Progress in Research for Seizure Prediction
In the 16th Judith Hoyer Lecture on Epilepsy at the annual meeting of the American Epilepsy Society in New Orleans, Michael Privitera, MD, of the University of Cincinnati Gardner Neuroscience Institute spoke about advances in research on seizure prediction. Work in seizure prediction is of vital importance as patients report that knowing when a seizure will occur would significantly improve their quality of life. In addition it is hoped that seizure prediction can lead to improved medical and neuromodulatory treatment and might also help decrease the risk of sudden unexpected death in epilepsy (SUDEP).
A major challenge in predicting seizures is that there are multiple factors that increase seizure tendency, or the risk of a seizure, and a seizure occurs when these multiple factors together increase total tendency above a certain seizure threshold that seizure occurs. Among these factors are individual triggers of which stress is the most commonly reported, heart rate changes, EEG changes (either scalp or intracranial), circadian and other multi-day rhythms, and fluctuations in stress hormones.
Many patients report that they “knew a seizure was coming” before one occurred, but retrospective reports may not be accurate. In prospective studies in which patients’ twice-daily predictions of seizure likelihood were compared with actual seizure occurrences, it was found that about 20% of patients are accurate predictors who had an 8 to 10 times higher likelihood of having a seizure when they reported a seizure was “highly likely”. This self-prediction combined with self-reported stress levels led to even better prediction accuracy. Based on this, a stress-reduction intervention was tested in a prospective controlled trial (Neurology. 2018;90:e963-e970); the group treated with a proven muscle relaxation technique had a 29% reduction in mean seizure frequency and the group treated with a control or ‘sham’ intervention of writing down the events of the day had a 25% reduction in mean seizure frequency, which may have been a placebo effect or a result of the self-reflection having a “mindfulness” therapeutic effect on stress and seizure that was unexpected.
Experience with neuromodulation devices that stimulate the vagus nerve in response to heart rate changes (VNS) or the brain directly in response to changes in intracranial EEG (RNS) have shown that seizures can be predicted through both of these measures. The long-term continual recording that occurs with RNS shows that more seizure-like activity occurs than is experienced clinically. Dr. Privitera brought up the possibility that some patients’ accurate self-prediction could be tied to subclinical EEG discharges, which future studies of RNS could elucidate. In other studies of long-term intracranial EEG recording, lessons learned include that prediction of seizure must be highly individualized and multivariate (Curr Opin Neurol. 2017;30(2):167-173). False positive alerts may reflect an increased risk of seizures that did not cross the seizure threshold and therefore are not really false positives. False positives in the deterministic sense may not be probabilistic false positives because the seizure tendency may have gone up with the seizure threshold being crossed. The field of seizure prediction will benefit from newer, more accurate devices, improvement in patient diary reporting methods, and machine learning methods to analyze the large and complex data sets.