Prognostication of neurological outcomes in comatose patients is non-trivial.  EEG is used widely for monitoring, but expert interpretation has shown only moderate interrater reliability. Quantitative (qEEG) features promise increased standardization and enable deeper information extraction from EEG.

The specific prognostic effects of spatial and temporal variables is a knowledge gap in literature. Prediction of long-term morbidity (as opposed to dichotomized good/bad) is nearly non-existent.

Coherence between EEG channels is a qEEG feature that has been used to study functional connectivity of brain function, which may provide insights into healing or deterioration in specific brain regions. 

To explore the prognostic value for neurological outcome of regional EEG coherence at specific time intervals post resuscitation.

We analyzed raw EEG of 351 patients from the I-CARE dataset, spanning 0-100 hours post resuscitation and rewarming. We calculated coherence for all electrode pairs across 5 frequency bands and grouped them spatially. We then performed factor analysis for dimensionality reduction and retained 9 factors. Factors had high contributions from the following variables:

• 0 : γ-coherence
• 1: left >> right central-central/occipital/parietal/temporal
• 2: non-γ, non-frontal
• 3: intra-frontal lobe
• 4: right-hemisphere
• 5: β, α coherences
• 6: intra-parietal lobe, parietal-occipital/central/temporal
• 7: frontal-other, θ-δ > α, right-left > right or left
• 8: occipital-occipital/parietal/central

High factor 3 and low factor 7 predicted dichotomized good outcome from 0-100 hours. Low factors 0 and 2 <60h predict good outcome.  High factor 6 <48h and low >60h predict good outcome.

Correlation with long-term morbidity varied. High factor 0 at 48-84h, high factor 5 >60h predicted moderate disability. High factor 2 was seen in mild, moderate, severe disability at >72h, <12h, 0-84h. High factor 4 at <12h, 48-84h predicted mild disability. 

Factor analysis indicated that coherence values autocorrelate by frequency and geometry.  Unique combinations provide specific insights at discretized time-points, including risk assessment of morbidity.