The short version
- Eleven measures of biological ageing run on the same 964 people in the Dunedin cohort showed low agreement with each other, indicating they may not measure the same aspects of ageing.
- Technical noise alone produced deviations of up to nine years between replicate runs of the same sample across six prominent epigenetic clocks.
- Retraining those clocks on principal components brought most replicate pairs to within 1.5 years, which shows the problem is fixable in research settings but is not fixed in a single consumer reading.
- Horvath's 2013 clock was built from 8,000 samples across 51 tissue types and 353 CpG sites, and GrimAge predicts time to death strongly in large validation cohorts, so the underlying science is sound even where the consumer product is not.
- A before-and-after biological age improvement of a few years after a supplement is well within the technical noise of the measurement and does not demonstrate an effect.
You spit in a tube, post it off, and six weeks later a dashboard tells you that you are 41.3 years old when your passport says 47. It is a wonderful feeling and an almost meaningless number. Here is why, from the same literature the tests are built on.
The science underneath is real
None of what follows is a claim that epigenetic clocks are fake. They are one of the more impressive things to come out of ageing research.
Horvath's 2013 multi-tissue predictor was built from 8,000 samples across 82 datasets and 51 healthy tissues and cell types, and estimates age from methylation at 353 CpG sites. It reads close to zero in embryonic stem cells, tracks cell passage number, and works across most human tissues — and in chimpanzees. Later clocks got more ambitious about outcomes rather than chronology: PhenoAge was trained on a composite of clinical measures to predict lifespan and healthspan, and GrimAge, built from methylation-based surrogates for seven plasma proteins plus smoking pack-years, predicts time to death with a Cox regression p-value on the order of 10 to the minus 75 in large validation data.
Those are serious research instruments. The question is what happens when you run them on one person.
Problem one: they are not measuring the same thing
Belsky and colleagues implemented eleven different quantifications of biological ageing — telomere length and erosion, three epigenetic clocks and their ticking rates, and three biomarker composites — in the same 964 middle-aged people from the Dunedin birth cohort, using repeated physiological and genomic data.
The result, in their own words, was contrary to expectation: low agreement between the different measures. Some were consistently related to physical functioning, cognitive decline and visible signs of ageing — the 71-CpG clock and the biomarker composites did best — but even for those, effect sizes were modest. Their conclusion is that the various proposed approaches may not be measuring the same aspects of ageing at all.
This is the single most important thing to know before buying a test. "Biological age" is not one quantity that different companies estimate with different precision. It is several different quantities wearing the same name.
Problem two: the noise floor is enormous
Higgins-Chen and colleagues examined technical reliability — what happens when you run the same sample twice.
Across six prominent epigenetic clocks, technical noise produced deviations of up to nine years between replicates. Not between people. Between two runs of the same biological material.
They also built the fix, retraining the clocks on principal components of the CpG-level data, which brought most replicate pairs to within 1.5 years. That is excellent methodology work and it is the reason the field can run clinical trials with these endpoints at all. It also tells you exactly how much confidence a single unreplicated consumer reading deserves.
Now apply that to the thing people actually do with these tests: take one, change something, take another six months later, and attribute the difference to the change. If your baseline reading carries several years of technical noise and so does the follow-up, a three-year "improvement" is well inside the range you would get from doing nothing at all and posting two tubes.
There is a disclosure worth noting here too, and the paper makes it: several authors have commercial relationships with a company selling epigenetic age products. The people most invested in these tests are also the ones publishing the reliability problem, which is to their credit and does not make the problem smaller.
What that means for anything sold alongside a test
This is where it stops being an academic point. A supplement company that shows you a biological age reduction after twelve weeks on its product has produced, at best, a number inside the noise band, from an unblinded sample of people who knew what they were taking, on a measure that does not agree with the other measures of the same construct.
We have written elsewhere about an NMN trial that reported a favourable blood biological age result from an online calculator while showing no significant difference in insulin resistance in the same participants. That pattern — soft endpoint moves, hard endpoint does not — is the one to watch for. It is the same reason we are unconvinced by senolytics marketing and by any GlyNAC claim built on hallmark scores rather than function.
The markers that do survive this scrutiny
Ask of any ageing marker: does it predict a hard outcome, is it reproducible on the same person, and does it move with something you can do?
Epigenetic clocks pass the first, fail the second badly in consumer form, and their response to intervention is largely unproven. VO2 max passes all three, which is why it, rather than a methylation score, is the number we would want you tracking. Grip strength and walking speed are free, repeatable, and predict function. A lipid panel, blood pressure and HbA1c predict the diseases that actually kill people and cost less than a spit kit.
What we would actually tell you
If you find the biology interesting and the money is genuinely spare, take the test — it is a curiosity, not a hazard. Just do not let the result change your behaviour, because a single reading is not precise enough to justify a change and a pair of readings is not precise enough to prove one worked.
If the money is not spare, spend it on things with a reproducible number attached. And be openly sceptical of anyone who sells you a supplement and the test that scores it. That is a closed loop with a commercial incentive at both ends, and we would say the same if the supplement were ours.
Good questions
Are biological age tests accurate?
Not at the individual level, in their consumer form. Technical noise alone produced deviations of up to nine years between two runs of the same sample across six prominent epigenetic clocks. Eleven different ageing measures applied to the same 964 people also agreed poorly with one another. The research versions are far more careful than the mail-order ones.
Why did two tests give me different biological ages?
Because they are probably not measuring the same thing. Telomere length, epigenetic clocks and biomarker composites were compared head to head in one cohort and showed low agreement, with the authors concluding they capture different aspects of ageing. Add several years of technical noise on top and two different results is the expected outcome, not an anomaly.
Can a supplement lower my biological age?
No supplement has convincingly demonstrated that, and the measurement is currently too noisy to support the claim. A few years of apparent improvement after twelve weeks sits inside the replicate error of the test itself. Be especially wary when the company selling the supplement also sells or scores the test.
Is it worth paying for an epigenetic age test?
As a curiosity, if the money is spare. As a decision tool, no. A single reading is not precise enough to justify changing anything, and a pair of readings is not precise enough to prove that a change worked. A lipid panel, blood pressure and HbA1c cost less and predict the diseases that actually shorten lives.
What should I track instead of a biological age score?
Markers that predict a hard outcome, reproduce on the same person, and move with effort. Cardiorespiratory fitness qualifies on all three. Grip strength and walking speed are free and repeatable. Standard bloods are cheap and interpretable by a doctor. None of them produce a satisfying single number, which is precisely why they are more honest.
Are epigenetic clocks useless then?
No, and we would not want that takeaway. They are strong research instruments that predict mortality convincingly in large cohorts, and reliability improves substantially with the principal-component retraining published in 2022. The gap is between what they can do across thousands of people and what a single tube can tell one person about themselves.
Sources
- Horvath S. DNA methylation age of human tissues and cell types. Genome Biol, 2013. View study
- Levine ME, Lu AT, Quach A, et al. An epigenetic biomarker of aging for lifespan and healthspan. Aging (Albany NY), 2018. View study
- Lu AT, Quach A, Wilson JG, et al. DNA methylation GrimAge strongly predicts lifespan and healthspan. Aging (Albany NY), 2019. View study
- Belsky DW, Moffitt TE, Cohen AA, et al. Eleven Telomere, Epigenetic Clock, and Biomarker-Composite Quantifications of Biological Aging: Do They Measure the Same Thing? Am J Epidemiol, 2018. View study
- Higgins-Chen AT, Thrush KL, Wang Y, et al. A computational solution for bolstering reliability of epigenetic clocks: Implications for clinical trials and longitudinal tracking. Nat Aging, 2022. View study
These statements have not been evaluated by the Food and Drug Administration. This product is not intended to diagnose, treat, cure, or prevent any disease.
