# The future is in the background: background EEG patterns, not acute seizures, predict epilepsy and neurodevelopmental outcomes in neonatal HIE

## TL;DR

This study shows that abnormal background EEG patterns in newborns with brain injury are better predictors of future epilepsy and developmental issues than seizures.

## Contribution

The study demonstrates that EEG background patterns are more predictive of outcomes than seizure burden in neonatal HIE.

## Key findings

- Severely abnormal EEG backgrounds strongly predict future epilepsy (RR = 7.22).
- Background EEG patterns are better predictors of neurodevelopmental outcomes than ictal burden.
- Machine learning models using background EEG features achieved high accuracy and AUC.

## Abstract

Hypoxic ischemic encephalopathy (HIE) is the most common neurologic emergency in the neonatal population, with a broad spectrum of potential neurodevelopmental outcomes. Additionally, HIE is the most common cause of seizures during the acute neonatal period. Unfortunately, predicting neurodevelopmental outcomes and epilepsy risk is difficult in this population, and seizure burden during the acute period has not consistently been correlated with outcomes in prior studies. We aimed to examine EEG background data to determine whether there is a relationship between background abnormalities, neurodevelopmental outcomes, and epilepsy risk, and whether this information is more informative for predicting outcomes compared to other clinical data points.

Patients were retrospectively recruited from level 3 Neonatal Intensive Care Units (NICU's) in Calgary, Alberta, from 2014 to 2020. All patients who met the criteria for therapeutic hypothermia after being classified as at risk for HIE were included in the study. Clinical information captured included measures from clinical examination, blood work, MRI (day 3–5, scored using Barkovich scoring system) and medications. Continuous video EEG (cvEEG) recordings were separated into day 1, 2, and 3, and separate classifications systems were used for background and ictal findings. Neurodevelopmental follow-up was completed at two years of age, and patients were also categorized as having no epilepsy, or either well-controlled or refractory epilepsy. Poisson regression models and relative risk were used to compare background and ictal scores to long term neurodevelopmental outcomes and future epilepsy risk. Three supervised learning algorithms were trained to predict neurodevelopmental outcomes based on clinical factors.

Two-hundred and six patients were eligible for the study. Among neonates with seizures, only 18% developed epilepsy, while 52% of those with severely abnormal EEG background patterns did. Total ictal burden was not significantly associated with epilepsy at follow up, and no antiseizures medications were significant predictors. In contrast, EEG background score was strongly associated with epilepsy risk (adjusted ß = 2.75, p = 0.002), with severely abnormal backgrounds conferring significantly increased risk (37.5% vs. 5.2%, RR = 7.22, 95% CI: 3.09–16.88). Similarly, ictal burden did not predict poor neurodevelopmental outcome or death, whereas background score was a strong predictor (adjusted ß = 1.74, p < 0.001; RR = 2.44, 95% CI: 1.70–3.50). Machine learning models identified background features as more predictive than ictal scores, with XGBoost achieving the best classification performance (accuracy 0.724) and random forest yielding the highest AUC (0.751).

In our cohort, EEG background patterns outperformed ictal burden in predicting both neurodevelopmental outcomes and future epilepsy risk. Although background patterns are not directly modifiable, they provide powerful, early markers of brain injury severity, offering clinicians a valuable tool for prognostication and family counseling at a critical juncture in care.