Using Quantitative EEG to Stratify Epilepsy Risk After Neonatal Encephalopathy: A Comparison of Automatically Extracted Features

Purpose: Neonatal encephalopathy (NE) is a commonly encountered, highly morbid condition with a pressing need for accurate epilepsy prognostication. We evaluated the use of automated EEG for prediction of early life epilepsy after NE treated with therapeutic hypothermia (TH). Methods: We conducted retrospective analysis of neonates with moderate-to-severe NE who underwent TH at a single center. The first 24 hours of EEG data underwent automated artifact removal and quantitative EEG (qEEG) analysis with subsequent evaluation of qEEG feature accuracy at the 1st and 20th hour for epilepsy risk stratification. Results: Of 144 neonates with NE, 67 completed at least 1 year of follow-up with a neurologist and were included. Twenty-three percent had seizures (N = 18) in the NICU and 9% developed epilepsy (N = 6). We found multiple automatically extracted qEEG features were predictive of epilepsy as early as the first hour of life, with improved risk stratification during the first day of life. In the 20th hour EEG, absolute spectral power best stratified epilepsy risk, with area under the curve ranging from 76% to 83% across spectral frequencies, followed by range EEG features including width, SD, upper and lower margin, and median. Clinical examination did not significantly predict epilepsy development. Conclusions and significance: Quantitative EEG features significantly predicted early life epilepsy after NE. Automatically extracted qEEG may represent a practical tool for improving risk stratification for post-NE epilepsy development. Future work is needed to validate using automated EEG for prediction of epilepsy in a larger cohort.