Babies born extremely early, before completing 28 weeks of pregnancy, carry a high chance of facing lasting issues with brain development and function. Monitoring brain activity soon after birth using amplitude-integrated EEG along with the original EEG waveforms (aEEG–EEG) offers promise for forecasting how these infants will develop over time. The goal here was to pinpoint specific qualitative and quantitative aspects of these early recordings that could indicate later neurodevelopmental results. Researchers reviewed data from extremely preterm newborns (delivered before 28 weeks and 0 days) who received standard two-channel aEEG–EEG tracking for the first 72 hours in the neonatal unit at Wilhelmina Children’s Hospital, Utrecht, Netherlands, spanning June 2008 through September 2018. Only those free from major genetic conditions, metabolic disorders, or serious congenital disabilities qualified. We extracted a wide range of characteristics from cleaned signals, covering qualitative elements in three main areas—overall background activity, cycles of sleep and wakefulness, and seizure-like events—and quantitative measures across four groups: frequency spectrum details, voltage levels, brain connectivity signals, and periods of signal interruption. Advanced machine learning techniques for both continuous prediction and category sorting tested how well these features could forecast 13 separate developmental measures (involving thinking skills, movement abilities, and behavior concerns) when children reached 2–3 years and again at 5–7 years. All evaluations adjusted for influencing variables including exact weeks of gestation, mother’s schooling, how sick the baby was, total morphine given, serious brain damage, and any drugs for seizures, calming, or anesthesia. The analysis covered 369 eligible infants and generated 339 distinct aEEG–EEG variables (9 qualitative plus 330 quantitative). Predictive regression models built with machine learning reached meaningful but limited success (correlations between 0.13 and 0.23) across nine of the 13 endpoints. Classification models, however, performed reasonably well at spotting children who later showed intellectual challenges versus those with the strongest results at school-entry age. Balanced accuracy hit 0.77 [95% CI: 0.62–0.90; P = 0.0020] for overall intelligence scores and 0.81 [0.65–0.96; P = 0.0010] for language-based intelligence scores. These same accuracy levels held steady even after limiting inputs strictly to the quantitative variables. Results show that brain monitoring data collected right after birth can help automatically flag extremely preterm babies at risk for difficult developmental trajectories. Such tools could become useful, understandable aids for clinicians making decisions and shaping care plans.
