Predicting functional outcome after stroke by modelling baseline clinical and CT variables

Transcription

1 Age and Ageing 2010; 39: doi: /ageing/afq027 Published electronically 15 March 2010 The Author Published by Oxford University Press on behalf of the British Geriatrics Society. All rights reserved. For Permissions, please Predicting functional outcome after stroke by modelling baseline clinical and CT variables JOHN M. REID 1,GORD J. GUBITZ 2,3,DINGWEI DAI 6,DAVID KYDD 2,4,GAIL ESKES 2,5,YVETTE REIDY 2, CHRISTINE CHRISTIAN 2,CARL E. COUNSELL 7,MARTIN DENNIS 8,STEPHEN J. PHILLIPS 2,3 1 Department of Neurology, Aberdeen Royal Infirmary, Foresterhill, Aberdeen AB25 2ZN, UK 2 Queen Elizabeth II Health Sciences Centre and Faculty of Medicine, Dalhousie University, Halifax, NS, Canada 3 Department of Medicine (Division of Neurology), Dalhousie University, Halifax, NS, Canada 4 Department of Diagnostic Imaging (Division of Neuroradiology), Dalhousie University, Halifax, NS, Canada 5 Department of Psychiatry, Dalhousie University, Halifax, NS, Canada 6 Center for Paediatric Clinical Effectiveness (CPCE), The Children's Hospital of Philadelphia, 3535 Market Street, Philadelphia, PA , USA 7 Division of Applied Health Sciences, University of Aberdeen, Polwarth Building, Foresterhill, Aberdeen AB25 2ZN, UK 8 Department of Clinical Neurosciences, Western General Hospital, Edinburgh, UK Address correspondence to: J. M. Reid. Tel: (+44) ; Fax: (+44) johnmreid@doctors.net.uk Abstract Background: we aimed to assess whether the performance of stroke outcome models comprising simple clinical variables could be improved by the addition of more complex clinical variables and information from the first computed tomography (CT) scan. Methods: 538 consecutive acute ischaemic and haemorrhagic stroke patients were enrolled in a Stroke Outcome Study between 2001 and Independent survival (modified Rankin scale 2) was assessed at 6 months. Models based on clinical and radiological variables from the first assessment were developed using multivariate logistic regression analysis. Results: three models were developed (I III). Model I included age, pre-stroke independence, arm power and a stroke severity score (area under a receiver operating characteristic curve, AUC = 0.882) but performed no better than Model II, which comprised age, pre-stroke independence, normal verbal component of the Glasgow coma score, arm power and being able to walk without assistance (AUC 0.876). Model III, including two radiological variables and clinical variables, was not statistically superior to model II (AUC 0.901, P = 0.12). Model II was externally validated in two independent datasets (AUCs of and 0.787). Conclusion: this study demonstrates an externally validated stroke outcome prediction model using simple clinical variables. Outcome prediction was not significantly improved with CT-derived radiological variables or more complex clinical variables. Keywords: acute stroke, elderly, outcomes, prognosis Introduction Models that accurately predict long-term outcome early in the course of stroke, which may assist clinical management, can be used to adjust for case mix between cohorts and for stratification in randomised trials. Several models based on clinical features alone or in combination with radiological features have been described. One such model (the six simple variable [SSV] model) [1] based on easily collected clinical variables performed as well as more complex models to predict independent survival at 6 months. Variables such as ischaemic lesion volume [2 4], the presence or extent of infarction obtained from cranial magnetic resonance imaging (MRI) [2] or computed tomography (CT) [3 6] are predictors of outcome after ischaemic stroke. In most studies, neither CT- nor MRI-derived variables significantly improved the performance of models comprised only of clinical predictors [3, 4, 7, 8]. One study found that diffusion-weighted imaging (DWI) lesion volume significantly improved outcome model performance [9], although this was not considered to be clinically significant. The SSV model has been validated in our stroke population [10]; however, this model was developed from a sub-acute stroke population, without use of cranial imaging data [1]. In this 360

2 Predicting functional outcome after stroke study, we hypothesised that the performance of models comprising simple clinical variables could be improved by information from the first CT scan (e.g. presence of infarction or leukoaraiosis) or more complex clinical variables. Methods Study population Patients admitted consecutively with a diagnosis of stroke from 2001 to 2002 at the Halifax Infirmary were enrolled in the Stroke Outcomes Study (SOS). Written consent for study participation was obtained from each patient or his/ her surrogate decision maker, which was approved by the Capital Health Research Ethics Board, Halifax, Nova Scotia, Canada. All patients underwent cranial imaging urgently on arrival in the emergency department. Patients were simultaneously entered into a stroke registry containing information from the first clinical assessment conducted at the time of admission by the on-call neurologist including gender, age, pre-stroke functional status, Oxfordshire Community Stroke Project (OCSP) [11] classification subtype, stroke severity score (which assesses neurological symptoms, signs and functional impairment) [12], Glasgow Coma Score (GCS) [13] and the presence of atrial fibrillation(af)onthe ECG. The information for each patient is linked using a unique identifier. Patients were followed up by telephone interview 6 months post-stroke when functional status was assessed by an assessor trained in administering the modified Rankin scale (mrs) [14]. The main outcome measure was independent survival (good outcome defined as mrs 2). Patient information from the discharge summary was available at the time of follow-up. CT scan interpretation Images from each patient s first CT scan were digitised, given an identifier code and analysed using the Consortium for the Investigation of Vascular Impairment of Cognition (CIVIC) scale [15]. CT slice thickness was 5 mm in the posterior fossa and 10 mm in the remainder of the brain. The CIVIC scale captures details of each discrete lesion seen on the CT image and whether an infarct is acute or not, and provides an assessment of leukoaraiosis. A leukoaraiosis score of 0 6 was calculated from the sum of periventricular leukoaraiosis only (one point) or extending to the cortex (two points) for three hemispheric areas (frontal, parieto-occipital and corona radiata) [15]. CT scans were read by a neuroradiologist (DK), stroke neurologist (SP) and stroke fellow (JR). The inter-rater reliability of the CIVIC scale was examined by analysing 134 scans for agreement on any focal abnormality using Fleiss kappa [16]. Only patient age, identifier code and gender were available to the CT reader. Predictor variables The modelling included the following variables derived from the first medical assessment: age, gender, stroke severity score [12], stroke subtype (either intraparenchymal haemorrhage as determined by CT or MRI, or OCSP [11] subtype for ischaemic stroke [total anterior circulation stroke {TACS}, partial anterior circulation stroke, posterior circulation stroke and lacunar stroke, or uncertain]), dysarthria, AF, history of previous transient ischaemic attack (TIA) or stroke, use of tissue plasminogen activator (tpa), time from symptom recognition to presentation (0 3, 3 6 or >6 h), stroke localisation (infratentorial, left or right hemisphere or other), stroke on waking and SSV variables (pre-stroke functional status, living alone prior to stroke, normal verbal component of the GCS, able to lift both arms off the bed and able to walk without assistance of another person). The following CIVIC scale variables were tested: any abnormality, any focal abnormality, number of focal lesions (acute and non-acute), acute infarction, acute infarction or haemorrhage, and leukoaraiosis. Thus, 16 clinical variables and six radiological variables were tested ensuring an event per variable ratio 10 [1]. In an additional analysis, the variable infarct volume was tested and calculated using an approximate method previously used to estimate infarct volume using DWI images [17]. The majority of these variables were binary (coded as 1 = yes, 0 = no); age, stroke severity score, leukoaraiosis and number of focal lesions were coded as continuous variables, and stroke subtype and localisation as categorical variables. Statistical techniques for generating the models The data are presented as mean ± standard deviation unless otherwise stated. Comparisons between groups were made using chi-square, Mann Whitney and Student s t-test where appropriate, with significance of P < To formulate the models to predict good outcome (mrs 2) at 6 months post-stroke, univariate analysis was performed comparing patients with and without good outcome in the (SOS) training dataset. Significant variables (P < 0.05) were subjected to multivariate logistic regression analysis. A stepwise selection procedure was applied; the probability for entry of a variable was set at 0.01 and for removal at 0.1. Three models were produced by entering either all the clinical variables, all clinical variables excluding the stroke severity score or including all the CIVIC and clinical variables. For each model, we checked the statistical assumptions of linearity (for age) [18, 19] and looked for interactions between age, sex and the other predictor variables in the model by using a pooled interaction test [18]. Validation of models Internal validation was completed for all three models using bootstrap techniques [20], with re-sampling occurring 500 times. External validation of Model II was performed using the OCSP (n = 530) dataset and the Lothian Stroke Register (LSR, n = 1,330) [1]. The former was a community-based incidence study and the latter a hospital-based cohort study of stroke patients from in-patient and out-patient services. Models I and III could not be externally 361

3 J. M. Reid et al. Table 1. Baseline characteristics of the study population and association with good outcome (modified Rankin scale [mrs] 2) at 6 months N (%) mrs 2 mrs > 2 Odds ratio... All Women 252 (47%) 95 (42%) 157 (51%) 0.7 ( )* Median age a 74 (61 80) 68 (57 76) 77 (67 83) 0.95 ( )*** Living alone pre-stroke 135 (25%) 59 (26%) 76 (25%) 1.1 ( ) Independent pre-stroke 437 (81%) 226 (99%) 209 (67%) 53 (14 447)*** Prior stroke/tia 148 (27%) 46 (20%) 98 (32%) 0.54 ( )* Atrial fibrillation 78 (15%) 19 (8%) 59 (19%) 0.39 ( )** Presenting <3 h 182 (34%) 56 (25%) 126 (41%) 0.48 ( )*** Stroke on waking 135 (25%) 53 (23%) 82 (26%) 0.84 ( ) Median stroke severity score a 6(5 8) 5 (4 6) 7 (6 9) 0.49 ( )*** Total anterior circulation stroke 75 (14%) 5 (2%) 70 (23%) 0.07 ( )*** Partial anterior circulation stroke 166 (31%) 72 (32%) 94 (30%) 0.96 ( ) Lacunar stroke 117 (22%) 68 (30%) 49 (16%) 2.1 ( )** Posterior circulation stroke 92 (17%) 54 (24%) 38 (12%) 2.1 ( )* Ischaemic stroke of uncertain type 18 (3%) 7 (3%) 11 (4%) 0.80 ( ) Haemorrhagic stroke 70 (13%) 22 (10%) 48 (15%) 0.58 ( )* Left hemisphere localisation 205 (38%) 69 (30%) 136 (44%) 0.56 ( )** Right hemisphere localisation 219 (41%) 93 (42%) 126 (41%) 1.01 ( ) Infratentorial localisation 80 (15%) 50 (22%) 30 (10%) 2.6 ( )*** Other localisation 34 (6%) 16 (7%) 18 (6%) 1.2 ( ) Dysarthria 287 (53%) 72 (32%) 215 (69%) 0.20 ( )*** Verbal Glasgow coma score = (65%) 198 (87%) 149 (48%) 7.1 ( )*** Able to lift both arms off bed 353 (66%) 204 (89%) 149 (48%) 9.2 ( )*** Able to walk without assistance 150 (28%) 114 (50%) 36 (12%) 7.6 ( )*** Treated with tpa 29 (5%) 9 (4%) 20 (6%) 0.60 ( ) CIVIC scale Any abnormality present 414 (82%) 147 (72%) 267 (93%) 0.31 ( )*** Any focal lesion present 335 (67%) 112 (55%) 223 (75%) 0.41 ( )*** Mean number of focal lesions a 1.1 ± ± ± ( )*** Acute infarct present 151 (30%) 44 (22%) 107 (36%) 0.49 ( )** Acute infarct or haemorrhage 198 (39%) 59 (29%) 139 (47%) 0.41 ( )*** Mean leukoaraiosis score a 1.3 ± ± ± ( )*** Medians are expressed with interquartile range. TIA, transient ischaemic attack; tpa, tissue plasminogen activator. a Odds ratio is per unit variable. *P < 0.05, **P < 0.001, ***P < tested since no database containing the appropriate variables was available. Model discrimination was assessed using the area under a receiver operating characteristic (ROC) curve (AUC), computed by a non-parametric method [21]. An AUC of 1 implies perfect discrimination, whereas an AUC of 0.5 implies the model performs no better than chance. The best model was defined as the model with the largest AUC [21, 22] or, if there were no statistically significant differences in the areas, the model with the simplest variables. Calibration was assessed using calibration curves (observed vs predicted outcome among patients grouped by deciles of predicted probability of a good outcome), with a 45 line indicating perfect calibration. The confidence intervals of AUCs and the observed probability of good outcome were calculated with bootstrapping. Comparisons of the mean AUC values from different settings were conducted using analyses of variance (ANOVA). All analyses were conducted using SAS software, version 9.1 (SAS Institute Inc., Cary, NC, USA). Patients with missing data points were excluded from analyses pertaining to that variable. Results Of 598 patients admitted to the stroke service between 2001 and 2002, 38 refused consent, 13 had repeat admissions (second admission excluded) and nine were lost to followup, leaving a final study group of 538 patients (Table 1). There were no significant differences in baseline clinical variables between patients who refused consent or were lost to follow-up (not shown). Moreover, 34 and 51% of patients presented within 3 and 6 h of symptom recognition, respectively. Five hundred and two (93%) patients had CT scans available for reading using the CIVIC scale, with the remainder having had an MRI only (n = 29) or CT scans from a referring hospital unavailable for analysis (n = 7). The Fleiss kappa for the three-way inter-rater agreement on the variable any focal abnormality was moderate (0.52, 95% confidence interval [CI] ). Of the 502 CT scans, eight (2%) were technically inadequate, 414 were abnormal, 335 had a focal abnormality, 151 had acute infarcts (31% of adequate scans, including nine acute haemorrhagic infarcts) and 56 (11%) had acute intracerebral haemorrhages. At dis- 362

4 Predicting functional outcome after stroke Table 2. Models for predicting alive and independent at 6 months Variables β (SE) Odds ratio (95% CI)... Model I (all clinical variables) Intercept (1.156) Age a (0.009) 0.95 ( ) Pre-stroke independence (0.758) ( ) Arm power (0.314) 4.06 ( ) Stroke severity score a (0.084) 0.58 ( ) Model II (clinical variables excluding stroke severity score) Intercept (0.982) Age a (0.009) 0.95 ( ) Pre-stroke independence (0.754) ( ) Verbal GCS (0.280) 3.05 ( ) Arm power (0.304) 5.84 ( ) Able to walk unaided (0.268) 2.48 ( ) Model III (all CIVIC and clinical variables) Intercept (1.496) Age a (0.010) 0.96 ( ) Pre-stroke independence (1.031) ( ) Arm power (0.331) 3.41 ( ) Stroke severity score a (0.094) 0.53 ( ) Leukoaraiosis score a (0.084) 0.77 ( ) Focal CT abnormality (0.272) 0.42 ( ) SE, standard error; β, regression coefficient; CI, confidence interval; GCS, Glasgow coma score. a Per unit variable. charge and 6 months post-stroke, 36 and 42% of patients had a good outcome (mrs 2), with a mortality of 17 and 24%, respectively. Univariate analysis demonstrated several significant clinical and radiological predictors of outcome at 6 months (Table 1). Predictive model development and internal validation ThethreemodelsareshowninTable2andtheirROC curves in Figure 1. Model I: all clinical variables Multivariate analysis using all clinical variables produced a model containing four variables that independently predicted good outcome at 6 months: age, pre-stroke independence, being able to lift both arms off the bed and stroke severity (Model I, AUC of 0.882). Entering stroke subtype (haemorrhagic stroke or OCSP subtype) as individual binary variables, rather than as a categorical group, produced no new independently predictive variables. Model II: all clinical variables excluding stroke severity score Stroke severity score alone had an AUC of 0.826, and as it may have made stroke subtype (which is linked to stroke severity) or some of the clinical variables redundant as predictors, multivariate analysis was performed excluding it. Figure 1. Receiver operating characteristic curve for Models I III predicting independent survival (modified Rankin scale score 2) at 6-month follow-up. This generated Model II consisting of five independent predictive variables with a similar model performance to Model I (AUC of 0.876). Modelling restricted to the subgroup of patients with ischaemic stroke (n = 468) produced identical predictor variables (AUC of 0.882, data not shown). Model III: all clinical and CIVIC variables Multivariate analysis performed with individual CIVIC variables entered separately in addition to all clinical variables produced models with AUCs from to All of the CIVIC variables tested in this way were independent predictors of outcome. On performing multivariate analysis including all the CIVIC and clinical variables, only leukoaraiosis score and any focal abnormality remained as significant independent predictors, producing a model comprising six variables (Model III, AUC 0.901, Table 2 and Figure 1), which was not statistically superior to Models I or II (P = 0.12, ANOVA). Infarct volume was tested in ischaemic stroke patients and was found to be a negative predictive variable of good outcome in univariate analysis (P = ), but was not an independent predictor in multivariate analysis (data not shown). External validation of Model II The simplest model (Model II) was externally validated with an AUC of in the OCSP dataset and an AUC of in the LSR dataset. The calibration curves for Model II using these two datasets showed good calibration, although it appeared to perform better in the hospital-based cohort (LSR, data not shown). 363

5 J. M. Reid et al. Discussion The main finding of this study is that Model II comprising five simple clinical variables performed as well in the original dataset as two other models that included a stroke severity score and information from the first CT scan. Model II contained five of the six variables from a previously published SSV model [1], which has also been externally validated [10, 23], endorsing the strength of these predictor variables. Unlike the SSV model, living alone pre-stroke was not a predictor of poor outcome at 6 months in our population. In the SSV derivation study, living alone pre-stroke did not predict 12-month outcome post-stroke [24]. In another study [25], having a companion at home was an independent predictor of good outcome (Barthel Index 50)at1week. Therefore, pre-stroke living arrangements may be important for short-term outcome post-stroke but is not a robust predictor of medium- to long-term outcome. Model II was externally validated in two independent sub-acute stroke cohorts (AUC ). The fivevariable Model II described here performed less well than the SSV model in the LSR cohort (AUC of vs 0.840, respectively) [1]. However, our population included a high proportion of hyper-acute stroke patients, whereas the validation cohorts mainly comprised milder post-acute patients [1]. Resolving which model best predicts outcome will require testing in other populations. The OCSP classification was not an independent outcome predictor in multivariate analysis despite variables such as TACS being a strong negative predictor of a good outcome (Table 1). This may be because the simple clinical variables account for the more predictive features of a TACS. The stroke severity score [12] alone was a strong predictor of good outcome at 6 months (AUC 0.826), but this finding could not be validated as a suitable external database was not available. Adding a stroke severity score into outcome prediction models may merely introduce unnecessary complexity. Although other stroke severity scales such as the National Institutes of Health Stroke Scale (NIHSS) have been found to be independent predictors of stroke outcome [4, 7], models that can be used without special training may find wider application. The investigators from the Third International Stroke Trial plan to directly compare outcome models that use simple clinical variables with those based on the NIHSS [23]. As yet, clinical stroke outcome models are not accurate enough to reliably predict outcome after stroke. This would require further testing to demonstrate if they were better than clinical judgment. Currently, they are best applied for case-mix adjustment between cohorts and for stratification in randomised trials. In common with several prior studies using CT and MRI [4, 7, 8], we found that radiological variables from the initial brain scan did not significantly improve the performance of outcome models derived from clinical variables. This suggests that detailed acute radiological data add little predictive information once the diagnosis of stroke is established, although cranial imaging helps exclude some stroke mimics. However, our study may have lacked power to detect statistically significant differences in models with the addition of CT data. Further, we detected acute infarction in less than one-third of CT scans, probably because our study population included both ischaemic and haemorrhagic strokes, and because 51% of the patients presented within 6 h of symptom recognition and were urgently scanned before ischaemic changes were apparent. The inter-rater reliability between three raters in this study was moderate (kappa 0.52), in keeping with the inter-observer agreement (kappa ) for the detection of any early CT sign of ischaemia reported in a recent systematic review [26]. Had the CT scans only been read by a neuroradiologist, detection of early ischaemic changes may have been improved. However, this would not reflect routine clinical practice where CT brain scans are not always interpreted by a neuroradiologist acutely. Leukoaraiosis was found to be an independent predictor of poor outcome, perhaps reflecting the development of post-stroke dementia [27]. Further, leukoaraiosis volume at the time of acute ischaemic stroke is a predictor of infarct growth [28], which may predict poor outcome. Our study has some limitations. Some variables known to be associated with outcome after stroke [29] (e.g. glucose) were not collected. This study was conducted in a tertiary referral stroke centre and may not be representative of all stroke patients. The definition of the best model (see Methods ) in this study favoured Model II. One disadvantage of this study is the lack of suitable datasets to externally validate Models I and III. These models could have superior external validation than Model II. Advantages of this study are the small number of patients lost to follow-up, high consent rate and it is one of a few studies to externally validate a stroke outcome model. This study describes an externally validated stroke outcome prediction model based on five simple clinical variables that is adequate for case-mix adjustment and for stratifying patients in randomised trials. Following a clinical diagnosis of stroke, it is difficult to improve on outcome prediction models based on simple clinical variables, either with more complex clinical variables or radiological data. Statistical models are not yet precise enough to reliably predict outcome for each individual patient. Key points Simple clinical variables are the best predictors of stroke outcome. CT-derived radiological data are predictive of outcome but add no significant improvement to outcome model performance. This study describes an outcome prediction model that is externally validated. 364

6 Predicting functional outcome after stroke Conflict of interest No conflicts noted. Sources of funding The SOS is a research project funded by the Capital Health Research Fund and the Nova Scotia Health Research Foundation. J.M.R. was supported by the Dalhousie Internal Medicine Research Foundation and by unrestricted educational grants from Merck Frosst and Hoffman-La Roche. Acknowledgements The authors thank Kara Thompson and Adam Webber for statistical assistance. References 1. Counsell C, Dennis M, McDowall M, Warlow C. Predicting outcome after acute and subacute stroke: development and validation of new prognostic models. Stroke 2002; 33: Thijs VN, Lansberg MG, Beaulieu C et al. Is early ischemic lesion volume on diffusion-weighted imaging an independent predictor of stroke outcome? A multivariable analysis. Stroke 2000; 31: Saver JL, Johnston KC, Homer D et al. Infarct volume as a surrogate or auxiliary outcome measure in ischemic stroke clinical trials. The RANTTAS Investigators. Stroke 1999; 30: Johnston KC, Wagner DP, Haley EC Jr, Connors AF Jr RANTTAS Investigators. Randomized Trial of Tirilazad Mesylate in Acute Stroke. Combined clinical and imaging information as an early stroke outcome measure. Stroke 2002; 33: Wardlaw JM, Lewis SC, Dennis MS, Counsell C, McDowall M. Is visible infarction on computed tomography associated with an adverse prognosis in acute ischemic stroke? Stroke 1998; 29: Barber PA, Hill MD, Eliasziw M et al. ASPECTS Study Group. Imaging of the brain in acute ischemic stroke: comparison of computed tomography and magnetic resonance diffusionweighted imaging. J Neurol Neurosurg Psychiatry 2005; 76: Hand PJ, Wardlaw JM, Rivers CS et al. MR diffusion-weighted imaging and outcome prediction after ischemic stroke. Neurology 2006; 66: Schiemanck SK, Kwakkel G, Post MW et al. Predicting longterm independency in activities of daily living after middle cerebral artery stroke. Does information from MRI have added predictive value compared with clinical information? Stroke 2006; 37: Johnston KC, Wagner DP, Wang XQ et al. GAIN, Citicoline, and ASAP Investigators. Validation of an acute ischemic stroke model: does diffusion-weighted imaging lesion volume offer a clinically significant improvement in prediction of outcome? Stroke 2007; 38: Reid JM, Gubitz GJ, Dai D et al. External validation of a six simple variable model of stroke outcome and verification in hyper-acute stroke. J Neurol Neurosurg Psychiatry 2007; 78: Bamford J, Sandercock P, Dennis M, Burn J, Warlow C. Classification and natural history of clinically identifiable subtypes of cerebral infarction. Lancet 1991; 337: Gasecki AP, Eliasziw M, Ferguson GG, Hachinski V, Barnett HJ. Long-term prognosis and effect of endarterectomy in patients with symptomatic severe carotid stenosis and contralateral carotid stenosis or occlusion: results from NASCET. NorthAmericanSymptomaticCarotidEndarterectomyTrial (NASCET) Group. J Neurosurg 1995; 83: Teasdale G, Knill-Jones R, van der Sande J. Observer variability in assessing impaired consciousness and coma. J Neurol Neurosurg Psychiatry 1978; 41: van Swieten JC, Koudstaal PJ, Visser MC, Schouten HJ, van Gijn J. Interobserver agreement for the assessment of handicap in stroke patients. Stroke 1988; 19: Rockwood K, Macknight C, Wentzel C et al. The diagnosis of "mixed" dementia in the Consortium for the Investigation of Vascular Impairment of Cognition (CIVIC). Ann NY Acad Sci 2000; 903: Fleiss JL, Nee CM, Landis JR. Large sample variance of kappa in the case of different sets of raters. Psychol Bull 1979; 86: Sims JR, Gharai LR, Schaefer PW et al. ABC/2 for rapid clinical estimate of infarct, perfusion, and mismatch volumes. Neurology 2009; 72: Harrell FE, Lee KL, Mark DB. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med 1996; 15: Hosmer DW, Lemeshow S. Applied Logistic Regression. New York: John Wiley & Sons Efron B, Tibshirany R. An introduction to the bootstrap. New York: Chapman and Hall, Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 1982; 143: DeLong ER, DeLong DM, Clark-Pearson DL. Comparing the area under two or more correlated receiving operating characteristic curves: a nonparametric approach. Biometrics 1988; 44: SCOPE (Stroke Complications and Outcomes Prediction Engine) Collaborations. IST. Lewis SC. Sandercock PA. Dennis MS. Predicting outcome in hyper-acute stroke: validation of aprognosticmodelinthethirdinternational Stroke Trial (IST3). J Neurol Neurosurg Psychiatry 2008; 79: Counsell C, Dennis M, McDowall M. Predicting functional outcome in acute stroke: comparison of a simple six variable model with other predictive systems and informal clinical prediction. J Neurol Neurosurg Psychiatry 2004; 75: Jorgensen HS, Reith J, Nakayama H, Kammersgaard LP, Raaschou HO, Olsen TS. What determines good recovery in patients with the most severe strokes? The Copenhagen Stroke Study. Stroke 1999; 30: Wardlaw JM, Mielke O. Early signs of brain infarction at CT: observer reliability and outcome after thrombolytic treatment systematic review. Radiology 2005; 235:

7 J. Benito-León et al. 27. Leys D, Henon H, Mackowiak-Cordoliani MA, Pasquier F. Poststroke dementia. Lancet Neurol 2005; 4: Ay H, Arsava EM, Rosand J et al. Severity of leukoaraiosis and susceptibility to infarct growth in acute stroke. Stroke 2008; 39: Counsell C, Dennis M. Systematic review of prognostic models in patients with acute stroke. Cerebrovasc Dis 2001; 12: Received 27 October 2009; accepted in revised form 20 January 2010 Age and Ageing 2010; 39: doi: /ageing/afq028 Published electronically 17 March 2010 The Author Published by Oxford University Press on behalf of the British Geriatrics Society. All rights reserved. For Permissions, please journals.permissions@oxfordjournals.org Low morale is associated with increased risk of mortality in the elderly: a population-based prospective study (NEDICES) JULIÁN BENITO-LEÓN 1,2,ELAN D. LOUIS 3,4,5,6,JESÚS RIVERA-NAVARRO 7,MARÍA JOSÉ MEDRANO 8 SATURIO VEGA 9,FÉLIX BERMEJO-PAREJA 1,2 1 The Department of Neurology, University Hospital 12 de Octubre, Madrid, Spain 2 Centro de Investigación Biomédica en Red sobre Enfermedades Neurodegenerativas (CIBERNED), Madrid, Spain 3 The G.H. Sergievsky Center, College of Physicians and Surgeons, Columbia University, New York, NY, USA 4 Department of Neurology, College of Physicians and Surgeons, Columbia University, New York, NY, USA 5 Taub Institute for Research on Alzheimer's Disease and the Aging Brain, College of Physicians and Surgeons, Columbia University, New York, NY, USA 6 Department of Epidemiology, Mailman School of Public Health, Columbia University, New York, NY, USA 7 Department of Social Sciences, University of Salamanca, Salamanca, Spain 8 Unit of Vascular Risk Factors, National Center for Epidemiology, ISCIII, Madrid, Spain 9 Arévalo Health Center, Arévalo, Ávila, Spain Address of the correspondence to: J. Benito-León, Avda. de la Constitución 73, portal 3, 7 Izquierda, E Coslada, Madrid, Spain. Tel: (+34) ; fax: (+34) jbenitol@meditex.es Abstract Objective: the study aimed to assess the association between morale and mortality. Design: we used data from the Neurological Disorders in Central Spain (NEDICES), a population-based study. Subjects: 2,516 older persons (mean age 75.7 years) participated in the study. Methods: Cox models were used to estimate risk of mortality. Morale was assessed using the Philadelphia Geriatric Center Morale Scale. Results: 489 (21.8%) participants died over a median follow-up of 5.9 years (range years), including 253 (21.8%) deaths among 1,163 participants with low morale scores, 168 (19.3%) among 870 participants with moderate scores and 68 (14.1%) among participants with high scores. In an unadjusted Cox model, relative risk (RR) of mortality in participants with low morale scores = 1.69 (P < 0.001) and RR in participants with moderate scores = 1.47 (P < 0.01) were compared to the reference group (participants with high scores). In a Cox model that adjusted for a variety of demographic factors and comorbidities, RR of mortality in participants with low morale scores = 1.35 (P <0.05) and moderate scores = 1.16 (not significant) were compared to the reference group. Conclusion: low morale may be an independent predictor of mortality in the elderly. By assessing morale, practitioners might be better positioned to identify patients with poorer prognoses. Keywords: elderly, epidemiology, quality of life, morale, mortality 366