In ds006136 SEEG, slow encoding evoked-potential gain did not predict full recall (12 patients; AUC 0.515, 95% CI 0.498–0.532; p = 0.13)

In 12 patients with load-3 object working memory recorded by stereo-electroencephalography (SEEG), the single-trial gain of the stimulus-evoked potential during encoding did not reliably predict full (3/3) recall. Across 134 bipolar middle and middle frontal gyrus (MTG/MFG) channels, the mean subject AUC in the 0–800 ms encoding window was 0.515 (bootstrap 95% CI 0.498–0.532; exact sign-flip p = 0.13); 9 of 12 subjects had AUC above 0.5. The delay-onset window gave AUC 0.514 (95% CI 0.502–0.525; p = 0.051; Holm-adjusted p = 0.10), and the pre-stimulus fixation control gave AUC 0.521 (95% CI 0.502–0.542; p = 0.073; Holm-adjusted p = 0.10). Exploratory results were also near chance for reliable-response channels (61 of 134; AUC 0.507, p = 0.51), MTG (AUC 0.508, p = 0.45), and MFG (AUC 0.520, p = 0.11). These results do not show that the slow evoked potential predicts recall beyond the pre-stimulus signal; small effects (AUC below about 0.53) are not excluded. The analysis used a 30 Hz low-pass filter, included only bipolar MFG/MTG channels available in this dataset, and compared full recall with a partial-recall class that combines 0/3, 1/3, and 2/3 recall. Trial alignment across channels was not verified, and the three objects were presented together, so item-level memory could not be tested.

working memory · stereo-electroencephalography · evoked potential · recall · signal detection · null result · OpenNeuro ds006136