The patient bootstrap interval for AUC gain from low frequency SEEG power above contact order included zero in 19 external patients after HUP training

We tested whether low frequency SEEG power adds clinical onset-mark ranking above contact order. The models used 36 HUP patients for training and 19 NIH and Johns Hopkins patients for testing. The primary reference was the stored signal. Mean patient AUC was 0.707 for contact order and 0.712 with power. The mean gain was 0.0046. Its 95 percent patient bootstrap interval was -0.0775 to 0.0656. Each draw sampled training patients, fitted both models again, and sampled test patients. The exact two-sided sign-flip p value was 0.894, conditional on fixed models and null sign symmetry. The evidence does not establish a gain or equal performance. Within contact name groups, the planned added model had lower AUC. Its mean change was -0.0780, with a fixed-model interval from -0.1366 to -0.0308 and a five-test Holm p value of 0.011. These secondary results exclude training uncertainty. The package preserves 91 input power files, source metadata, pinned raw retrieval recipes, code, patient results, and bootstrap arrays. Independent checks passed. A complete offline repeat matched all nine numerical result files byte for byte. This is a secondary analysis of previously tested public cohorts. Clinical marks are annotations. Patient resampling does not estimate uncertainty across new centers.

stereoelectroencephalography · epilepsy · seizure onset zone · contact order · low frequency power · area under the curve · patient bootstrap · external validation