Recruitment order repeated across seizures in 20 HUP patients, but the earliest pair did not beat tie-aware chance

Question: for a responsive stimulator, how fast does a seizure spread across an SEEG implant, and do the same pairs lead in every seizure? Data: per-seizure 20-100 Hz log-power maps (1-s windows, -40 to +20 s around the marked onset) of OpenNeuro ds004100 from sci publication 3b7b4a7a; 20 patients with 2-5 seizures, 77 seizures. The plan was written before the analysis. A pair is recruited at the first 2 consecutive windows with robust z >= 3 against [-40,-15) s. Results (patient-level, 95% CI): the recruitment order repeats across seizures (mean Spearman rho 0.37, CI 0.29-0.45, 20/20 patients > 0). It stays above a within-shaft shuffle (excess 0.14, CI 0.10-0.19, 20/20, p = 5e-5). SOZ pairs are recruited earlier (AUC 0.71, CI 0.63-0.80, 18/19 patients). These results hold for z = 2.5 and z = 4 and for line length. Counted from the mark, 25% of pairs are recruited at a median of 4.7 s and 50% at 15 s or later. About 5% of pairs cross the threshold 5-14 s before the mark, and in 68/77 seizures the first crossing is before -5 s. Thus the first pair is not a reliable onset marker with this detector. The pre-specified hit@3 test (first pair of a later seizure among the 3 earliest of seizure 1) used a chance level that ignores rank ties. With a tie-aware chance level it is not significant (0.14 vs 0.085, p = 0.10). A post-hoc onset-anchored rule gives hit@3 0.37 vs 0.14 (p = 5e-4); this needs a new pre-specified test. Limits: one center, clinical labels, 1-s resolution, window ends at +20 s. The package has the code, the input features with S3 ETags, the per-seizure tables, the figures and the reproduction steps.

stereoelectroencephalography · seizure onset zone · ictal recruitment · 20-100 Hz power · recruitment order · responsive neurostimulation