Seizure-end amplitude change is consistent within patients, while discharge slowing depends on detector threshold and channel choice in HUP SEEG

We tested whether seizure-end dynamics are stable within patients in HUP stereo-EEG data. We analyzed derived per-seizure offset tables from AutoSci publication f087d7e7 (OpenNeuro ds004100 v1.1.3), retaining seizures with at least 8 discharges in the final 20 s and patients with at least 2 seizures. For each metric, we estimated ICC(1), used 10,000 permutations of seizure-to-patient labels, applied Holm correction, and calculated patient-bootstrap confidence intervals. In the primary detector threshold K = 5, top-channel variant (118 seizures, 32 patients), ICCs were 0.32 for discharge slowing (log2 ISI ratio; 95% CI 0.12–0.43; Holm p = 0.0006), 0.26 for rho(ISI, time) (p = 0.003), and 0.28 for log2 amplitude ratio (p = 0.003). Slowing ICC remained 0.32 after regression on seizure duration, baseline SD, and line-length ratio, and was 0.25 (p = 0.017) when retaining one seizure per channel; it was absent in the 20–40 s pre-offset control window (ICC 0.08, p = 0.21). Amplitude ICC was significant in all four variants (K3, K5, K8, and second-ranked channel; 0.24–0.32). Slowing ICC was weaker at K3 (0.13, p = 0.051) and K8 (0.18, p = 0.033), absent on the second-ranked channel (0.06), and lost in the one-seizure-per-channel controls for K3, K8, and the second-ranked channel. Repeated channel choice may therefore explain part of the slowing consistency. Patient-mean slowing did not differ between Engel 1 and Engel 2–4 outcomes (p = 0.86). Limits include one centre, clinical offset annotations, a simple detector, 2–6 seizures per patient, and wide ICC confidence intervals.

stereo-EEG · seizure offset · intracranial EEG · within-patient consistency · intraclass correlation · discharge slowing · amplitude change · epilepsy