The unpredictability of spontaneous seizures is a major burden for epilepsy patients, affecting their independence, safety and psychological wellbeing. In recent years, several studies utilised continuous electroencephalogram (cEEG) to demonstrate the existence of robust cycles of interictal epileptiform activity (IEA) in patients with epilepsy. These cycles can be observed both daily and hourly, and might be useful to predict future seizure probability.
This retrospective study aimed at evaluating the feasibility of forecasting seizure days in adults with focal epilepsy.
The investigators evaluated retrospective cEEG data recorded with an implanted RNS system device in adults with medically refractory focal epilepsy followed between 2004 and 2018. Patients were split into two cohorts: 1) experienced 20 or more electrographic seizures (n=18; development cohort) and 2) experienced 20 or more self-reported seizures (n=157; validation cohort). The seizure risk was evaluated using the IEA recorded by the implanted device and by applying point process statistical models. The primary outcome of the study was the percentage of patients with forecasts showing an improvement over chance (IoC).
The models were able to generate daily seizure forecasts for the following calendar day with an IoC in 83% and 66% of patients in the development and validation cohorts, respectively. It was shown that the forecasting could be extended up to 3 days with an IoC in 11% and 39% of patients in the two respective cohorts. IoC was achieved in 100% of patients in the development cohort when testing forecasting of 1 hour.
The results of this study could serve as a basis for future prospective clinical trials that have the potential to establish how epilepsy patients can benefit from seizure prediction over longer periods of time. A reliable seizure prediction method could mitigate uncertainty and enable novel seizure prevention strategies or warning systems in the future.
The data reported corroborate the emerging opinion that epileptic seizures are not completely random events. This retrospective study further showed that epileptic seizure probability can be forecasted several days in advance by utilising IEA cycles recorded with an implanted device.