This abstract has been invited to present during the Better Patient Outcomes through Diversity Platform poster session
Rationale: Epilepsy, one of the most common neurological disorders, is characterized by spontaneous seizures that seem to occur randomly. Yet, studies using chronic electroencephalography (cEEG) have revealed daily (circadian) and multi-day (multidien) cycles of brain epileptic activity. Leveraging these cycles for clinical seizure forecasting depends critically on their prevalence, strength, and patterns, which are currently unknown. Methods: Here, using long-term cEEG (median 5.9 y) from participants in the RNS System clinical trials (N=256), we identify “seizure chronotypes,” common temporal patterns of seizure timing across individuals. Empirical circular distributions were obtained and their skewness measured as the Phase-Locking Value (PLV). We applied generalized linear models (GLM) to model these distributions. Results: Multidien and circadian seizure cycles were highly prevalent (60% and 89%, Figure1) and had similar strengths (PLV = 0.34 ± 0.18 and 0.34 ± 0.17, Figure1). Across subjects, we identified five circadian peak times: morning, mid-afternoon, evening, early night, and late night, and multidien periodicities centered around four common peaks: 7, 15, 20, and 30 days. Independent of multidien period length, clinical (self-reported) seizures consistently occurred during the rising phase of cycles of interictal epileptiform activity. Incorporating these potential biomarkers as variables in a GLM demonstrated above-chance predictive value in a majority of individuals and at the group level. Conclusions: Our findings reveal the high prevalence and characteristic patterns of seizure cycles at circadian and multidien timescales with similar strength. Further, we confirm the robustness of these organizing principles of seizure timing, using a statistical model that extracts commonalities across subjects. These findings suggest that seizure forecasting using these cycles may be possible and have broad applicability.
Figure 1. Prevalence and strength of seizure cycles at multiple timescales. Distribution of phase-locking values (PLV, y-axis; dots are individual subjects) for seizures in relation to six underlying cycles (x-axis): a) circannual, b) monthly, c) lunar, d) weekly, e) multidien, f) circadian. Seizure cycle strength (effect sizes) can be interpreted as weak (PLV ≤ 0.2), moderate (PLV > 0.2 to ≤ 0.4), strong (PLV > 0.4 to ≤ 0.6) and very strong (PLV > 0.6). Violin contours highlight the shape of the distribution. Funding: Please list any funding that was received in support of this abstract.: Wyss Center for neuro-engineering Ambizione fellowship, Swiss National Science Foundation