Neurological prognostication of patients in post-traumatic coma remains challenging due to the paucity of reliable markers in the acute phase. We aimed to assess whether early EEG features reflecting multiple levels of brain connectivity related to consciousness and their temporal evolution were associated with awakening and long-term functional outcomes in severe traumatic brain injury (TBI). We retrospectively included 95 severe TBI patients who underwent >48 h of continuous EEG (cEEG) monitoring between 2015 and 2024. Quantitative and functional connectivity EEG features representing brainstem–forebrain, thalamocortical, and cortico-cortical connectivity were extracted at six time points within the first 48 h of recording. The primary outcome was functional independence at 12 months, defined as a score on the Glasgow Outcome Scale score ≥4; the secondary outcome was consciousness recovery at ICU discharge. We trained multiple XGBoost machine learning models to predict recovery using various combinations of EEG and International Mission for Prognosis and Analysis of Clinical Trials in TBI (IMPACT) score features and compared their predictive performances, as measured by the area under the receiver operating characteristic curve (AUROC). The best-performing model achieved a median AUROC of 0.70 (IQR: 0.59–0.80) for 12-month outcome, combining EEG and IMPACT features, but did not significantly differ from the EEG and age model (AUROC: 0.69; IQR: 0.61–0.76). The EEG-only model outperformed the IMPACT model (AUROC: 0.65 vs. 0.56), whereas the IMPACT model without age showed no predictive value above chance (AUROC: 0.49; IQR: 0.41–0.57). Prediction of awakening at ICU discharge was predominantly influenced by age. Neither EEG alone nor IMPACT without age effectively predicted awakening at ICU discharge (AUROC: 0.52 IQR: 0.43–0.59 and 0.54 IQR: 0.46–0.62, respectively). The combined EEG and IMPACT model (AUROC: 0.63; IQR: 0.53–0.69) was not significantly superior to EEG-and-age alone (AUROC: 0.60; IQR: 0.53–0.67). Early EEG connectivity features demonstrated clinically relevant predictive value for long-term outcome following severe TBI, surpassing most IMPACT score features, except age, whose predictive contribution may be influenced by self-fulfilling prophecy. These findings indicate that EEG features from cEEG monitoring provide complementary prognostic information during the acute phase of severe TBI when clinical predictors alone may be insufficient.
Anzalone et al. (Fri,) studied this question.