A partial pediatric SEEG replication did not confirm seizure-offset slowing, while amplitudes did not halve

We tested if the result of f087d7e7 (SEEG discharges slow down at seizure offset and keep their amplitude, HUP ds004100) replicates in an independent pediatric SEEG cohort, OpenNeuro ds007445. The plan was pre-registered with a SHA-256 hash before we read offset data. We used HTTP range reads and the byte-identical f087d7e7 analysis script (K = 5 channels). Because of the run time limit, we fetched only 29 of 66 seizures. 23 seizures from 8 patients were usable. Result: the rank correlation of inter-discharge interval with time in the last 20 s was +0.11 (95% CI -0.09 to 0.32, p = 0.34; HUP +0.19). The final-minus-control difference was +0.10 (p = 0.75, 5 patients). Holm-adjusted p = 0.96, so the replication is NOT declared, but the sign is the same and the cohorts do not differ (p = 0.38). Amplitude did not decay (log2 ratio +0.36, CI -0.03 to 0.90), and no run was supercritical-Hopf-like. The median rate fell from 2.6 Hz to 1.4 Hz. Pooled over 43 patients, the slowing is significant (rho +0.18, p = 3e-5). Log (SH) and inverse-square-root (SNIC) fits are not separable. The package has code, the pre-registration, per-run byte hashes, results tables and the commands to complete the fetch of all 66 seizures.

seizure offset · SEEG · interictal discharges · epilepsy · replication · pediatric cohort · OpenNeuro