Early identification of epilepsy surgery candidates: A multicenter, machine learning study.

Area of Science:

• Neurology
• Artificial Intelligence
• Medical Informatics

Background:

• Epilepsy surgery is underutilized, delaying critical interventions for patients.
• Automated identification of surgical candidates can facilitate earlier treatment.
• Developing predictive models is key to improving access to epilepsy surgery.

Purpose of the Study:

• To develop and validate site-specific machine learning (ML) algorithms for identifying potential epilepsy surgery candidates.
• To enable earlier detection of patients eligible for resective epilepsy surgery.
• To assess the performance of ML models in diverse pediatric and adult epilepsy centers.

Main Methods:

• A multicenter, retrospective, longitudinal cohort study was conducted.
• ML algorithms were trained using diverse data sources including clinical notes, EEG/MRI reports, and patient demographics.
• Site-specific algorithms were developed for pediatric and adult epilepsy centers, distinguishing surgical cases from controls.

Main Results:

• Pediatric surgical candidates were identified up to 2.0 years before presurgical evaluation (AUC=0.76).
• Adult surgical candidates were identified up to 1.0 year before presurgical evaluation (AUC=0.85).
• By the time of presurgical evaluation, ML models achieved high accuracy (AUC=0.93 for pediatrics, AUC=0.95 for adults).

Conclusions:

• Site-specific ML algorithms effectively identify epilepsy surgery candidates early in the disease course.
• These algorithms demonstrate utility in diverse clinical settings, potentially increasing epilepsy surgery utilization.
• Early identification through ML can streamline the pathway to surgical intervention for epilepsy patients.