A proposal for a machine-learning algorithm for the prediction of seizure recurrence risk at 2 years after discontinuation of anti-seizure medications

Background This article presents a novel method to create a scale for predicting the risk of seizure recurrence 2 years after discontinuation of anti-seizure medications (ASMs). The approach enables the development of a straightforward, easily calculable score scale with a clear-cut threshold to effectively identify high-risk patients. Methods Clinical and demographic features of epileptic patients who had discontinued their ASMs were incorporated into a customized R algorithm. This algorithm evaluated each covariate and its corresponding weight to determine the combination of weighted scores creating the scale with the highest accuracy for predicting seizure recurrence risk. Moreover, the algorithm generated a threshold that differentiates between low- and high-risk patients. Results In our dataset, the 10 covariates whose combination was associated with the scale with the maximized Youden index (0.51) selected by the algorithm were: “duration of epilepsy”, “duration of the seizure-free period on therapy”, “duration of ASM tapering”, “development delay”, “age at ASM withdrawal over 50 years”, “gender”, “age at ASM withdrawal over 40 years”, “structural etiology of epilepsy”, “failure of previous ASM discontinuations”, “age at seizure onset”; The sensitivity and specificity of the scale were 0.87 and 0.64 respectively. Scale scores above 5 identify high-risk patients. Conclusions This work does not aim to propose a definitive scale to predict the risk of seizure recurrence 2 years after discontinuation of ASMs. Rather, it introduces a data-driven tool to support the development of a scale with a clear-cut threshold, easily applicable in the clinical practice and understandable to patients.