Abstracts

Rationale: There is resurgent interest in the role played by autonomic dysfunction in seizure generation. Advances in wearable sensors make it convenient to track many autonomic variables in patient populations. The purpose of this study is to assess peri-ictal changes in surrogate measures of autonomic activity in epilepsy patients. Methods: All procedures received prior IRB approval. One patient admitted for presurgical evaluation using intracranial EEG (iEEG) was monitored for four days with additional sensors for surface EEG (fronto-central), EKG, submental EMG and a wrist-worn device (Empatica) that measured variables relevant to autonomic function (AF), specifically electrodermal activity (EDA), heart rate (HR), blood volume pulse (BVP) and skin temperature (ST). Six electro-clinical seizures were identified, all during sleep, and corresponding one-hour preictal segments were extracted for analysis. Six hour-long interictal segments, 4 during wakefulness and 2 during sleep, were also identified. To verify that the labeled samples of different states were distinct, they were compared using electrophysiologically derived features. Since the surface EEG  turned out to be unreliable, sleep-related spectral band power features S1 (Delta/Theta power ratio) and S2 (Alpha+Sigma+Beta+Gamma)/(Delta+Theta) power ratio) of the iEEG, and the mean-squared value of the submental EMG, were estimated in 1-second long non-overlapping windows. A high frequency signal power feature (HF) was also estimated from iEEG in the seizure onset zone. These features were compared for interictal and preictal samples using ANOVA. The mean value of each AF variable was computed in successive 2-min epochs and compared for interictal sleep, interictal wake, and preictal periods using ANOVA. A naive Bayes classifier was designed and tested using ten-fold cross-validation to test whether preictal and interictal epochs could be distinguished using AF variables alone. Results: Sleep features S1, S2, and EMG power differed significantly for the manually labelled vigilance states: high S1, low S2, and low EMG corresponded to sleep while low S1, high S2, and high EMG marked wakefulness. Furthermore, S1, S2, and HF differed significantly for preictal and interictal sleep (p < 0.001). The AF variables showed relevant changes as well: EDA increased drastically, while HR and BVP experienced marked variability, in the ictal compared to the preictal period. EDA and HR were significantly different for preictal and interictal segments (ANOVA; p < 0.001); the change in ST did not reach significance (p = 0.052). The naive Bayes classifier labeled preictal epochs with 90% sensitivity and 96% specificity based on AF variables. Conclusions: Appreciable preictal changes in EDA, ST, and HR were documented in a patient monitored during Phase II presurgical evaluation. Recruitment is ongoing and data collected from two additional patients are being analyzed. This early result suggests that autonomic measurements, which can be conveniently measured using noninvasive devices, may have some predictive value for epileptic seizures in certain individuals. Funding: This study was made possible by: scholarship support from the Higher Committee for Education in Iraq to AA; an Alpha Omega Alpha Postgraduate Award to MFV; a seed grant from EpiC, the University of Kentucky Epilepsy Research Center, to MBO, SS and MV; and National Science Foundation Grant No. 1539068 to SS.