Neurophysiologic Biomarkers of Invasive Neuromodulation Therapy for Epilepsy

INTRODUCTION: Epilepsy affects 65 million people globally, with drug-resistant epilepsy (DRE) developing in 30% of patients. Neuromodulation therapies, such as deep brain stimulation (DBS), responsive neurostimulation (RNS), and vagus nerve stimulation (VNS) are attractive treatment options, but assessing efficacy is time consuming. Predictive biomarkers of treatment response may expedite assessment and pave the way for closed-loop strategies that optimize outcomes. This systematic review identifies scalp and intracranial electroencephalography (EEG)-derived biomarkers associated with clinical efficacy in invasive neuromodulation for DRE and explores the impact of preprocessing and methodologic variations on biomarker efficacy. MATERIALS AND METHODS: A systematic review followed Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 guidelines, searching PubMed from 2004 to 2025 for human studies on EEG biomarkers post invasive neuromodulation for people with DRE. Inclusion criteria required ≥four participants, single neuromodulation modalities, and EEG-based measures linked to seizure outcomes. Exclusion criteria eliminated multimodal therapies, craniotomies, and non-English studies. Owing to heterogeneity, qualitative synthesis identified trends, supplemented by an exploratory logistic regression analysis assessing the preprocessing, methodologic, and biomarker type (node-level, edge-level, graph-theory metrics) impacts on biomarker efficacy. RESULTS: This review included 30 studies (2005-2024), covering DBS (n = 8), RNS (n = 7), and VNS (n = 15). Ten of 30 studies were of focal DRE; three of 30 studies were generalized DRE, and half of the included studies did not specify whether participants had focal or generalized epilepsy. Biomarkers spanned node-level (eg, interictal epileptiform discharges, spectral power, n = 14), edge-level (eg, phase-amplitude coupling, coherence, n = 13), and graph-theory metrics (n = 3). Key findings were as follows: 1) interictal epileptiform discharge (IED) rate reductions commonly correlate with seizure frequency reduction; 2) short-term changes in the aperiodic parameters appear to be predictive of longer-term clinical outcome; and 3) reduced synchronization is repeatedly associated with improved seizure outcomes across RNS, VNS, and some DBS contexts. Four studies reported early measures (intraoperative or ≤six months post implant) that predicted longer-term seizure outcomes. Exploratory logistic regression suggested nonsignificant trends favoring larger sample sizes, higher sampling rates, data normalization, and edge-level metrics; use of artifact-removal algorithms tended to reduce the likelihood of a reported significant association. All regression estimates were imprecise (nonsignificant, wide 95% CIs) CONCLUSION: In conclusion, EEG-derived biomarkers, particularly IEDs, aperiodic measures, and synchronization measures, show promise for predicting neuromodulation efficacy in DRE. Nonsignificant trends revealed through logistic regression analysis suggest that methodologic and preprocessing variations may influence biomarker predictive efficacy. We would recommend more comprehensive reporting of the signal processing protocols and biomarker performance to enable more robust causal inference.