Multimodal wearable sensors inform cycles of seizure risk

Objective Seizure unpredictability is a major source of disability for people with epilepsy. Recent work using chronic brain recordings has established that for many individuals with epilepsy seizure risk is not random, but corresponds to circadian and multiday (multidien) cycles in brain excitability. Here, we aimed to evaluate whether multimodal wearable device recordings can characterize cycles of seizure risk, and compare wearables performance with concurrent chronic brain recordings. Methods Fourteen subjects underwent long-term ambulatory monitoring with a multimodal wrist worn device (measuring heart rate, heart rate variability, accelerometry, tonic and phasic electrodermal activity, temperature) and an implanted responsive neurostimulation system (measuring interictal epileptiform abnormalities (IEA) and electrographic seizures). Wavelet time-frequency analyses identified circadian and multiday cycles in wearable and brain recordings. Circular statistics assessed seizure phase locking to cycles in physiology. Results Ten subjects met inclusion criteria. The mean recording duration was 232 days. Seven subjects had reliable electrographic seizure detections (mean 76 seizures). Seizure phase locking to multiday cycles occurred in six (IEA), five (temperature), four (heart rate, phasic electrodermal activity), and three (accelerometry, heart rate variability, tonic electrodermal activity) subjects. Seizure phase locking to residual HR multiday cycles (HR after regression of correlated physical activity (ACC)) increased to six subjects. Interpretation Long timescale cyclical changes in wearable recordings are common in epilepsy, and seizures occur at preferred phases of these cycles for many individuals. Broadly accessible wearable technology can provide new insights into the chronobiology of epilepsy with implications for seizure forecasting.