Seizure Detection as a Digital Biomarker

Seizure detection is the automated identification of seizure events from wearable sensors. The clinical problem it addresses is specific and well documented: seizure diaries, which remain the basis for most treatment decisions and trial endpoints in epilepsy, miss a substantial proportion of events. People are frequently unaware of their own seizures, particularly those occurring in sleep or with impaired awareness, and carers are not always present.

Detection systems currently target generalised tonic clonic seizures, the type with the most distinctive physiological signature and the highest risk. They combine movement, which shows characteristic rhythmic activity, with autonomic signals such as electrodermal activity and heart rate, which surge markedly during and after these events.

This is the most regulated application in the neurology section of this library, and also the narrowest.

Wrist worn multimodal devices record accelerometry together with autonomic signals and apply an algorithm that flags candidate events, usually alerting a caregiver. Detection of generalised tonic clonic seizures using a wrist accelerometer was demonstrated in a prospective multicentre study, and multimodal detectors combining movement with electrodermal activity have been assessed in multicentre clinical evaluations.

The field has published standards for testing and clinical validation of seizure detection devices, which specify how sensitivity and false alarm rate should be measured and reported. That matters because sensitivity alone is meaningless here: a device can detect almost everything if it also alarms constantly, and the false alarm rate determines whether a family can live with it.

Reference standard is video electroencephalography monitoring, which is why most validation happens in epilepsy monitoring units.

The main use is safety alerting at home, particularly overnight, where the risk of sudden unexpected death in epilepsy is highest and where an unwitnessed seizure may go unnoticed until morning. The second use is objective seizure counting, which improves on diary data for both clinical management and trial endpoints.

In research, detection is used to characterise true seizure frequency, to test whether an intervention reduces events the person may not report, and to study circadian patterns of seizure occurrence. Because no seizure specific questionnaire exists in this library, the honest position is that the clinical counterpart is the seizure diary itself, and that diary is exactly the instrument this measure exists to improve on.

Coverage is narrow. Cleared systems target generalised tonic clonic seizures. Focal seizures, absence seizures and other non convulsive types have far weaker physiological signatures at the wrist and are not reliably detected, so a device cleared for convulsive seizures does not provide general seizure monitoring, and assuming otherwise is dangerous.

False alarms remain a practical burden. Vigorous activity, tooth brushing and some sleep movements can trigger alerts, and repeated false alarms erode trust and adherence.

Detection is also not prediction. These systems recognise a seizure as it happens or immediately after; they do not forecast one. Validation comes largely from monitoring units, where behaviour differs from home, so real world performance can differ from published figures.

• Beniczky S, et al. Detection of generalized tonic-clonic seizures by a wireless wrist accelerometer: a prospective, multicenter study. Epilepsia. 2013. pubmed.ncbi.nlm.nih.gov
• Onorati F, et al. Multicenter clinical assessment of improved wearable multimodal convulsive seizure detectors. Epilepsia. 2017. pubmed.ncbi.nlm.nih.gov
• Beniczky S, et al. Standards for testing and clinical validation of seizure detection devices. Epilepsia. 2018. pubmed.ncbi.nlm.nih.gov
• Elger CE, Hoppe C. Diagnostic challenges in epilepsy: seizure under-reporting and seizure detection. Lancet Neurol. 2018. pubmed.ncbi.nlm.nih.gov