Scientists have proposed a new model for predicting seizures based on EEG data. It will accurately identify newborns who need EEG monitoring and avoid complex and unnecessary procedures in other cases.
Researchers at Stanford University have developed a first-of-its-kind neonatal seizure prediction model to identify high-risk infants who require EEG monitoring in the NICU. The results of the work were published in The Lancet Digital Health.
|The results of more than 42 thousand EEGs of 1117 newborns were analyzed. Hypoxic ischemic encephalopathy was identified in 150 participants. To create the prediction model, the researchers used standardized EEG features from electronic health records.
Initially, the authors used a traditional logistic regression model, but its accuracy in predicting seizures was only 84%. After applying a machine learning algorithm, it was possible to build a new model that, with high accuracy, made it possible to identify children at high risk of seizures and, therefore, predict individual needs for EEG monitoring. The accuracy of this model for predicting seizures in neonates with hypoxic ischemic encephalopathy was greater than 90%.
The scientists emphasized that the model can be tuned to avoid missing seizures, providing a sensitivity of up to 97% in the general population of newborns and 100% among newborns with hypoxic ischemic encephalopathy. The diagnostic service is available as an online tool.
The researchers expect that incorporating this model into actual clinical practice could significantly improve the quality and effectiveness of care during these critical early days of children’s lives.