Cenobamate modulates EEG cortical activity and connectivity in individuals with drug-resistant epilepsy: a pharmaco-EEG study

Abstract

Objective: Quantitative electroencephalography (qEEG) metrics are demonstrated to correlate with and predict clinical response in individuals with epilepsy. Cenobamate is an effective anti-seizure medication recently approved as an add-on therapy for individuals with epilepsy, but its effects on qEEG are unknown. We aimed to evaluate the modulation of qEEG metrics induced by cenobamate and its relationship with clinical response.

Methods: We performed a prospective study with a cohort of 18 individuals with epilepsy (8 women, 47 ± 16 years old) and 25 healthy subjects (HS). They underwent a 19-channel EEG before and 6 months after cenobamate administration. Power spectral density (PSD) and phase locking value (PLV) for delta, theta, alpha, beta, and gamma frequency bands were calculated. Correlation analysis and analysis of covariance exhibited significant cenobamate-induced changes in qEEG and their relationship with seizure frequency changes. A regression analysis was performed to evaluate the association with clinical responders.

Results: A total of 11 out of 16 individuals with epilepsy (69%, with 2 dropping out) were cenobamate responders (≥50% seizure frequency reduction). Cenobamate did not modify any PSD parameter but induced significant changes in PLV levels (p < 0.01). A decrease in PLV correlated with seizure reduction (p < 0.03). Regression analysis showed a strong association between PLV modulation and cenobamate responsiveness (a sensitivity of 0.75, a specificity of 0.84, and an accuracy of 0.81).

Conclusion: Cenobamate induces an EEG connectivity modulation that is highly associated with cenobamate clinical response.

Significance: Connectivity analysis of pharmaco-EEG can provide new hints toward the development of innovative biomarkers and precision medicine in individuals with epilepsy.

Keywords: EEG connectivity; cenobamate; drug-resistant epilepsy; pharmaco-EEG; response biomarker.

Figure 1

Study design. PwE, people with epilepsy; EEG, electroencephalogram; HS, healthy subjects; T0, before cenobamate administration; T1, 6 months after cenobamate administration.

Figure 2

Correlation analysis between EEG PLV changes and seizure response after CNB. Scatterplot and regression line of seizure reduction percentage (SF_T0−T1, dependent variable) over percentage variation of PLV connectivity T1 vs. T0 (PLV_T1−T0, independent variable) for each frequency band. * p-value of correlation coefficient < 0.05; PLV, phase locking value; PLV_T1−T0, percentage variation of PLV connectivity T1 vs. T0; SF_T0−T1, percentage variation of seizure frequency T1 vs. T0; T0, before CNB administration; T1, 6 months after CNB administration.

Figure 3

PSD and PLV levels in PwE and HS. Boxplot and violin plot of PLV and relative PSD distribution for HS, PwE at T0, and PwE at T1 for each frequency band. CNB induced a significant overall PLV modulation between T0 and T1 as shown by ANCOVA with SF_T0−T1 as covariate (not shown in the figure). * post-hoc analysis of ANOVA/ANCOVA models showing a p-value <0.05; ANCOVA, analysis of covariance; HS, healthy subjects; PLV, phase locking value; PSD, power spectral density; PwE, people with epilepsy; T0, before CNB administration; T1, 6 months after CNB administration.

Figure 4

Regression model predicting clinical responsiveness. ROC curve of the regression model. Top right panel: Five most relevant features of the model; bar lengths are proportional to the magnitude of linear coefficients; predictors on the left side of the dotted line predicted a higher probability of clinical CNB response, the contrary for predictors on the right side of the dotted line. Bottom left panel: Confusion matrix of the regression model. Bottom right panel: Predictive performance of the predictive model. Acc, accuracy; CNB, cenobamate; NPV, negative predictive value; PLV, phase locking value; PLV_T0, PLV levels at T0; PLV_T1−T0, percentage variation of PLV connectivity T0 vs. T1; PPV, positive predictive value; PSD, power spectral density; PSD_T0, PSD levels at T0; ROC AUC, area under the receiver operating characteristic curve.