Network connectivity predicts effectiveness of responsive neurostimulation in focal epilepsy

Responsive neurostimulation is a promising treatment for drug-resistant focal epilepsy; however, clinical outcomes are highly variable across individuals. The therapeutic mechanism of responsive neurostimulation likely involves modulatory effects on brain networks; however, with no known biomarkers that predict clinical response, patient selection remains empiric. This study aimed to determine whether functional brain connectivity measured non-invasively prior to device implantation predicts clinical response to responsive neurostimulation therapy. Resting-state magnetoencephalography was obtained in 31 participants with subsequent responsive neurostimulation device implantation between 15 August 2014 and 1 October 2020. Functional connectivity was computed across multiple spatial scales (global, hemispheric, and lobar) using pre-implantation magnetoencephalography and normalized to maps of healthy controls. Normalized functional connectivity was investigated as a predictor of clinical response, defined as percent change in self-reported seizure frequency in the most recent year of clinic visits relative to pre-responsive neurostimulation baseline. Area under the receiver operating characteristic curve quantified the performance of functional connectivity in predicting responders (>= 50% reduction in seizure frequency) and non-responders (<50%). Leave-one-out cross-validation was furthermore performed to characterize model performance. The relationship between seizure frequency reduction and frequency-specific functional connectivity was further assessed as a continuous measure. Across participants, stimulation was enabled for a median duration of 52.2 (interquartile range, 27.0-62.3) months. Demographics, seizure characteristics, and responsive neurostimulation lead configurations were matched across 22 responders and 9 non-responders. Global functional connectivity in the alpha and beta bands were lower in non-responders as compared with responders (alpha, p(fdr) < 0.001; beta, p(fdr) < 0.001). The classification of responsive neurostimulation outcome was improved by combining feature inputs; the best model incorporated four features (i.e. mean and dispersion of alpha and beta bands) and yielded an area under the receiver operating characteristic curve of 0.970 (0.919-1.00). The leave-one-out cross-validation analysis of this four-feature model yielded a sensitivity of 86.3%, specificity of 77.8%, positive predictive value of 90.5%, and negative predictive value of 70%. Global functional connectivity in alpha band correlated with seizure frequency reduction (alpha, P = 0.010). Global functional connectivity predicted responder status more strongly, as compared with hemispheric predictors. Lobar functional connectivity was not a predictor. These findings suggest that non-invasive functional connectivity may be a candidate personalized biomarker that has the potential to predict responsive neurostimulation effectiveness and to identify patients most likely to benefit from responsive neurostimulation therapy. Follow-up large-cohort, prospective studies are required to validate this biomarker. These findings furthermore support an emerging view that the therapeutic mechanism of responsive neurostimulation involves network-level effects in the brain. To prognosticate outcomes with neurostimulation for epilepsy, Fan et al. investigate functional network connectivity measured non-invasively with magnetoencephalography as a novel biomarker for effectiveness of responsive neurostimulation (RNS) therapy. Resting-state functional connectivity in alpha and beta frequency bands predicted response to subsequent RNS therapy and correlated with seizure frequency reduction.