Predicting seizure recurrence after status epilepticus: a multicenter exploratory machine learning approach.

Area of Science:

• Neurology
• Artificial Intelligence in Medicine
• Predictive Analytics

Background:

• Predictive tools for seizure recurrence after status epilepticus (SE) are limited.
• Status epilepticus (SE) without prior seizures (de novo SE, dnSE) affects a significant patient population.
• Identifying patients at risk for recurrent seizures post-SE is clinically important.

Purpose of the Study:

• To evaluate the efficacy of various machine learning (ML) models in predicting seizure recurrence after a first episode of de novo SE (dnSE).
• To compare the predictive performance of ML algorithms against traditional logistic regression models.

Main Methods:

• A multicenter retrospective cohort study involving patients aged ≥16 years with dnSE.
• Development and validation of ML models (k-NN, Naïve Bayes, ANN, SVM, Decision Tree, Random Forest) and logistic regression using clinical and neurophysiological data.
• 70% of data for training, 30% for validation; predictive capability assessed using Area Under the Receiver Operating Characteristic Curve (AUROC).

Main Results:

• 35.2% of patients experienced seizure recurrence within two years post-SE.
• Progressive symptomatic SE etiology, non-convulsive SE with coma, and out-of-hospital SE were associated with recurrence; acute symptomatic SE was protective.
• The Random Forest algorithm achieved the highest predictive capability (AUROC 0.687), outperforming logistic regression (AUROC 0.594).

Conclusions:

• The Random Forest algorithm demonstrates superior predictive performance for seizure recurrence after dnSE compared to logistic regression.
• While AI shows potential, current predictive accuracy for seizure recurrence post-dnSE requires further improvement through additional research and data.
• Over one-third of patients with dnSE experience seizure recurrence within two years, highlighting the need for better predictive strategies.