Sex-specific aging clocks reveal differences in disease-risk associations

Researchers reported September 16 in Nature Medicine that biological-aging clocks trained separately for females and males reveal different associations with disease and mortality. The study adds evidence that the reference population behind an aging score matters when researchers use it to assess future health risks.

The MULTI Consortium developed 38 clocks spanning 15 organ systems, using imaging, blood proteins and metabolites. Its observational and computational analyses drew on UK Biobank and additional aging and Alzheimer’s research cohorts. Outcomes included incident disease, all-cause mortality and cognitive decline.

The investigators also examined brain-aging estimates in 1,055 participants with baseline MRI scans from the A4 Alzheimer’s prevention trial. These analyses concerned associations with subsequent cognition; they do not establish that changing an aging score improves health.

### What an aging score can tell researchers

An aging clock turns measurements into an estimated age or an age-related score. Its usefulness depends on what information went into the model, which people supplied the training data and how reliably the result relates to outcomes in other populations.

Those methodological questions predate this week’s publication. In an August 2025 Nature Aging analysis, Junhao Wen developed 11 organ-related protein clocks using 2,448 plasma proteins from 43,498 UK Biobank participants. That work examined how age-bias correction, sample size, underlying illness in training participants and the organ specificity of proteins affect interpretation and generalizability.

The earlier analysis also found that combining information across organs improved predictions of systemic disease categories and mortality. It provides useful context for evaluating any new clock: a compelling association is only one part of the evidence needed to judge a measurement tool.

For prevention research, the practical question is whether an additional score supplies dependable information beyond what researchers or clinicians already know. A model can be informative about population patterns while still leaving substantial uncertainty about an individual’s future.

### Why the Alzheimer’s trial context matters

The original A4 trial, published in July 2023 in the New England Journal of Medicine, randomized 1,169 adults aged 65 to 85 to solanezumab or placebo. Participants had elevated brain amyloid but no cognitive impairment. Its primary endpoint measured change in a cognitive composite over 240 weeks.

Solanezumab did not significantly slow cognitive decline. The between-group difference was −0.30 points, with a 95% confidence interval from −0.82 to 0.22. Brain amyloid increased less in the treatment group, illustrating why a biomarker difference must be considered separately from a demonstrated cognitive benefit.

Reusing trial data to investigate risk patterns can generate valuable hypotheses. But a subsequent biomarker analysis does not overturn the original randomized treatment comparison. Any proposal to select future patients using such a marker would need its own prospective evaluation.

### Limits before clinical use

The new study’s genetic analyses were restricted to European ancestry, repeated biological measurements were limited, and sex was modeled as a binary variable. The authors also call for external validation at biobank scale and caution that separate models are not universally superior to pooled approaches.

The editorial implication is a concrete standard for future claims: ask whether a clock predicts meaningful outcomes in the population where it will be used, and whether acting on its result improves those outcomes. Those are separate tests. Neither an impressive model nor a statistically significant association can substitute for both.