Temporal patterns of neurostimulation efficacy in drug-resistant epilepsy: A systematic review and meta-analysis

Objective: To systematically evaluate the temporal efficacy trajectories and relative performance of four neurostimulation modalities (open-loop Vagus Nerve Stimulation [ol-VNS], closed-loop VNS [cl-VNS], Deep Brain Stimulation [DBS], and Responsive Neurostimulation [RNS]) in patients with drug-resistant epilepsy (DRE). Methods: Following PRISMA guidelines, PubMed, Cochrane Library, and Web of Science were searched for studies published during the modern neuromodulation era (2014–2024). We included original studies of patients without prior resective surgery who received single-device implantation. Data were mapped into four standardized follow-up windows (3, 6, 12–18, and ≥21 months). Pooled responder rates (RRs, ≥50% seizure reduction) were calculated using a random-effects model. A mixed-effects meta-regression was conducted to explore the interaction between device modality and time. Results: A total of 94 unique cohorts were included. Overall RRs demonstrated a robust, time-dependent improvement: 38.7% (95% CI: 29.9%–48.4%) at 3 months, 48.2% (95% CI: 42.3%–54.1%) at 6 months, 59.3% (95% CI: 56.2%–62.2%) at 12–18 months and peaking at 61.8% (95% CI: 57.6%–65.7%) at ≥ 21 months. Meta-regression (QM = 96.3, p < 0.0001) confirmed a highly significant main effect for follow-up duration. While numerical differences existed among devices (with cl-VNS, DBS, and RNS showing higher overall efficacy baselines compared to ol-VNS), the interaction terms between device type and follow-up duration were universally non-significant (all p > 0.05), indicating that the current data do not provide statistical evidence of diverging temporal trajectories across modalities. Significance: Across all modalities, neurostimulation produced a robust, time-dependent improvement in responder rates, with progressive gains consolidating over 1–2 years. This temporal pattern is consistent with the hypothesis of delayed network remodeling, though the present data cannot confirm a shared mechanism. Numerical differences in baseline responder rates across devices are more likely attributable to patient selection and indication bias than to intrinsic hardware superiority. © 2026