Abstract
<title>Abstract</title> <p>Background Responsive neurostimulation (RNS) delivers on-demand intracranial electrical stimulation triggered by detected ictal activity. Whether pre-stimulation ictal intracranial EEG (iEEG) features evolve longitudinally across months of chronic therapy, and whether those changes track the degree of therapeutic challenge faced by each patient, remains poorly characterised at the cohort level. Objective To extract, model, and report longitudinal trajectories of spectral band power and inter-channel phase-locking value (PLV) across a complete, openly available 8-patient RNS-iEEG cohort, and to examine associations between pre-stimulation features and clinician-prescribed stimulation amplitude as a proxy for seizure burden. Methods We analysed 355 weekly sessions (range 35–68 per patient; total 6,019 recordings) from all eight patients in the ds007095 OpenNeuro dataset, recorded on the NUEHILL BCM210A RNS device at 200 Hz. For each 90-second recording, the first 60-second pre-stimulation epoch was used to compute Welch power spectral density (0.64 s Hann window, 78.1% overlap) and per-band PLV (Hilbert-transform phase-locking between two iEEG contacts) across five canonical frequency bands (δ 1–4 Hz, θ 4–8 Hz, α 8–13 Hz, β 13–30 Hz, γ 30–80 Hz). Session-level features were obtained by averaging across all within-session recordings. Linear OLS slopes characterised longitudinal trajectories per patient; Spearman rank correlations (pooled, n = 355 sessions) assessed feature-amplitude associations. Results Substantial inter-patient heterogeneity was observed in trajectory slopes. Sub-08 (24F) showed the strongest broadband power increases (delta slope + 2.5×10⁻³ log₁₀ V²/Hz/day, p < 0.001), while sub-04 (41M) showed the most consistent decreases across all bands (delta slope − 1.5×10⁻³, p < 0.001), suggesting divergent ictal network dynamics. Pooled Spearman analysis identified delta power as the feature most strongly correlated with stimulation amplitude (ρ = 0.43, p < 0.001), followed by theta power (ρ = 0.33), beta-band PLV (ρ = 0.32), and gamma power (ρ = 0.26). Broadband PLV showed no significant association with amplitude (ρ=−0.02, p = 0.75). Conclusions Pre-stimulation delta and theta band power show robust positive associations with stimulation amplitude across the full 8-patient cohort, suggesting that patients with greater ictal low-frequency power require more intensive therapeutic stimulation. The marked between-patient slope heterogeneity underscores the necessity of individualised longitudinal modelling. These findings are exploratory; validation in larger cohorts with formal seizure-frequency outcome data is required before clinical translation.</p>