Abstracts

Rationale: Clinically validated automated mobile seizure detection devices have been poorly tested in outpatient settings to date. Detectors relying on motion-related signals to recognize convulsive seizures (CSs) are likely to yield higher false alarm rates (FAR) in real life than in a clinical environment, due to a wider range of possible physical activities. Embrace is a wrist-worn device combining Accelerometer and Electrodermal Activity to detect and alert to CSs. Retrospective analysis combining machine learning classifiers trained on outpatient data (Onorati et al, 12th European Congress on Epileptology, 2016), and different sensitivity levels for active and rest periods (Caborni et al, 32nd International Epilepsy Congress, 2017), yielded a FAR of less than 1 false alarm (FA) per day. The aim of this work is to investigate Embrace real-time performances when allowing user-adjustable sensitivity level in outpatient settings. Methods: Ten users were enrolled to test sensitivity levels of Embrace for 1 week. They were asked to switch the levels at their discretion, using the smartphone app paired with their Embrace. They were told that sensitivity level ‘Sens1’ was suggested during active periods, to reduce FAs triggered by non-seizure events, while ‘Sens2’ was suggested during rest time, to increase the probability of detecting seizures during sleep, a potentially more dangerous time. They were instructed to meticulously report each CS. Sensitivity (Se) was computed as the ratio of CSs that triggered an alert to the total number of CSs. FAs were counted by subtracting the number of detected CSs from the alert number. FAR was obtained as the number of FAs divided by the total recorded hours, normalized by 24 hrs. Rest periods were labeled off-line with a proprietary algorithm. Results: Overall, 1,288 hrs of data were recorded (median hrs/patient=144, min=36, max=187): 711 hrs at ‘Sens1’ and 577 hrs at ‘Sens2’. The users set the level to ‘Sens1’ for 56% of active time and to ‘Sens2’ for 50% of rest time, which is not as ideal as recommended, but reflects real use. Overall, 4 CSs were reported by 3 users, including 3 nocturnal seizures. The global real-time performances were Se=4/4 CSs and FAR=0.48 per 24 hrs (26 FAs overall). At ‘Sens1’, Se=1/1 CSs and FAR=0.44, whereas at ‘Sens2’, Se=3/3 CSs and FAR=0.54, with an increase of 22.7%. During rest, no FAs were triggered. During active periods, FAR=0.52 at ‘Sens1’ and FAR=0.67 at ‘Sens2’. Off-line simulations using only ‘Sens1’ showed that Se=2/4, with 2 nocturnal CSs missed, and FAR=0.32, while simulations using only ‘Sens2’ provided Se=4/4 and FAR=0.93. Conclusions: We evaluated real-time performance of the Embrace system when users were allowed to switch sensitivity levels. Comparison with standard mode (level ’Sens1’ only) showed that a user-adjustable sensitivity level can improve sensitivity for nocturnal CS detection (Se from 2/4 to 4/4) without making unbearable FARs (FAR from 0.32 to 0.48). Future work is underway evaluating a larger pool of users and longer monitoring periods.  Funding: No funding