Two people the same age can sit at opposite ends of health: one with clean checkup numbers, one already managing hypertension and fatty liver. The difference may not show up in a standard panel at all, but in the metabolic rhythm recorded in blood.

A team led by Prof. Bu Jun at Renji Hospital, Shanghai Jiao Tong University School of Medicine, working with Prof. Qian Kun at Shanghai Jiao Tong University, published the work in Science Translational Medicine, where it made the cover. The cover theme: "Ticking Metabolites."

What the study did

The cohort covered 18,476 participants from six centers in China, aged 1 to 96, followed for about six years.

The method combined a home-grown high-throughput solid-phase mass spectrometry platform with ensemble machine learning. The team mapped plasma metabolic profiles across the full lifespan and narrowed them to 56 representative age-related metabolic features.

Sample size, age span and follow-up length are what give this weight. Metabolomics studies are common; longitudinal cohorts covering the whole lifespan and tracking outcomes are not.

The core metric: mAA

From that base the team built "metabolic age acceleration" (mAA), which quantifies how far a person's metabolic age deviates from their chronological age.

The associations are direct. As the cohort aged, mAA correlated with hypertension, fatty liver, diabetes and higher all-cause mortality risk. In the all-cause mortality analysis, each standard deviation increase in mAA corresponded to roughly a 50 percent higher risk.

A second metric, "age deviation," was developed on a disease cohort. Values were higher in patients with more severe and systemic disease, and predicted prognosis and mortality in multimorbidity.

What it means clinically

The practical contribution is turning "speed of aging" into something quantifiable: one blood draw, one number representing deviation from peers.

The value to an individual is lead time. If mAA runs well above chronological age, the metabolic state is accelerating — and lifestyle or chronic-disease interventions at that point have more room to work than after a lab value goes abnormal.

The study also found associations between lifestyle, chronic disease management and metabolic aging, adding real-world evidence that proactive management is not just plausible but measurable.

Limits worth stating

First, this is observational. It establishes correlation, not that lowering mAA extends life; that requires intervention trials.

Second, six years of follow-up is short for a process that unfolds over decades.

Third, the 56 features were derived from a Chinese cohort. Cross-population validity needs more work.

Fourth, the metric depends on a specific mass spectrometry platform. Standardization across platforms is a real operational problem.

The data is open

The team published the database and model framework at pu-lab.sjtu.edu.cn/metsclock — usable for bioinformatics and clinical research groups.

AI's role here is not discovering the secret of aging. It is picking the few dozen features with real signal out of tens of thousands of metabolites, then building a reproducible scoring model. Unglamorous work, and the precondition for precision health management to actually function.