Seizure prediction in 1117 neonates leveraging EMR-embedded standardized EEG reporting

Abstract

AbstractBackgroundAccurate prediction of seizures can help direct resource-intense continuous EEG (CEEG) monitoring to high-risk neonates. We aimed to use data extracted from standardized EEG reports to generate seizure prediction models for vulnerable neonates.MethodsIn 2018, we implemented a novel CEEG reporting system in the electronic medical record (EMR) that incorporated standardized terminology. We developed seizure prediction models using logistic regression, decision tree, and random forest models for neonates and specifically, neonates with hypoxic-ischemic encephalopathy (HIE), using EEG features on day 1 to predict future seizures.FindingsWe evaluated 1117 neonates, including 150 neonates with HIE, with CEEG data reported using standardized templates. Implementation of a consistent EEG reporting system, which documents discrete and standardized EEG variables, resulted in >95% reporting of key EEG features. Several EEG features were highly correlated, and patients could be clustered based on specific features. However, no simple combination of features adequately predicted seizure risk. We therefore applied computational models to complement clinical identification of high-risk neonates. Random forest models incorporating background features performed with classification accuracies of up to 90% for all neonates and 97% for neonates with HIE, and recall (sensitivity) of up to 97% for all neonates and >99% for neonates with HIE.InterpretationUsing data extracted from the standardized EEG report on the first day of CEEG, we predict the presence or absence of neonatal seizures on subsequent days with classification performances of >90%. This information, incorporated into routine care, can guide decisions about the necessity of continuing CEEG beyond the first day and thereby improve the allocation of limited CEEG resources. Additionally, this analysis illustrates the benefits of standardized clinical data collection which can drive learning health system approaches to personalized CEEG utilization.FundingChildren’s Hospital of Philadelphia, The Hartwell Foundation, NINDS, Wolfson FoundationResearch in contextEvidence before this studyWe searched the literature on EEG-based seizure prediction among neonates in PubMed from January 1, 1946, to June 1, 2022, using combinations of the keywords “seizure,” “prediction,” “EEG,” “neonatal,” and “hypoxic-ischemic encephalopathy.” We used no language restrictions. Prior studies relied on manual review of EEG reports to forecast seizures in neonates using regression models. These studies were limited in sample size as they required manual review of reports and manual data entry. No studies were identified using automated collection of EEG data from routine care, and none used machine learning-based modeling techniques.Added value of this studyWe built seizure prediction models based on standardized EEG features reported in the EMR which could predict seizures in neonates, and particularly those with HIE, with greater than 90% accuracy. Furthermore, these models could be tuned to not miss seizures, performing with recall (sensitivity) of up to 97% in the overall neonatal cohort and >99% among neonates with HIE, while still maintaining precision (positive predictive value) of up to 92% and 97%, respectively. Previous studies have built seizure-prediction models using EEG data, but most have used features derived from manual scoring of EEG tracings or computational analysis of the raw EEG recordings. While these studies are informative, they are not easily scalable for incorporation into routine clinical practice. To our knowledge, this is the first study reporting a seizure-prediction model based on standardized reports already documented in the EMR that can be used for clinical decision support to improve care for critically ill neonates. Prediction models developed in our study are available at http://neopredict.helbiglab.io.Implications of all the available evidenceContinuous EEG monitoring is currently the standard of care for critically ill children at increased risk of seizures. While effective for seizure detection, long-term monitoring is resource-intensive and can have physical and psychosocial consequences, such as skin breakdown and reduced bonding. Accurately predicting which neonates are likely to seize after an initial shorter period of monitoring would help allocate resources towards neonates at highest risk of seizures and avoid unnecessary use of limited EEG monitoring resources in neonates at low risk of seizures. Furthermore, the ability to directly extract these predictors from the EMR will allow for automated predictions and dashboard development for use at scale and in real-time in clinical care.