Hippocampal high-gamma load effects did not repeat across sessions in 14 patients

We tested whether per-patient working-memory load effects in hippocampal stereo-EEG (SEEG) repeat across sessions. We analyzed archived per-trial features from OpenNeuro ds004752 (version 1.0.1): 14 epilepsy patients, 48 sessions, and 1,908 correct, clean trials. Features were Spearman correlations between memory set size (4, 6, 8) and response time or maintenance-minus-fixation log power. For high gamma (70–150 Hz), the session-stratified group effect was a small decrease (weighted z = −0.074, SE = 0.024). Between-patient spread was not greater than sampling noise (I² = 0; 95% upper limit for tau = 0.103), and split-half reliability was near zero (r = 0.01). Gamma, beta, and the spectral exponent showed the same lack of measurable spread and repeatability. With about 75 trials per half, the noise SD of a per-patient z was about 0.13; about 213 trials per half would be needed for r = 0.7 if high-gamma tau were at its upper limit. The underlying task-related feature changes were more stable across halves (r = 0.37–0.79). Session-stratified response-time effects varied between patients (I² = 0.48, p = 0.025), but split-half reliability was not significant (r = 0.40, p = 0.074). Alpha showed a candidate split-half correlation (r = 0.55, p = 0.024) that did not survive Holm correction (p = 0.19). Calibration found that the planned pooled-trial Q test was not valid for response time (false-positive rate 0.19); session-stratified estimates were calibrated. These results do not support using the load contrast as an individual marker with 1–4 sessions and this trial count. Limits include 14 patients, one hippocampal shaft per patient, short analysis windows, and epilepsy patients.

hippocampus · stereo-EEG · working memory · high gamma · test-retest reliability · measurement noise · epilepsy