Automated assessment of EEG background for neurodevelopmental prediction in neonatal encephalopathy

Abstract

Objective: Assess the capacity of brain state of the newborn (BSN) to predict neurodevelopment outcomes in neonatal encephalopathy.

Methods: Trends of BSN, a deep learning-based measure translating EEG background to a continuous trend, were studied from a three-channel montage long-term EEG monitoring from a prospective cohort of 92 infants with neonatal encephalopathy and neurodevelopmental outcomes assessed by Bayley Scales of Infant Development, 3rd edition (Bayley-III) at 18 months. Outcome prediction used categories "Severe impairment" (Bayley-III composite score ≤70 or death) or "Any impairment" (score ≤85 or death).

Results: "Severe impairment" was predicted best for motor outcomes (24 h area under the curve (AUC) = 0.97), followed by cognitive (36 h AUC = 0.90), overall (24 h AUC = 0.84), and language (24 h AUC = 0.82). "Any impairment" was best predicted for motor outcomes (12 h AUC = 0.95), followed by cognitive (24 h AUC = 0.85), overall (12 h AUC = 0.75), and language (12 and 24 h AUC = 0.68). Optimal BSN cutoffs for outcome predictions evolved with the postnatal age. Low BSN scores reached a 100% positive prediction of poor outcomes at 24 h of age.

Interpretation: BSN is an excellent predictor of adverse neurodevelopmental outcomes in survivors of neonatal encephalopathy after therapeutic hypothermia, even at 24 h of life. The trend provides a fully automated, objective, quantified, and reliable interpretation of EEG background. The high temporal resolution supports continuous bedside brain assessment and early prognostication during the initial dynamic recovery phase.