Comparison of two simple models for prediction of short term mortality in patients after severe traumatic brain injury

Abstract

IntroductionThe subscale motor score of Glasgow Coma Scale (msGCS) and the Abbreviated Injury Score of head region (HAIS) are validated prognostic factors in traumatic brain injury (TBI). The aim was to compare the prognostic performance of a HAIS-based prediction model including HAIS, pupil reactivity and age, and the reference prediction model including msGCS in emergency department (ED), pupil reactivity and age. MethodsSecondary analysis of a prospective epidemiological study including patients after severe TBI (HAIS > 3) with follow-up from the time of accident until 14 days or earlier death was performed in Switzerland. Performance of prediction, based on accuracy of discrimination [area under the receiver-operating curve (AUROC)], calibration (Hosmer-Lemeshow test) and validity (bootstrapping with 2000 repetitions to correct) for optimism of the two prediction models were investigated. A non-inferiority approach was performed and an a priori threshold for important differences was established. ResultsThe cohort included 808 patients [median age 56 {inter-quartile range (IQR) 33–71}, median motor part of GCS in ED 1 (1–6), abnormal pupil reactivity 29.0%] with a death rate of 29.7% at 14 days. The accuracy of discrimination was similar (AUROC HAIS-based prediction model: 0.839; AUROC msGCS-based prediction model: 0.826, difference of the 2 AUROC 0.013 (−0.007 to 0.037). A similar calibration was observed (Hosmer-Lemeshow X2 11.64, p = 0.168 vs. Hosmer-Lemeshow X2 8.66, p = 0.372). Internal validity of HAIS-based prediction model was high (optimism corrected AUROC: 0.837). ConclusionsPerformance of prediction for short-term mortality after severe TBI with HAIS-based prediction model was non-inferior to reference prediction model using msGCS as predictor.