Your age may not tell the whole story about your health, but your blood could

By Pooja Toshniwal PahariaReviewed by Susha Cheriyedath, M.Sc.Oct 18 2026

What can the chemicals circulating in your blood reveal about how you age? Scientists examined metabolic patterns across 96 years of life to investigate their relationship with health and longevity.

Study: Metabolomic aging and mortality clocks based on ensemble learning identify systemic aging risk across the human life span in 18,476 individuals. Image Credit: luchschenF / Shutterstock

In a recent study published in the journal Science Translational Medicine, researchers developed two metabolomic aging metrics to examine relationships with chronic disease and mortality risk using metabolic data from over 18,000 individuals aged 1 to 96 years. The findings highlight the potential of plasma metabolomic profiling for future population-level risk assessment and improved risk stratification.

With advancing age, physiological functions gradually begin to decline. Greater biological age has also been associated with an increased risk of chronic diseases and mortality. Yet metabolomic signatures across the lifespan are unclear. Understanding metabolomic changes associated with aging could inform therapeutic development and targeted interventions to prevent age-related physiological deterioration.

About the study

In the present study conducted across six centers in Shanghai, China, the researchers analyzed plasma samples from 18,476 Shanghai Pudong Cohort participants aged 12 months to 96 years using nano-assisted laser desorption/ionization mass spectrometry (LDI-MS).

The study first included 5,252 healthy individuals with a mean age of 55 years to develop the metabolomic age model. The researchers then tested the model in a healthy validation group of 5,456 participants and examined it in a disease cohort of 7,768 participants, with mean ages of 55 and 68 years, respectively.

Standardized questionnaires captured participants' diet, lifestyle, sleep, mental health, and family medical history where these data were available. Baseline assessments were conducted in 2019, chronic disease status was reassessed in 2023, and mortality was tracked through January 2025.

In 2023, 3,745 healthy-cohort participants and 4,658 disease-cohort participants completed repeat surveys and updated chronic disease assessments. Of these, 1,500 underwent repeat plasma metabolomic profiling.

In the healthy cohort, the team estimated metabolomic age using a stacking multialgorithm ensemble model and calculated metabolomic age acceleration (mAA), which indicates the extent to which metabolomic age differs from a person's actual age.

For participants with existing health conditions, the researchers used Cox regression models to develop an age deviation metric that measured how metabolite-based mortality risk differed from what would be expected for a person's age.

Pathway enrichment analysis identified biological pathways associated with aging. Feature selection combined random forest, mutual information, and LASSO regression to identify aging-associated metabolites. Additional statistical analyses reduced prediction bias and assessed sex-specific patterns.

The team tested the ensemble approach by retraining it on UK Biobank (UKB) proteomic, National Health and Nutrition Examination Survey (NHANES) physiological biomarker, and human gut microbiome data.

Results

A panel of 56 age-related metabolites retained most of the age-predictive information provided by all 232 metabolite features. Higher mAA values were linked to several lifestyle, environmental, and family-history factors, including active smoking, secondhand smoke exposure, hair dye use, history of shift work, and parental history of premature cardiovascular disease. Higher mAA remained associated with greater mortality risk after accounting for established health measures.

Higher mAA was associated with greater consumption of organ meat, sugar, processed meat, seafood, and fried foods, whereas mixed grains and exercise were linked to lower values. Repeated analyses linked increased physical activity and reduced fried-food intake to more favorable changes in mAA.

Separately, among participants with existing hypertension, those who achieved blood pressure below 140/90 mmHg by 2023 showed more favorable changes in the mortality-linked age deviation metric than those with uncontrolled hypertension. These observational findings do not establish whether individual foods or lifestyle changes directly affect biological aging.

Participants with accelerated mAA were more likely to develop chronic conditions such as diabetes, hypertension, and fatty liver disease, and had a higher risk of death from any cause. The researchers also found that age deviation was associated with overall health status, the presence of diseases affecting multiple body systems, and the burden of multiple conditions, suggesting that it could offer a practical way to assess health risks across populations.

Both metrics also improved the ability to distinguish individuals at higher and lower mortality risk beyond frailty assessments, cardiovascular health scores, and routine blood biomarkers, though the improvements in risk discrimination were modest.

The metabolomic alterations implicated biological pathways associated with threonine, serine, glycine, butanoate, cysteine, and methionine metabolism. Metabolites positively linked with aging included 3-methylindole, creatinine, and sebacic acid, while leucine increased after midlife and was associated with greater mAA. Meanwhile, l-asparagine, glyceric acid, and 4-hydroxyproline, a collagen-related metabolite, demonstrated negative associations with age.

Serine and L-asparagine showed similar patterns in both sexes, whereas glyceric acid and L-cysteine showed greater sex differences, and the mortality-linked age deviation metric was higher in men than in women in the disease cohort.

Among the metabolites, 3-methylindole had the most statistically significant positive association with mortality risk, whereas 4-hydroxyproline had the most statistically significant inverse association. In general, metabolites linked to higher mortality risk also showed associations with higher body mass index (BMI), greater frailty, and poorer cardiovascular health.

Age prediction was strong when the ensemble approach was retrained on NHANES and UKB data but moderate for gut microbiome data. These tests assessed the adaptability of the modeling approach, rather than independently validating the original plasma clock.

In participants with pre-existing disease, the original metabolomic age model corresponded less closely to chronological age, supporting the development of a separate mortality-linked measure. Among participants with comorbidities, age deviation was consistently associated with routine blood biomarkers.

Higher creatinine, uric acid, and urea nitrogen levels were linked to greater age deviations, whereas hemoglobin, hematocrit, and erythrocyte counts showed opposite associations. Conditions such as chronic kidney disease, gout, and stroke were individually associated with higher age deviation, which also increased as the number of existing diseases rose.

Implications

The findings suggest that metabolomic clocks could help identify people at risk of chronic disease and death. If mechanistic investigations and large-scale studies involving diverse populations and longer follow-ups confirm these findings, these metrics could complement epigenomic and proteomic clocks. Epigenomic clocks track age accurately but may be less responsive to current physiological changes, while proteomic clocks provide organ-level information but can require complex assays.

Rapid plasma metabolomic profiling may offer a simpler approach to large-cohort analysis. Since metabolites reflect ongoing biochemical activity, examining them alongside environmental exposures, genetic variations, and pathological changes could provide a more comprehensive picture of health, provided future studies establish causal links. 

The predominantly Chinese study population, limited repeat sampling, and need for independent validation of mortality prediction restrict immediate clinical application. Metabolite identification and proposed pathways also require further experimental confirmation. The researchers did not adjust exploratory metabolite associations for multiple comparisons.