Biological Age Tests Explained: GrimAge, DunedinPACE and What They Can't Tell You

Key Takeaways

The sales page is always the same. A photograph of a kit. A spit tube or a cheek swab. A promise that in three weeks you will learn your real age, and that if you follow the programme and retest, you can watch it fall. The number arrives with a decimal point and a graph.

What the sales page does not tell you is that the field's own standards body wrote, in 2024, that these tools are not yet validated for use in individuals; that six of the best-known clocks differ by up to nine years when the same blood sample is run twice; that a cheek swab and a blood draw from the same person can disagree by decades; and that the number will be higher if you had the flu last month. None of this makes the science bad. The science is extraordinary. It makes the product something other than what it is sold as.

This guide explains what the clocks measure, which ones predict anything, where the noise comes from, and what to ask before you pay. If you want the intervention side first, it is in How to Actually Lower Your Biological Age, and the protocol that puts both together is here.

What a Methylation Clock Is

DNA does not change with age, but the chemical tags on it do. Methyl groups attach to cytosines at specific positions — CpG sites — and the pattern of which sites are methylated drifts in a remarkably consistent way across a lifetime. In 2013 two groups independently showed that a weighted sum of methylation at a few hundred such sites could predict calendar age to within a few years.

The multi-tissue clock used 353 sites and was trained on roughly 8,000 samples from 51 healthy tissue types. It read embryonic stem cells as age zero and cancer tissue as, on average, 36 years older than the patient. The blood clock used a different set of sites in 656 people aged 19 to 101, and found that the rate of ageing it measured was influenced by sex and by genetic variants. These are the first-generation clocks, and they established the field.

The key insight came next. If a clock says you are 55 and you are 50, the five-year gap — epigenetic age acceleration — might mean something. A 2015 meta-analysis of four older cohorts found that every five years of acceleration carried 21% higher mortality; a 2016 meta-analysis in 13,089 people across three ancestries confirmed that acceleration predicted death independently of age, smoking, education and all the usual risk factors.

The Generations, and Why They Differ

The difference between generations is not refinement; it is a change of target. A clock trained to predict calendar age is, at the limit, perfectly accurate and perfectly useless — it tells you what your passport already says. The second-generation clocks were trained to predict outcomes. GrimAge was built from methylation proxies for plasma proteins associated with mortality, plus a methylation signature of smoking pack-years, and its association with time to death in validation cohorts had a p-value with 75 zeros after the decimal point.

The practical consequence showed up in a 490-person Irish cohort followed for a decade, tested on nine clinical phenotypes. Horvath and Hannum acceleration predicted none of them. PhenoAge predicted four, and lost them after adjustment. GrimAge predicted eight of nine, and remained significant for walking speed, polypharmacy, frailty and death after full adjustment. In 9,537 Scots followed for 13 years, GrimAge acceleration predicted incident COPD with a hazard ratio of 2.2, diabetes 1.5 and heart disease 1.4.

DunedinPACE answers a different question. Instead of "how old do you look", it asks "how fast are you ageing right now", calibrated against two decades of biomarker change in a single New Zealand birth cohort. It reports a ratio — 1.0 is average — and it was built with low-reliability methylation probes deliberately excluded. It adds predictive information beyond GrimAge, and it is the clock that moved in the caloric-restriction trial when the others did not.

The Noise Nobody Puts on the Box

Here is the finding that should be on every product page. In 2022 a group ran the same blood samples repeatedly through the methylation array and computed six well-known clocks on each replicate. Technical noise alone produced deviations of up to nine years between runs of the same sample. The same group showed that retraining the clocks on principal components of the methylation data — averaging across many correlated sites instead of trusting individual ones — brought replicate agreement to within about 1.5 years, and made intervention effects and longitudinal trends detectable where they had been buried.

The noise has several sources, and each is a question to ask a vendor:

Which clock? First-generation clocks are the noisiest. A vendor that reports a Horvath age to one decimal place is reporting noise to one decimal place.

Which tissue? In 2025, 83 people aged 9 to 70 gave buccal, saliva, dried-blood-spot and venous-blood samples. For some clocks, average within-person differences between oral and blood estimates approached 30 years, and oral and blood values were poorly correlated even after correcting for cell composition. GrimAge2's authors report that it applies to saliva, and that is a genuine advance — but it does not make a cheek-swab result interchangeable with a blood result from a different clock.

Which pipeline? In 2,053 people with 146 technical replicate pairs, 41 methylation predictors were run through 101 different pre-processing pipelines. Most predictors reached excellent agreement only with appropriate normalisation, and pipeline choice materially changed downstream mortality associations. Two laboratories can return different ages from one sample.

Technical versus biological reliability. A 2026 study separated the two: technical reproducibility across 18 methylation biomarkers was mostly excellent, but biological reliability across repeat samples taken days apart under varying conditions — after meals, under stress — was substantially lower, and the clocks with poor biological reliability gave variable and sometimes misleading intervention responses. A clock can be perfectly repeatable on the same tube of blood and still tell you something different on Tuesday than it did on Friday.

Biological Age Moves With Your Health, Which Is the Point and the Problem

Biological age is supposed to respond to what happens to you; that is what makes it a biomarker. But it responds faster than most buyers assume. A 2023 study across mice and humans showed that epigenetic, transcriptomic and metabolomic ages rose sharply during major surgery, pregnancy and severe COVID-19, and returned toward baseline on recovery. In COVID-19 specifically, one study of 413 patients found progressive acceleration with severity that partly reversed in late recovery; another found no acceleration at all. The disagreement is itself informative: the clocks are sensitive to illness and to how it is measured.

The consequence for a consumer is simple. A single reading taken within weeks of an infection, an operation or a very bad month is inflated by an unknown amount. If you test, test when you are well, and never interpret one result.

The Measures That Are Not Methylation

Blood-chemistry age. Long before methylation clocks, biological age was computed from routine laboratory values. In 9,389 Americans followed for 18 years, the Klemera-Doubal method — a weighted combination of ordinary biomarkers — predicted mortality significantly better than chronological age, and once it was in the model, chronological age added nothing. The PhenoAge algorithm itself is a clinical-chemistry calculation (albumin, creatinine, glucose, CRP, white cell counts and a few more) before it is a methylation clock. The authors of the BioAge toolkit note that in multiple studies blood-chemistry measures have outperformed methylation measures for predicting healthy lifespan. They cost a fraction, and every input is a number your physician can act on. The Biohacker's Blood Panel covers the inputs.

Telomere length. Weakly correlated with epigenetic age, and a weaker mortality predictor: in two Scottish birth cohorts, a standard deviation of epigenetic age carried 22% higher mortality while a standard deviation of telomere length carried 11% lower, with the telomere result not reaching significance. The two are independent, and the methylation-based estimate of telomere length outperforms the measured value.

Proteomic organ clocks. The newest generation estimates the age of individual organs from thousands of plasma proteins. In 5,676 people, about one in five had at least one organ ageing markedly faster than the rest; accelerated heart ageing carried 250% higher heart-failure risk, and brain and vascular ageing predicted Alzheimer's as strongly as a leading blood biomarker. These are research tools for now, but they point to where the field is going: not one age, but a profile.

Glycan age. Patterns of sugar chains on IgG antibodies explain up to 58% of the variance in chronological age in 5,117 Europeans, more than telomere length does, and track inflammation. The assay is precise; much of the published validation comes from the company that sells it.

How to Read a Result, If You Buy One

1. Find out which clock it is. If the answer is Horvath or Hannum only, the number is too noisy to track. Ask for GrimAge2, PhenoAge in principal-component form, or DunedinPACE.
2. Blood beats saliva beats cheek. If you plan to retest, use the same tissue every time.
3. One lab, one platform, forever. Switching vendors resets your baseline.
4. Ignore the decimal. The noise floor for the better clocks is around 1.5 years; for the older ones, far more.
5. Do not test when ill, injured, pregnant or recovering. Six weeks clear.
6. Treat the first result as a baseline, not a verdict. Information lives in the trend across 12-month intervals, and only when the change exceeds the clock's known noise.
7. Run a clinical panel alongside. It is the cheaper, better-validated biological age, and it tells you what to do.

What the Field Says About Itself

The Biomarkers of Aging Consortium, formed by the researchers who built most of these tools, published a framework in Cell in 2023 noting "the current lack of standards and consensus on the properties of a reliable aging biomarker", followed by a validation roadmap in Nature Medicine and, in 2024, a survey of experts identifying six barriers to clinical translation. Among the core recommendations: validate biomarkers that are robust and responsive at the level of individuals, and link them to clinically actionable insight. Both are stated as requirements not yet met.

That is the honest position. The clocks are among the most important measurement tools gerontology has produced. They have transformed what can be learned from a cohort of ten thousand people. They are not yet instruments for managing one person — and the vendors selling them as such are ahead of the evidence.

Frequently Asked Questions

Can I reverse my biological age? You can change a clock reading. Whether that is the same thing is the open question, covered in the intervention review. The randomised evidence shows effects of months, not years.

Which test is the best value? A comprehensive clinical blood panel, interpreted as biological age with a free algorithm, outperforms most methylation clocks for mortality prediction and tells you what to change. If you want a methylation clock, DunedinPACE or a principal-component GrimAge, from blood, is the defensible choice.

My result says I am 12 years younger. Is that real? Possibly. It is also within the range that tissue choice, clock choice and pipeline can produce. Repeat it in a year, same everything, and see whether it holds.

My result says I am 12 years older. Should I worry? First confirm you were well when sampled, and that the sample was blood. Then run a clinical panel — the actionable risks will show up there. Do not change medication on a clock result.

How often should I test? Twelve months, if at all. The largest randomised supplement effect on a clock was three months of ageing over three years; a quarterly test cannot see that against the noise.

Where do I start? With the Biological Age Optimization Protocol, which puts the behaviours with evidence first and the testing rules second.

Related Research

• How to Actually Lower Your Biological Age: What the Trials Show
• The Biological Age Optimization Protocol: Test, Intervene, Retest
• The Biological Age Optimization Protocol
• The Epigenetic Clock: How Peptides and Interventions Change DNA Methylation Age
• The Biohacker's Blood Panel: 40 Biomarkers Worth Tracking