Developing an Electroencephalogram-based Model to Predict Awakening after Cardiac Arrest Using Partial Processing with the BIS Engine.

Summary

Using 48-hour EEG processed through a virtualized BIS Engine, a compact neural network leveraging four subparameters (inverse BSR, mean spectral power density, gamma power, theta/delta power) predicted recovery of command-following after cardiac arrest with AUC 0.86 and high sensitivity, outperforming qualitative EEG scoring. Gamma power emerged as a novel correlate of recovery potential.

Key Findings

• A 3-layer neural network using four BIS subparameters achieved AUC 0.86, accuracy 0.87, sensitivity 0.83, specificity 0.88.
• Model outperformed the modified Westhall qualitative EEG framework for predicting command-following recovery.
• Gamma band power was identified as a novel positive correlate of recovery potential after cardiac arrest.
• Hourly-averaged features from 48-hour frontotemporal EEG were sufficient to drive high performance.

Clinical Implications

If externally validated, BIS-based EEG subparameters could enable earlier, accessible prognostication after cardiac arrest using equipment already present in many ICUs and ORs, informing counseling and care pathways. Use should remain adjunctive within multimodal prognostication.

Limitations

• Single-center retrospective design without external, multi-center validation.
• Dependence on proprietary BIS subparameters; potential confounding from sedation and targeted temperature management.

Future Directions

Prospective multi-center validation, integration with clinical variables and multimodal predictors, and assessment of real-time bedside deployment and impact on decision-making and outcomes.