Machine learning models for decision support in epilepsy management: A critical review

Area of Science:

• Neurology
• Medical Informatics
• Artificial Intelligence

Background:

• Epilepsy management presents significant clinical challenges, including trial-and-error medication selection and long delays for surgical intervention.
• Approximately one-third of patients experience drug-resistant epilepsy (DRE), highlighting the need for improved treatment strategies.
• Current diagnostic and treatment pathways for epilepsy require enhanced decision support tools.

Purpose of the Study:

• To review the potential application of machine learning (ML) predictive models as clinical decision support tools in epilepsy management.
• To assess the use of ML in predicting anti-seizure medication (ASM) responsiveness and identifying patients with DRE.
• To evaluate ML's role in identifying surgical candidates and predicting epilepsy surgery outcomes.

Main Methods:

• A comprehensive literature search was conducted on Medline and Embase for studies published between 2000 and 2020.
• Included studies focused on ML applications in predicting ASM responsiveness, DRE, surgical candidacy, and surgical outcomes in epilepsy.
• Data sources included clinical, neuropsychological, imaging, electroencephalography, and health claims data.

Main Results:

• 24 relevant articles were identified, with a focus on predicting surgical outcomes (13 studies).
• A variety of ML algorithms were employed, with some models demonstrating high predictive performance (AUC > 0.9).
• Most studies were single-center with small sample sizes; limitations include lack of standardization and prospective evaluation.

Conclusions:

• The clinical utility of ML models for decision support in epilepsy management requires further determination.
• Future research should prioritize larger studies with external validation and prospective evaluation of ML models.
• Standardization in reporting model performance is crucial for advancing the field.