To assess seizure precipitants and the value of electronic diary (E-diary) data in seizure prediction among patients with medication-resistant epilepsy (MRE).

Accurate seizure prediction would improve treatment and safety for people with epilepsy. Small studies utilizing a prospective written or E-diary approach have demonstrated an ability of select patients to predict occurrence of seizures with a common trigger being self-reported stress. As part of a multicenter, double-blind, randomized control trial to assess stress reduction for MRE, we evaluated factors associated with successful seizure self-prediction.

Patients were followed for 8-week baseline and 12-week treatment periods. Twice daily, participants rated the likelihood of a seizure in the next 24 hours on a 5-point scale from “very unlikely” to “almost certain”, along with mood, premonitory symptoms, stress ratings and seizure counts. A model of seizure prediction utilizing a gradient boosting algorithm calculated false detection and true positive rates of seizure occurrence based on these data. Seizure threshold scores were calculated to generate a relationship between these rates and a ROC curve was generated. Relative influence scores were generated to identify variables most predictive of seizure occurrence.

66 participants contributed data (3126 seizures) and had >85% diary adherence. E-diary data increased the true positive rates of seizure occurrence over a wide range of false detection rates. Subjective predictability increased following a seizure, suggesting some patients may be predicting “clusters” of seizures. Neither self-reported stress nor other reported precipitants were predictive factors in the model.  

Some patients with MRE are able to predict seizures to some extent. Prediction improved following seizures, suggesting some patients may be predicting “clusters” of seizures. Future studies should focus on understanding factors that drive seizure prediction and combine E-Diary with EEG and circadian data.