Seizure Forecasting and Detection with Wearable Devices and Subcutaneous EEG – Outcomes from the My Seizure Gauge Trial
Benjamin Brinkmann1, Ewan Nurse2, Pedro Viana3, Mona Nasseri4, Levin Kuhlmann5, Philippa Karoly6, Tal Pal Attia1, Nicholas Gregg1, Boney Joseph1, Matthias Dumpelmann8, Mark Cook7, Gregory Worrell9, Andreas Schulze-Bonhage10, Mark Richardson11, Dean Freestone2
1Mayo Clinic, 2Seer Medical, 3King's College London, 4University of North Florida, 5Information Technology, Monash University, 6University of Melbourne, 7Neurology, University of Melbourne, 8Freiburg University, 9Neurology, Mayo Clinic College of Medicine, 10Universitaetsklinikum Freiburg, 11Institute of Psychiatry
To develop the ability to forecast seizures in patients with epilepsy without intracranial devices. 
Seizure forecasting has been established using intracranial EEG, but  minimally invasive devices may permit seizure forecasting and provide accurate seizure records.
Patients were recruited for ultra-long monitoring with a wearable device (Empatica E4, Fitbit Charge HR, or Fitbit Inspire) and concurrent ambulatory EEG monitoring (UNEEG SubQ, EpiMinder, NeuroPace RNS) at three sites. Wearable and EEG data from enrolled patients was recorded for 8 months or more. Self-reported electronic seizure diaries and periodic mood and symptom surveys were recorded by participants as well. Recorded data were analyzed to assess the ability to detect seizures, to identify circadian and multi-day cycles, and to forecast and detect seizures

Thirty-nine patients with epilepsy and one volunteer have recorded over 16000 days (43.8 years) of ambulatory wearable and EEG data, including over 1700 seizures. Nine patients left the study before completion due to device malfunctions, complications, poor adherence, or unanticipated seizure freedom. Analysis in this cohort has established the following: 

  • Seizure forecasting significantly better than chance in 6 patients with EEG confirmation of seizures using the wrist-worn Empatica E4 device for 6-12 months. Results from a large-scale data science contest on will be presented.  

  • Seizure forecasting using the subscalp EEG significantly greater than chance in 5 of 6 patients with at least four seizures recorded over at least three months using a Bidirectional LSTM neural network.  

  • Heart rate circadian and multi-day cycles were significantly phase-locked with self-reported seizure likelihood in 10 of 19 patients 

  • Tonic and phasic electrodermal activity, heart rate, and actigraphy cycles are significantly correlated with iEEG-confirmed seizures in 11 patients measured over at least 8 months. 

This project has established the feasibility of forecasting seizures using long-term cycles, wearable devices, and subcutaneous EEG.