Biological Age Tests: What They Measure, Are They Worth It

A biological age test does not measure age. It reads chemical tags on your DNA at a few hundred specific positions and feeds them into a statistical model that returns a single number expressed in years. The science behind those models is real and has taught us a great deal about ageing. The consumer product built on top of it is not ready, and the reason is less about biology than about metrology: run the same blood sample twice and the answer can move by years, which means the cheerful "I reversed three years" after a good winter of training is often just the instrument talking to itself.

Methylation, without the jargon#

Every cell in your body carries the same genome, yet a liver cell and a neuron behave nothing alike. The difference lives in annotations layered on top of the sequence: which genes are switched on, which are muted, and how firmly. One of the most important annotations is a methyl group — a tiny chemical tag attached to a cytosine sitting next to a guanine. Those positions are called CpG sites, and the human genome has roughly 28 million of them.

The rough rule is that heavier methylation in a gene's regulatory region means a quieter gene. Methylation is not damage and not mutation: the letters of the genome are untouched. It is a volume control, and it moves — with cell type, with age, with smoking, with inflammation, and with which cells happened to end up in the tube of blood you sent off.

The observation that launched the field is that these changes are not random. As people get older, some CpG sites reliably gain tags and others reliably lose them, consistently enough that a model can look at the pattern and estimate how old someone is. That is all an epigenetic clock does: take a few hundred selected sites, weight them, output a number.

Which means the number is not a reading from a device that detects some underlying essence of ageing. It is the output of a regression model, and a regression model returns whatever it was trained to return. Everything that follows — why some clocks are more useful than others — comes down to what each one was asked to predict.

Three generations of clocks#

Generation one: predicting the birth certificate#

Two 2013 papers defined the field. Steve Horvath published a model built on 353 CpG sites and trained across dozens of tissues and cell types, which is why it works almost anywhere in the body rather than only in blood. Gregory Hannum's group published a blood-based model at nearly the same time.

Both were trained to predict chronological age, and both do it impressively well, typically within a few years. The commercial idea grew from the residual: if the model says 47 and your passport says 42, those five extra years are supposedly accelerated ageing. That interpretation is not baseless — the residual does show modest associations with mortality and disease.

But there is a built-in ceiling. The training procedure explicitly minimises the gap between prediction and chronological age. Anything that makes you differ from your age-matched peers is, from the model's perspective, prediction error to be squeezed out. A clock trained on chronological age is optimised against the very signal we want it to carry.

Читайте также: The Horvath Clock: What It Is and What the Test Can Tell You

Generation two: train on health instead#

The obvious fix is to stop using the birth certificate as the target. In 2018 Morgan Levine and colleagues built DNAm PhenoAge in two stages. First they combined clinical measures — albumin, creatinine, glucose, C-reactive protein, blood indices and others — with chronological age into a composite "phenotypic age" that captures mortality risk. Then they trained a methylation model to predict that composite. The result substantially outperformed earlier measures for all-cause mortality, cancers, healthspan, physical functioning and Alzheimer's disease.

In 2019 Ake Lu and colleagues took the idea further with GrimAge. They first used methylation to estimate plasma levels of seven ageing-related proteins, added a methylation-based estimate of smoking pack-years, and combined those surrogates into a predictor of lifespan. Its association with time to death, time to coronary heart disease and time to cancer stands out among epigenetic clocks — though the smoking component means that for smokers, a substantial part of what GrimAge "discovers" is something they already knew.

A different question: how fast, not how old#

DunedinPACE, published by Daniel Belsky and colleagues in 2022, answers something else entirely. It comes out of the Dunedin cohort of New Zealanders born in 1972–1973, followed for decades, in whom 19 indicators of organ-system integrity were measured repeatedly. That let the researchers estimate, for each individual, how quickly those systems were deteriorating — a personal rate of ageing measured over twenty years — and then compress that trajectory into a single blood test.

The output is a pace, not an age. Roughly 1.0 means one year of biological ageing per calendar year; 1.2 means twenty per cent faster; 0.85 means slower. Notably, the authors deliberately restricted the input data to exclude probes with poor test–retest reliability. The field's reliability problem was recognised from inside.

Clock	What it outputs	What it was trained on	Practical value to an individual
Horvath (2013)	Age in years, almost any tissue	Chronological age, 353 CpG sites	Scientifically foundational, weak basis for personal decisions
Hannum (2013)	Age in years from blood	Chronological age in adults	Accurate at guessing your birthday, limited health signal
PhenoAge (2018)	Phenotypic age in years	Composite of clinical markers reflecting mortality risk	Stronger link to outcomes; interpreting one person's number is still contested
GrimAge (2019)	Lifespan estimate in years	Surrogates for 7 plasma proteins plus smoking pack-years	Best mortality prediction; heavily influenced by smoking history
DunedinPACE (2022)	Rate of ageing, 1.0 is average	Decline in 19 organ-system indicators over 20 years	Conceptually closest to tracking change, but no intervention evidence

The reliability problem sits at the centre#

For a test to be useful to a population, it needs to correlate with outcomes across many people. For a test to be useful to you, it needs something stricter: measure the same material twice and get roughly the same answer. Without that, the difference between your two results is uninterpretable, because you cannot tell whether you changed or the assay did.

In 2022 Albert Higgins-Chen and colleagues addressed this directly in Nature Aging, and the phrasing in their own abstract is blunt: methylation data can be surprisingly unreliable, and technical noise produced deviations of up to nine years between replicates for six prominent epigenetic clocks. Replicates — essentially the same material run twice.

Hold that against the marketing. A test that promises to detect the three years you clawed back through sleep, exercise and diet is claiming a resolution finer than its own demonstrated error. The effect being sold is smaller than the noise floor.

The important part of that paper is constructive, not dismissive. The authors showed that computing principal components across CpG-level data before feeding the model largely fixes the problem: retrained principal-component versions of the same six clocks agree between most replicates within about 1.5 years. That is a different instrument, and it matters enormously for clinical trials of ageing interventions. But the method is recent and not every commercial service has adopted it, which is precisely why asking which version you are buying is a reasonable thing to do.

Technical noise is compounded by biological noise. The clocks read a mixture of white blood cells, and the proportions of that mixture shift with a cold, a bad week of sleep, a hard training block or seasonal allergy. Sample before a holiday and after it and part of what you have measured is your immune state, not your rate of ageing.

Why "I got younger" is a claim about the test#

Suppose the reliability issue were solved and the number were trustworthy. A deeper problem remains: nobody has shown what a change in it means.

Every clock described above was validated as a prognostic marker. People with worse scores die earlier and get sick more often, on average, in observational data. It does not follow that pushing the score down pushes the risk down. This is the classic surrogate endpoint trap: a marker can travel alongside a process without being a lever on it.

There is also a purely statistical route to self-deception: regression to the mean. People tend to retest after a bad first result, having also changed their habits. Extreme values drift back toward the middle on repeat measurement even when nothing whatsoever has changed — and the improvement is easy to credit to the new routine.

Finally, the number does something to you. Telling a healthy 44-year-old that they are biologically 52 is not neutral information. It may prompt useful changes, or it may launch a cascade of unnecessary scans, panels and supplements. That mechanism is familiar from screening in general: testing an asymptomatic person can generate harm entirely on its own.

Читайте также: Health Check-Ups by Age: Which Tests Are Worth It

Telomeres and the other contenders#

Telomeres are repetitive caps on chromosome ends that shorten as cells divide. It is an elegant image of a biological clock, and at population level the association with age and disease is genuine. For an individual it is close to useless, for three reasons. The spread of telomere length among healthy people of the same age dwarfs the age-related trend, so your value tells you little about you. Commercial assays reproduce poorly — the same disease as the epigenetic clocks, in a more advanced stage. And there is no lever: no intervention has been shown to improve health outcomes by lengthening telomeres, whatever the supplement aisle implies.

Blood biomarkers and composite indices. An older and more modest idea: take ordinary lab results — albumin, creatinine, CRP, glucose, blood counts, lung function — and collapse them into one biological age figure. This is, as noted, exactly how the clinical phenotype underlying PhenoAge was constructed. The approach is honest in that every input is cheap, interpretable and attached to an intervention with evidence behind it. Its weakness is that summing understandable numbers does not produce a more accurate number, and the tidy single score hides which input is actually off. If your CRP is elevated, the useful information is "your CRP is elevated", not "your biological age is 49".

What these tests cost#

After "does it work", the next question is what it costs — and here the market is wider than most people expect, because "biological age test" covers products that differ tenfold in price and considerably more in what they actually run.

What you're buying	Typical price	What actually happens
Consumer epigenetic clock, one-off kit	$200–500	Saliva or blood spot posted to a lab, run on a methylation array, several clock scores returned
Epigenetic clock on subscription	$100–200 per retest, $300–600 a year	The same test on a schedule — usually quarterly — with a dashboard and trend line
Telomere length test	$90–300	qPCR estimate of average telomere length in white blood cells
Blood-biomarker 'biological age' index	$150–350	An ordinary lab panel collapsed into a single age-like score
Longevity clinic panel including a clock	$500–3,000+	Broad bloodwork, sometimes imaging and fitness testing, with a clock bundled in
A standard clinical panel	$30–80, often covered by insurance or a GP	Lipids, HbA1c, CRP, blood count, blood pressure — the inputs with outcome evidence

Three things are worth knowing before reading those numbers as a quality ladder.

Price does not track accuracy. The expensive and the cheap epigenetic tests typically run the same commodity methylation array. What the higher price usually buys is a longer report, more clock variants, a nicer dashboard and a coaching call — not a more reliable measurement. The reliability problem described above is a property of the platform and the clock algorithm, and paying more does not fix it.

The subscription model deserves particular scepticism. Selling quarterly retests only makes sense if the thing being measured changes meaningfully over a quarter and the instrument can detect that change. Neither holds. Set the retest interval against the technical noise Higgins-Chen and colleagues documented — deviations of up to nine years between replicates of the same sample — and a three-month cadence is, in most cases, a paid subscription to measurement noise with a trend line drawn through it.

Bundled clinic panels hide the cheap part. In a $2,000 longevity workup, the components with the strongest evidence — lipids, HbA1c, blood pressure, a fitness assessment — are the ones that would have cost under a hundred anywhere else. The clock is the expensive garnish, and it is also the only item on the list without an intervention attached to it.

The comparison that settles it for most people is not "is $400 a lot of money" but "what else does $400 buy". A full lipid panel, HbA1c, CRP, a thyroid check and a blood count, in most countries, come to a fraction of a single epigenetic kit — and every one of those numbers is reproducible, interpretable, and attached to a treatment that has been shown to change outcomes. A VO2max test, or a year of the gym membership that would improve it, sits in the same price band and has considerably better evidence behind it.

None of which makes the science worthless. If you are curious, can afford it, and treat the result as an expensive novelty rather than a health decision, there is no harm in it. The harm starts when the number substitutes for the measurements that would actually have told you something.

What to measure instead#

The irony is that the measurements most tightly tied to how you are actually ageing are cheap, available and validated over decades.

Blood pressure. The best value in medicine: instant, free, and hypertension remains the leading modifiable risk factor for stroke and heart attack while producing no symptoms whatsoever.

Lipids and HbA1c. Two tests that, combined with age, blood pressure and smoking status, produce a ten-year cardiovascular risk estimate — a single number backed by decades of randomised evidence and connected to treatments that work.

Waist circumference. A crude but durable marker of visceral fat that the bathroom scale misses entirely. Requires a tape measure.

VO2max. Maximal oxygen uptake is among the strongest known predictors of all-cause mortality, and unlike a methylation score it is directly trainable. Outcome link and lever, both present.

Читайте также: VO2max and Lifespan: What the Evidence Shows

Grip strength. A hand dynamometer is inexpensive, and grip predicts mortality and later-life independence better than many laboratory panels. It responds to resistance training.

Gait speed. Walk four metres at your normal pace and time it. One of the most robust functional markers of ageing, and it requires a corridor.

Читайте также: Strength Training for Beginners: How to Start

Notice the property shared across that list. For each item we know not only that it tracks with outcomes, but what to do when the value is poor, and that doing it improves outcomes. Epigenetic clocks currently have only the first half of that pair.

There is also a middle path worth knowing about. The blood-chemistry version of PhenoAge — the formula that came before the methylation clock of the same name — runs on an ordinary panel with no sequencing at all. It gives you a biological age figure with a mortality validation behind it, and, unlike a consumer kit, it shows you which individual markers produced the number:

From blood workPhenoAge — the Levine formula, validated against mortalityFrom lifestyleNo blood test needed. This is not a biological age

Your age

years

Albumin

g/L ⇄

Creatinine

µmol/L ⇄

Fasting glucose

mmol/L ⇄

CRP (high-sensitivity)

mg/L ⇄

Lymphocytes

%

MCV — mean cell volume

fL

RDW — red cell distribution width

%

Alkaline phosphatase

U/L

White blood cells

×10³/µL

Your PhenoAge

38.96.1 years younger than your birth certificate · Chronological age: 45

Your markers are well above average for your age. That is a good result — but remember PhenoAge is a surrogate measure, not a guarantee.

What drives the gap

How far each marker moves the result compared with a healthy reference panel

• CRP (high-sensitivity)+0.7 yrs

Читайте также: Biological Age Test: Calculate PhenoAge From Blood Work

The verdict#

Epigenetic clocks are a real scientific achievement. They gave ageing research something it did not previously have: a quantitative, lifestyle-sensitive, outcome-linked measure that can be applied to a living person. In trials of geroprotective compounds, in epidemiology, in studying the biological imprint of childhood adversity or air pollution, they are among the most valuable tools available — and the Higgins-Chen work shows their reliability can be dramatically improved rather than merely lamented.

But there is a gap between "useful for studying groups" and "useful to you", and this technology has not yet crossed it. While rerunning one sample can shift the answer by years, while no evidence connects a shifted score to better health, and while the interpretation depends on which model version your provider happens to license, buying the test purchases an interesting fact rather than a medical decision.

The reasonable stance is to follow the field with genuine interest and to run your own health on measurements that carry both predictive power and a lever. Blood pressure, lipids, HbA1c, waist, fitness, strength and walking speed are duller than epigenetics. They are also the ones that currently answer the question people are really asking when they buy a biological age test: what should I do to age more slowly.