MRI-Estimated Brain Age Predicts Memory Decline in Healthy Older Adults

A brain that appears older than a person’s chronological age was associated with lower cognitive performance and greater memory decline over approximately five years in a large cohort of cognitively healthy older adults, according to an analysis of the population-based AGES-Reykjavik Study. Researchers found that a larger brain age gap, or BAG, was linked to poorer baseline memory, processing speed and executive function after accounting for chronological age, sex and education. The clearest longitudinal finding concerned memory: evidence that BAG predicted faster decline in processing speed or executive function was less definitive, although post hoc subgroup analyses suggested differences in executive-function trajectories among brain-age groups.

The study included 4,606 participants with high-quality structural MRI and cognitive data at baseline. They were members of an Icelandic cohort established to investigate risk factors for disease and disability in aging. Their mean age was 76.4 years, and 58% were women.

A subset of 2,561 participants returned for follow-up cognitive testing after a mean interval of 5.2 years. Researchers estimated brain age from T1-weighted MRI using an automated whole-brain analysis pipeline, then subtracted each participant’s chronological age from the estimate. A positive value meant that the brain appeared older than the participant, while a negative value meant that it appeared younger.

The longitudinal analysis therefore asked whether an MRI-based measure of brain aging could provide information about later cognitive change beyond calendar age alone. At baseline, each additional year of advanced brain age was associated with lower standardized memory scores by a model coefficient of −0.047, lower processing-speed scores by −0.044 and lower executive-function scores by −0.037. All three associations were statistically significant, with p values below 0.001.

The models accounted for chronological age, sex and years of education. Researchers also generated residualized cognitive scores, removing the modeled effects of those covariates before examining the relationship between the remaining score variation and BAG. The baseline BAG values ranged from −18.6 to 11.3 years, with a mean of −2.0 years and a standard deviation of 4.3 years.

BAG was negatively correlated with chronological age, with r = −0.598 and p below 0.001, making adjustment for age important in interpreting the results. The mean BAG was slightly higher among participants who returned for follow-up testing than among those who did not, −1.7 years versus −2.0 years, a difference that reached statistical significance. Because the longitudinal sample was therefore not necessarily representative of everyone included at baseline, the follow-up findings should be understood as applying to participants who were assessed twice rather than automatically to the entire cohort.

Among the 2,561 participants assessed twice, average cognition declined in all three domains. The mean change was −0.237 standard deviations for memory, −0.318 for processing speed and −0.232 for executive function, with all changes statistically significant at p below 0.001. Mixed-effects models were then used to test whether baseline BAG predicted change over time.

These models included fixed effects for BAG, assessment time and their interaction, as well as chronological age, sex, education and the number of days between assessments. Subject-specific random effects accounted for correlation between each participant’s baseline and follow-up scores. This design allowed the researchers to distinguish a BAG-related difference in cognitive level from a BAG-related difference in the rate of decline.

For memory, the analysis found both a significant main effect of BAG and a significant interaction between BAG and assessment time. The main effect, with a coefficient of −0.039, indicated poorer memory among participants with older-estimated brains at both time points. The interaction coefficient was −0.0071, with a 95% confidence interval from −0.0142 to −0.0001 and p = 0.046.

This indicated that higher baseline BAG was associated with a steeper decline in memory over follow-up, independently of the measured demographic and timing factors. The result supports the use of BAG as a prospective marker of memory vulnerability in this cohort, although it does not establish that brain aging causes the observed change. Processing speed and executive function showed a different pattern.

A larger BAG remained associated with lower performance in both domains across the study period: the longitudinal main-effect coefficients were −0.044 for processing speed and −0.028 for executive function, with p values below 0.001. However, neither BAG-by-time interaction was statistically significant. The interaction was −0.0026 for processing speed, with a 95% confidence interval of −0.0086 to 0.0031 and p = 0.402, and −0.0051 for executive function, with a 95% confidence interval of −0.0120 to 0.0019 and p = 0.156.

Thus, the primary models supported a BAG-related difference in the level of performance, but did not establish that BAG significantly changed the rate of decline in either domain. Secondary post hoc analyses divided participants into five BAG groups: ≤ −10 years, −10 to ≤ −5 years, −5 to ≤ 0 years, 0 to ≤ 5 years and 5 to ≤ 10 years. The groups with positive BAG values showed steeper declines in memory and executive function than the groups with negative or zero BAG values.

Pairwise comparisons were significant for memory, with adjusted p values from 0.001 to 0.022, and ranged from 0.001 to 0.051 for executive function. The highest group therefore represented brains estimated to be 5 to 10 years older than chronological age, not a category beginning only above five years. Although the higher-BAG groups also showed steeper absolute declines in processing speed, the analysis did not identify additional group differences in that domain at follow-up relative to baseline.

These subgroup findings were exploratory and depended on dividing a continuous brain-age measure into categories. The study offers prospective evidence that MRI-estimated brain age contains information about cognitive vulnerability beyond a person’s calendar age, but several limitations restrict how far the findings can be generalized. BAG was a global measure and could not identify which brain regions were responsible for particular cognitive outcomes.

Although participants were considered healthy, subtle prodromal pathology could not be completely excluded. The models did not assess interactions with other established risk factors for cognitive decline, including diabetes and smoking, and the analysis used structural MRI alone rather than diffusion imaging, functional MRI or other modalities. The relatively short follow-up period also limits interpretation of trajectories, while the relatively preserved cognitive functioning of the cohort may have reduced the ability to detect subtle differences in processing-speed decline.

Moreover, the cohort was relatively homogeneous and geographically localized to Iceland, so validation in more ethnically and geographically diverse populations would be needed before the measure could be treated as a broadly established clinical biomarker. For now, the results identify brain age gap as a possible non-invasive marker of vulnerability to memory decline among older adults, rather than as a diagnostic test or a demonstrated method for predicting dementia. The authors propose that future studies should examine longer follow-up, conversion to mild cognitive impairment or dementia, and whether interventions targeting physical activity, vascular health or cognitive training can alter brain-age trajectories.

Those questions remain open: the present analysis shows associations between older-estimated brains and cognitive outcomes, not that changing brain age would prevent cognitive decline. Its strongest longitudinal finding is that a larger BAG predicted steeper memory loss over roughly five years, while evidence for domain-specific changes in processing speed and executive function was less definitive.

Subject of Research: MRI-estimated brain age and five-year cognitive decline in healthy older adults

Article Title: Advanced brain age and longitudinal cognitive decline in healthy older adults

Article References: Kristinsson, S., Newman-Norlund, R., Bonilha, L., Bakos, L., Rorden, C., Sigurdsson, S., Gudnason, V., & Fridriksson, J. (2026). Advanced brain age and longitudinal cognitive decline in healthy older adults. GeroScience. https://doi.org/10.1007/s11357-026-02598-w

Image Credits: AI Generated

DOI: 10.1007/s11357-026-02598-w

Keywords: brain age gap, MRI, healthy aging, cognitive decline, memory, executive function, processing speed, Reykjavik Study, neuroimaging, longitudinal cohort, Advanced, brain