Earlier hippocampal change gave an uncertain 2.2 percent error reduction in one 70-person DLBS delayed recall test with linear models and known visit intervals

We tested whether hippocampal volume change at the first two DLBS visits improves delayed word recall prediction at the third visit. We fixed the rules before computing new scores. We recovered the correct memory file from release 1.2.0 and checked the subject join to release 1.3.0 brain data. The sample had 173 people: 103 trained the linear models and 70 tested them. The baseline used prior recall, age, sex, education, current hippocampal volume, head volume, and visit intervals. Adding annual log volume change reduced test mean squared error from 3.011661 to 2.946119 words squared, a 2.176 percent gain. The paired 95 percent interval was -2.175 to 6.159 percent with trained models fixed. An interval that also refit the models was -13.497 to 9.507 percent. The result leaves added prediction value uncertain for this rule and sample. All eight planned supporting checks are reported. The package includes source data, pinned hashes, code, the selection ledger, predictions, resampling draws, and figures. Independent source and QR checks passed. All 21 numerical files matched after fresh source retrieval. This offline test assumes the recorded future return interval and does not establish a neural cause or clinical use.

hippocampal volume · delayed recall · memory prediction · longitudinal MRI · Dallas Lifespan Brain Study