Table 2 Machine learning algorithms for the response to AEDs treatment
Study | Study design | Prediction target | Training and testing cohorts | Algorithms | Predictive features | Performance | External validation |
|---|---|---|---|---|---|---|---|
Devinsky et al. 2017 [18] | Retrospective | Choice of AEDs for individual patients | 40, 000 patients for training and 10, 000 patients for testing | Machine learning algorithm | About 5 000 features | AUC of 0.72 | Yes |
An et al. 2017 [19] | Retrospective | Prediction of DRE | 175, 735 were training cohort and the other 117, 157 were test cohort | Multivariate logistic regression, support vector machine and random forest | 1 270 features | AUC of 0.76 | No |
Petrovski et al. 2009 [20] | Prospective | Prediction of AEDs treatment outcomes | 115 patients with newly diagnosed epilepsy | K-nearest neighbors | 279 candidate genes | Accuracy of 83.5% and sensitivity above 80% | Yes |
Yao et al. 2019 [22] | Retrospective | Prediction of AEDs treatment outcomes | 287 patients with newly diagnosed epilepsy | Decision tree, random forest, support vector machine, XGBoost and logistic regression | Demographic features, medical history, EEG and MRI | F1 score and AUC value showed good performance | No |
Zhang et al. 2018 [23] | Retrospective | Prediction of efficacy of levetiracetam | 46 patients with newly diagnosed epilepsy | Support vector machine | Clinical features and sample entropy | 75.0% accuracy in the training set and 90% in the test set | No |