Biological age is an estimate of how much age-related change your body has accumulated, expressed in years so you can compare it against the calendar. The idea behind it is sound and well evidenced: two 50-year-olds can carry very different amounts of accumulated damage, and that difference predicts disease and death better than a birth date does.
The part that consumer marketing skips is what kind of thing the number is. Biological age is not measured the way blood glucose is measured. It is predicted by a statistical model that reads a set of biological markers and outputs a number trained to match some target — sometimes your calendar age, sometimes a panel of clinical biomarkers, sometimes time to death. Change the training target and the number changes, on the same sample, on the same day. That single fact explains almost every confusing thing about the category.
The Verdict
What a methylation clock actually reads
DNA methylation is a chemical modification: a methyl group attached to a cytosine base, usually where a cytosine sits next to a guanine — a CpG site. There are tens of millions of these sites in the human genome, and their methylation status influences whether nearby genes are switched on or off.
Methylation at certain sites changes with age in a sufficiently regular pattern that a model trained on thousands of samples can predict age from a few hundred to a few thousand of them. That is the whole mechanism. The lab measures methylation at the selected sites, feeds the values into a trained model, and the model returns a number. Nothing in the process examines your organs, your fitness, or your disease risk directly. It compares your marker pattern to patterns the model has seen before.
This has an important consequence. The clock inherits every property of its training data. If it was trained on blood samples, it is a blood clock. If the training population was predominantly of one ancestry or one age band, its accuracy outside that band is weaker. And if it was trained to predict calendar age, it was explicitly optimized to disregard the health-related variation that a buyer is paying to detect.
Three generations of clocks, answering three different questions
"DNA methylation age" names a family of algorithms rather than one test. The generation matters more than the brand printed on the kit.
| Clock | Generation | Trained to predict | Output shape | What that means in practice |
|---|---|---|---|---|
| Horvath (2013) | First generation | Chronological age across many tissues | Reports an age in years | Trained to guess your calendar age, so it is designed to ignore much of the health variation you want to detect |
| Hannum (2013) | First generation | Chronological age in blood | Reports an age in years | Blood-specific, same structural limitation as Horvath |
| PhenoAge (2018) | Second generation | Nine clinical biomarkers plus age | Reports an age in years | Moves with metabolic and inflammatory status, which makes it responsive but also volatile |
| GrimAge (2019) | Second generation | Time to death, smoking pack-years, plasma proteins | Reports an age in years | The strongest common clock for mortality prediction; heavily weighted by smoking exposure |
| DunedinPACE (2022) | Pace of aging | Rate of change across 19 organ-system markers | Reports a speed, where 1.0 is average | Answers a different question — how fast, not how old — and is better suited to tracking an intervention |
The counterintuitive result sits in the first row. A clock that guesses your calendar age very accurately is, by construction, a poor progress tracker — accuracy at predicting your birth date means insensitivity to how healthy you are. Second-generation clocks were built to fix exactly that by training against clinical biomarkers and mortality instead. DunedinPACE went further and changed the output type altogether, reporting a rate rather than an age.
The four consumer methodologies compared
Methylation is the most common approach but not the only one. A number from any one of these cannot be compared against a number from another.
| Method | Sample | Typical price | What it reads | Evidence status |
|---|---|---|---|---|
| DNA methylation clocks | Blood or saliva | $200–$500 | Methyl tags at CpG sites across the genome | Strongest research base of the four. Second-generation clocks predict mortality better than chronological age alone. |
| Blood-panel composites | Standard blood draw | $0–$600 | An algorithm over routine labs: HbA1c, CRP, albumin, creatinine, white cell count and similar | Transparent and directly actionable. The number depends entirely on which formula the vendor selected. |
| Telomere length | Blood | $100–$300 | Average length of chromosome-end caps | Weak at the individual level. Measurement error between draws can exceed a decade of real change. |
| Glycan age | Blood | $300–$500 | Sugar structures attached to circulating IgG antibodies | Tracks chronic inflammation. Smallest published dataset of the four. |
| Organ-specific proteomic age | Blood | Research settings mostly | Plasma protein signatures attributed to individual organs | Genuinely promising and still emerging. Not a settled consumer product. |
Telomere testing warrants a specific caution. Telomere length varies substantially between cell types and between blood draws taken days apart, and published measurement error on consumer assays can exceed the difference between a 40-year-old and a 55-year-old. It is the cheapest entry point to the category and the hardest result to act on.
What actually determines biological age
Biological age is set by cumulative damage and the body's capacity to repair it — not by any single system. Twin and family studies put the heritable contribution to lifespan at roughly 20–30%, which means the large majority of the variance is environmental and behavioural. That split is the reason the number is worth measuring at all: most of what sets it is, in principle, modifiable.
Four inputs carry most of the weight, and they are the same four that dominate the outcome literature. Metabolic status — insulin resistance, visceral fat, and glycaemic control — feeds directly into every composite and second-generation clock. Chronic inflammatory burden raises hs-CRP and IL-6, which are inputs to several algorithms and drivers of glycan-based measures. Cumulative exposures, of which smoking is by far the largest single contributor to GrimAge. And cardiorespiratory fitness and muscle mass, which act less on the clocks directly and more on the outcomes the clocks were trained to predict.
What determines the number on your report is a narrower question, and the difference matters. Your report reflects those four drivers filtered through whichever algorithm the vendor chose, plus the state of your body in the week you gave the sample. That second component is not small — which is what the next section is about.
What moves your number that has nothing to do with aging
Before treating any change as real, rule out the boring explanations. Ranked roughly by how much each can distort a result.
| Confounder | Size of effect | Mechanism | How to control it |
|---|---|---|---|
| Blood cell composition | Large | Methylation is read from a mixture of cell types. A shift in the proportion of neutrophils to lymphocytes changes the readout on its own. | Prefer a test that reports immune cell deconvolution alongside the age |
| Recent infection or vaccination | Moderate to large | Acute immune activation shifts both cell proportions and inflammatory markers | Wait 2–4 weeks after illness or vaccination |
| Which clock the lab ran | Very large | Horvath and PhenoAge answer different questions and routinely differ by years on the same sample | Compare only like against like |
| Vendor and array version | Large | Different platforms, normalization pipelines, and reference sets produce non-comparable numbers | Never compare a result from one company against another |
| Sample handling and transit | Moderate | Collection quality and time to processing affect the assay | Post early in the week, follow the kit instructions exactly |
| Time of day and fasting state | Small to moderate | Affects the biomarkers feeding composite scores more than the methylation ones | Match conditions between baseline and retest |
| Acute stress, sleep debt, hard training | Moderate | Raises inflammatory markers that feed second-generation clocks | Do not test during an unusual week |
The cross-vendor row is the one people most often ignore. Two companies running different arrays, different normalization pipelines, and different reference populations do not produce comparable numbers, even if both are running a clock with the same published name. If you intend to track change over years, pick one vendor and stay with it. Switching vendors resets your baseline entirely.
What the evidence supports, and what marketing adds
Three claims are worth separating, because they carry very different weights of evidence.
- Well supported. Methylation-based estimates predict mortality and disease incidence across large populations better than chronological age alone, particularly the second-generation clocks. This has been replicated in multiple cohorts.
- Partly supported. Health behaviors associate with clock readings in observational data — smoking, obesity, poor metabolic control and heavy alcohol use all track with older-looking readings. Association is not the same as demonstrated causation, and observational studies cannot rule out that a third factor drives both.
- Not established. That deliberately lowering your clock reading causes better health outcomes. This is the claim most implied by consumer marketing and the one with the least support. The clock is a surrogate, and a surrogate that predicts an outcome does not automatically become a treatment target — moving the marker and improving the outcome are separate propositions, and the trials to prove the second have not been done.
Randomized evidence that any intervention shifts a clock reading is thin: the studies that exist are generally small, short, and use varied clocks, which makes them hard to pool. The most rigorous is CALERIE, a randomized trial of sustained caloric restriction, which found a small but statistically real slowing of DunedinPACE over two years. A few percent of pace, over two supervised years, is the correct scale for expectations.
How to use a result without over-reading it
- Take two baselines, not one. Sample twice, four to six weeks apart, under matched conditions, and average them. A single first reading is usually taken during a curious, stressed, or unwell week and is the most common source of a fake improvement later.
- Fix the conditions. Same vendor, same time of day, same fasting state, and no testing within two to four weeks of illness, vaccination, a hard training block, or a stretch of poor sleep.
- Retest no more than once every 9–12 months. Shorter intervals mostly report measurement noise and the random state of your week.
- Read the markers underneath. Fasting insulin, HbA1c, ApoB, hs-CRP, blood pressure, and VO2 max each name a problem and a fix. The composite score does not.
- Ignore the reversal arithmetic. Claims of a decade erased almost always compare a poor-conditions baseline against a good-conditions retest.
- Take the conventional markers to a clinician, not the headline age. No major guideline incorporates biological age testing into screening or treatment decisions, so there is no protocol to apply to the number itself. Our guide to reading a biological age report works through the sections in the order that makes them interpretable, including how large a gap has to be before it is worth raising.
Safety notes
A biological age result should never delay a clinical evaluation. It does not screen for cancer, cardiovascular disease, or any specific condition, and a favorable number carries no reassurance about symptoms you actually have. Be cautious with interventions marketed on the strength of moving this number — high-dose supplement stacks, off-label senolytics, plasma exchange protocols, and prescription drugs used outside a trial or specialist supervision all carry real risk and lack outcome evidence for this use. Anyone with a chronic condition, on medication, pregnant, or under active treatment should involve a physician before changing anything on the basis of a score.
Next
- How to lower biological age — what the evidence supports
- Elysium Index review
- TruDiagnostic review
- Biological age hub — the four methodologies compared
Related
- Best biological age test — five providers on one rubric
- Biomarker guides — the underlying labs a biological age score summarizes
- Normal vs optimal ranges
- Supplement guides — what the evidence supports and what it does not
Frequently Asked Questions
What is biological age, in plain terms?
Biological age is an estimate of how much age-related change your body has accumulated, expressed in years so it can be compared against the calendar. It is not a measured quantity like your height or your blood glucose. It is the output of a statistical model that reads a set of biological markers and predicts something — sometimes your calendar age, sometimes your clinical health, sometimes your mortality risk. Different models trained on different targets will disagree about the same person, and that disagreement is a property of the models, not a laboratory error.
How is biological age actually measured?
Most consumer tests read DNA methylation, the pattern of methyl groups attached to cytosine bases at specific CpG sites across the genome. That pattern changes with age in a regular enough way that a model trained on thousands of samples can predict age from it. The sample is usually blood or saliva, processed on an array or by targeted sequencing. Other approaches exist: composites computed from a routine blood panel, telomere length, glycan profiles on antibodies, and research-stage organ-specific protein signatures. They are not interchangeable.
Why do two tests give me different biological ages?
Because they are answering different questions. A first-generation clock like Horvath was trained to predict your calendar age, so it is built to be accurate at that and correspondingly insensitive to health differences. PhenoAge was trained against clinical biomarkers, and GrimAge against time to death, so both respond to health status instead. Layer on differences in laboratory platform, normalization pipeline, and reference population, and a spread of six to ten years between vendors is unremarkable. Neither lab made a mistake.
How accurate is biological age testing?
Useful at the population level, noisy at the individual level. Second-generation clocks predict mortality across large cohorts better than chronological age does, which is a real and replicated finding. On a single person, test–retest variation on the order of a few years is common, which is roughly the size of the change a year of serious effort might produce. Principal-component versions of the standard clocks were developed specifically to reduce that noise. This gap between population validity and individual precision is the central practical limitation of the category.
What is the difference between an age clock and a pace-of-aging measure?
An age clock estimates a state — how old your biology looks right now. A pace measure such as DunedinPACE estimates a rate — how fast you appear to be aging per calendar year, where 1.0 is the population average. The distinction matters for anyone tracking an intervention. A state estimate has to overcome its own baseline noise before a change becomes visible. A rate estimate is built to describe change and is less dependent on whether your first reading happened to be taken during a bad month.
Is a biological age result a diagnosis?
No, and no consumer methylation test is regulated as a diagnostic device. A result that comes back older than your calendar age indicates that your marker pattern resembles that of older people in the training data. It does not identify a condition, name a cause, or tell you what to change. The clinically useful information sits one layer down, in the individual markers — fasting insulin, HbA1c, ApoB, blood pressure, hs-CRP — which name specific problems with specific treatments. Any concerning result should route to a clinician looking at those markers, not to a supplement.
Should I get tested at all?
Testing is defensible if you are running a defined multi-year intervention and want an aggregate outcome measure alongside bloodwork, if a single number genuinely motivates you and you know that about yourself, or if you are already deep in the data and treating the score as one input. Skip it if you have never run a comprehensive blood panel — that panel costs less and tells you what to fix. Skip it if you will only ever test once, because a single reading with no comparison point carries almost no information for an individual.
If my biological age is lower than my real age, what does that mean?
It means your marker pattern resembles that of younger people in the model’s training set, under the conditions you happened to test in. It is a reasonable signal that nothing major is going wrong, and it is close to unactionable. There is no next step it points to, and a retest a year later will very likely land inside the noise band. Readers in this position generally get more from measurements that produce a target: VO2 max, a DEXA body composition scan, a coronary artery calcium score, or an ApoB level.