Machine learning-based models for predicting glioma-associated epilepsy: a systematic review and meta-analysis

Area of Science:

• Neurology
• Artificial Intelligence
• Medical Informatics

Background:

• Glioma-associated epilepsy (GAE) is a frequent and debilitating complication in glioma patients.
• Predicting seizures in this population is difficult due to complex tumor-host interactions.
• Machine learning (ML) models offer advanced pattern detection capabilities for high-dimensional data.

Purpose of the Study:

• To systematically review and meta-analyze the predictive performance of ML-based models for GAE.
• To evaluate the diagnostic accuracy of ML models in predicting GAE.

Main Methods:

• A comprehensive systematic review following PRISMA guidelines was conducted across four major databases.
• Included studies developed ML models for GAE prediction.
• Pooled estimates for AUC, accuracy, sensitivity, specificity, and DOR were calculated.

Main Results:

• Thirteen studies involving 3,253 patients were analyzed.
• Pooled AUC was 0.87 (95% CI: 0.83-0.91) and pooled accuracy was 0.82 (95% CI: 0.76-0.88).
• Pooled sensitivity was 0.77 (95% CI: 0.64-0.87), specificity was 0.93 (95% CI: 0.86-0.96), and DOR was 40.1 (95% CI: 17.1-94.0).

Conclusions:

• ML-based models exhibit strong diagnostic performance for predicting GAE.
• Clinical integration can aid risk stratification and optimize therapeutic strategies.
• Addressing limitations like heterogeneity and lack of external validation is crucial before real-time implementation.