Why two biological age tests give you two different answers

TL;DR

• Two tests disagreeing is the category behaving normally - different molecular inputs, different training targets, different reference populations, different noise.
• The training target decides what a clock is good at - a 2025 Nature Communications comparison of 14 clocks against 174 disease outcomes in 18,859 people put second-generation clocks well ahead of first.
• Technical noise was real and largely fixed - Nature Aging 2022 found 3–9 year deviations between replicate runs; principal-component versions bring most within about a year.
• Biological noise was not fixed - clocks move across ordinary days with no intervention, which limits what any pair of readings can show.
• "Years younger" depends on an undisclosed reference distribution - change the comparison population and the number changes without you changing.
• Do not average two disagreeing results - they are different constructs, and the mean of incomparable quantities has no interpretation.

Educational content, not medical advice. No test, provider or company is named, ranked or compared. No doses, schedules or protocols appear, and nothing here positions any intervention as a means of changing a result.

---

Two tests, two ages: the thing nobody warns you about

The experience is common enough to be predictable. Someone buys a biological age test, gets a number, finds it interesting, and later buys a second one - different provider, different method, sometimes months later, sometimes from the same blood draw. The second number is years away from the first.

The instinct at that point is binary: one of these is right and one is wrong, and the job is to work out which. That instinct is reasonable and it is also the wrong frame, because it assumes the two tests were measuring the same thing. They were not, and the ways in which they differ are systematic rather than random. Once you can see the structure, disagreement stops being a scandal and becomes exactly what you would expect.

This does not mean the field is fraudulent or that the underlying science is empty. It is not. Aging biology is producing genuinely important work, some of it replicated at very large scale. What it means is narrower and more useful: a single consumer result is a fragile object, and two consumer results disagreeing is the normal behaviour of the category, not a defect in your particular sample.

The rest of the library covers the adjacent questions. What a biological age test actually measures covers the families of test and what each one is built from. Epigenetic clocks covers the molecular detail of the methylation clocks specifically. Function as a biological-age marker covers the cheap, well-validated physical measures that predict outcomes as well as most clocks and get a fraction of the attention. Organ-specific aging clocks covers the newer proteomic tests that report separate ages per organ. "Fitness age" on a wearable covers the consumer-device score that is routinely mistaken for a biological-age result and is computed from something else entirely. This page owns one question only: where does the difference between two results come from, and what should you do about it.

Where a difference between two results actually comes from

It helps to stop thinking of "the difference" as one thing. It is a sum of several independent contributions, and they are not all the same kind of problem.

Almost every argument about "which test is accurate" collapses if the arguers have not first agreed which rows they are talking about. Rows one to three are the tests being different by design. Rows four and five are presentation choices that obscure the difference. Rows six to eight are the measurement being unstable, which is a different and more serious matter.

Different clocks were trained on different things and predict different outcomes

This is the deepest of the design differences and the one most worth understanding, because it explains disagreement that no amount of laboratory improvement will remove.

A biological age model is built by taking some biological measurement, fitting a statistical model against some target, and expressing the output in years. The unit is a presentation choice, not a discovery. Years are intuitive; nothing about the underlying model requires that its output be expressed in them.

What the model was fitted against - its training target - determines what it is good at, and the generations of epigenetic clock differ on precisely this point.

First-generation clocks were trained to predict chronological age. The model was rewarded for guessing your birthday correctly from your methylation pattern, and it does that impressively well. Which creates an awkward property: if a model's job was to output your actual age and it output something else, that deviation is partly meaningful biology and partly model error, and separating the two is not straightforward. The gap you are being sold as a finding is, in part, the residual of a model that missed.

Second-generation clocks changed the target. Instead of predicting age, they were trained against mortality and clinical-biomarker composites - rewarded for ranking people by how likely they were to die or become ill, not by how old they were. They predict health outcomes considerably better as a result. A 2025 comparison in Nature Communications evaluated fourteen epigenetic clocks against ten-year onset of 174 disease outcomes in 18,859 individuals and found second-generation clocks substantially outperforming first-generation ones for disease prediction - which is exactly what the training-target argument predicts.

Third-generation measures changed the question again, estimating the rate of multi-system physiological decline rather than a static level. A pace expressed as years per year is not the same kind of statement as an age expressed in years, and the two cannot be meaningfully subtracted from one another.

Clinical-biomarker composites take a different route entirely, combining routine lab values - albumin, creatinine, glucose, inflammatory markers - into an age-like index. They are cheap, interpretable, and built from measurements a clinician already knows how to act on. They also reflect current physiological state, which moves with acute illness and recent events.

Proteomic organ clocks are newer still, inferring separate ages for separate organs from organ-enriched plasma proteins. The biology behind them is real and has replicated at scale; the individual-level performance is the least characterised of any family.

So a model optimised to reproduce your birth certificate and a model optimised to rank mortality risk will disagree about you, and both can be behaving exactly as designed. The useful question is not which number is right. It is what was this particular model rewarded for getting right, and is that the thing I care about.

There is a further layer that reports rarely surface: the reference distribution. "You are 8 years younger" means your measurement sits where the model expects a person eight years younger to sit - in some population. Which population? The training cohort? A vendor's customer base, which is self-selected, health-interested and unrepresentative? A general-population reference from a different country? The answer materially changes the number, and it is frequently not stated. Two tests using identical methods but different reference populations will return different ages for the same person, and neither is lying.

Test-retest variability: what the same sample does on the same platform

This is where disagreement stops being about design and starts being about reliability. It is the part vendors answer least often, and it turns on a distinction almost never made in consumer-facing material.

Technical reliability is reproducibility across replicate runs of the same sample. Split one tube, run it twice, compare. This has been measured, and the results were uncomfortable enough to motivate a methodological fix. Work published in Nature Aging in 2022 showed that running the same sample twice through six widely used epigenetic clocks could produce deviations of roughly three to nine years between technical replicates - with wider deviations reported elsewhere. Reproducibility also varied with laboratory handling: slide position within the array, and DNA extraction protocol. That is variation caused by the lab, not by the person.

The field's response was direct and largely successful. Principal-component versions of the major clocks were developed to address exactly this, aggregating across correlated CpG sites rather than depending on individual noisy ones. PC versions bring most technical replicates to within about a year of each other. This is a genuine advance, and it has a practical consequence for a reader comparing two results: "this clock" and "the reliable version of this clock" are not the same measurement. Whether a given result came from a PC version or not materially changes how much weight it can bear, and consumer reports do not reliably say.

Biological reliability is a different and harder problem: stability across repeated collections from the same person over hours or days, during which nothing that would count as an intervention has happened. The person ate meals, slept badly, moved through a circadian cycle, had an ordinary week.

This is where the remaining problem sits. Recent work - including a 2025 preprint explicitly framed around biological versus technical reliability of epigenetic clocks - makes the point directly: most clocks now achieve excellent technical reliability, but their biological instability limits prognostic and interventional utility. Fixing the assay did not fix the measurement, because the residual variability was never coming from the laboratory.

The consequence, stated plainly:

If a measure can move by several years between a Tuesday and a Thursday with no intervention, then a several-year difference between two tests - or between two time points - cannot be interpreted as a finding about you.

That sentence is the practical core of this page. It is not an argument that the clocks are meaningless. It is an argument about what a single reading, or a pair of readings, in one person can support. Within-person short-term variability sets a floor on detectable change, and a claimed difference smaller than that floor cannot be distinguished from noise no matter how good the underlying science is.

Why "your biological age is 8 years younger" is a number, not a finding

Suppose the two tests agreed. Suppose both said you were eight years younger than your chronological age. What would that entitle you to conclude?

Less than it feels like, for three reasons that compound.

Population-level association is not individual-level prediction. Second-generation clocks show real, well-replicated associations with mortality and morbidity across large cohorts. People whose clock reads older than their chronological age have, on average, worse subsequent outcomes. The word doing the work is average. A marker can show a robust, highly significant association across thousands of people and still discriminate weakly for any single person, because the outcome distributions for "reads old" and "reads young" overlap heavily even when their means differ reliably. This is standard statistics rather than a criticism unique to aging clocks - the same applies to cholesterol and blood pressure. The difference is that clinical medicine generally handles it explicitly, with absolute risk framing and decision thresholds, while consumer biological age reporting typically presents a population-level association as a personal fact. The general discipline is covered in what a study actually measured, and applied to routine bloodwork in optimal versus normal lab ranges.

Precision is not accuracy. A result reported to one decimal place is not more informative for being reported to one decimal place. Point estimates for one person carry uncertainty that is usually neither displayed nor propagated into the report - a criticism made directly in a 2025 npj Aging discussion of whether the field needs aging clocks at all, which identified ignored prediction uncertainty as a core structural weakness. A number with no error bar invites a confidence the underlying model does not have.

A gap is defined relative to a reference, and the reference is a choice. "Eight years younger" is not a property of your body. It is a statement about where your measurement sits in a distribution someone else selected. Change the reference population and the number changes without anything about you changing at all.

And there is a fourth point which is uncomfortable but honest: the science underneath this is genuinely contested, and this page is not going to resolve it in either direction. There are serious researchers who regard aging clocks as among the most important measurement advances in the field, and serious researchers who argue that the constructs are underdefined, the clinical validation inconsistent, and the uncertainty systematically unreported. Both camps are engaging with the same data. Anyone telling you the question is settled - in either direction - is telling you something the literature does not currently support.

What these tests are genuinely useful for

Scepticism about consumer inference is not the same as dismissal, and it would be dishonest to leave it there.

In research, they are doing real work. Large samples and control groups absorb the individual-level noise that makes a single consumer result fragile. Clock-based measures have produced replicated population findings about aging, exposure and disease risk that would have been hard to get any other way. Nothing in this article is an argument against that.

As a prompt to look harder at inputs, they can earn their place - particularly with clinical-biomarker composites, where you can see which input moved. If an index is being driven by an abnormal inflammatory marker or a glucose value, that is a visible, interpretable, clinically actionable finding, and it is a finding about the input rather than about the composite. A test that sends you to look at a lab value you were ignoring has done something useful, even if the headline number has not.

As an entry point to thinking about aging measurement, they are effective, and a person who buys one and ends up understanding the difference between association and prediction has got value for money in an unexpected form.

What they are not currently validated for is the use most people buy them for: tracking whether something you did worked. Demonstrating that requires showing not only that the marker correlates with outcomes, but that a change in the marker corresponds to a change in an actual clinical outcome - surrogate validation, not correlation. That work is largely not done, no aging clock is currently accepted as a validated surrogate endpoint, and researchers in the field say plainly that the question remains open. A number that moves after three months of effort is, on current evidence, at least as likely to reflect measurement variation, an acute physiological state, or regression to the mean as it is to reflect a changed trajectory. What the randomised evidence actually shows on that specific question is covered in can anything actually move an epigenetic clock.

It also follows - and this page states it deliberately - that a lower number is not itself an outcome worth pursuing. The things with genuine outcome evidence behind them are unglamorous and thoroughly familiar, and they are worth doing on their own merits rather than as a strategy for improving a score. Treating the score as the goal inverts the relationship between measurement and health.

Why a result is hard to interpret - and what to do with one you already have

This site does not name, rank, review or recommend any commercial testing service, and does not distinguish between providers. What it can usefully do is describe the properties of the category, because those properties - not the identity of whoever produced the number - are what set the limits on what any single result can carry.

Test-retest variability is usually unreported. How far a result moves when the same person is measured twice - technically, across replicate runs, and biologically, across days - is the figure that sets the floor on interpretation, and it is rarely stated anywhere on a consumer report. Its absence does not mean it is small. It means the reader has no way of knowing.

Clocks were trained against different targets, and the report rarely says which. Chronological age, mortality, phenotypic composites, rate of decline: four different questions, and a result is an answer to whichever one its model was fitted against. The related detail - whether a methylation result came from a principal-component version of a clock, which is roughly the difference between reproducibility within a year and within several - is similarly not something a report reliably discloses.

The reference distribution is a choice, and it is usually undisclosed. "Years younger" is a statement about where a measurement sits inside some population. Change the population and the number changes without anything about the person changing.

And a single number carries no action. Even a result that is internally sound is not a diagnosis, is not an individual risk figure, and does not arrive attached to a next step. That is not a fault in any particular result - it is what the category currently supports, and it is why the useful destination for a result is a clinician who can put it beside your history rather than a spreadsheet where it becomes a trend line.

If you already have two results that disagree, the sequence is short.

Do not average them. They are different measurements of different constructs against different references. The mean of two incomparable quantities is not a better estimate; it is a third number with no interpretation.

Work out which rows of the table above apply. Different molecular input? Different training target? Different reference population? One a PC version and one not? Different sample date, with something happening physiologically on one of them? Most disagreements are substantially explained by two or three of these, and the exercise is genuinely clarifying.

Consider what was going on when each sample was taken. Acute illness, an inflammatory flare, a recent hard training block, dehydration, poor sleep, a new medication. These move the underlying biology, and the report does not record them.

Do not treat either number as a diagnosis, and do not act on either alone. No aging clock is a diagnostic test, none was built or authorised as one, and no result described here can be converted into an individual risk figure.

If something concerns you, take it to a clinician - and take the actual reports. The useful move is almost always conventional: established tests appropriate to the concern, interpreted against your history, symptoms, medications and family history. Conventional assessment is far better validated than any aging clock, and a clock result is something that should prompt that assessment rather than replace it. If the conventional workup is clean, that is meaningful information about the clock result - not the other way round.

Consider whether the money would tell you more somewhere else. If the goal is information that changes decisions, a standard blood panel, blood pressure, body composition and a measured VO2max or grip strength will usually tell you more, cost less, and be interpretable by a clinician. Those measures are covered in function as a biological-age marker and biomarker monitoring.

The reasonable posture, given where the evidence actually stands, is curiosity without urgency. Two tests disagreeing is not a crisis and it is not a scandal. It is the category behaving normally, and knowing why is worth more than either number.

Considerations for longevity, rest and energy peptides

Because two biological age tests can disagree by years for reasons that have nothing to do with you, a clock result is a poor scorecard for judging whether a peptide or supplement did anything. That matters for the compounds below, which come up often in longevity, sleep and energy conversations. None has been shown to change a biological age result, and none is presented here as a way to do so. If you want to explore one, Vivre Labs offers access, and we recommend starting with baseline bloodwork.

• MOTS-c. A small peptide encoded in mitochondrial DNA, studied for its effects on metabolism. The evidence is cell and animal work; there is no completed interventional human trial. It is investigational and not FDA-approved. An FDA advisory committee (PCAC) recommended it for the 503A bulks list in July 2026; that vote is advisory only, and it has not been added to the list. See the MOTS-c page.
• Epithalon (epitalon). Claims about telomerase and epigenetic age rest mostly on small Russian studies and animal work. It is investigational; PCAC recommended epitalon in July 2026, again advisory only and not yet acted on. See the epithalon page.
• DSIP (delta sleep-inducing peptide). Discussed for sleep, but the human sleep data are small, old and inconsistent. It is investigational, and PCAC voted 6-7 against recommending it (as emideltide) for the 503A list in July 2026. See the DSIP page.
• SS-31 (elamipretide). FDA-approved in September 2025 only for Barth syndrome, a rare inherited mitochondrial disorder. Any other use, including for energy or ageing, is off-label. See the SS-31 page.
• NAD+. A coenzyme involved in cellular energy metabolism. Small human studies show supplementation raises blood NAD+ levels, which is not evidence of a change in biological age or in any ageing outcome. See the NAD+ page.
• The cellular energy and longevity protocol. Combines SS-31, MOTS-c and epithalon, and carries the evidence limits of each component. Details are on the cellular energy and longevity protocol page.

Measure against something stable. The practical lesson of this page is that a biological age result can move for reasons unrelated to anything you did, so it should not be the scorecard. If you and a physician decide to try any of these, anchor the decision to a stable, conventional baseline biomarker panel taken before you start and repeated afterwards, read by the physician against your history, symptoms and medications. Those markers have established clinical meaning and known variability, which no aging clock yet has.

Explore it with verified supply. Every Vivre Labs lot is independently tested and the certificate of analysis is published, so you know exactly what is in the vial.

Frequently asked questions

Which biological age test is the most accurate? "Accurate" needs unpacking, because the tests answer different questions and no single answer wins across all of them. For predicting the birthday you already know, first-generation methylation clocks are impressively precise and correspondingly uninformative. For association with health outcomes across populations, second-generation clocks trained against mortality and clinical composites outperform first-generation ones. For estimating rate rather than level, pace-of-aging measures are conceptually better suited. For technical reproducibility, principal-component versions of the major clocks are clearly superior. And for actually predicting mortality and disability, functional measures - VO2max, grip strength, gait speed - hold up as well as anything at a fraction of the cost. Asking which is "most accurate" without specifying the criterion produces an answer that cannot be right.

Should I average two disagreeing results? No. Averaging assumes the two numbers are noisy estimates of the same underlying quantity, and in most cases they are not - they are different constructs, from different molecular inputs, trained against different targets, compared against different reference populations. The average of two incomparable quantities is a third number with no interpretation. The more useful exercise is decomposing where the difference came from.

Does a big gap between two tests mean one of them is faulty? Usually not. Most of a gap is explained by design differences that are working as intended, plus a component of measurement variability that has been quantified for methylation clocks and is not trivial. A gap of several years between two different methods is within the range you would expect from the published reliability and design literature. That is a statement about how much weight a single result can bear, not an accusation about any particular test.

Are these tests worth buying at all? That depends entirely on what you expect from them. As a curiosity, or as a prompt to look harder at inputs you were ignoring, one may be. As a measurement you intend to act on, or to repeat after an intervention to see whether the intervention worked, the reliability and validation picture says the money is buying a number that cannot support that use. The honest position is that the underlying science is real, actively contested, and moving - and that consumer inference is currently running well ahead of it.

Related reading

• What a biological age test actually measures - the families of test and what each was built from and trained against
• Epigenetic clocks - the molecular detail behind the methylation-based measures
• Function as a biological-age marker - VO2max, grip strength and gait speed, and how they compare on evidence
• Organ-specific aging clocks - the proteomic tests that report separate ages per organ
• Telomere length testing - a third measurement family, and why its inter-lab variability is larger than most buyers expect
• Can anything actually move an epigenetic clock? - what the randomised evidence shows about change over time
• What a study actually measured - the methods-literacy companion to this page
• Optimal versus normal lab ranges · Biomarker monitoring: what to track and why
• Part of Testing & Biological Age: The Measurement Guide

---

Educational content, not medical advice. No commercial biological age test, provider, laboratory or testing service is named, recommended, ranked or compared anywhere on this page, and nothing here should be read as an assessment of any specific offering. Biological age estimates are research-derived measures, not diagnostic tests; none is currently a validated surrogate endpoint for an individual's response to an intervention, and no result described here can be converted into an individual risk figure. No compound, supplement or protocol is positioned as a means of altering a result, and a lower result is not presented as a health outcome. Any test result that concerns you should be discussed with your own clinician alongside conventional assessment. Claims for the Malaysian market are subject to MAB requirements.

References

• Horvath and Hannum first-generation DNA methylation clocks - trained to predict chronological age. · PhenoAge and GrimAge second-generation clocks - trained against mortality and clinical-biomarker composites. · DunedinPACE - third-generation pace-of-aging measure estimating rate rather than level.
• Higgins-Chen AT, et al. "A computational solution for bolstering reliability of epigenetic clocks: implications for clinical trials and longitudinal tracking." Nature Aging, 2022. (Technical replicate deviations of approximately 3–9 years across six major epigenetic clocks; principal-component versions bringing most technical replicates to within roughly 0–1.5 years.)
• References on CpG-level reliability and on slide-position and DNA-extraction effects on methylation clock reproducibility, including reported technical-replicate deviations approaching a decade.
• bioRxiv (2025) - Biological versus Technical Reliability of Epigenetic Clocks: most clocks achieve excellent technical reliability, but biological instability limits prognostic and interventional utility. Preprint; not peer-reviewed at time of writing.
• "An unbiased comparison of 14 epigenetic clocks in relation to 174 incident disease outcomes." Nature Communications, 2025. (18,859 individuals; ten-year onset of 174 disease outcomes; second-generation clocks outperformed first-generation clocks for disease prediction.)
• "Do we actually need aging clocks?" npj Aging, 2025. (Critique of abstract construct definitions, inconsistent clinical validation, and unreported prediction uncertainty; surrogate-endpoint status not established.)
• Oh HS, Wyss-Coray T, et al. "Organ aging signatures in the plasma proteome track health and disease." Nature, 2023;624(7990):164–172, and subsequent UK Biobank replication in Nature Medicine, 2025 - proteomic organ-specific clocks as a distinct family with the least-characterised individual-level performance.
• Epistemic-uncertainty literature challenging the reliability of aging clocks for detecting rejuvenation effects in individuals.
• General methodological references on the distinction between population-level association and individual-level prediction, on surrogate endpoint validation, and on reference-distribution dependence of standardised scores.
• General references on VO2max, grip strength and gait speed as predictors of mortality and disability.