Abstracts

Rationale: A seizure likelihood marker could enable improved seizure monitoring and thereby facilitate adjustment of treatments based on daily seizure risk assessments. Specifically, electrodermal activity (EDA), an autonomic marker for sympathetic skin activity, exhibits unique properties in the setting of seizures and thus may serve as a digital predictive biomarker for seizure likelihood assessment. Here, we tested differences of patient-specific EDA 24-hour-modulation-amplitudes between patients with and without seizures.

Methods: We included patients wearing an E4 wearable biosensor (Empatica®, Milan, Italy), which records temperature and EDA, with either no seizure or with one or more seizures (focal impaired awareness or generalized tonic-clonic seizure) during continuous video-EEG monitoring at Boston Children’s Hospital. Because EDA relates to thermoregulation, we analyzed surface body temperature recorded at the wrist or ankle (TEMP). We modeled the 24-hour pattern of EDA and TEMP recordings with nonlinear mixed-effects harmonic models and determined the modulation amplitude by calculating the difference between peak and trough from the resulting curve per patient (Figure 1). For comparing the modulation amplitude between the 2 groups, we ran generalized estimating equations (GEEs) and determined the sensitivity and specificity of the amplitudes differentiating between patients with recorded seizures and no-seizure patients by logistic regression analysis. Seizures were confirmed based on video-EEG review.

Results: We included 61 patients with no seizure (sex: 25 males and 36 females; age: median (p25-p75), 12.8 (9.9-15.3) years) and 37 patients with a seizure (sex: 22 males and 15 females; age: median (p25-p75), 9.2 (5.6-14.4) years) during the recording. We found lower EDA modulation amplitudes (t = 2.14, p = 0.04, d = 0.45) and higher TEMP modulation amplitudes (t = 2.13, p = 0.04, d = 0.44) in patients with seizures as compared to patients without seizures. Including EDA and TEMP modulation amplitudes into logistic regression analysis revealed a preliminary sensitivity of 0.30 and a specificity of 0.92 (see Figure 1 for receiver operating characteristic curve generated from logistic regression probability values).

Conclusions: Taking the limitations of our retrospective analysis into account, i.e., short recordings, high inter-individual variability, and unequal group sizes, the results seem promising to follow up on the analysis of expressions of 24-hour modulation of neurophysiological signals in autonomic signals continuously recorded from wearables. The next steps include the addition of further clinical variables and validation of EDA modulation as a predictive biomarker for seizure likelihood.

Funding: Please list any funding that was received in support of this abstract.: This study was supported by the Epilepsy Research Fund. SV was supported by Deutsche Forschungsgemeinschaft, Grant/Award Number: VI 1088/1-1.