Latent trajectory analysis of immediate, 30-minute and one-week delayed recall as a sensitive cognitive marker of early Alzheimer’s disease pathology in cognitively unimpaired individuals

Background: Accelerated long-term forgetting - disproportionate memory loss over days despite normal short-delay retention - has been suggested as an early cognitive manifestation of preclinical Alzheimer’s disease (AD). Existing retention-ratio approaches overlook the full acquisition-to-consolidation pattern. We aimed to characterize distinct learning-retention profiles and explore their associations with early AD pathology in cognitively unimpaired adults. Methods: A total of 322 cognitively unimpaired older adults (mean age 67, 60% female, 57% APOE-ε4 carriers, 95% with family history of AD) completed the Free and Cued Selective Reminding Test face-to-face (16-word list: three immediate recall trials, a 30-minute delayed recall). An additional, unannounced one-week delayed recall was administered by phone. We used growth mixture modeling across these five timepoints to identify distinct trajectory classes and explored their associations with AD biomarkers (CSF Aβ42/Aβ40, p-tau181/Aβ42; plasma p-tau217) using general linear models, and retrospective longitudinal MRI (bilateral total medial temporal lobe and hippocampal subfield volumes derived from a dedicated high-resolution inversion-recovery acquisition processed using the ASHS pipeline, three acquisitions across 6.6 years) employing linear mixed-effects models, adjusted for age, sex, and education, and intracranial volume for volumetric outcomes. Associations were corrected for multiple comparisons within outcome families by the Benjamini-Hochberg procedure (pFDR). Results: Three distinct trajectory classes were identified: High (n = 201; 62.4%), Intermediate (n = 81; 25.2%), and Low performers (n = 40; 12.4%). Intermediate performers showed preserved learning and 30-minute recall but disproportionate forgetting at one week compared to High performers. Plasma p-tau217 levels were significantly associated with classes (pFDR = .021), with Low performers showing higher concentrations than Intermediate (β = 0.49; p = .012), and High performers (β = 0.50; p = .006). CSF p-tau181/Aβ42 followed the same pattern (pFDR = .021), with Low performers differing from High performers (β = 0.58; p = .006) and Intermediate performers not (β = 0.22; p = .161), while CSF Aβ42/Aβ40 was not associated with class. Longitudinal models revealed significant class × time interactions for bilateral total medial temporal lobe volume (pFDR = .034); Low performers showed significantly greater atrophy than High performers (β = -0.056; p = .015), corresponding to 2.3× greater annual atrophy. In particular, CA1 subfield revealed the strongest effect (pFDR = .019); both Low (β = -0.048; p = .004) and Intermediate performers (β = -0.029; p = .036) showed significantly greater annual decline than High performers (3.0× and 2.2× faster annual atrophy, respectively). Conclusions: Learning and forgetting trajectories including delayed recalls beyond standard intervals relate to early AD pathology, in particular plasma p-tau217 elevation, and accelerated hippocampal atrophy. Further studies exploring the ability of combined metrics of plasma p-tau217 and long-term recall trials to predict short-term cognitive decline are warranted. Keywords: accelerated long-term forgetting, preclinical Alzheimer’s disease, memory consolidation, growth mixture modeling, free and cued selective reminding test, plasma p-tau217, hippocampal volume.