Machine learning model to predict the efficacy of antiseizure medications in patients with familial genetic generalized epilepsy.

Abstract

OBJECTIVE:

This study aimed to establish a machine learning model that can predict the efficacy of antiseizure medications (ASMs) in patients with familial genetic generalized epilepsy (GGE).

METHODS:

We prospectively followed up patients with familial GGE for at least 3 years between January 2007 and January 2017. We collected and analyzed the patients' demographic characteristics, medical history, and related auxiliary examinations. The results of the epileptic seizures were divided into two categories: seizure-free and drug-resistant epilepsy. We selected and trained thirteen classification models, i.e., random forest classifier, logistic regression, gradient boosting classifier, light gradient boosting machine, ridge classifier, linear discriminant analysis, support vector machine-linear kernel, extra tree classifier, Ada boost classifier, naive Bayes classifier, decision tree classifier, K neighbors classifier, and quadratic discriminant analysis, to get the best performing classification model.

RESULTS:

A total of 854 patients with familial GGE were included in the study after excluding 89 who were lost to follow-up. Among them, 631 patients with familial GGE became seizure-free, and 223 developed drug-resistant epilepsy with a 74.89% remission rate. Among the 13 models, the random forest classifier model was the most effective with an accuracy of 91.23% and an F1 score of 84.21%. Among the 18 patient characteristics, the most effective indicators of the final treatment results were the number of seizure types experienced, response to the first drug, prior treatment duration and number of pre-treatment seizures.

SIGNIFICANCE:

The random forest classifier model can be used to early predict the results of ASM treatment based on the clinical data of patients with familial GGE. This finding can help clinicians make timely adjustments to treatment strategies and improve patients' prognosis.

• Authors: Junhong Wu, Yan Wang, Ling Xiang, Yixue Gu, Yin Yan, Lulin Li, Xin Tian, Wei Jing, Xuefeng Wang
• Journal: Epilepsy research (Epilepsy Res) Vol. 181 Pg. 106888 (03 2022) ISSN: 1872-6844 [Electronic] Netherlands
• PMID: 35176621 (Publication Type: Journal Article, Research Support, Non-U.S. Gov't)
• Copyright: Copyright © 2022 Elsevier B.V. All rights reserved.

Topics

• Bayes Theorem
• Epilepsy
• Epilepsy, Generalized (drug therapy, genetics)
• Epileptic Syndromes
• Humans
• Machine Learning