Two-Hourly Prediction of Hypoxic-Ischemic Injury Timing Using EEG During the Latent Phase in Near-Term Fetal Sheep

Hypoxia-ischemia (HI) around the time of birth can cause severe neuronal damage, leading to lifelong disability. Early intervention with therapeutic hypothermia in the initial 6-hour latent phase of injury significantly reduces neuronal damage. However, identifying the precise timing of HI onset and its subsequent evolution remains a significant clinical challenge. The development of tools to predict the timing of HI injury progression has the potential to revolutionize neonatal care, optimising the application of hypothermia whilst aiding further phase-specific therapeutic research.In this study we utilised an extensively validated near-term fetal sheep model-a well-established translational analogue for neonatal HI- to explore the feasibility of predicting injury timing using 2-hourly EEG recordings after HI, during the early-, mid-, and late-latent sub-phases, respectively. We demonstrate the effectiveness of machine learning (ML) classifiers, trained on spectral features extracted from 5-minute EEG windows, in predicting the three 2-hourly sub-phases within the entire latent phase. Notably, the KNN classifier achieved a cross-validated accuracy of 83.4±4.5% with a One-vs-the-Rest AUC of 0.94±0.05 in correctly identifying the timing of EEG sections.Clinical relevance-This study shows the feasibility of real-time EEG based prediction of timing after HI, highlighting potential for the development of a clinical decision support tool to optimise the therapeutic application of hypothermia.