Circular mixture model-based characterization of time-of-day seizure patterns during presurgical monitoring

Abstract Epilepsy is one of the most prevalent neurological diseases. Despite advancements in treatment, Drug-resistant epilepsy continues to pose significant challenges. Identifying temporal patterns in seizure occurrence may provide insight into the mechanisms underlying seizure timing. This study analyzes seizure patterns in patients undergoing presurgical evaluation using the long-term electroencephalogram EPILEPSIAE database. After applying exclusion criteria, 280 patients were analyzed, encompassing 1813 annotated seizure onsets. Seizure timing was treated as a circular variable, and each patient’s data were fit to a von Mises Mixture Model, with each component representing a cluster of seizure onset times. Clustering strength was evaluated using the Bayesian Information Criterion by comparing each model against a uniform one. To address the increased false positive rate associated with small sample sizes, primary results are reported for patients with more than 8 seizures. In this subgroup, model-supported evidence of time-of-day seizure onset clustering was found in 30% of patients. Results from patients with fewer seizures are reported separately as exploratory analyses due to reduced model stability at low sample sizes. These findings demonstrate that, for some patients, structured time-of-day patterns can be detected even during presurgical monitoring, where medication tapering, sleep disruption, and the hospital environment may perturb seizure timing, suggesting that rhythmic influences on seizure occurrence may persist despite substantial environmental perturbations.