Neither single frequencies (1–45 Hz) nor a full-spectrum ridge model predicted session-0 free recall from task-period SEEG in 261 RAM patients

Question: does any part of the intracranial spectrum show why some patients recall more words than other patients? Earlier work (0f2a7cb1) tested 6 fixed spectral features and found no relation. This work tests all 89 frequencies from 1 to 45 Hz (0.5 Hz steps). Data: archived 60 s task-period bipolar SEEG spectra of 261 RAM patients (c3188d78; OpenNeuro ds004789 FR1 and ds004809 catFR1, session 0) and session-0 recall rates (0f2a7cb1). Method: relative spectrum (log power minus its 1-45 Hz mean), mean over pairs. Covariates: site, sampling rate, sex, log number of pairs, dataset and age. The plan was written and hashed before the analysis. Primary test: partial Spearman rho at each frequency, with a family-wise maximum-|rho| permutation test (10000 permutations). Secondary test: cross-validated ridge regression on the full spectrum, with a permutation test. Results: the largest |rho| was 0.11 at 5 Hz (95% CI of rho -0.23 to +0.01; family-wise p 0.42; threshold 0.18). The ridge model did not predict recall rate (CV R2 -0.037; covariates only +0.006; p 0.95). Absolute power, FR1 only and 5 band means also gave no significant result. Without the age covariate, 12 Hz power seemed to predict lower recall (rho -0.16). A post-hoc check agrees with an age confound: age correlates with 12 Hz power (rho +0.32) and with recall rate (rho -0.23). Limits: epilepsy patients aged 18-64 y; one 60 s task segment; mean over all implanted pairs, so regional effects can be missed; the test cannot detect single-frequency effects with |rho| below about 0.18. The package has the code, the copied input spectra with hashes, all results and reproduction steps.

SEEG · intracranial EEG · free recall · memory · spectral power · epilepsy · ridge regression · RAM study