Dynamic Glasgow Coma Scale Trajectories Improve Mortality Prediction in Traumatic Brain Injury: A Multicenter Intensive Care Unit Cohort Study

Abstract

Traumatic brain injury (TBI) is a leading cause of death and disability. While the Glasgow Coma Scale (GCS) guides initial assessment, single values miss evolving neurological change. In this multicenter ICU cohort integrating NSICU, MIMIC-IV, and eICU databases, we analyzed adults (≥18 years) with TBI who had ≥3 GCS measurements within the first 120 ICU hours. Using 12-hourly measures, latent class growth modeling identified four dynamic GCS trajectories (Stable High, Rapidly Improving, Persistently Moderate, Persistently Low), and we quantified cumulative neurological burden with a mean threshold-based area-under-the-curve (TBM-AUC) summarizing time above prespecified GCS thresholds. Among 3,132 patients, mortality increased monotonically across trajectories, highest in the Persistently Low group (adjusted hazard ratio [HR] 4.95, 95% confidence interval: 3.14–7.81 vs. Stable High). Lower TBM-AUC was strongly associated with mortality; most pronounced at threshold 13 (HR 0.34). Age-stratified analyses showed a trajectory-by-age interaction (p = 0.013), with Persistently Low conferring the greatest risk in both younger and older adults. Adding trajectory class to baseline predictors improved discrimination (AUC: 0.820–0.861, p < 0.001) with consistent gains in integrated discrimination improvement, net reclassification improvement, and median risk score across Boruta-, LASSO-, and best-subset–based models. Dynamic GCS trajectories and TBM-AUC provide prognostic information beyond conventional assessments and may enhance risk stratification and clinical decision-making in neurocritical care; prospective validation is warranted.