EEG-Derived Proxies of Cortical Excitability in Epilepsy: Group Discrimination, Temporal Stability and Medication Sensitivity

Rationale: Reliable electroencephalography (EEG) biomarkers of cortical excitability could improve diagnosis and longitudinal monitoring in epilepsy, yet it remains unclear which metrics best balance sensitivity across individuals with intra-individual stability over time. Methods: We analyzed scalp EEG recordings from the open-access Temple University Hospital EEG Epilepsy Corpus, comprising 1,404 recordings from 96 individuals with neurologist-confirmed epilepsy and 85 healthy controls across multiple sessions. Eight global measures were computed: aperiodic exponent and offset, sample entropy, detrended fluctuation analysis exponent and derived index, spatial gamma-band phase consistency, and absolute and relative alpha power. Group differences were assessed by permutation tests with false discovery rate correction at recording, session, and subject levels. Associations with antiseizure medication burden, temporal stability, and cross-metric correlation structure were evaluated as secondary analyses. Results: Aperiodic parameters showed the most robust case-control separation, remaining significant after subject-level averaging (exponent: median difference = 0.20, q = 0.010; offset: median difference = 0.25, q = 0.011). Entropy and alpha power distinguished groups at the recording and session levels, while gamma-band phase consistency was significant at the session level only; none of these survived subject-level averaging, suggesting greater state-dependency. Higher medication burden was associated with reductions in alpha power and detrended fluctuation analysis, and adjusting for it substantially attenuated group differences, though residual effects in the aperiodic exponent persisted. Cross-metric correlation structure was preserved between groups but modestly reorganized by medication burden. Conclusions: Aperiodic spectral parameters are the most robust EEG markers of epilepsy, reflecting stable trait-like network properties. Complexity and synchrony measures capture complementary, state-sensitive dimensions. Medication burden substantially influences multiple metrics, underscoring the need to account for pharmacological effects when interpreting EEG biomarkers in epilepsy.