Abstracts

Rationale: Epilepsy is a disorder characterized by paroxysmal transitions between multistable states. Dynamical systems have been useful for modeling the paroxysmal nature of seizures. At the same time, intracranial EEG recordings have recently demonstrated that an electrographic measure of epileptogenicity, interictal epileptiform activity, exhibits cycling patterns ranging from ultradian to multidien rhythmicity, with clinical seizures locked to specific phases of these latent cycles. However, the factors that contribute to variations in characteristic periodicities of seizure cycles remain elusive._x000D_
Methods: We developed a Bayesian switching linear dynamical system (SLDS) to estimate latent cycles in epilepsy from patient-reported clinical seizures via a Markov chain Monte Carlo algorithm that employs particle Gibbs with ancestral sampling. The SLDS was applied to 1012 people with self-reported clinical seizures from SeizureTracker.com to estimate underlying latent multidien cycles. Unsupervised learning was performed on spectral features of latent cycles to investigate factors that contribute to different characteristic periodicities in people with epilepsy.  _x000D_
Results: Two clusters of patients were identified with different spectral properties of multidien cycles.  Patients in the first cluster exhibited stronger tendency towards shorter cycles, greater spectral spread, and greater spectral entropy, indicating a tendency toward less well-defined and shorter multidien cycles. In contrast, patients in the second cluster exhibited longer cycles, lower spectral variance, and lower spectral entropy, indicating a tendency toward more well-defined and longer multidien cycles. Adults were more likely to belong to the first cluster whereas children were more likely to belong to the second cluster, indicating a
tendency toward shorter multidien cycle periodicities among adults and longer multidien cycle periodicities among children. People whose seizures were frequently triggered by tiredness, stress, or hormonal fluctuations were more likely to belong to the first cluster, indicating a tendency toward shorter multidien cycle periodicities in this group. People who reported their seizures to be triggered by stress, mood, and tiredness had more power in short cycle bands, whereas those whose seizures were triggered by overheating and light had more power in longer cycle bands.

Conclusions: This work contributes to knowledge of cycling in epilepsy by investigating how multidien seizure cycles vary across people with epilepsy. In addition, the switching linear dynamical system developed in this work provides a statistical approach that may be used in future work to estimate seizure cycling within a nonlinear dynamical systems framework. _x000D_
Funding: ETW was supported by a fellowship from the Gulf Coast Consortia on the NLM Training Program in Biomedical Informatics and Data Science T15LM007093. SC is supported by the National Institute of Neurological Disorders and Stroke, National Institutes of Health (5R25NS070680-12). The contents are solely the responsibility of the authors and do not necessarily represent the views of the NIH.