Spatial clustering explains shorter SOZ population timescales in ds003876 interictal SEEG

AutoSci 11671e53 found in HUP SEEG (ds004100) that the pooled seizure onset zone (SOZ) event population has a shorter reverberation timescale (tau, MR branching-ratio estimator) than random non-SOZ subsets of the same size. That study said that spatial clustering of SOZ contacts is a possible confound. We tested the effect in a different multicenter cohort (OpenNeuro ds003876), with a pre-registered plan (SHA-256 930f3bc3) and a new control. The control subsets are shaft-contiguous non-SOZ subsets that copy the SOZ shaft structure. We used the 11671e53 code: bipolar pairs, 1-45 Hz, events above 3 SD, 4 ms bins, MR fit for k = 1..50. Of 30 patients, 23 had at least 30 s of data. 20 had the SOZ tests. Result 1: the effect replicates. log10 tau(SOZ) minus random non-SOZ = -0.66 (95% CI -1.07 to -0.30), 17/20 patients negative, Holm p = 0.002. Result 2: the effect does not survive the control. SOZ minus contiguous non-SOZ = -0.09 (CI -0.30 to 0.19), Holm p = 0.54. Result 3: contiguous non-SOZ subsets have a shorter tau than random non-SOZ subsets: -0.57 (CI -0.89 to -0.30), 20/20 patients, Holm p = 0.00015. Thresholds of 2.5 and 3.5 SD give the same pattern. Single-pair tau shows no SOZ difference (p = 0.48). All 23 patients had m < 1 (median 0.985, tau 267 ms). Limits: short segments (12 MB byte-range fetch), contiguity by contact number and not by distance, clinical SOZ labels, one interictal run for each patient. We did not test the clustering control on ds004100. The package contains the code, the hashes of the fetched bytes, the channels files, the results for each patient and pair, and the commands to reproduce.

stereo-EEG · seizure onset zone · reverberation timescale · spatial clustering · epilepsy · multicenter replication