Mortality-trained MRI brain age acceleration mediates sex-specific associations between health-related phenotypes and incident dementia

Background Brain age acceleration (BAA) has emerged as a promising neuroimaging marker of brain health and neurodegeneration. However, most existing brain age models are trained to predict chronological age, which may not be the most clinically relevant target. In this study, we instead developed a brain age model trained on all-cause mortality risk. Methods Using multimodal MRI data, including T1-weighted, T2-FLAIR, and diffusion tensor imaging, from more than 46,000 UK Biobank participants, we constructed sex-specific brain age models within a Cox-LASSO framework to prioritize neuroimaging features associated with mortality. Residual-based BAA was then derived to reflect deviation from expected brain health after accounting for chronological age. External validation was performed in the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and West China Health and Aging Cohort Study (WCHAC) cohorts. Results BAA was consistently associated with higher all-cause mortality risk in both sexes, with hazard ratios of approximately 1.10 to 1.11 per year increase, and was also related to multiple chronic diseases and neuropsychiatric outcomes. Higher BAA was significantly associated with dementia and Alzheimer’s disease. Sex-stratified analyses revealed distinct patterns. In males, BAA was broadly associated with cardiometabolic, inflammatory, and socioeconomic phenotypes, whereas in females the main associations were concentrated in smoking and central adiposity. Longitudinal causal mediation analysis, with exposures measured at baseline, BAA assessed at the imaging visit, and dementia ascertained thereafter, showed that BAA mediated part of the associations between several modifiable phenotypes and incident dementia. In males, leukocyte count showed the largest mediated proportion at 48.4%, while smoking, liver enzymes, and cardiorespiratory measures showed smaller but interpretable mediated effects. In females, smoking pack-years and waist to hip ratio showed significant mediated pathways. Conclusions These findings suggest that a mortality-trained brain age model may better capture clinically relevant brain aging. BAA may serve as a sex-specific neuroimaging marker linking systemic health to dementia risk and may have value in risk stratification and prevention. Graphical Abstract Created in BioRender. Lin, L. (2026) https://BioRender.com/qwg707i