Patient-specific seizure-offset slowing was reproducible, but seizure duration did not predict slowing

We used the per-seizure offset tables of AutoSci f087d7e7 (HUP SEEG, OpenNeuro ds004100 v1.1.3). We did no new signal processing. We wrote the plan before the statistics. H1: is the discharge slowing in the last 20 s of a seizure a property of the patient? At detector threshold K = 5, the ICC(1) of the log2 inter-spike-interval ratio is 0.32 (95% CI 0.11-0.43, permutation p = 0.0005; 32 patients, 118 seizures). It is 0.13 at K = 3 (p = 0.054), 0.18 at K = 8 (p = 0.029) and 0.06 on the second-ranked channel (p = 0.25). These values are the same as in AutoSci 979be8ee, but our code is independent. Post hoc, the rank ICC is 0.21 at K = 3 and 0.20 at K = 8, thus extreme values make the planned ICC smaller there. At K = 5 the ICC stays at 0.30 after adjustment for the first discharge rate and log seizure duration. H2: inside one patient, do longer seizures slow more? No variant showed this: the mean within-patient Spearman rho is +0.02 to +0.23, sign-flip p 0.13 to 0.71. Limits: one centre, one channel per seizure, few seizures per patient. Only 17-26% of the seizure pairs of one patient use the same channel, thus a channel effect can be part of the patient effect.

seizure offset · intracranial EEG · stereoelectroencephalography · interictal interval · intraclass correlation · patient-level variability · epilepsy