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

Area of Science:

• Neuroscience
• Critical Care Medicine
• Biomedical Engineering

Background:

• Accurate prognostication for comatose cardiac arrest survivors is critical but challenging.
• The Bispectral Index (BIS) monitor, typically used for anesthesia depth, may offer novel electroencephalogram (EEG) subparameters for predicting neurological recovery.

Purpose of the Study:

• To determine if internal EEG subparameters from the BIS monitor can predict recovery of consciousness in comatose cardiac arrest survivors.
• To compare the prognostic performance of a BIS-derived EEG model against the modified Westhall qualitative EEG scoring framework.

Main Methods:

• A retrospective cohort study involving 315 comatose cardiac arrest patients.
• A three-layer neural network trained on 48-hour continuous EEG recordings, processed using virtualized BIS Engine emulation.
• The model utilized hourly averaged BIS subparameters and was validated against the modified Westhall score.

Main Results:

• A neural network trained on four BIS subparameters (inverse burst suppression ratio, mean spectral power density, gamma power, theta/delta power) achieved maximum prognostic accuracy.
• The model demonstrated superior performance in predicting recovery of consciousness compared to qualitative EEG assessment (AUC 0.86, accuracy 0.87).
• Gamma band power emerged as a novel correlate of recovery potential, previously unreported in this context.

Conclusions:

• Four internal EEG features from the BIS Engine, processed by a neural network, significantly outperform current standards in predicting consciousness recovery after cardiac arrest.
• These repurposed BIS subparameters show promise for improved patient assessment and prognostication in the critical care setting.
• The findings highlight the potential of leveraging anesthesia monitoring technology for post-cardiac arrest neurological prognostication.