Adding contact number gave an uncertain AUC gain for a fixed HUP power model in 19 external SEEG patients

We tested whether scaled contact number adds value to two low frequency SEEG power features. The fixed models trained on 36 HUP patients from OpenNeuro ds004100 version 1.1.3. They ranked source onset marks in 19 external patients from ds003876 version 1.0.2. Mean patient AUC changed from 0.7298 to 0.7675. The paired gain was 0.0376. A joint 95 percent patient bootstrap interval was -0.0313 to 0.1041, with the three observed centers fixed. The exact sign-flip p value was 0.3318 under independent, symmetric zero effects and fixed HUP fits. The mean change was 0.1069 at Johns Hopkins and -0.0128 at NIH. These site values are descriptive. The evidence does not establish a reliable gain or equivalence. This is a secondary analysis of previously studied public cohorts. The package preserves source hashes, all patient scores, all fits, all resampling draws, and repeat commands. Three fresh raw-data checks passed. A full analysis repeat gave identical files. The result concerns source clinical marks, short records, and a fixed model. It does not test treatment benefit.

stereoelectroencephalography · epilepsy · seizure onset zone · contact number · spectral power · feature transfer · patient-level AUC · external validation