To propose a new data-driven approach to defining seizure clusters, and to compare the performance of current seizure cluster definitions.

Seizure clusters are often encountered in people with severe or poorly controlled epilepsy. There is no uniform consensus on how a seizure cluster should be defined. This study proposes a new data-driven approach to defining seizure clusters and evaluates current definitions for accuracy in identifying seizure clusters.

Clinical and electrographic seizure diary data collected by SeizureTracker.com and the Responsive Neurostimulation (RNS) system were analyzed. Hurst statistics were used to quantitate seizure clustering in clinical seizures. Conventional definitions were evaluated for sensitivity and specificity for identifying seizures exhibiting statistical evidence of clustering. The association between time since the last seizure cluster and the number of subsequent clustered seizures was evaluated for clinical and electrographic seizures.

A total of 564,510 clinical seizures from 1,826 people with self-reported epilepsy in SeizureTracker.com and 360,120 electrographic seizures from 14 people with epilepsy implanted with the RNS system were analyzed. Empirical seizure cluster percentiles based on the Hurst exponent are provided and proposed as a standardized measure of seizure cluster tendency. Conventional definitions varied widely in ability to identify clusters, with significant variability in the ISI threshold needed to identify statistical clusters. The number of seizures in a given cluster followed a power law in relation to time since the last seizure cluster for clinical and electrographic seizures.

We propose a data-driven approach to identifying seizure clusters which accounts for individual variability in baseline seizure frequency and avoids prechosen thresholds. These results provide unique insight into the prognostic significance of current seizure cluster definitions and proposes a new individualized approach to defining seizure clusters.