Math distractor performance did not predict list recall in 18 SEEG patients; a held-out decoder showed a small association

In an exploratory offline analysis of 18 patients performing delayed free recall (OpenNeuro ds004789, FR1, session 0), list-level math-distractor performance did not predict recall rate: mean Spearman rho was +0.084 for correct-problem count (95% bootstrap CI -0.058 to +0.222; exact Wilcoxon p = 0.30), +0.054 for accuracy (17 patients; CI -0.078 to +0.181; p = 0.46), and -0.098 for median response time (CI -0.228 to +0.046; p = 0.15). A channel-by-frequency decoder tested on held-out lists had a small positive association with recall (mean rho +0.163, CI +0.057 to +0.274; 13/18 patients positive; p = 0.016). Spectral tilt had mean rho +0.120 (CI +0.018 to +0.224; p = 0.054). The decoder and math count were not correlated (rho -0.011; p = 0.97). The decoder-minus-math difference was not significant (mean +0.079; p = 0.44), and the decoder association after partial control for math count and list number was smaller and not significant (mean +0.085; p = 0.18). The study was exploratory, was not preregistered, and did not correct for multiple tests; no table test met a Bonferroni threshold across eight tests. Results concern one task, session, and subset of 18 patients; they do not show that the decoder is better than the math measure and cannot exclude a small math effect.

delayed free recall · SEEG · math distractor · memory decoding · list-level prediction · leave-one-list-out · exploratory analysis