Related Experiment Video
Updated: Sep 23, 2026
Network Analysis of Foramen Ovale Electrode Recordings in Drug-resistant Temporal Lobe Epilepsy Patients
Published on: December 18, 2016
Predicting levetiracetam neuropsychiatric adverse effects using graph-theoretical electroencephalographic network
Marco Sferruzzi1, Lorenzo Ricci1,2, Margherita Anna Grazia Matarrese3
1Department of Medicine and Surgery, Research Unit of Neurology, Neurobiology, Neurophysiology, University Campus Bio-Medico di Roma, Rome, Italy.
Objective:
Levetiracetam (LEV) is an antiseizure medication approved for focal and generalized epilepsy. Despite its efficacy, neuropsychiatric adverse effects (NP-AEs) may lead to treatment discontinuation. Quantitative electroencephalographic (qEEG) graph-theoretical analysis has emerged as a valuable tool for identifying neurophysiological markers of treatment response. We investigated whether baseline network metrics could identify drug-naive people with focal epilepsy (PwE) at risk of developing clinically significant LEV-related NP-AEs.
Methods:
We conducted a multicenter retrospective longitudinal study across three Italian centers. From an original LEV monotherapy cohort of 134 PwE, we identified all 31 eligible patients who developed clinically significant LEV-related NP-AEs within the first month after LEV initiation (AE+) and matched them by age and sex with 31 patients without NP-AEs (AE-). All participants underwent resting-state EEG within 30 days before LEV. We compared relative power spectral density (PSD) and weighted phase lag index (wPLI) connectivity-derived graph-theoretical metrics (clustering coefficient, global efficiency, path length, modularity, node strength) using linear mixed-effects models. Baseline variables showing significant between-group differences were included in an elastic-net penalized logistic regression model. Model performance was evaluated using nested cross-validation and permutation testing (1000 permutations).
Results:
Relative PSD showed no differences between groups. The AE+ group showed higher baseline theta-band wPLI-derived clustering coefficient (p = .032), characteristic path length (p = .033), and node strength (p = .03) than AE-. In the elastic net model, theta-band graph-theoretical metrics predicted NP-AE occurrence with an accuracy of .71 (95% confidence interval [CI] = .60-.82) and an area under the curve of .71 (95% CI = .57-.83). Permutation testing confirmed that model discrimination exceeded chance level (p = .003).
Significance:
Baseline theta-band qEEG functional network organization emerged as a significant neurophysiological predictor of LEV-related NP-AEs. These findings suggest that pretreatment EEG may capture a neurophysiological vulnerability phenotype associated with LEV tolerability, supporting the potential role of network-level biomarkers in guiding individualized treatment decisions.
