Tailoring antiseizure treatment with a wearable device: A proof-of-concept study in absence epilepsy

Abstract

Objective

Typical absence seizures are underreported. We aimed to improve patient care using a wearable electroencephalograph (wEEG) at home and assess a machine learning (ML) pipeline for absence detection.

Methods

Patients with typical absences used a wEEG device 12–24 h 1 week after antiseizure medication (ASM) adjustments. Three-hertz generalized spike–wave discharges (SWDs) ≥ 3 s were used as absence surrogates. After manual inspection, we used the results to guide medical treatment. The outcomes were seizure freedom, number of consecutive measurements without relapse, and side effects. Afterward, we used the ML pipeline on the recordings, and a neurologist reviewed the output. Review time and diagnostic performance were compared with manual inspection.

Results

Nineteen patients (12 female, median age = 24 years) were followed for a median of 5 months (range = 1–12). The median recording time for each session was 21.3 h (range = 10–24). Fifteen patients (79%) were seizure-free during the last measurement, including seven of 11 (63%) diagnosed with refractory epilepsy. Ten patients relapsed after a median of 1–2 recordings (range = 1–6) without 3-Hz SWDs. Side effects occurred in 21% of patients. Manual file inspection identified 806 3-Hz SWDs of ≥3 s. The ML pipeline reduced a neurologist's median review time for 24-h wEEG from 27 (range = 10–45) to 4.3 min (range = .1–10), with a sensitivity, precision, F1-score, and false positives per hour of .8, .95, .87, and .007, respectively.

Significance

Home-based wEEG allows patient monitoring after ASM adjustments, improving absence seizure management. The ML-based pipeline performed well and was crucial in reducing review time.