Abstracts

Rationale: Accurate seizure detection is paramount in clinical decision making for patients with epilepsy. Evaluation by inpatient video-EEG recording in the epilepsy monitoring unit (EMU), the current gold standard, is highly resource intensive, of limited availability and cannot be undertaken over long periods. The current seizure recording method in the community relies on patients completing a seizure ‘diary,’ either paper or electronic. This method is inaccurate because many conditions can mimic seizures and over 50% of seizures are not recognised by patients. A number of wearable devices have been tested for non-invasive and ambulatory seizure detection. Only one device has received regulatory approval. We reviewed current evidence on the performance of these devices in detecting different seizure types. Methods: We performed a systematic review of Pubmed, EMBASE and Web of Science to identify published studies between 2000 and 2019 that reported the sensitivity (percentage) of a particular device in detecting a type of epileptic or psychogenic non-epileptic seizure (PNES), as compared to the EMU diagnosis. Search terms included ‘device’, ‘seizure’ and ‘detection’. Studies that tested devices that were invasive or restricted the patient’s mobility, were excluded. For each included study, details on the parameters measured, number of events, sensitivity, false alarm rates (FAR) and study conclusions, were extracted. Results: We reviewed 291 publications, of which 23 met our criteria (Figure 1). The median number of patients included in these studies was 14 (range 5 to 71); median number of events captured was 22 (range 4 to 51); median assessment period in the EMU 4 days (range 4 to 5). A variety of consumer and research devices were tested. 19 studies detected motor seizures only, 2 non-motor seizures only, and 2 both seizure types (Table 1). 12 detected generalised tonic clonic seizures (GTCS), of which 7 studies measured 3D accelerometry (ACM), with sensitivities ranging from 31% to 100%, and FARs 11.8/24 hours (h) to 0.12/24h. The other 5 studies measured electromyographic (EMG) signals, with sensitivities ranging from 63% to 100% and FARs 4.03/24h to 0.96/24h. 1 study detected other types of motor seizures using 3D ACM (sensitivity 91%; FAR 25/24h). 3 studies detected GTCS and focal to bilateral tonic clonic. Of these, 1 measured EMG signals only (sensitivity 76%; FAR 2.52/24h) and the other two studies measured 3D ACM and electrodermal activity (EDA) (sensitivity 94.55% and 88%; FARs 0.2/24h and 1/24h, respectively). 3 studies detected GTCS and convulsive PNES using 3D ACM with sensitivities ranging from 95.2% to 100% and FARs 5/24h to 7/24h. In the 2 studies detecting non-motor seizures only, HR and photoplethysmography were measured in one (sensitivity 57%; FAR 46.1/24h) and EEG signals in another (sensitivity 98.4%; FAR 5.52/24h). In the 2 studies that included both motor and non-motor seizures, one study measured heart rate, peripheral capillary oxygen saturations and EDA (sensitivity 100%; FAR 0.24/24h), and the remaining study measured cerebral oxygenation, displaying low sensitivity (24%) and no report of FAR. Conclusions: While wearable devices hold promise for non-invasive detection of epileptic seizures, firm conclusion of their performance from this review is limited by their short-term in-hospital evaluation in low number of patients and seizures. Prospective studies in larger patient populations over longer periods are needed to determine their utility in routine epilepsy management. Funding: No funding