In 66 SEEG patients, SOZ extent did not predict seizure freedom; failure tracked smaller labels in ds003876

We tested three pre-registered hypotheses. We used the clinical channel labels from 66 SEEG-only patients in OpenNeuro ds004100 and ds003876 (31 seizure-free, 35 not seizure-free). H1: the SOZ fraction of good SEEG contacts is lower in seizure-free patients. H2: the number of SOZ shafts is lower. H3: the fraction of SOZ contacts that are resected or ablated is higher (n = 50). The tests were dataset-stratified permutation tests with Holm correction. No hypothesis was supported. The H1 AUC was 0.32 (95% CI 0.19-0.46), the H2 AUC was 0.34 (0.22-0.47) and the H3 AUC was 0.59 (0.42-0.75; Holm p = 0.42). For H1 and H2, the data showed the opposite direction: failure went with a smaller labeled SOZ. This effect came from ds003876 (H1 AUC 0.14). In ds004100 there was no effect (AUC 0.44 and 0.52). When we removed the MIAMI RNS cohort post hoc, the effect became smaller (H1 AUC 0.37, CI 0.22-0.53). Engel class I as success gave the same result. Limits: we used channel metadata only, outcome definitions differ between sites, and the SOZ labels and treatment choices are not independent. The package contains the plan with its hash, the code, all input files with SHA-256 values, the results for each subject, and the reproduction steps.

stereo-EEG · seizure onset zone · epilepsy surgery · surgical outcome · OpenNeuro · clinical labels · metadata analysis