Your biological age can change within hours. The problem with epigenetic clocks

A meal, acute stress or a short-term environmental exposure can shift the output of some epigenetic clocks, even when the laboratory measurement itself is highly precise.

Your biological age can change within hours. The problem with epigenetic clocks

Table of contents

    An epigenetic clock does not measure age like a thermometer

    Biological age tests based on DNA methylation often reduce complex biology to one compelling number: 35 years, 42 years, 51 years. It is therefore tempting to treat that number as a relatively stable property of the body and compare repeated measurements in search of signs of accelerated aging or rejuvenation.

    A new analysis published in Aging Cell suggests that the output of an epigenetic clock may be much more sensitive to a person’s short-term physiological state than its laboratory precision would imply.

    The researchers examined 18 DNA methylation-based aging biomarkers. Most performed very well when exactly the same sample was measured again. The problem became more apparent when new samples were collected from the same person after events such as a meal, acute stress or a short-term environmental exposure.

    That distinction is important. A test can be technically highly reproducible while still being biologically unstable.

    This also does not mean that a person can literally age or become younger by several years within a few hours. A more cautious interpretation is that some clocks respond to transient physiological changes that alter the estimated biological age.

    Study details

    • Publication title: Biological Versus Technical Reliability of Epigenetic Clocks and Implications for Disease Prognosis and Intervention Response.
    • Authors: Raghav Sehgal, Daniel S. Borrus, John Gonzalez, Yaroslav Markov, Albert Higgins-Chen, and the Alzheimer’s Disease Neuroimaging Initiative.
    • Affiliation: Department of Psychiatry and Department of Pathology, Yale University School of Medicine, United States.
    • Publication: 2026, Aging Cell, volume 25, issue 8.
    • DOI: 10.1111/acel.70635.
    • PMID: 42525215.
    • PMCID: PMC13418614.
    • Full text: PubMed Central and PubMed.
    • Study type: a reliability and reproducibility analysis of DNA methylation-based aging biomarkers using data from multiple independent datasets.
    • Biomarkers analyzed: 18 epigenetic clocks covering chronological age, mortality-risk and pace-of-aging measures.
    • Short-term stability assessment: repeated samples were analyzed under conditions including a standardized meal, acute psychosocial stress, air pollution exposure and altitude change.
    • Samples in short-term analyses: after a meal, N=34 with four measurements during one day; acute stress, N=34 with four measurements during one day; altitude change, N=21 with four measurements across seven days; air pollution exposure, N=16 with three measurements across two days.
    • Primary finding: most clocks showed very strong technical reproducibility but substantially weaker biological stability between separate samples collected from the same person over short time intervals.
    • Funding: the publication lists support including a National Institute on Aging grant, Yale University, the Gruber Foundation and an Impetus Grant. Alzheimer’s Disease Neuroimaging Initiative data were also supported by additional public and private sources.
    • Conflicts of interest: Raghav Sehgal and Albert Higgins-Chen are co-inventors of Systems Age, which is covered by a patent application. The paper also reports consulting relationships with selected companies in longevity and diagnostics. The remaining authors reported no conflicts of interest.

    The study therefore does not primarily ask which clock captures the one true biological age most accurately. Instead, it asks whether these measures are stable enough to be used reliably in research, disease prediction and intervention studies.

    A meal or acute stress can shift the result

    The main problem did not appear to be the DNA methylation technology itself. It appeared when biology changed between blood draws.

    In the standardized meal experiment, samples were collected four times from 34 participants during a single day. A similar setup was used in an acute psychosocial stress experiment. In both cases, the stability of many clocks dropped substantially.

    The researchers assessed reliability using the intraclass correlation coefficient, or ICC. In simple terms, ICC helps distinguish how much of the observed variation comes from stable differences between people and how much comes from fluctuations between repeated measurements.

    Across the combined short-term exposure datasets:

    • most clocks showed only moderate to good biological reliability, while none reached the range the authors classified as excellent,
    • GrimAge, GrimAge II, DNAmEMRAge and DunedinPACE generally appeared toward the lower end of the reliability range,
    • principal-component-based clocks such as PCGrimAge and PCPhenoAge, together with SystemsAge, performed better in this respect, although they still did not match the stability seen in technical replicates,
    • the largest variability appeared after acute exposures, particularly after meals and stress,
    • altitude-related changes were relatively more stable than responses to exposures occurring over the course of a single day.

    This means that two different readings from the same epigenetic clock do not automatically indicate a real change in the underlying rate of aging. Part of the difference may reflect the person’s immediate physiological state.

    Technical precision is not the same as biological stability

    One of the most important aspects of the paper is the distinction between two concepts that are often treated as if they were the same.

    Technical reproducibility

    If the same DNA sample is analyzed multiple times, the result should be similar. On this measure, most of the clocks performed well.

    The authors examined, among other things:

    • 196 people with duplicate technical measurements on the EPIC 850K platform,
    • 128 people with duplicate measurements on the 450K platform,
    • the influence of a sample’s position on a laboratory slide,
    • different DNA extraction procedures.

    Under these conditions, many clocks showed very high technical reproducibility.

    Biological reproducibility

    The more difficult question is whether the result remains similar when a new sample is taken from the same person a few hours later.

    Here, performance was substantially weaker.

    Importantly, technical and biological reliability were essentially unrelated. Across the 18 clocks, the correlation between them was only about r=0.017.

    In other words, the fact that a laboratory can reproduce the result from the same sample with high precision does not guarantee that a newly collected sample from the same person will produce a similar biological age estimate.

    This matters especially for commercial longevity testing, where someone may test before changing their lifestyle and then repeat the test several months later, interpreting the difference as evidence that a diet, training plan or supplement protocol has worked.

    How to interpret a biological age result

    The study does not show that epigenetic clocks are useless. Some of these biomarkers have documented associations with mortality risk, disease outcomes and functional decline. What it does show is that a single result should not be treated as a perfectly precise reading of one objective biological age.

    The authors also performed additional analyses suggesting that reliability matters when these clocks are used to predict later outcomes or evaluate interventions.

    More reliable clocks produced more stable estimates of associations with later cognitive test performance. A similar pattern appeared when the researchers reanalyzed clock responses in a dietary intervention.

    In practice, several implications follow:

    • A small difference between two tests does not necessarily represent a real change in aging biology. Part of the difference may reflect short-term physiology or the natural variability of the biomarker.
    • Conditions at the time of sampling may matter. When results are compared over time, it is sensible to standardize factors such as time of day, fasting status and exposure to acute stress as much as possible.
    • A trend across repeated measurements may be more informative than a single before-and-after comparison. Repeated testing can reduce the risk of overinterpreting a temporary fluctuation.
    • Not all epigenetic clocks are equally stable. A biomarker’s popularity does not necessarily mean that it is the best tool for detecting small longitudinal changes.
    • A change in the score is not automatically proof of slower or faster biological aging. For such a conclusion to be convincing, the observed change should clearly exceed the usual variability of that specific biomarker.

    This is particularly relevant in consumer longevity testing. A result showing someone to be “three years younger” can look impressive, but without knowing the normal test-retest variability of that clock, it is difficult to determine how much biological meaning should be attached to the difference.

    Important limitations

    The study highlights a significant methodological problem, but it also has limitations of its own.

    Most importantly, the authors analyzed existing datasets from different experiments, rather than one large study designed from the start to compare all 18 clocks under identical conditions.

    Several additional points deserve attention:

    • The short-term exposure datasets were relatively small. The meal and stress experiments each included 34 participants, the altitude experiment 21, and the air pollution experiment 16.
    • The results do not mean that all epigenetic clocks are unreliable. Stability varied substantially between individual biomarkers.
    • Short-term variation is not necessarily just measurement noise. It may reflect genuine biological changes, for example in immune-cell composition or the body’s response to acute stress.
    • It remains unclear how much of this short-term variation represents meaningful biology and how much makes long-term aging harder to measure. This is a key question for future research.
    • The study evaluates reliability, not the existence of one objectively true biological age. Different clocks were developed for different purposes and may capture different aspects of aging biology.

    The most useful conclusion is therefore not that “epigenetic clocks do not work.” It is that their results should be interpreted in the context of each clock’s natural variability and the conditions under which the sample was collected.

    In a longevity field where a change of several biological years is sometimes presented as evidence that an intervention has worked, that is an important distinction.

    Sources