6 - Development and Validation of the EEG-Based Neonatal Encephalopathy Score (ENES) for Real-Time and Objective Assessment of HIE Severity

Publication Number: 1005.6

Srinivas Kota, University of Texas Southwestern Medical School, Dallas, TX, United States; Saeed Montazeri, University of Helsinki, Espoo, Uusimaa, Finland; Lina Chalak, Childrens Health Dallas UT Southwestern, DALLAS, TX, United States; Sampsa Vanhatalo, Children's Hospital, Helsinki, Uusimaa, Finland

Background: Hypoxic-ischemic encephalopathy (HIE) after perinatal asphyxia is a leading cause of neonatal mortality and morbidity. The treatment protocols are largely based on clinically assessed HIE grades, which can be challenging in infants with mild HIE (HIE 1) that may show delayed HIE progression. We hypothesized that the bedside (a)EEG data could be used to train an EEG-based Neonatal Encephalopathy Score (ENES) that provides a continuous assessment of HIE severity within the first day of life.

Objective: To develop a fully automated ENES algorithm for monitoring HIE severity, and to validate its performance on independent patient cohorts.

Design/Methods: We developed a deep learning-based algorithm using (a)EEG data and postnatal age in hours from an infant cohort with only perinatal asphyxia (PA), mild (HIE1), moderate (HIE2), and severe (HIE3). The ENES output was made continuous, ranging from 0 (PA) to 3 (HIE 3). The model was trained using one cohort from Helsinki, and it was validated in three ways: i) cross-validation with the training cohort (N=91), ii) independent validation with a cohort from Helsinki (N=92), and iii) external validation with a cohort from Texas (N=45).

Results: The ENES demonstrated strong differentiation of HIE grades across the first 24hrs of life, in both datasets from Helsinki. In comparison, aEEG and BSN trends showed an expected recovery over time, and they could not differentiate HIE severities throughout the first day of life. External validation using an independent dataset of 44 term newborns with mild to severe HIE at UT Southwestern confirmed the robustness and generalizability of the ENES across study sites and EEG recording systems (figure 1A). Inspection of individual ENES time courses showed that ENES can effectively reflect the progression of an infant's HIE from mild to severe during the first day of life (figure 1B).

Conclusion(s): ENES is the algorithm that is developed to indicate the HIE severity, and it clearly outperforms the previously existing trends, aEEG and BSN. Neurophysiologically, the findings indicate that that EEG signal contains significant information about an infant's encephalopathy. Clinically, ENES holds promise for facilitating timely and objective identification of newborns for neuroprotective and other early interventions.

Figure 1
(A) Median ENES at 8-12 hours postnatal age compared across datasets I-III. Each dot represents a subject. (B) Individual ENES trend (semi-transparent lines) and median +- IQR (solid lines and the highlighted areas) colored per HIE group up to 24 hours postnatal age in datasets II and III.