Abstracts

Rationale: Detecting and predicting seizures from peripheral recordings of autonomic nervous system (ANS) activity offers the potential for developing non-stigmatizing, easy-to-use devices for pediatric patients. However, relevant biomarkers that differentiate between inter- and pre-ictal states are still missing. Central nervous system physiological changes may precede epileptic seizures long before their abrupt onset. It has been suggested that these early warning signs are signatures of network dynamics approaching a state transition. Theoretical considerations of seizure onset as a dynamic state transition exhibited by the ANS dictate an increase in information shared and transferred between subsystems of the ANS towards seizure onset. Here, we empirically tested these predictions by monitoring the information shared and transferred between subsystems of the ANS during the transition from normal to epileptic seizure in a large, long-term data set. Methods: People with epilepsy (PWE) were equipped with a wireless multi-sensor device (Empatica® E4, Milan, Italy) during long-term, continuous video-EEG monitoring at Boston Children’s Hospital. We analyzed all epileptic seizures that a patient had during this monitoring period, including subclinical seizures, if they were at least two hours away from a preceding seizure. Data analysis was performed on 30-second data segments. A segment was considered pre-ictal if it occurred between 61 minutes and one minute prior to a seizure, or as inter-ictal if it occurred at least two hours away from any seizure. Continuous electrodermal activity (EDA), temperature (TEMP), heart rate (HR), mean and variance, as unimodal measures as well as mutual information (MI), and transfer entropy (TE) were calculated for all possible modality pairs per segment and averaged over inter- or pre-ictal periods, respectively. Multivariate repeated measure analyses of variance (ANOVA) were used for statistical analysis. Results: Sixty-seven PWE (9.49±5.92 years) were included. While TEMP, F(1, 67) = 27.074, p < .001, ηp2 = .288, and HR, F(1, 67) = 11.547, p = .001, ηp2 = .147 were higher in pre- than in inter-ictal periods, mean EDA did not differ significantly. Signal variance did not exhibit a significant difference between inter- and pre-ictal periods. Significant period by modality pair interactions for MI, F(2, 134) = 4.831, p = .009, ηp2 = .067, showed that MI increased for all modality pairs but to a different extent. The interaction for TE, F(1, 134) = 5.958, p < .001, ηp2 = .082, showed that all modality pairs except for HR-TEMP, showed higher TE in pre- than in inter-ictal periods. Conclusions: Theory-informed measures of ANS activity within and across modalities differ between inter- and pre-ictal periods. The overall increase in interactions between ANS subsystems is in line with the expectations that network dynamics approach a state transition at seizure onset. Markers related to system dynamics show the potential for a differentiation between inter- and pre-ictal data and might therefore be potential biomarkers for timely seizure detection and prediction. Further investigation of these transitions may serve as groundwork for improving seizure forecasting algorithms and diagnostics in the clinic and in broad ambulatory settings. Funding: Funded by ERF