Predicting antiseizure medication response in newly diagnosed epilepsy using quantitative EEG and machine learning

Area of Science:

• Neurology
• Biomedical Engineering
• Data Science

Background:

• Predicting long-term epilepsy outcomes is challenging using only clinical data and visual EEG interpretation.
• Machine learning (ML) offers potential for pattern recognition in EEG data for outcome prediction.
• ML application for predicting anti-seizure medication (ASM) response in newly diagnosed epilepsy is underexplored.

Purpose of the Study:

• Develop and evaluate ML models to predict ASM response in newly diagnosed epilepsy patients.
• Enhance personalized treatment strategies and facilitate early identification of drug resistance.
• Compare predictive performance using clinical data, EEG features, and combined data.

Main Methods:

• Retrospective cohort study of 94 adult epilepsy patients with new diagnoses and pre-treatment EEG.
• Exclusion of patients with structural brain lesions on MRI.
• Three ML approaches: clinical variables only, EEG features only, and combined data.
• Models used: Logistic Regression, XGBoost, Random Forest; evaluated at epoch and patient levels.
• Responders defined as seizure-free for one year during the second year post-ASM initiation.

Main Results:

• 77.7% of patients achieved seizure freedom.
• Clinical features alone yielded moderate prediction (XGBoost AUROC: 0.69).
• EEG features improved prediction (Random Forest AUROC: 0.68 at patient level).
• Combined clinical-EEG model significantly enhanced prediction (Random Forest AUROC: 0.81).
• Key EEG predictors: beta/gamma band power spectral density and sample entropy.

Conclusions:

• Quantitative EEG analysis with ML shows promise for predicting epilepsy prognosis.
• Integrated clinical and quantitative EEG data improve ASM response prediction.
• ML models can support personalized epilepsy treatment and early detection of drug resistance.
• Further validation in larger, diverse populations is required for clinical implementation.