Rationale:
Artificial intelligence (AI) combined with accelerometer (ACC) and electrodermal activity (EDA)-based wristbands has shown accurate detection of generalized tonic-clonic seizures (GTCS) in both EMU [1] and real-world settings [2]. Recently, the EpiMonitor system received FDA clearance for GTCS detection, employing a previously validated algorithm together with a next-generation ACC and EDA wristband. Because detection performance is contingent on both the algorithm and the quality of the sensor signals, it is essential to establish that the system maintains substantial equivalence when a new wearable component is introduced.
[1] Onorati F, et al. Prospective Study of a Multimodal Convulsive Seizure Detection Wearable System on Pediatric and Adult Patients in the Epilepsy Monitoring Unit. Front Neurol. 2021;12:724904.
[2] Chen W, et al. Real-world Evaluation of an FDA-cleared Wrist-worn Seizure Detector. American Epilepsy Society Annual Meeting, 2024.
Methods:
Patients (n = 35) at risk of GTCS wore the EpiMonitor device in real-life settings. When a likely GTCS was detected, the wristband triggered an alert to caregivers, who could confirm the event as a GTCS or mark it as a false alarm (FA); missed GTCS could also be manually logged.
Outpatient seizures were validated by combining patient/caregiver reports with blinded manual bio-signal review of device-triggered (confirmed or unconfirmed) and manually reported events by two independent raters.
Endpoints evaluated were positive percentage agreement (PPA) and daily false alarm rate (FAR), with PPA as the ratio of true positive GTCS to total GTCS, corrected for multiple GTCS per patient, and overall FAR as the total false alarms divided by the total days (day = 24 hours of recorded data). Confidence intervals (CI) used non-parametric bootstrapping.
Results:
The analyzed dataset consisted of a total of 2491 days from 35 patients (age 9-78, mean 20.9), during which 107 GTCS from 35 patients occurred (GTCS per patient range 1-16, median 2). 103 out of 107 GTCS were detected, providing a corrected PPA=0.94 (CI PPA=[0.89, 0.98]). The overall FAR per day was 0.24 (CI FAR=[0.12, 0.39]). The distribution of individual FAR showed that 11.4% of patients had no FAs, 45.7% of patients had FAR within 0.1 per day, i.e., one false alert every 10 days, and 68.6% of patients had FAR within 0.2 per day, i.e, one false alert every 5 days (see Figure 1). Peaks in FAs were observed during daytime, with only 8.3% of false alerts reported during nighttime (Figure 2).
Conclusions:
GTCS detection using an FDA-cleared algorithm combined with a next-generation ACC and EDA-based wrist-worn device demonstrated robust and generalizable performance in real-world settings, with PPA and FAR remaining within the ranges established by predecessor wristbands. These findings support the continued reliability of the algorithm across hardware generations and reinforce the clinical utility of wearable seizure detection as a scalable tool for real-world epilepsy management.Funding:
Funded by Empatica