Forecasting seizure clusters from chronic ambulatory electrocorticography

Area of Science:

• Neurology
• Epilepsy Research
• Computational Neuroscience

Background:

• Seizure clusters pose significant risks, including morbidity and sudden unexpected death in epilepsy.
• Accurate forecasting of seizure clusters is crucial for timely intervention.
• Current methods for predicting seizure clusters are limited.

Purpose of the Study:

• To evaluate the performance of a novel algorithm designed for forecasting seizure clusters.
• To assess the feasibility of predicting periods of heightened seizure risk.

Main Methods:

• Analysis of chronic ambulatory electrocorticography data from 10 subjects with drug-resistant epilepsy.
• Pseudoprospective study design using training (85%) and validation data splits.
• Application of a novel algorithm based on the Kolmogorov-Smirnov statistic and individualized autoregressive models for forecasting seizure cluster probability.

Main Results:

• The algorithm demonstrated forecasting capabilities for seizure clusters at .5-, 1-, and 2-day horizons with mean absolute scaled errors (MASE) of .74, .78, and .83, respectively.
• Electrocorticography data yielded a mean of 394 seizure events per subject over an average recording span of 38 months.
• Seizure clusters exhibited quasiperiodic patterns, supporting the feasibility of forecasting.

Conclusions:

• The novel forecasting algorithm shows promise in predicting seizure clusters.
• Forecasting seizure clusters could enable chronotherapy, optimizing treatment timing during vulnerable periods.
• Further validation in larger epilepsy cohorts is warranted.