Biological age test

What a biological age test is

A biological age test turns a lab measurement into an estimated age or pace of aging.[1][2][3][4][5] One kind of test, called an epigenetic clock, reads DNA methylation, which are chemical marks on DNA that change with age.[1] The result comes from a statistical model.

What the clocks measure

The Horvath, Hannum, PhenoAge, GrimAge and DunedinPACE clocks were trained on four different targets: calendar age, a blood-chemistry composite, time to death and a rate of change.[1][2][3][4][5] Because the clocks were trained on different targets, their results are different quantities and cannot be swapped for one another.

Two clocks trained on the same target, such as the Horvath and Hannum clocks, are still different models, built from different training data and, in these two cases, different tissues.[1][2]

The Horvath clock, published in 2013, reads DNA methylation at 353 sites and was trained to estimate calendar age using 7,844 non-cancer samples from 82 datasets covering 51 tissues and cell types.[1]

The Hannum clock, published in 2013, was trained to estimate calendar age from DNA methylation in whole blood from 656 people aged 19 to 101.[2] The Horvath and Hannum clocks were trained on calendar age, so health outcomes were not part of what they were trained to estimate.[1][2]

The PhenoAge clock first calculates a phenotypic age from 9 routine blood measures plus age, then uses DNA methylation at 513 sites to estimate that phenotypic age.[3]

The GrimAge clock combines DNA methylation estimates of 7 blood proteins and of smoking pack-years and was trained to estimate time to death.[4] Because the GrimAge clock includes an estimate of smoking, part of what it reports is smoking history.

The DunedinPACE clock gives a pace of aging, scaled so that 1 equals the average rate of one year of biological aging per calendar year, and it was trained on the rate of change in 19 biomarkers over 20 years in one birth cohort of 1,037 people.[5] The authors of the DunedinPACE paper describe their birth cohort as relatively modestly sized and drawn from a single country, and say the cohorts they used were mostly of white European descent.[5]

How repeatable one result is

In a 2022 study funded by the US National Institutes of Health, 36 blood samples were each measured twice, and across six epigenetic clocks the two results from the same sample differed by up to 9 years.[6]

In the body of that paper, the largest difference between two results for the same sample was 8.6 years for the clock with the widest gap, and the median difference for each clock ranged from 0.9 to 2.4 years.[6]

Versions of the clocks reworked by the same researchers agreed within 1.5 years for most pairs.[6] The researchers' own clocks are licensed to a company, which also supplied some of the study's data.[6]

A 2026 analysis that pooled four sets of repeat measurements found that the majority of 18 clocks gave highly consistent results, with an intraclass correlation above 0.9, when one sample was measured again.[7] In four short-term datasets in the same 2026 analysis, in which the same people gave blood samples at different times, most clocks reached only moderate to good consistency, with intraclass correlations of about 0.4 to 0.7, and no clock reached the excellent range.[7]

The authors of the 2026 analysis wrote that the biological events they examined had no control groups examined at the same time, so it remains unclear whether the instability is due to the event or to taking blood at different times.[7] Two authors of the 2026 analysis are named co-inventors of a clock it tested.[7] The authors of the 2026 analysis also wrote that their technical estimates do not capture other sources of variability, such as differences between laboratories or between lots of reagents.[7]

In one 2024 study, white blood cells from a single healthy 52-year-old man were sampled every 3 hours for 72 hours, and 13 of 17 epigenetic clocks showed significant oscillation over the day, 8 of which stayed significant after correction for multiple testing.[8] In that man, the average epigenetic age estimated by the Horvath clock oscillated from 55.3 years at about 11:30 PM to 58.5 years at about 11:30 AM.[8] The authors wrote that using a single individual prevented them from making generalizations about differences between people.[8]

How the sample type changes a result

In a 2025 study of 83 people aged 9 to 70, researchers ran epigenetic clocks on cheek cells, saliva and blood fractions from the same people.[9]

Of the six clocks that report an age in years, three gave a cheek cell result 21 to 40 years higher than the result from blood, two gave a result 1 to 6 years higher, and one gave a result 8 to 11 years lower.[9] PedBE, one of the two clocks that read 1 to 6 years higher, is a clock built for children, and its results come from the children's samples.[9] The Horvath clock, which read lower on cheek cells, was trained on cheek, saliva and blood samples.[9]

The authors concluded that applying blood-derived epigenetic clocks to oral-based tissues may not yield comparable estimates of epigenetic age.[9]

In a study of 107 people aged 5 to 74 with matched saliva and blood samples, funded by the Max Planck Society, blood-derived clocks agreed moderately between the two sample types for newer measures and poorly for first-generation clocks.[10] The authors wrote that the similarity between sample types may be adequate for population research but may not suit clinical or commercial uses.[10]

Links with death and disease

A 2016 meta-analysis of 13 cohorts with 13,089 people and 2,734 deaths found that the epigenetic age acceleration measures it considered were associated with time to death, independent of chronological age.[11] One author of the 2016 meta-analysis is a named inventor on a patent application for measures of epigenetic age acceleration.[11]

A 2019 meta-analysis pooled 6 studies for each of two epigenetic clocks, with 17,988 participants across both analyses.[12] In that meta-analysis, a 5-year higher epigenetic age relative to calendar age was associated with an 8% higher risk of death from any cause for the Horvath clock and a 15% higher risk for the Hannum clock.[12]

The authors reported that the funnel plots were asymmetrical, indicating positive publication bias, and said further research is needed to determine how far DNA methylation age can be used as a clinical biomarker.[12] That review searched for studies up to September 2, 2018, covered first-generation clocks only, and reported no funding and no competing interests.[12]

In a 2026 study of 1,083 adults aged 60 to 80 in the Berlin Aging Study II, followed for an average of 7.4 years, a pace of aging measure, DunedinPACE, had the strongest association with death among the 14 aging biomarkers the study compared.[13] One author of the Berlin study is a cofounder, stockowner, board member and consultant of a company, and another is its chief executive and a stockowner.[13]

In a 2025 study of 18,859 people in Scotland followed for 10 years, 14 epigenetic clocks were compared with 174 incident diseases and death from any cause, and 176 pairings of a clock and a disease showed a significant association after correction for multiple testing.[14] In 32 of those 176 pairings, adding the clock to a model of standard risk factors increased classification accuracy by more than 1%.[14]

One author of the Scotland study is an inventor of DunedinPACE, which is licensed to a company, and of the other authors one advises a company and one is employed by it.[14] The Scotland study was supported by grants that include the US National Institute on Aging and the Wellcome Trust.[14]

Using a result for one person

In a 2025 perspective, four authors who write about epigenetic clocks argued that, as currently understood, these clocks should not be used to make decisions about an individual.[15] A link seen across thousands of people does not tell one person what will happen to them.[15]

The same four authors wrote that epigenetic clocks do not have clear cut-off values of aging acceleration that would indicate an intervention is needed.[15]

The authors of the Scotland study concluded that second- and third-generation epigenetic clocks show promise for disease risk prediction, particularly for respiratory and liver conditions.[14]

In searches of PubMed and Europe PMC on October 8, 2026, we found no study that checked a consumer biological age test's result against what later happened to the people who took it.

What changing a score has and has not shown

In a 2-year trial that randomized 220 healthy adults without obesity, calorie restriction slowed one measure, DunedinPACE, by about 2% to 3% and did not change two others, PhenoAge and GrimAge.[16] The authors said it is unclear whether the change in DunedinPACE will translate into less illness or death over the long term.[16]

The analysis was supported by US National Institutes of Health grants, and two of the authors are named inventors of DunedinPACE, which was licensed to a company.[16] The authors described the treatment effect sizes as small.[16]

In searches of Europe PMC and ClinicalTrials.gov on October 8, 2026, we found no published trial showing that lowering a biological age score led to fewer illnesses or deaths.

A 2026 analysis of 51 interventions in longitudinal intervention datasets, with 3,128 samples and 16 epigenetic biomarkers, found that 19 interventions significantly decreased epigenetic age and 26 had no significant effect, and the others significantly increased it, in tests not corrected for multiple testing.[17] The authors of that analysis wrote that the absence of an empirically defined minimal clinically important difference for DNA methylation clocks remains a key barrier to using them as surrogate endpoints.[17]

The authors also wrote that second-generation biomarkers, including PCGrimAge, SystemsAge and DunedinPACE, showed the strongest and most consistent responsiveness.[17] Three authors of that analysis are employees of a company and developed one of the clocks it tested, and two others are named co-inventors of a patented clock the paper names among the most responsive.[17]

Two authors of a 2022 article on endpoints for aging trials wrote that a treatment that slows a measure of biological age that is not a true surrogate is likely to be widely used but ineffective or even harmful.[18] One of those two authors has been paid as a consultant by a company.[18]

The authors of a 2025 perspective wrote that a difference in estimated epigenetic age of a year or two, the size often reported by intervention studies, may fall within the margin of error introduced by these sources of noise.[15]

Telomere length tests

In a 2018 meta-analysis of 121,749 people with 21,763 deaths, a one standard deviation shorter telomere length in white blood cells was associated with a hazard ratio for death of 1.09 (95% confidence interval 1.06 to 1.13).[19]

The authors noted between-study heterogeneity and said the hazard ratio was smaller in people over 80.[19]

A 2021 critical appraisal of 25 studies that compared qPCR telomere measurement with other methods found that only two had tested how closely repeat measurements agreed, with intraclass correlations of 0.89 and 0.92.[20] The authors concluded that the current literature on the relation between telomere measurement methods is lacking in validity and scientific rigor, and the appraisal was funded by US National Institutes of Health institutes with no competing interests declared.[20]

In a secondary analysis of the 2-year CALERIE trial in healthy adults aged 21 to 50 without obesity, calorie restriction produced no significant difference in telomere length change across the first year or from baseline to 24 months, and the authors said it is unclear whether this reflects long-term effects, a hormesis-like adaptation, or measurement error and insufficient statistical power.[21]

How strong the evidence is

Evidence that getting a biological age result from a lab measurement such as DNA methylation improves health: no support found. In searches of PubMed, Europe PMC and ClinicalTrials.gov on October 8, 2026, we found no completed, published randomized trial that tested whether getting an epigenetic age result improves a health outcome.

We found one small randomized pilot, in 34 adults with metabolic syndrome, that compared education about epigenetic aging with no such education during a 4-week diet.[22][23] The pilot measured how closely people followed the diet and how acceptable they found the testing.[22] No health outcome was compared between the groups in that pilot.[22]

On October 8, 2026, ClinicalTrials.gov listed two randomized trials of telomere length information, one of personal telomere length disclosure with distress as the main outcome and one of telomere education in cardiac rehabilitation with exercise capacity as the main outcome, and neither had posted results.[24][25]

Evidence level for how consistently these tests measure: emerging. The repeat-measurement studies are small method studies, and in two of them the authors license a clock they tested or are named co-inventors of one.

Evidence level for links with death and disease: emerging. The findings come from observational cohort studies, which show associations.

Guidelines and expert statements

In searches of PubMed and Europe PMC on October 8, 2026, we found no clinical practice guideline from a medical society that recommends biological age testing.

A 2025 expert consensus statement, written for research trials, listed DNA methylation and epigenetic clocks among 14 biomarkers of aging on which experts reached consensus.[26] Several members of that panel declare ties to companies, one as a paid consultant to a company focused on measuring and developing interventions for biological age.[26] The panel reached its judgments by voting, and it wrote that "all biomarkers were deemed precise and reliable, with lower agreement for telomere length".[26]

In a 2025 expert consensus process on biomarkers of aging for use in intervention studies, telomere length did not reach consensus, with moderate agreement of 51% to 69%.[26]

A 2026 narrative review of biological age in orthopedics states: "Molecular aging clocks are not currently recommended for routine care."[27]

Separately, in lung disease, a 2026 perspective describes a consensus statement on telomere length testing in lung transplant evaluation, and a 2023 European Respiratory Society statement on familial pulmonary fibrosis lists telomere length measurement among its questions; both concern testing for disease, not biological age.[28][29]

A 2024 paper by a consortium of researchers on biomarkers of aging states: "no consensus exists on how biomarkers of aging should be validated before their translation to the clinic."[30] The authors of that consortium paper declare patents, licenses, stock or company roles for many of the authors.[30]

Questions to ask a provider

Questions worth asking before you decide on a test:

• Which clock does the test report: Horvath, Hannum, PhenoAge, GrimAge, DunedinPACE or one the company built?
• What sample does the test use, and was this clock built and checked on that kind of sample?
• If the lab measured my sample twice, how far apart could the two results be?
• Does the result change with the time of day the sample is taken?
• How large a change between two tests is more than measurement variation?
• Has acting on this result been shown to improve health?
• Which lab runs the test?
• Who can see my data, and how long is it kept?

Talk to a licensed clinician about whether a biological age test makes sense for you.

Limits and possible harms

Researchers writing in 2025 said that no direct evidence yet documents harm from individual use of epigenetic clocks.[15] They also warned that results can be misread, and pointed to research on consumer genetic tests, where they say misunderstanding can lead to unwarranted anxiety, fatalism, false reassurance or disengagement from legitimate medical care.[15]

In searches of PubMed and Europe PMC on October 8, 2026, we found no published study that measured anxiety or distress in people who had received their own epigenetic age result.

One uncontrolled follow-up survey in a UK research project, in which 91 of 178 adults answered a year later, asked about personalized health reports that combined epigenetic clock estimates, polygenic risk scores and lifestyle feedback and recorded self-reported outcomes such as feeling healthier, and its authors say the contribution of epigenetic age cannot be isolated.[31]

A qualitative interview study of 17 cancer caregivers who were shown mock data for four commercially available stress biomarkers, one of them telomere length, reported in its abstract that the caregivers "generally accepted the biomarker data" and that for some people biomarker data may actually increase their subjective distress.[32] A survey of 425 adults aged 50 and older about a hypothetical epigenetic dementia risk test found that risk worry exceeded stigma.[33] A randomized comparison in 180 young adults of three ways to present cardiovascular risk, one of them "Fitness Age", assigned a low or high result based on self-reported lifestyle and found that high-risk results were perceived as less credible but more worrying.[34]

On October 8, 2026, ClinicalTrials.gov listed a completed randomized trial of personal telomere length disclosure in 263 women aged 50 to 65, with distress as the main outcome, and no results had been posted.[24]

Authors of a 2025 perspective wrote that excluded or missing probes could bias decisions based on a false negative, such as appearing deceptively younger, or a false positive, such as appearing older than the underlying biological state.[15] The same authors warned that adopting epigenetic clocks as clinical biomarkers could open the door to their use in justifying unnecessary, unproven, or even harmful interventions.[15]

If you have a medical emergency, call 911. Problems with medical devices, including imaging equipment, can be reported to FDA through MedWatch.

Provider directory

Find a Provider

Some facilities require a clinician's order. Requirements vary.