LAUSR

Significance Dynamical systems have been used to model the paroxysmal nature of seizures in epilepsy. In parallel, seizures have been discovered to occur with cyclical periodicity in epilepsy, occurring at specific phases of latent epileptogenic activity. Many questions regarding drivers of interindividual heterogeneity in seizure cyclicity remain unanswered. Here, we demonstrate the application of a switching linear dynamical system statistical model, with covariate selection, to the estimation of latent seizure cycles in epilepsy. We apply our model to 1,012 people with epilepsy to characterize heterogeneity in seizure chronotype. Spectral and unsupervised clustering analyses validate the ability to capture latent cycling patterns and reveal that the periodicity of multidien cycles varies according to age and susceptibility to seizure triggers.