Authors :
Presenting Author: Stephanie Dailey, BA – Boston Children's Hospital, 300 Longwood Ave, Boston, MA 02115, USA
Tanuj Hasija, PhD, MSc – Paderborn University, Paderborn, Germany
Rasmus Jakobsmeyer, Ph.D. – Paderborn University
Maurice Kuschel, MSc – Paderborn University, Paderborn, Germany
Dennis Reinsch, M.Sc – Paderborn University
Henric Menne, B.Sc Candidate – Paderborn University
Nele Weike, M.Sc – Paderborn University
Jana Schulz, M.Sc – Paderborn University
Bo Zhang, PhD – Boston Children’s Hospital, Harvard Medical School, Boston, MA, USA
Michele Jackson, BA – Boston Children's Hospital, 300 Longwood Ave, Boston, MA 02115, USA
Claus Reinsberger, MD, PhD – Paderborn University/Mass General Brigham
Tobias Loddenkemper, MD – Boston Children's Hospital, 300 Longwood Ave, Boston, MA 02115, USA
Rationale:
Wearable devices are an increasingly important tool in seizure detection and prediction for patients with epilepsy (PWE). Cardiac data has garnered attention in epilepsy monitoring, so understanding its wearable signal variation across different device types is crucial. We conducted a multisite study to compare blood volume pulse (BVP) features in different movement states between wearable device types within both healthy controls and pediatric patients admitted for long-term epilepsy monitoring (LTM).
Methods:
Healthy controls from Paderborn University (UPB) wore an electrocardiogram (ECG), the Empatica E4 (Milan, Italy) and Emotibit (California, USA) on the dominant hand, and the EmbracePlus (Milan, Italy) on the non-dominant hand, with recordings taken between 12/2024 and 1/2025. Wearable data were recorded while the participant was resting (supine) and standing Patients included from Boston Children’s Hospital (BCH) were between 1 month and 21 years old and were admitted for LTM between 11/2023 and 2/2025 and wore the Empatica E4 and EmbracePlus (Milan, Italy) on the same wrist. Wearable device recording segments were selected for asleep and awake stages, with an epileptologist confirming the sleep state via electroencephalogram.
The signal duration for analysis ranged from 7-9 minutes. Time stamps from all compared devices were synchronized and aligned at UPB and BCH according to previously published protocols. BVP feature analysis was conducted on seven features using Kubios HRV Scientific software (V 4.1.2.1), and feature values were averaged over each segment duration per participant and device. Mean values and standard deviations across those at each site were computed, and the Pearson correlation coefficient between ECG and all wearable features at UPB and the two BCH devices was calculated. We performed a one-way repeated measures ANOVA to compare BVP features recorded from the three UPB devices and a paired t-test for the same features from the two BCH devices.
Results:
We included 17 controls from UPB (52.9% female, median age: 25.0) and 7 pediatric patients from BCH (14.3% female, median age: 13.2). Of these 7, 6 had a confirmed diagnosis of epilepsy following the LTM. Those with complete data from all devices were included in the analysis, with multiple recordings per patient from BCH averaged. BCH awake data showed more variability between devices. However, no significant differences were found between the features extracted from each device type across different conditions within each site (p >0.05, Table 1). Most BVP features from the UPB devices were correlated with the ECG features, and features were correlated between the two devices in the BCH asleep state BCH (Table 2).
Conclusions:
This investigation will enable us to understand the signal recording differences between devices better and develop device-agnostic seizure monitoring for PWE. Future investigations will compare additional autonomic signals and expand to a larger study population to confirm this pilot study’s findings.
Funding:
The Epilepsy Research Fund and Paderborn University Research Award supported this study.