Abstracts

Rationale: Patients with medication-resistant epilepsy undergo intracranial EEG (iEEG) monitoring to map epileptic networks during pre-surgical evaluation. Multiple seizures are evoked in the epilepsy monitoring unit by reducing medications. However, this process poses a significant risk of injury [1] particularly when seizures are convulsive and prolonged, which is more common with medication withdrawal. The goal of this study is to quantify pre-ictal changes in brain dynamics that predict seizure severity, introducing a window to intervene and reduce morbidity associated with these events.
_x000D_ Methods: We retrospectively analyzed pre-ictal and ictal iEEG recordings from 61 people with medication-resistant epilepsy undergoing pre-surgical evaluation. Overcoming the limitations of the clinical standard for a seizure severity scale [2], we developed a new metric that objectively combines seizure semiology, duration, and spread to quantify the severity of individual seizures. In the one-hour interval prior to each seizure, we computed time-varying spectral and network features from iEEG as well as corresponding abnormality features by incorporating a normative iEEG atlas. To identify an interval prior to seizure onset where severity can be predicted, we integrated time-varying iEEG features into a multivariate regression model and evaluated the association of pre-ictal iEEG features with seizure severity.
_x000D_ Results: A total of 222 seizures were evaluated in the quantitative seizure severity scale and event duration varied between 4.9 and 782 seconds. A total of 141 seizures had focal onset that remained focal and 81 seizures had focal onset that spread bilaterally and associated tonic-clonic semiology. Seizure severity correlated with clinical seizure type and seizure duration. The seizure severity scale was sensitive to medication tapering strategies, as more severe seizures occurred at lower medication levels (Pearson r = -0.38, p < 1e-2). Seizure severity also correlated with age (Pearson r = 0.58, p < 1e-4) and patients who were seizure free following surgery had less severe seizures than patients who were not seizure free (Mann-Whitney test, p < 0.05). A time-varying multivariate regression model of spectral, network, and abnormality features exhibited peak correlations between predicted and true seizure severity scores between 37.73 and 33.07 minutes before seizure onset (Spearman r = 0.80 at maximum, 95% CI = [0.74, 0.85]), indicating that pre-ictal dynamics may vary with seizure severity on this timescale.