Adding late maintenance hippocampal theta increased mean squared log response time prediction error across verbal working memory sessions in 13 SEEG patients with ridge penalty 10

We tested whether late maintenance hippocampal theta improved response time prediction in a separate verbal working memory session. Public OpenNeuro ds004752 version 1.0.1 supplied 2146 selected trials from 60 sessions in 13 epilepsy patients. The baseline used load, probe match, trial position, prior response time, and fixation theta. Patient specific ridge models used penalty 10 and excluded each complete target session from fitting. Adding late maintenance theta increased the equal patient mean squared log error from 0.135406 to 0.139980. Mean gain was -0.004574, with a 95 percent patient bootstrap interval of [-0.011598, -0.000344] under independent patient sampling. The exact mean sign flip p value was 0.053467 under a symmetric null gain assumption. Nine patients had negative gains and four had positive gains. One patient had a large negative gain. Reference checks changed the mean gain and did not establish a general theta prediction result. The package preserves fixed input features, code, all predictions and models, source versions and hashes, raw retrieval recipes, independent calculation checks, and an exact offline repeat. Raw window checks covered all 62 input sessions. One fresh full recording reproduced all 16 stored arrays. This is a new outcome analysis of a shared clinical cohort. It does not establish a neural cause, clinical value, or absent theta information.

hippocampal theta · verbal working memory · response time prediction · stereoelectroencephalography · ridge regression · session-held-out validation