are by definition inconsistent with a diagnosis of lacunar infarction, and, in view of the pathophysiological mechanisms operating, are unlikely to be features of boundary zone infarction. Thus, the effects of these two factors are consistent with clinical observations and with the good prognoses we observed for these subtypes of stroke. The finding of a prevalence of atrial fibrillation of 4% and the association of this dysrhythmia with an increased risk of early and late mortality were similar to reports in the Oxfordshire Community Stroke Project. 33 The mechanism underlying this excess risk of death is not clear from our data, as it appears independent of the severity of the stroke and cardiac failure. Although studies have consistently emphasized the adverse effect of increasing age on survival after stroke, Howard et al 5 suggested that this relates to other variables that are more common in the elderly. In a multicenter hospital-based study, they found that after controlling for other variables, age only provided prognostic information for patients who had had a cerebral infarct, suggesting that age may be related to complicating factors that cause death in older patients, for example, pneumonia. Although the numbers were small, Bamford et al also found a positive association between increasing age and an increasing proportion of deaths due to immobility rather than direct neurological damage. Premorbid disability may be a confounding variable in other studies that have examined the relevance of age to the outcome from stroke. In the Oxfordshire study, for example, those patients who had been dependent before stroke had a higher 3-day case-fatality rate (33%) than those who had been independent (7%). When the analysis was restricted to those patients who had sustained a cerebral infarct, the age-adjusted risk of death was doubled among those patients who had been dependent before their stroke. The trend of increasing case-fatality rate with increasing age was only present in those who were independent before stroke. In another study Bonita et al 2 used residence as a surrogate measure of disability and found that those patients who lived in institutions at the time of a stroke had a very poor prognosis independent of a number of other variables. This study confirms that the effect of age on survival after stroke is complex. While loss of consciousness and incontinence continued to have an effect on mortality during year after cerebral infarction, two other factors the presence of a partner and peripheral vascular disease were also important. The willingness and ability of caregivers to respond successfully to their role are known to influence patients' emotional reactions to stroke, their success in rehabilitation, and whether they can remain outside institutional care. This social support may also improve the chances of survival by offsetting the complications of immobility. Whereas other studies have identified hypertension, coronary heart disease, and diabetes mellitus as important factors that influence survival after stroke, 7 our data suggest that peripheral vascular disease (as indicated by a history of intermittent claudication) is a better marker of the health of the vascular tree. Although the elderly appear to be at greater risk for the complications of stroke, we could not confirm that premorbid disability and social isolation, which are more common among the elderly, have independent significance with regard to survival. Many algorithms designed to predict death during the acute phase of stroke have included neurological signs such as pupillary reaction, gaze paresis, and extensor plantar responses; various measures of disability; and some biochemical markers. Although a number of predictive equations that use several variables generate values that correlate with outcome, no model yet examined has been shown to be sufficiently accurate, that is, to account for a sufficient proportion of the variation in outcome, to be used as a basis for clinical decisions about individual patients with stroke. In light of these difficulties, the importance of simple measures as opposed to complex models has recently been emphasized. 3 ' 4 The implication from this study is that measures which reflect the severity of neurological damage, such as loss of consciousness, paresis, and incontinence, which are easily assessed at the bedside, may help to direct management. Furthermore, the influence of comorbidity, particularly cardiac failure, atrial fibrillation, and peripheral vascular disease, in determining survival after stroke is clear. Patients who have none of these six features are likely to survive and may be a subgroup in which treatment to prevent and minimize disability will yield the greatest dividends. On the other hand, patients who have one or more of these features have a measurably greater risk of death and would be the best groups in which to test new treatments for hyperacute stroke. Finally, since age is not an independent risk factor for death after stroke, this factor alone should not be used to deny the elderly treatment and therapy. Appendix Criteria for Subtypes of Stroke. Cerebral infarction was denned as a stroke for which necropsy confirms cerebral infarction; CT scan performed

9 Downloaded from by on August 2, 28 within 28 days of the onset of symptoms shows an area of attenuation or a normal appearance in the vascular territory that corresponds to the recent symptoms and signs; or magnetic resonance imaging shows a slight hypointensity (dark) with or without mass effect on T r weighted images and a bright area of hyperintensity with or without mass effect on T 2 - weighted images. Cases of cerebral infarction were further divided into four groups: a. Large-artery (occlusive) infarction was defined as a stroke presumably due to in situ thrombosis of a large or medium-sized cerebral artery. b. Embolic infarction was defined as a stroke for which cerebral imaging or necropsy showed a hemorrhagic component to the cerebral infarction or when cerebral infarction occurred in the presence of clinical data highly suggestive of embolism, such as an acute myocardial infarction in the previous 3 months; the presence of valvular heart disease; recent mechanical or traumatic manipulation to the heart, neck, carotid, or vertebrobasilar vessels; complicated (ulceration or hemorrhage) internal carotid atheroma; or a history of paroxysmal tachyarrhythmias or bradyarrhythmias. c. Lacunar infarction was defined as a stroke with clinical symptoms and signs compatible with one of the five major recognized lacunar syndromes (pure motor, pure sensory, ataxic hemiparesis, dysarthria-clumsy hand, and sensorimotor), when cerebral imaging or necropsy confirmed that the cause was occlusion of a small penetrating cerebral vessel. d. Boundary zone infarction was defined as stroke for which infarction was documented within a characteristic "watershed" boundary zone between the territories of the main cerebral arteries, usually the parietal-occipital regions, but also the basal ganglia, cerebellum, and spinal cord. 2. Primary intracerebral hemorrhage was defined as a stroke in which a focal collection of hemorrhage arose from within the brain parenchyma and traumatic intracerebral hemorrhage, hemorrhagic transformation of a cerebral infarct, and hematoma associated with a ruptured aneurysm were excluded by necropsy and/or other investigations. 3. Subarachnoid hemorrhage was diagnosed in a patient who presented with the characteristic features of an abrupt onset of severe headache and/or loss of consciousness (with or without focal neurological signs), and lumbar puncture, CT scan, or necropsy showed hemorrhage in the subarachnoid space. 4. Undetermined stroke was defined as a stroke for which a patient had not undergone a CT scan within 28 days of the onset of symptoms, or in cases in which reports for cerebral imaging or necropsy could not be retrieved. Appendix 2 Definitions of Vascular Risk Factors Hypertension was defined as two or more readings of blood pressure of 6 (systolic) or 95 (diastolic) before the onset of stroke or the use of antihypertensive medication at any time before or at the time of stroke. Myocardial infarction was defined as at least two of the following: a history of anterior chest pain, specific cardiac enzyme levels more than twice the upper limit of normal, and the development of pathological Q waves on a standard 2-lead electrocardiogram. Peripheral vascular disease was defined as a history of intermittent claudication (WHO criteria: calf pain that was atherosclerotic in origin induced by exercise and relieved within minutes by rest). Diabetes mellitus was defined as a history of diabetes that was confirmed in the patient's medical records, or the patient was taking insulin or an oral hypoglycemic agent or had a random blood glucose concentration of mmol/l. Smoking status was defined as "never," "current," or "exsmoker (>2 months)," with notation made as to the number of cigarettes smoked per day. Anderson et al Predictors of Survival After Stroke 943 Appendix 3 Diagnostic Criteria for Causes of Death Stroke. Patients were classified as having died directly from a stroke if this condition was the immediate or underlying cause of death and clinical features or autopsy confirmed that death was the result of either cerebral edema or direct brain stem compression. Pneumonia. The primary cause of death was bronchopneumonia if clinical and radiological findings included fever and pulmonary consolidation, autopsy confirmed confluent bronchopneumonia, or the patient died from respiratory failure having remained in a coma longer than 2 weeks after the onset of the stroke without clear evidence of another cause. Coronary heart disease. Deaths were ascribed to coronary heart disease if the patient had a definite acute myocardial infarction or a coronary death as defined by criteria adopted by WHO for the MONICA project. 23 The MONICA protocol provides rigid definitions for "definite myocardial infarction" and "possible myocardial infarction" that include symptoms of infarction that are typical or atypical (by the site of pain) together with specified serial changes in the electrocardiogram and/or enzyme levels and necropsy evidence of fresh myocardial infarction, recent coronary occlusion, or evidence of chronic ischemic heart disease. A diagnosis of sudden cardiac death was made when a patient with established coronary artery disease was unexpectedly found dead within hour of being clinically stable. Either well-established coronary symptoms in life or the presence at necropsy of subtotal coronary stenosis or healed myocardial infarction was accepted as sufficient evidence to make a diagnosis of chronic ischemic heart disease. Pulmonary embolism. Death due to pulmonary embolism was diagnosed if there was evidence (by venography, ventilation-perfusion lung scanning, pulmonary angiography, or necropsy) that this was the immediate or underlying cause of death. Other. This category included other diseases that led to death. Acknowledgments This study was supported by grants from the National Health and Medical Research Council, the Australian Brain Foundation, and the Medical Research Foundation of Royal Perth Hospital. It would not have been possible without the cooperation of numerous doctors and the patients and their families. We are indebted to the other members of the study team: Professor P. Burvill and Dr G. Johnson (Department of Psychiatry and Behavioral Science, University of Western Australia); Associate Professor T. Chakera (Department of Radiology, Royal Perth Hospital); and L. Nussbaum, R. Tuohy, J. Forrester, F. Gout, J. Lim, J. Linto, and S. Forbes (Department of Public Health, University of Western Australia). References. Sandercock PAG, van den Belt AGM, Lindley RI, Slattery J. Antithrombotic therapy in acute ischaemic stroke: an overview of the completed randomised trials. J Neurol Neurosurg Psychiatry. 993;56: Sheikh K, Brennan PJ, Meade TW, Smith DS, Goldenberg E. Predictors of mortality and disability in stroke. / Epidemiol Community Health. 983;37: Wade D, Skilbeck CE, Wood VA, Langton Hewer R. Long-term survival after stroke. Age Ageing. 984;3: Silver FL, rris JW, Lewis AJ, Hachinski VC. Early mortality following stroke: a prospective review. Stroke. 984;5: Howard G, Walker MD, Becker C, Coull B, Feibel J, McLevoy K, Toole JF, Yatsu F. Community hospital-based stroke programs: rth Carolina, Oregon, and New York, III: factors influencing survival after stroke: proportional hazards analysis of 429 patients. Stroke. 986; 7:

10 944 Stroke Vol 25, October 994 Downloaded from by on August 2, Chambers BR, rris JW, Shurvell BL, Hachinski VC. Prognosis of acute stroke. Neurology. 987;37: Fullerton KJ, MacKenzie G, Stout RW. Prognostic indices in stroke. Q J Med. 988;66: Allen CMC. Predicting the outcome of acute stroke: a prognostic score. J Neurol Neurosurg Psychiatry. 984;47: Lefkovits J, Davis SM, Rossiter SC, Kilpatrick CJ, Hooper JL, Green R. Acute stroke outcome: effects of stroke type and risk factors. Aust NZJ Med. 992;22: Bamford J, Dennis M, Sandercock PAG, Burn J, Warlow C. The frequency, causes and timing of death within 3 days of a first stroke: the Oxfordshire Community Stroke Project. J Neurol Neurosurg Psychiatry. 99;53: Dennis M, Burn JPS, Sandercock PAG, Bamford JM, Wade DT, Warlow CP. Long-term survival after first-ever stroke: the Oxfordshire Community Stroke Project. Stroke. 993;24: Bamford J, Sandercock P, Dennis M, Burn J, Warlow C. Classification and natural history of clinically identifiable subtypes of cerebral infarction. Lancet. 99;337: Barer DH, Mitchell JRA. Predicting the outcome of acute stroke: do multivariate models help? Q J Med. 989;7: Gladman JRF, Harwood DMJ, Barer DH. Predicting the outcome of acute stroke: prospective evaluation of five multivariate models and comparison with simple methods. J Neurol Neurosurg Psychiatry. 992;55: Kernan WN, Horwitz RI, Brass LM, Viscoli CM, Taylor KJW. A prognostic system for transient ischemia or minor stroke. Ann Intern Med. 99;4: Daverat P, Castel JP, Dartigues JF, Orgogozo JM. Death and functional outcome after spontaneous intracerebral hemorrhage: a prospective study of 66 cases using multivariate analysis. Stroke. 99;22:l Sacco RL, Wolf PA, Kannel WB, McNamara PM. Survival and recurrence following stroke. Stroke. 982;3: Katz S, Branch LG, Branson MH, Papsidero JA, Beck JC, Greer DS. Active life expectancy. N Engl J Med. 983;39: Mor V, Murphy J, Masterson-Allen S, Willey C, Razmpour A, Jackson ME, Greer D, Katz S. Risk of functional decline among well elders. J Clin Epidemiol. 989;42: Vogt TM, Mullooly JP, Ernst D, Pope CR, Hollis JF. Social networks as predictors of ischemic heart disease, cancer, stroke and hypertension: incidence, survival and mortality. J Clin Epidemiol. 992;45: Bonita R, Ford MA, Stewart AW. Predicting survival after stroke: a three-year follow-up. Stroke. 988; 9: Anderson CS, Jamrozik KD, Burvill PW, Chakera TMH, Johnson GA, Stewart-Wynne EG. Ascertaining the true incidence of stroke: experience from the Perth Community Stroke Study, Med J Aust. 993;58: WHO MONICA Project Principal Investigators. The World Health Organisation MONICA Project (Monitoring Trends and Determinants in Cardiovascular Disease): a major collaboration. J Clin Epidemiol. 988;4: Anderson CS, Jamrozik KD, Burvill PW, Chakera TMH, Johnson GA, Stewart-Wynne EG. Determining the incidence of subtypes of stroke: results from the Perth Community Stroke Study, Med J Aust. 993;58: Teasdale G, Murray G, Parker L, Jennet B. Adding up the Glasgow Coma Scale. Ada Neurochir Suppl (Wien). 979;28: Demeurisse G, Demol O, Robaye E. Motor evaluation in vascular hemiplegia. Eur Neurol. 98;9: Wade DT, Collin C. The Barthel ADL Index: a standard measure of physical disability? Int Disabil Studies. 988;: Collin C, Wade DT, Davies S, Home V. The Barthel ADL Index: a reliability study. Int Disabil Studies. 988;: Holbrook M, Skilbeck CE. An activities index for use with stroke patients. Age Ageing. 983;2: Gong G. Cross-validation, the jackknife, and the bootstrap: excess error estimation in forward logistic regression. J Am Stat Assoc. 986;8: EGRET Statistical Package. Seattle, Wash: Statistics and Epidemiology Research Corp; Sacco RL, Ellenberg JH, Mohr JP, Tatemichi TK, Hier DB, Price TR, Wolf PA. Infarcts of undetermined cause: the NINCDS Stroke Data Bank. Ann Neurol. 989;25: Sandercock P. Bamford J, Dennis M, Burn J, Slattery J, Jones L, Boonyakarnkul S, Warlow C. Atrial fibrillation and stroke: prevalence in different types of stroke and influence on early and long term prognosis (Oxfordshire Community Stroke Project). Br Med J. 992;35: Glass TA, Matchar DB, Belyea M, Feussner JR. Impact of social support on outcome in first stroke. Stroke. 993;24:64-7.