The research, published in European Geriatric Medicine, leveraged the A/T/N classification framework, a widely adopted scheme in which ‘A’ denotes amyloid-beta pathology, ‘T’ denotes phosphorylated tau, and ‘N’ denotes neurodegeneration, typically measured as atrophy of the medial temporal lobe on CT or MRI. In this study, cerebrospinal fluid levels of amyloid-beta 42 and phosphorylated tau defined the A and T markers, while radiologists rated medial temporal atrophy using the Scheltens visual scale, with age-adjusted cut-offs determining whether a score was abnormal. Episodic memory was assessed with the Rey Auditory Verbal Learning Test, a 15-item word-list task that measures both learning across five trials and free recall after a 30-minute delay.

The cohort was deliberately naturalistic rather than curated. Unlike highly selected research samples such as the Alzheimer’s Disease Neuroimaging Initiative, the MemClin project enrolled all patients referred for neuropsychological examination across participating clinics, capturing the messy heterogeneity of real clinical practice. The final sample comprised 141 patients with Alzheimer’s disease dementia, 403 with mild cognitive impairment, and 132 with subjective cognitive impairment, with ages ranging from roughly 36 to 94 years. Diagnoses were made through multidisciplinary consensus meetings in which clinical presentation remained primary and biomarkers played a supportive role, mirroring the way most memory clinics actually operate.

Because many patients scored zero on delayed recall—a floor effect expected in a memory-clinic population—the team employed weighted least-squares regression rather than ordinary linear models, assigning observation-specific weights to stabilize variance. Six regression models tested whether age moderated the relationship between each biomarker and each memory measure, controlling for sex and education, with a Bonferroni-corrected significance threshold of p less than 0.008. The results were unambiguous for two of the three biomarkers. Abnormal amyloid-beta 42 interacted significantly with age on delayed recall, with the detrimental effect of amyloid abnormality shrinking as age increased (β = 0.14, p < 0.001). Medial temporal atrophy showed a parallel interaction (β = 0.13, p = 0.002). Both models explained about 22 percent of the variance in delayed recall performance.

Phosphorylated tau, by contrast, did not survive the statistical correction, though its interaction pattern was borderline significant and trended in the same direction. The authors suggest this may reflect the comparatively stronger specificity of phosphorylated tau as an Alzheimer-specific marker, one whose relationship to cognition may be less entangled with age than amyloid or atrophy. Previous work has indicated that phosphorylated tau levels are less strongly related to age than amyloid-beta 42 or total tau, lending plausibility to that interpretation, although the researchers caution that a non-significant interaction should not be read as proof that tau is entirely age-independent.

To pinpoint where the biomarker-cognition link begins to fail, the team stratified the sample into two-year age bands and re-ran the association between abnormal amyloid status and memory performance repeatedly across those strata. The attenuation accelerated sharply at the upper end of the age distribution: for participants aged 80 to 82 and older, abnormal amyloid-beta 42 no longer showed a statistically meaningful association with episodic memory performance, with p-values exceeding 0.36, while the association remained robust in younger bands. Medial temporal atrophy followed the same trajectory. In other words, the diagnostic sensitivity of these markers appears to erode earlier than the traditional ‘oldest old’ threshold of 85 years, a finding the authors describe as unexpected from a clinical standpoint.

The biological explanation likely lies in the sheer prevalence of Alzheimer pathology in advanced age. Autopsy and imaging studies have shown that abnormal amyloid can be detected in up to 40 percent of cognitively healthy elderly individuals, and that by the time symptoms emerge, amyloid burden has largely saturated. Neuropathological research has also demonstrated that the correlation between Alzheimer-type pathology and dementia weakens with advancing age, as vascular disease, hippocampal sclerosis, TDP-43 proteinopathy, inflammatory processes, and individual differences in cognitive reserve increasingly shape clinical outcomes. The landmark 90+ Study illustrated this vividly: roughly half of its participants without dementia nonetheless met criteria for Alzheimer pathology at autopsy. In the oldest old, medial temporal atrophy may similarly reflect a mixture of age-related processes rather than Alzheimer-specific neurodegeneration, diluting its predictive value.

The clinical implications are provocative. The authors raise the question of whether lumbar puncture and cerebrospinal fluid assessment are justified in patients older than 80, given the weak association between the biomarkers and core clinical measures such as learning and free recall. They are careful, however, to draw boundaries around that claim. The finding should not be interpreted as questioning the broader utility of CSF biomarkers, which may remain important for diagnostic evaluation, prognosis, and determining eligibility for emerging disease-modifying therapies, including anti-amyloid immunotherapies. Nor should the exploratory age-stratified analyses be treated as confirmatory; small subgroup sizes, the cross-sectional design, and the risk of type 1 error all temper the conclusions, and the authors frame these results as hypothesis-generating pending large-scale longitudinal validation.

The study also carries methodological caveats that the researchers confront directly. Participants excluded for missing data differed in age from those included—excluded dementia patients were older, while excluded MCI and SCI patients were younger—raising the possibility of selection effects, although the pattern of CSF testing being more common in younger, diagnostically challenging patients arguably makes the sample representative of real practice. Visual atrophy ratings were based on CT in 60 percent of cases and MRI in 40 percent, a combination supported by evidence of comparable inter-rater reliability. Biomarkers were evaluated individually rather than in combination, and only verbal learning and free recall were examined, leaving recognition memory, cued recall, and executive functions for future study.

What emerges is a nuanced portrait of biomarker diagnostics at the frontier of human longevity. In a naturalistic cohort spanning the full cognitive-impairment continuum, the two biomarkers most proximal to memory circuitry—amyloid and medial temporal atrophy—lost traction against advancing age, while phosphorylated tau held its pattern more steadily. If replicated longitudinally, these findings could reshape diagnostic algorithms for the fastest-growing segment of the dementia population, prompting clinicians to weigh clinical presentation more heavily and biomarkers more cautiously once patients cross their ninth decade. For now, the message is one of calibrated skepticism: the molecular signature of Alzheimer’s disease does not translate into memory impairment with equal fidelity at every age, and medicine’s most trusted biomarkers may need an age-adjusted interpretation of their own.

Subject of Research: Age-related weakening of the association between Alzheimer's disease ATN biomarkers and episodic memory in memory clinic patients

Article Title: The associations between ATN biomarkers and episodic memory diminish as age increases

Article References: The associations between ATN biomarkers and episodic memory diminish as age increases. (n.d.). https://doi.org/10.1007/s41999-026-01606-8

Image Credits: AI Generated

DOI: 10.1007/s41999-026-01606-8

Keywords: Alzheimer's disease, ATN biomarkers, amyloid-beta 42, phosphorylated tau, medial temporal atrophy, episodic memory, cerebrospinal fluid, cognitive aging, memory clinics, RAVLT, mild cognitive impairment, diagnostic biomarkers