445 Performance of the IMPACT and CRASH Prognostic Models for Traumatic Brain Injury in a Contemporary Multicenter Cohort: A TRACK-TBI Study

Abstract

INTRODUCTION: The International Mission on Prognosis and Analysis of Clinical Trials in Traumatic Brain Injury (IMPACT) and Corticosteroid Randomisation After Significant Head Injury (CRASH) prognostic models for mortality and functional outcome in traumatic brain injury (TBI) were developed using historical data from 1984-2004. METHODS: The prospective, 18-center Transforming Research and Clinical Knowledge in Traumatic Brain Injury study enrolled participants aged = 17-years who presented to Level 1 trauma center and received head computed tomography (CT) scan <24-hours of TBI. Data was extracted from participants meeting IMPACT (Glasgow Coma Scale (GCS) 3-12; completed 6-month Glasgow Outcome Scale-Extended (GOSE); N = 441) and CRASH criteria (GCS 3-14; completed 2-week and 6-month GOSE; N = 831). Original model coefficients were applied, and performances were assessed using discrimination (area under the receiver-operating characteristic curve (AUC)) and calibration (intercept; slope; [95% confidence interval]). RESULTS: IMPACT and CRASH models discriminated mortality (AUC = 0.77-0.81, 0.90-0.91, respectively) and unfavorable outcome (0.77-0.81; 0.83, respectively). The IMPACT Lab model overestimated mortality (intercept = -0.79 [-1.05, -0.53]; slope = 1.37 [1.05-1.69]) and acceptably estimated unfavorable outcome (intercept = 0.07 [-0.14, 0.29]; slope = 1.19 [0.96-1.42]). The CRASH CT model overestimated mortality (intercept = -1.06 [-1.36, -0.75]; slope = 0.96 [0.79-1.14]) and unfavorable outcome (intercept = -0.60 [-0.78, -0.41]; slope = 1.20 [1.03-1.37]). Calibration performance differences were observed across the 5 leading enrollment centers (IMPACT Lab model: intercept -1.66 to -0.35, slope 1.07-1.86; CRASH CT model: intercept -1.81 to -0.67, slope 0.67-1.56) and unfavorable outcome (IMPACT Lab model: intercept -0.57 to 0.34, slope 1.28-2.89; CRASH CT model: intercept -1.15 to -0.10, slope 1.02-1.44). CONCLUSIONS: IMPACT and CRASH models discriminated mortality and unfavorable outcome. Observed overestimations of mortality and unfavorable outcome underscore the need to develop prognostic models that incorporate contemporary changes in TBI casemix. Investigations to elucidate the relationships between increased survival, functional outcome, treatment intensity, and center-specific practices will be highly relevant.