Biological Age Tests (Epigenetic Clocks): Are They Worth Taking?

The question 'is it worth it' comes in several versions. If you are interested in the science of ageing and want to see your result as an experiment, the test is a legal and safe way of collecting data about yourself (it uses saliva or a drop of blood). But if you hope that one number will tell you whether your lifestyle 'works', or help you choose supplements, exercise or therapy, the current state of knowledge does not justify that. Below we break the topic into parts: how such a clock is built, how successive generations of clocks differ, how much a result can fluctuate, what is known about changing the result with interventions, and what to look for if you decide to test anyway.

It also helps to separate two concepts that marketing blends into one. Biological age in the broad sense is any measure of how 'worn' the body is compared with peers: it could be aerobic fitness, grip strength, blood parameters or a biomarker panel. An epigenetic clock is only one such measure, based on one type of data, namely which stretches of DNA carry methyl groups. We cover this further in our entries on biological age clocks, epigenetic age acceleration and organ biological age.

DNA methylation is the chemical tagging of cytosines (usually at sites called CpG) that does not change the genetic code but influences how readily a gene is read. With age, the methylation pattern at thousands of sites changes predictably enough to be used as a statistical 'readout' of age. An epigenetic clock is simply a mathematical model (most often regularised regression) that predicts a number, interpreted as age, from the methylation level at selected CpG sites.

The first widely known multi-tissue clock was described by Steve Horvath in 2013. It was built on about 8,000 samples from 82 datasets covering 51 healthy tissues and cell types, and is based on 353 CpG sites. The author showed, among other things, that the resulting age is close to zero in embryonic and induced pluripotent stem cells, correlates with cell passage number in culture, and that age acceleration is largely heritable. In an analysis of more than 6,000 cancer samples, all 20 cancer types considered showed significant age acceleration, on average 36 years. These were strong arguments that the signal is biologically meaningful and not just a statistical curiosity.

What matters is what the clock actually measures. The first clocks were trained to reproduce calendar age as accurately as possible. Their output is therefore, by design, 'calendar age plus noise', and the interesting information lies in the difference: whether someone scores higher or lower than their birth date would predict (so-called epigenetic age acceleration). Whether that difference has clinical meaning depends on how the clock was built and what it was validated against, which we return to below.

The first generation (including Horvath's 2013 clock) predicts calendar age. The second generation, which includes PhenoAge and GrimAge, was trained on indicators linked to disease and mortality, so it predicts health events better than merely 'guessing the birth date'. The third generation, represented by algorithms such as DunedinPoAm and the later DunedinPACE, does not try to answer 'how old are you biologically' but 'how fast are you ageing right now'. For a patient that is a completely different question and so a different interpretation of the number.

GrimAge, described in 2019, is a composite indicator built from seven methylation-based estimators of plasma proteins (including plasminogen activator inhibitor 1, PAI-1, and growth differentiation factor 15, GDF-15) and a methylation-based estimate of smoking pack-years. In large validation datasets it predicted time to death significantly better than earlier clocks (Cox regression p-value on the order of 2.0E-75), as well as time to coronary heart disease and cancer. That is important information, but remember that part of GrimAge's predictive power comes from the fact that it explicitly includes a smoking signature, a factor the patient already knows about.

DunedinPoAm was built differently. Researchers measured 18 biomarkers of organ-system integrity in 954 members of the Dunedin cohort between ages 26 and 38 (12 years of follow-up), composed a 'pace of ageing' measure from the rate of change, and then trained an algorithm to predict it from blood methylation. Validation in several cohorts and in the CALERIE calorie-restriction trial gave the authors proof of principle that pace of ageing can be estimated from a single blood sample. The authors compare it to a speedometer, as opposed to the odometer that calendar age represents. The newer DunedinPACE develops that idea and is the one that appears in some commercial offers.

For a tool meant to track change over time, the key question is repeatability. If the same sample, measured twice, gives different results, you cannot tell real change from noise. This is where the promise of 'measure now, measure in six months, see whether you got younger' most often falls apart.

This result has two sides. The bad news: classic clocks, in the form usually deployed in commercial kits, can produce a result that is 'several years different' without any biological change. The good news: the problem has been recognised and methods that improve repeatability exist. From a consumer's perspective the conclusion is simple: before you trust a change of '2.1 years' between two measurements, ask the provider which clock is used, whether it is a version robust to technical noise, and what the documented difference between repeats of the same sample is.

On top of that sits a second, biological layer of variability. In an analysis of more than 4,000 samples it was shown that up to 53% of the variation in methylation age may be attributable to blood cell proportions, in particular the balance between naive and memory T cells. The influence of cell composition on age acceleration itself was weaker (up to 21% of variance explained), and after accounting for it the associations with disease risk weakened only moderately. The practical meaning: a cold, an infection or an acute inflammatory reaction just before sampling can shift blood composition, and therefore the result, with no relation to your 'true' pace of ageing.

The most common marketing promise is: improve your lifestyle and the test result will fall. Data on this are still sparse and mostly come from small studies, often unblinded and with short follow-up. It is worth looking at one of the most frequently cited, because it shows both the potential and the limitations of this approach.

What follows from this? First, the result is promising, but it is a small sample, men only, a short time and one clock, plus a very complex intervention, so it is unknown which element mattered. Second, what was measured was a change in a number, not whether someone actually gets ill less often or lives longer; improvement of a clock score is not proof of improved health. That distinction is at the core of the whole discussion about ageing biomarkers: for a clock to serve as a surrogate endpoint in trials (that is, to stand in for waiting years for diseases and deaths), it must be shown that its change under an intervention predicts improvement in real health events, and that evidence does not yet exist for epigenetic clocks.

It is reasonable to ask yourself: what will change in my decisions if the result comes out 'good', and what if it comes out 'bad'? For most people the answer is: nothing, because the most important actions are already known. Do not smoke, maintain a healthy body weight, exercise regularly (including strength and endurance training), sleep well, keep blood pressure, lipids and glucose under control and use screening tests appropriate for your age. None of these steps requires an epigenetic clock, and their effectiveness is far better documented.

There are situations where a test has a cognitive or research justification: someone is taking part in a scientific study, wants to collect data deliberately for a personal experiment (remembering its limitations), works in the longevity industry or simply likes data about themselves. That is perfectly fine as long as we do not confuse curiosity with diagnostics. On the other hand the psychological risk is real: a result 'eight years older than your ID' can cause needless anxiety, while 'eight years younger' can give false security and lead to neglecting tests with proven value.

On price and availability. Tests are offered mainly by foreign companies, often as a kit mailed to you for collecting saliva or capillary blood; they usually cost at least a few hundred zloty, and packages with several clocks and a report can be noticeably more expensive (prices and scope change, so we do not give specific amounts). It is worth checking which clock is used, whether the provider discloses measurement reliability, where genetic and methylation data are processed and what happens to the sample after analysis. Methylation data are a person's biological data and deserve the same caution as a genetic test result.

If you are interested in biological age rather than epigenetic curiosity, first consider measures with much stronger data on survival and that you can improve: aerobic fitness (VO2max), muscle strength (for example grip strength), blood pressure, HbA1c and lipid profile. In these areas there are clear thresholds, guidelines and interventions of proven effect, and their change in response to training or treatment is real and verifiable.

At this point an argument often appears: 'but clocks do predict mortality'. That is true, and it is the strongest argument for the scientific value of these tools. But predicting risk in cohorts of thousands of people is something different from usefulness for an individual patient who wants to know what to do. A good comparison is LDL cholesterol: we know that a high level raises risk, we know how to lower it, and each intervention has a proven effect on cardiovascular events. For epigenetic clocks the 'we know what to do' part is still open.

Epigenetic clocks are one of the more interesting achievements of ageing biology in recent years and will undoubtedly play a role in clinical trials of geroprotective interventions. The problem lies not in the science but in its premature transfer to consumer offers: a single number from one sample looks certain, but the data on repeatability, cell composition and the lack of validation for interventions say otherwise. If you have the budget and the curiosity, a test is an acceptable experiment; if the budget is limited, every zloty spent on measuring blood pressure, lipids, HbA1c, checking fitness and strength training will give more health than one number from an epigenetic laboratory.