Abstracts

Rationale: Seizure detection may provide timely warning to patients with epilepsy (PWE), and may facilitate optimal treatment. ECG is a promising seizure detection modality with advantages including ability to monitor long term, less complex setup, and less stigmatization. Most ECG-based methods for seizure forecasting and detection, e.g., responsive VNS systems, use heart rate-related variables. A recent study found full ECG signals contain more information about pre-ictal states than heart rate variables alone, and may also supplement SUDEP risk assessment. We aim to identify pre-ictal states based on ECG features and validate data-analytic findings based on reported ECG changes in the literature. Methods: We used multiday ECG data from ten patients to train deep neural networks to differentiate pre- from interictal states. Algorithm performance was assessed on out-of-sample test data from the same patients (70% training, 30% testing). We compared models using full ECG signal, presented by its power spectrum, to models using heart rate variables alone. We then used layer-wise relevance propagation (LRP) to determine the most relevant ECG signal features. Based on these data-driven results, we performed a literature review according to PRISMA guidelines using key words related to ECG, SUDEP and epilepsy. Inclusion criteria were patients with all epilepsy types and ECG. Exclusion criteria were literature not written in English, publications focused on heart rate variability, case reports, literature reviews, and animal studies. We identified 502 abstracts, screened these based on our criteria, reviewed 110 full manuscripts, and included 24 papers in the review. Results: ECG signal in the frequency range <= 40 Hz were most informative of preictal state based on data from 10 patients. In this series, full ECG performed better than models using heart rate variables alone. Based on our literature review, patients with epilepsy are prone to ECG abnormalities beyond heart rate. In the literature, PWE had abnormal QTc intervals, ST segment abnormalities, elevated T Waves, early repolarization, increased P Wave dispersion and PR intervals during interictal periods when compared to controls. When comparing pre- to ictal period information in the literature, arrhythmias, QTc prolongation and ST segment changes were most common. While most of these changes may be benign, one study found pathologic ictal QTc interval prolongation in 23% of patients, while another found serious ictal ECG changes in 6%, with ST segment changes among the potentially fatal changes. Conclusions: Low frequency ECG changes may assist in the detection of the preictal state, based on data from 10 patients. The literature review supports that PWE have distinct ECG baseline abnormalities when compared to healthy controls. Comparison of pre- to interictal and ictal periods points to ECG changes beyond heart rate. Better characterization of these low-frequency ECG changes may improve seizure detection and forecasting, and help identify patients at risk for SUDEP. Funding: Supported in part by ERF and NARSAD Young Investigator Grant of the Brain & Behavior Research Foundation.