Longer-horizon seizure prediction based on continuous EEG data that incorporated circadian and multi-day (multidien) cycles is feasible, a retrospective analysis found.
“This study shows that seizure probability can be forecasted days in advance by leveraging multidien interictal epileptiform activity cycles recorded with an implanted device,” wrote Maxime O. Baud, MD, PhD, of University of Bern, Switzerland, and co-authors in Lancet Neurology. “These results highlight the feasibility of seizure forecasting over horizons longer than previously possible.”
The study used data from two groups of adults who had drug-resistant focal epilepsy and a responsive neurostimulator system (RNS) device implanted for clinical indications: a development cohort (n=18; electrographic seizure data) and a validation cohort (n=157; self-reported seizure data).
The device allowed underlying patterns in interictal epileptiform activity to be identified. With ictal data, the information was used to develop and test individual patient models relating interictal epileptiform activity to cyclic biorhythms and predict seizures.
For the primary outcome — the percentage of patients with forecasts showing improvement over chance — data based on multidien cycles enabled seizure forecast above chance 24 hours in advance in the majority of patients: 15 of 18 (83%) in the development cohort and 103 of 157 (66%) in the validation cohort.
The data also forecasted seizures to 3 days in advance in 2 of 18 (11%) and 61 of 157 (39%) in the two cohorts, respectively. Hourly seizure forecasts showed above-chance performance in all 18 patients (100%) in the development cohort.
“Our results support the evidence base that seizures are not entirely random events. Use of cyclical patterns of brain activity to forecast seizures hours to days in advance might enable novel seizure warning systems and prevention strategies,” Baud and colleagues wrote. “This study will serve as a basis for prospective clinical trials to establish how people with epilepsy might benefit from seizure forecasting over long horizons.”
In an accompanying editorial, Mark J. Cook, MD, of University of Melbourne in Australia, wrote, “Crucially, the study showed that cyclic seizure patterns could allow seizures to be predicted over much longer time frames than previously shown (days vs minutes or hours), and that the complex cycles can be simply extracted from accurate capture of the timing of events. The study confirms that the capture of event frequency and event times alone can lead to clinically useful seizure forecasting.”
“Most importantly, collecting data with intracranial devices no longer appears necessary for accurate prediction, which could be achieved by other less invasive methods of EEG capture,” Cook added.
The unpredictability of seizure occurrence is a top patient concern regardless of seizure type or frequency and also affects how implanted neurostimulation to abort seizures is applied. Monitoring devices such as the RNS system have provided data that allows detailed exploration of interictal discharge, biorhythm cycles, and seizures.
A review of seizure forecasting cited “an overwhelming body of evidence” showing non-random, time-specific patterns relating seizures to daily (circadian), shorter-than-daily (ultradian), and multidien cycles. While multidien periods range from 7 to 35 days from patient to patient, they are relatively stable within an individual.
A 2016 study of the RNS device found that epileptiform activity was strongly periodic over 24 hours, peaking at night, with differences in circadian influence by seizure onset zone (e.g., limbic versus neocortical onset). Prior work by Baud and co-authors in 37 patients showed that interictal epileptiform activity oscillated with circadian and patient-specific multi-day periods, with seizures occurring preferentially during the rising phase of multidien interictal epileptiform activity rhythms.
“Seizures were shown to be occurring at particular time points in these cycles, and this information was harnessed to create prediction algorithms,” Cook noted. “The possibility that these cycles might be a consequence of lifestyle patterns, or even the result of anticonvulsant therapies, was resolved by the identification of similar cycles in animal models of epilepsy.”
Building on their earlier work, Baud and colleagues analyzed RNS data from focal epilepsy patients age 18 or older obtained between January 2004 and May 2018. Each had 20 or more electrographic seizures (development cohort) or self-reported seizures (validation cohort) recorded or reported.
Forecasting models were validated by applying them to data from the separate validation cohort, which then were used to forecast self-reported seizures that had been logged daily by participants enrolled in the 9-year RNS system clinical trial, Baud and colleagues said.
“Prospective clinical trials are planned to assess directly the ways in which people with epilepsy might benefit from replacing a state of constant subjective uncertainty about upcoming seizures with a continuum of quantified uncertainty (a forecasted probability) at different horizons,” they noted. “The clinical utility of seizure forecasting has not been established by this study or previous studies.”
Limitations of the study include lack of information on medication non-compliance. Study conclusions should be considered hypothesis-generating rather than clinical evidence, the researchers said.
-
Longer-horizon seizure prediction based on continuous EEG data incorporating circadian and multi-day cycles is feasible, a retrospective study showed.
-
The study used data from patients who had drug-resistant focal epilepsy and a responsive neurostimulator system device implanted for clinical indications; similar results might be obtained with less invasive forms of data-gathering.
Paul Smyth, MD, Contributing Writer, BreakingMED™
The study had no funding.
Baud reports personal fees and grants from the Wyss Center for Bio and Neuroengineering and has a pending patent.
Cook holds two patents and has stock in Epiminder and Seer Medical.
Cat ID: 34
Topic ID: 82,34,730,34,192,925
