Biological age: what is worth measuring
Two people, both born in 1971. One spent last weekend hiking. The other plans the day around a single flight of stairs. The calendar insists they are the same age. Every system in their bodies disagrees.
Two people, both born in 1971. One spent last weekend hiking and recovered with a good night’s sleep. The other plans the day around a single flight of stairs. The calendar insists they are the same age. Every system in their bodies disagrees.
The gap between those two people has a name, it can be measured, and, unlike the calendar, it moves. This piece explains what “biological age” actually means and what is worth measuring. It also looks at how far the consumer epigenetic age tests can be trusted, and why midlife is the window where measurement pays best.
What does the calendar miss?
Chronological age counts orbits of the sun. It is perfectly accurate and almost uninformative, because it measures time rather than wear. Biological age is the attempt to measure the wear directly. It asks about the current condition of the body’s systems and, more importantly, the pace at which they are declining.
The calendar counts time; biology counts wear, and the two can drift a decade apart in either direction. That drift is the most consequential number to follow, because it is the one you can actually change.
Which raises the practical question: measured how?
What is worth measuring?
Ignore, for a moment, the exotic tests. The workhorse measurements are unglamorous and mostly familiar. What changes is how you read them.
Function comes first: grip strength, walking speed, how hard you can push yourself, and how quickly you recover afterwards. These are whole-system outputs, and they predict late-life independence with uncomfortable reliability. Then the quiet chemistry: fasting insulin, the ratio of triglycerides to HDL, and hs-CRP, a blood marker of background inflammation. None of these is a longevity test by itself. Read together, and repeated over time, they describe a trajectory.
A trajectory carries more information than any single reading: direction and rate of change beat position every time. A fasting insulin that reads 4, then 6, then 8 across six years never leaves the reference range. Yet it is telling you something no single “normal” result can: the slope is wrong. That drift is exactly the early signal a one-off test cannot show. We have written separately on why a normal fasting glucose can sit on top of established insulin resistance.
What about the newer biological-age clocks built on chemical marks on DNA? They are real science and powerful research tools. What a year-to-year change means for one individual is still being worked out. They may earn a place in the schedule; they have not replaced it. Because many readers now hold such a report in their hands, the section below sets out what is established and what is not.
The figure’s real lesson is how small the difference looks at 45, which is precisely when it is most worth acting on.
Are epigenetic age tests accurate?
They are accurate enough to be useful in research, and not yet standardised enough to stand alone in a clinic. The measurement itself is real chemistry. The links with health outcomes in large studies are real links. What does not follow is that the single number on a consumer report describes one person’s rate of ageing to the year. Nor does a change over twelve months necessarily mean what the report implies.
What does an epigenetic clock actually measure?
Cells attach small chemical tags to DNA that help decide which genes are switched on. Scientists call this tagging DNA methylation. The pattern of tags shifts in a partly predictable way as the years pass. An epigenetic clock is a statistical model built on that shift. It reads the tags at a chosen set of positions, weights each one, and converts the result into a number reported in years.
Two things follow from that definition. First, a clock is only as good as the outcome it was trained to predict. Second, its output is a statistical estimate, not a direct reading of anything a cell is doing. It is closer to a risk score than to a haemoglobin.
Are the clocks all one thing?
No. “Epigenetic age” is a category, not a test. The generations differ in what they were trained on, which is the single most useful thing to know when reading a report.
| Generation | Examples | Trained to predict | What the output represents |
|---|---|---|---|
| First. | Horvath (Genome Biology, 2013); Hannum (Molecular Cell, 2013). | Chronological age. | An estimated calendar age; the interesting part is the gap between estimate and true age. |
| Second. | PhenoAge (Aging, 2018); GrimAge (Aging, 2019). | Clinical and mortality-related measures, not calendar age alone. | An age-scaled risk estimate; links with mortality are generally stronger than for first-generation clocks. |
| Third. | DunedinPACE (eLife, 2022). | Rate of change across organ systems within one birth cohort. | A pace of ageing: biological years accrued per calendar year, rather than an age. |
Two reports from two companies can both be internally valid and still tell the same person different things. They were built to answer different questions. That is not a fault in either test. It is a reason not to average them, and a reason not to switch providers between measurements.
Where does the accuracy question bite?
- Agreement between clocks. Different clocks applied to the same sample do not return the same answer. They also read largely different parts of the genome.
- Repeatability. Technical reliability varies by clock and platform. Higgins-Chen and colleagues addressed this in Nature Aging in 2022, proposing statistical reconstructions to reduce the noise carried by individual probes. The point for a reader: a difference between two reports may be measurement variation rather than biology. We set out the same problem in distinguishing a real change from assay noise, and it applies with more force here than to most routine chemistry.
- What is in the tube. Blood-based clocks are influenced by the mix of white cell types in the sample, which itself moves with infection, stress and time of day. Saliva, cheek swab and blood are not interchangeable, and a clock trained in one tissue does not transfer cleanly to another.
- Population of origin. Most clocks were trained in groups that do not represent everyone now being tested. How well they calibrate outside those groups is still an open research question.
Where do they stand with regulators?
No methylation-based age measure is currently accepted by a major regulator as a validated stand-in for a clinical outcome. Consumer offerings are generally sold as wellness products or run as laboratory-developed tests. That route does not involve a regulator agreeing that the number predicts an outcome, or that moving the number moves the outcome. A 2026 commentary in Clinical Trial Vanguard on epigenetic clocks and the FDA makes the same observation about the American framework; we could verify only its headline claim. A February 2026 review in eBioMedicine addresses the same gap between what these measures are and what clinical use would require.
What would have to be true before a clock could guide care?
- Reliability. Technical variation on the same assay must be small next to the change being tracked. Otherwise a twelve-month “improvement” is unreadable.
- Standardisation. Results must mean the same thing across laboratories, platforms and populations.
- Prediction for individuals. A link with mortality across a study population is not the same as informing a decision about one person.
- Outcome tracking. A change in the marker must predict a change in the outcome that matters. This is the highest bar, and no ageing biomarker has cleared it yet. Researchers are working on exactly this.
None of this makes the field unserious. It places it where most of the evidence on modifiable ageing currently sits: coherent in mechanism, supported by population studies, and still being studied at the level of the individual number.
How should you read a result someone has already bought?
People arrive with these reports. The useful questions are practical rather than philosophical.
- Which clock was used? Was it trained on calendar age, on mortality-related measures, or on pace of change? The reading differs.
- What tissue was sampled, and by which laboratory? A retest should repeat both.
- Does the report separate the estimate from the gap between estimated and calendar age, or show only a headline number?
- Is there any statement of the assay’s repeatability? If not, a difference between two reports cannot be interpreted.
- What is the result being read alongside? A methylation age with no grip strength, fasting insulin or inflammation marker beside it is a number without a context.
Documented as one measurement among several, retested on the same assay over years, an epigenetic age has a defensible place in a long-term record. Treated as a verdict, it asks more of the evidence than the evidence can yet give. Outcomes vary between individuals.
Why is midlife the high-value window?
Age-related disease is not an event that begins with its diagnosis. Inflammation, insulin resistance, oxidative stress and declining cellular energy run silently for decades first, each feeding the others. Inflammation blunts insulin signalling. Poor glucose handling raises the oxidative load. Oxidative damage wears the cell’s energy factories, and struggling energy factories signal further inflammation. We have set out those four processes and the loop they form in more detail. The visible diseases are the late chapters. What midlife offers is lead time: the processes are already measurable, still quiet, and still steerable, and a small change of slope compounds for thirty years.
Most of that lead time is wasted on a single mistaken idea: that acting on longevity means acquiring things. A supplement stack, a device, a test with a number on it. The evidence points somewhere much less marketable.
What actually moves the number?
The strongest levers are behavioural and boring. Strength work, because muscle is the organ of independence and a glucose sink besides. Sleep. What and when you eat. And then measurement itself: the same few markers, on a schedule, year after year. The schedule is what turns numbers into trajectories, and trajectories are the entire game.
Put strength on the problem list. Muscle declines quietly from midlife unless actively opposed, and it is one of the few organs whose ageing is squarely in the person’s own hands. A grip measurement and a sit-to-stand count cost nothing. On current evidence they say more about the next twenty years than most scans, and more than most methylation reports.
- The calendar counts time; biology counts wear, and the gap is measurable and movable.
- Direction and rate of change beat any single reading, however wide-ranging.
- Epigenetic clocks are not one test. They differ in what they were trained on, they disagree on the same sample, and none is yet a validated regulatory endpoint.
- If a methylation age is measured, repeat it on the same assay and read it beside function and chemistry, not instead of them.
- Midlife is the window: the processes behind late-life disease are already running, and still steerable.
- The strongest tools are a schedule and a barbell, not a shopping list.
Wear, slope and lead time are the three ideas worth carrying out of this, and the knowledge check will show how firmly they have landed.
