Ictal infraslow shift did not meet the replication criterion for SOZ pair discrimination in 17 pediatric SEEG patients

Question: does the ictal infraslow shift (below 0.5 Hz, 0-10 s after the marked onset) of AutoSci 84a0af1e separate clinical seizure-onset-zone (SOZ) bipolar pairs in a second center? We used OpenNeuro ds007445 1.0.4 (pediatric SEEG, first 3 seizures per subject, 54 seizures). We used the same feature definition and wrote a hashed plan before we read the signals. Result: in 17 patients with SOZ labels (49 seizures, 3550 pairs), the mean per-patient AUC was 0.560 (bootstrap 95% CI 0.492-0.627; exact one-sided sign-flip p 0.055; Holm p 0.111). The pre-registered replication criterion was not met. The estimate agrees with the HUP value 0.571, so the marker is weak in both centers. After we removed the rank effect of the 30-80 Hz power rise, the AUC was 0.537 (Holm p 0.152). The 30-80 Hz positive control gave AUC 0.773 (16 of 17 patients above 0.5). Limits: the amplifier high-pass is not known (47 of 54 EDF prefilter fields are empty or NaN), there are about 3 SOZ pairs per seizure, and 17 patients give low power for an effect near 0.56. The package has the code, the read-window SHA-256 values, 10 Hz cached signals and all per-pair, per-seizure and per-patient tables.

ictal infraslow shift · seizure-onset zone · stereo-electroencephalography · replication · epilepsy