High-amplitude epochs explained 8% of the theta memory effect in 51 SEEG subjects, but did not reduce the high-frequency effect

In 51 patients with stereo-electroencephalography (SEEG) data from session 0 of a free-recall word-encoding task, we tested whether high-amplitude epochs retained by a 5-times-median artifact rule predicted forgetting and explained the subsequent memory effect (SME). We used archived per-word, sample/fs-timed band-power and root-mean-square (RMS) features for 54 subjects; 51 met the preregistered minimum of 15 recalled and 15 not-recalled words. The primary flag was a robust z score greater than 3 for ln(RMS), among epochs retained by the base rule. Across subjects, flagged-pair load was greater before forgotten words (mean Welch t = -0.759; Holm-adjusted p = 0.0013). Removing flagged epochs reduced the 3–8 Hz theta SME by 8.0% (95% bootstrap CI 1.9%–15.2%; Holm p = 0.034), while the clean theta SME remained significant (−0.405, p = 0.0001). The 70–150 Hz high-frequency activity SME did not decrease (all epochs +0.136; cleaned +0.149; Holm p = 0.098). A post hoc check found that ln(RMS) correlated with theta power (median r = 0.561); after controlling for mean theta power, the load effect was smaller and not significant (t = −0.300, p = 0.061). Thus, the amplitude proxy did not explain the FR1 SME to a large degree, and its association with forgetting was not specific evidence for interictal epileptiform discharges. The proxy is not a waveform-based IED detector; raw waveforms, other time windows, and validated IED detection were not tested. Full methods, sensitivity analyses, limitations, source hashes, code, and results are provided in the main text and supporting files.

stereo-electroencephalography · subsequent memory effect · word encoding · theta power · high-frequency activity · high-amplitude epochs · interictal epileptiform discharges · free recall