Biological age: what the tests measure and their limits

Two ages that aren't the same thing

Chronological age is the time that has passed since you were born. Biological age starts from an observation: people of the same age can have bodies that work very differently. In research, a “biological age predictor” is a measure that changes with age and that also adds information about the risk of health problems that calendar age alone doesn't provide.⁹

What doesn't exist is a single biological age for each person. The figure depends entirely on which markers are measured and how it's calculated, and there's no agreement on which ones should be used: with another test, or another tissue, you get another number.¹³ If a report tells a 56-year-old that they are “biologically” 65, the only thing it can mean is that what was measured looks more like what is usual in people older than them: a shorthand for talking about probabilities, not a date.¹³

How researchers try to measure it

Many measures have been proposed, from telomere length to blood proteins.⁹ Here we focus on three families:

• Epigenetic clocks. They measure DNA methylation, chemical marks that influence which genes are switched on.¹³ At each point in the DNA the change with age is small, but a mathematical model that combines dozens or hundreds of points estimates an age.¹
• Combinations of blood tests. They bring together several values from a routine blood test, such as glucose, albumin or C-reactive protein, into a single figure expressed in years.³
• Frailty indexes. They count how many health problems a person has accumulated — symptoms, diseases, limitations in daily life, test results — out of a total of at least 30 or 40 considered.¹⁰

The difference in purpose matters. The first clocks were tuned to predict calendar age accurately; later ones, to predict mortality or the pace of decline. When PhenoAge and the Horvath and Hannum clocks were put into the same model, only PhenoAge remained associated with higher mortality.³ And in 964 middle-aged adults from a New Zealand cohort, studied with eleven different measures of biological aging, agreement between them was low.⁸

It's also worth knowing what's inside each figure. One of the components of GrimAge, for example, is a DNA-based estimate of how much the person has smoked.⁴ Part of what these clocks capture are risk factors that are already known through other routes.

What they predict: associations in large groups

In cohort studies, people whose estimated age is higher than their calendar age have, on average, a higher risk of illness and death in the following years. In all of these figures, the effect of age itself has already been taken out:

• Phenotypic age, from blood tests. In 6,209 US adults followed for up to 12 years, each extra year was associated with a 9% higher risk of death.³
• PhenoAge. In five samples from US cohorts, each extra year was associated with a 4.5% higher risk of death.³
• GrimAge. In 6,935 people from five cohorts, followed for an average of 13.7 years, each extra year was associated with a 10% higher risk of death; it also predicted coronary heart disease.⁴
• DunedinPACE. In 2,471 people from the Framingham cohort, with a mean age of 66, a faster pace — one standard deviation above the mean — was associated with a 65% higher risk of death, and also with more cardiovascular disease and more disability.⁵
• Frailty index. In 754 people aged 70 or older who lived at home without disability, a higher index was associated with higher mortality.¹⁰

The problem with repeating the test

For a figure to be useful to you, it should come out similar if the measurement is repeated. With epigenetic clocks that isn't guaranteed, for three reasons.

Laboratory noise. In 36 blood samples from people aged 37 to 75, each analyzed twice, six well-known clocks gave median differences of 0.9 to 2.4 years between the two analyses of the same sample, and maximum differences of 4.5 to 8.6 years; every sample deviated notably on at least one clock.⁶ GrimAge was the most stable of those six,⁶ and DunedinPACE, designed to avoid this noise, also gave very consistent results when the analysis was repeated.⁵ There are versions of the clocks that greatly reduce the problem.⁶ A 2026 analysis found that, in general, the clocks give consistent results when the analysis is repeated in the laboratory, although details such as the sample's position on the chip or the method used to extract the DNA still change the result for some of them.⁷

Changes in the body itself. The same analysis compared blood samples from the same people taken hours or days apart: before and after a meal, a psychologically stressful situation induced in the laboratory, breathing polluted air or going to high altitude. No clock reached excellent stability, and some widely used ones, such as GrimAge and DunedinPACE, were among the least stable after stress.⁷ These are small studies, mostly with young adults, but the authors find it implausible that a meal or a bout of stress really ages or rejuvenates someone by several years: what moves is the measurement.⁷

The tissue sampled. The result also depends on the tissue analyzed.¹³ In the test data for the Horvath clock, the median error was 2.7 years in saliva, 3.7 in blood and 18 in muscle.¹

What a commercial test can tell you, and what it can't

The most common biological age tests sold to the public are epigenetic clocks.¹³ A commentary published in 2025 in a medical ethics journal points to several problems: many companies' algorithms are proprietary, so independent experts can't review them; the result may interest you, but it doesn't call for any medical action, and it doesn't mean you'll fall ill or die sooner than expected for your age.¹³ The authors call on companies to do a better job of explaining the reliability and limits of their tests.¹³

A large group of researchers in the field warned as early as 2019 that the accuracy of these clocks for an individual is unknown, as is their accuracy in populations different from those used to build them, and that there were no standards for how to communicate an estimated biological age.¹¹ In 2024, another large group of researchers working on biomarkers of aging found that there is no consensus on how to validate them before bringing them into clinical practice, and that, unlike markers for specific diseases, neither the US Food and Drug Administration (FDA) nor the European Medicines Agency has guidance on developing, measuring or validating them.¹²

From this we draw a practical reading, which is ours and not a finding of the studies: when you get a biological age result, ask which measure was used, on which tissue, how much it varies when repeated and which population it's compared with. If no one can tell you, the figure tells you very little.

If an app or a watch shows you an “age” or a similar score, that's an estimate too: an algorithm calculates it from a few sensors, as we explain in what a wearable really measures and what it estimates.

If a result worries you, talk it over with a health professional. A test figure isn't a diagnosis, either way.

Lowering the number isn't living longer

Many headlines talk about “rejuvenating” or “reversing” biological age. But a clock showing a lower figure doesn't prove that you'll live longer or in better health. Even if these clocks do reflect aging, it still has to be established whether they can be modified, whether any lifestyle change or treatment reverses them, and what effect that would have on long-term health.¹¹ Nor is it yet known what is cause and what is consequence in what they measure.⁹

Researchers in the field want to use these markers as stand-ins for the outcomes that matter — falling ill, losing independence, dying — to judge whether something extends healthy life; but there's still no agreement on how to validate them before using them that way.¹² Noise makes it harder still: in a short study, a two-year difference between groups can be masked by variations of several years in the measurement itself,⁶ and with the less reliable clocks, the estimated effect of a dietary change even reversed direction depending on which of the two analyses of the same sample was used.⁷

In addition, the first effects of interventions on epigenetic age were seen in mice.¹¹ Why a result in animals isn't enough to know what will happen in people is covered in the article on studies in animals and cells.