Seizure monitoring by combined diary and wearable data: A multicenter, longitudinal, observational study

Abstract Objective In patients with intractable epilepsy, accurate diaries of seizure occurrence and timing can substantially inform management. Wearable devices that provide confirmation of seizure occurrence complement such diaries, which are frequently incomplete and/or inaccurate. Here, we combined seizure diaries and longitudinally deployed wrist‐worn device recordings to evaluate whether wearable recordings contain information that can discriminate between days containing seizure‐related activity and those without. Methods Patients with focal seizures were prospectively enrolled in a clinical trial to test the effectiveness of eslicarbazepine acetate as an adjunct to levetiracetam or lamotrigine (phase IV clinical trial NCT03116828). One hundred two patients maintained a seizure diary and wore a biosensor for >30 weeks. Based on diaries, we labeled days as either “seizure” versus “no‐seizure” or “preseizure” versus “no‐preseizure.” We compared patterns obtained by harmonic 24‐h modeling between conditions. Best‐ranking wearable markers and seizure diary variables were fed into a fully connected neural network, with several hidden layers and depth as hyperparameters that classified between seizure day conditions. Results The final sample contained 70 patients (median age = 42.5 years, 43 female) with 5437 recorded patient‐days, including 557 seizure days and 537 preseizure days. Twenty‐four‐hour patterns in electrodermal activity and accelerometry differentiated no‐seizure versus seizure days, as well as no‐preseizure versus preseizure days ( p < .01). Classification between no‐seizure and seizure days (weighted F1 = .81, sensitivity = .82, specificity = .67) as well as between no‐preseizure and preseizure days (weighted F1 = .82, sensitivity = .80, specificity = .66) revealed good performance. Significance Wearable data capture seizure‐related differences with daily resolution, differentiating between days with lower and higher seizure susceptibility. Combining diary‐based clinical and wearable data bears the potential for developing a dynamic seizure detection and prediction system with daily resolution.