ORIGINAL ARTICLES Some prognostic models for traumatic brain injury were not valid Chantal W.P.M. Hukkelhoven a , Anneke J.J. Rampen a , Andrew I.R. Maas b , Elana Farace c , J. Dik F. Habbema a , Anthony Marmarou d , Lawrence F. Marshall e , Gordon D. Murray f , Ewout W. Steyerberg a, * a Center for Medical Decision Making Sciences, Department of Public Health, Erasmus MC–University Medical Center Rotterdam, P.O. Box 1739, 3000 DR Rotterdam, The Netherlands b Department of Neurological Surgery, Erasmus MC, Rotterdam, The Netherlands c Department of Neurological Surgery, University of Virginia, Charlottesville, VA, USA d Department of Neurosurgery, Virginia Commonwealth University, Richmond, VA, USA e Department of Neurological Surgery, University of California, San Diego, CA, USA f Department of Community Health Sciences, University of Edinburgh, Scotland Accepted 20 June 2005 Abstract Objective: Various prognostic models have been developed to predict outcome after traumatic brain injury (TBI). We aimed to deter- mine the validity of six models that used baseline clinical and computed tomographic characteristics to predict mortality or unfavorable outcome at 6 months or later after severe or moderate TBI. Study Design and Setting: The validity was studied in two selected series of TBI patients enrolled in clinical trials (Tirilazad trials; n 5 2,269; International Selfotel Trial; n 5 409) and in two unselected series of patients consecutively admitted to participating centers (European Brain Injury Consortium [EBIC] survey; n 5 796; Traumatic Coma Data Bank; n 5 746). Validity was indicated by dis- criminative ability (AUC) and calibration (Hosmer–Lemeshow goodness-of-fit test). Results: The models varied in number of predictors (four to seven) and in development technique (two prediction trees and four logistic regression models). Discriminative ability varied widely (AUC: .61–.89), but calibration was poor for most models. Better discrimination was observed for logistic regression models compared with trees, and for models including more predictors. Further, discrimination was better when tested on unselected series that contained more heterogeneous populations. Conclusion: Our findings emphasize the need for external validation of prognostic models. The satisfactory discrimination indicates that logistic regression models, developed on large samples, can be used for classifying TBI patients according to prognostic risk. Ó 2006 Elsevier Inc. All rights reserved. Keywords: Traumatic brain injury; Prognosis; Models, statistical; Validation studies; Glasgow Outcome Scale; Mortality 1. Introduction Traumatic brain injury (TBI) carries a high mortality and is a major cause of lifelong disability in a predominantly young population. Many studies have investigated the value of baseline clinical and computed tomographic (CT) char- acteristics for early prediction of long-term outcome. Inter- national guidelines contain a section dedicated to prognosis [1,2]. This section summarizes existing knowledge on the predictive value of age, Glasgow Coma Scale (GCS), pupils, hypoxia, and hypotension as well as individual CT characteristics (status of basal cisterns, shift, traumatic sub- arachnoid hemorrhage). A number of studies in TBI have focused on the development of prognostic models with which the risk for mortality or morbidity after TBI can be assessed using multiple patient characteristics. With early attempts, risk prediction was difficult on admission, but more recent studies have shown better results [3–10]. Risk assessment is important for various purposes: to inform relatives on realistic expectations, to support clinical decision making and resource allocation, or to classify pa- tients according to prognostic risk. The latter may be useful for comparing different patient series, to study treatment re- sults over time or for stratifying patients for randomized clinical trials (RCTs) [11]. The different aims of models * Corresponding author. Tel.: 131-10-408-7053; fax: 131-10-408- 9455. E-mail address: E.Steyerberg@erasmusmc.nl (E.W. Steyerberg). 0895-4356/06/$ – see front matter Ó 2006 Elsevier Inc. All rights reserved. doi: 10.1016/j.jclinepi.2005.06.009 Journal of Clinical Epidemiology 59 (2006) 132–143