P.091 Parameterized short-segment EEG improves neurological recovery prediction in patients with severe brain injury

Background: Predicting neurological recovery in patients with severe brain injury remains challenging. Continuous EEG monitoring can detect malignant patterns but is resource-intensive, and its role in long-term functional outcome prediction is unclear. This study evaluates the utility of parameterized short-segment EEG, acquired via EEG cap, in predicting neurological recovery. Methods: We analyzed short-segment high-density EEGs from 42 patients in the NET-ICU cohort with acute neurological injury. EEGs were pre-processed into standard clinical formats and parameterized using five visual EEG features associated with outcome prediction. Random Forest Classifier (RFC) models were trained and cross-validated to predict recovery of responsiveness (following 1-2 step commands during or after ICU admission) using: EEG features alone; clinician prediction combined with EEG features. Results: EEG-based prediction outperformed clinician bedside assessment (AUC ROC: 0.80 vs. 0.67) under the RFC model. Combining clinician Glasgow Outcome Scale–Extended (GOSE) scores with EEG features improved overall predictive performance (AUC ROC: 0.91). Conclusions: Standardized EEG features obtained using EEG caps can improve the accuracy of neurological recovery predictions in patients with acute severe brain injury. This suggests that automated extraction of background brain signals has the potential to provide clinically meaningful prognostic information in critical care settings, enhancing accessibility and resource efficiency.