A consumer test that returns a single number claiming you are 14 years younger than your birth certificate is reporting the output of a research instrument, not a medical verdict. Here is what that instrument actually measures, where it has held up, and where it has not.
Longevity
Published by Astra, which offers some of the treatments discussed. Educational, not medical advice.
An epigenetic clock is a statistical model. It takes methylation levels at a set of specific sites across the genome, DNA methylation being a chemical modification that changes with age without changing the underlying DNA sequence, and combines them into a single number using weights learned from a training dataset. The clock did not discover that you are aging faster or slower. It found a pattern in your methylation data that resembled the pattern seen in older or younger people in whatever dataset trained it.
The first clock built this way, and still the most cited, is Horvath's multi-tissue clock, trained across a wide range of tissue types and cell types so that it could estimate age from almost any human sample rather than blood alone.1 Hannum and colleagues built a parallel clock trained specifically on blood, using a different set of methylation sites and a different training population, and it produces a related but not identical estimate of age.2
Both were trained to predict chronological age itself: given a methylation profile, guess how old this person's birth certificate says they are. That target sounds circular, and in a sense it is. The interesting biology only shows up when a clock's guess disagrees with the calendar, and the open question from the start was whether that disagreement meant anything about health.
The next generation of clocks changed the training target, and this is the single most important technical shift in the field. PhenoAge was trained not on chronological age but on a composite of clinical biomarkers linked to mortality and morbidity, then mapped back onto methylation data, an approach explicitly framed as an epigenetic biomarker of aging for lifespan and healthspan.3 The clock is trying to predict clinical decline, using methylation as the measurement tool, rather than trying to predict the number on your driver's license.
GrimAge went further in the same direction. It was trained directly to predict lifespan and healthspan, incorporating methylation-based surrogates for smoking history and several plasma proteins associated with mortality risk, and the original paper reports that GrimAge strongly predicts both lifespan and healthspan in the cohorts studied.4 This is the clock most associated with strong mortality prediction in the epidemiological literature, and it is also the clock furthest removed from a simple 'how old do your cells look' framing. It is closer to a mortality risk score wearing methylation as its input.
DunedinPACE represents a further conceptual shift again. Rather than estimating a static age at a single point in time, it was built to measure the pace of aging, using longitudinal data from the same individuals tracked over years to see how fast their biological systems were declining relative to the passage of calendar time.5 A pace-of-aging measure and a point-in-time age estimate are answering different questions, and conflating them, as consumer marketing often does, obscures which question a given number is actually trying to answer.
Each generation improved on a real limitation of the last. But each generation is still a population-trained model applied to an individual, and none of them has been validated as a tool for telling one person that a specific intervention worked for them personally.
CALERIE was a multi-year, multicentre, randomized trial of sustained caloric restriction in healthy, non-obese adults, one of the most rigorous long-term human trials of an aging intervention that exists. A later analysis applied DNA methylation measures of biological aging to CALERIE's stored samples, comparing participants randomized to long-term calorie restriction against controls.6
The result was not a clean win. Some DNA methylation measures of biological aging shifted in the direction consistent with slower aging in the calorie-restricted group. Others did not move in a way that reached the same conclusion. This is exactly the kind of finding that gets simplified in marketing into 'calorie restriction reverses aging by X years,' when what actually happened is more interesting and more honest: different clocks, measuring different aspects of biological aging, gave different answers to the same intervention in the same people.
That split matters for anyone reading a biological age report. If the best-controlled long-term human trial of a real intervention produced a mixed result across different clocks, a single consumer number claiming a clean answer for one person, from one blood draw, is claiming more precision than the underlying science currently supports.
Before any of this can be used to judge whether an individual intervention worked, the clock has to give a consistent answer when applied twice to the same untreated person. This is test-retest reliability, and it is the least exciting and most load-bearing question in the whole field.
A dedicated paper addressed this directly, describing a computational solution built specifically to bolster the reliability of epigenetic clocks, stating plainly that reliability needed improving to make these tools usable for clinical trials and longitudinal tracking.7 The fact that this paper needed to exist, years after the original clocks were published, is itself informative. If the base measurement bounced around from noise alone by an amount comparable to the effect size an intervention is trying to detect, then a single before-and-after biological age reading on one person could look like improvement, or decline, purely from measurement noise.
This is not a reason to dismiss epigenetic clocks. It is a reason to be skeptical of any single consumer report presented as a definitive individual result, especially one taken once before an intervention and once after, with no accounting for the clock's own retest variability.
The study most responsible for the popular idea that biological age can be reversed is TRIIM, a small trial combining growth hormone with two other drugs, which reported reversal of epigenetic aging and immunosenescent trends measured by several clocks.8 It is a genuinely interesting pilot finding and it deserves credit for being one of the first studies to test an active intervention against these newer clocks at all.
It also enrolled nine men and had no control group. Nine participants with no placebo arm and no comparison group is a study design built to generate a hypothesis, not to confirm one. Every subsequent headline describing this as proof that biological age reversal is achievable in general is claiming more than a nine-person, uncontrolled pilot can support, however real and however interesting the observed methylation changes were in those nine men.
A separate, later diet-and-lifestyle intervention was run as an actual randomized controlled trial and reported a potential reversal of epigenetic age using diet and lifestyle changes.9 It is a step up in design from TRIIM: it had randomization. It was still small and short, a pilot in scale and duration rather than a definitive long-term outcomes trial. 'Pilot' and 'small' are not insults here; they are accurate descriptions that the trial's own authors would recognise, and they are the reason this finding needs a larger, longer follow-up before it settles anything.
Put the pieces together. Five generations of increasingly sophisticated clocks, each trained on a different target, from chronological age to mortality risk to pace of change over time. A best-in-class randomized trial that produced a split result across different clocks. A documented reliability problem serious enough to need its own fix. A famous reversal headline from nine uncontrolled men, and a more rigorous but still small and short randomized follow-up.
None of that adds up to a tool that can hand one person a single trustworthy number and call it their true biological age, let alone claim that number moved 14 years younger because of a specific product they bought. It adds up to a genuinely promising and improving field of population-level biomarker research that is, at present, being sold to individual consumers as something more finished than it is.
That is not a reason to ignore the field. Cardiorespiratory fitness, strength training, sleep, and other unglamorous levers all have their own, separately documented relationships with long-term mortality risk, and none of them require a methylation test to justify pursuing them. It is a reason to treat a consumer biological age report the way you would treat any single-measurement research instrument applied to an audience of one: interesting, worth tracking over time with the same test and the same lab, and not, on its own, a verdict.
A biological age number is the output of a statistical model trained on population data, reported back to one individual as if it were a diagnosis. The models are real and improving. The individual verdict is not what they were built to give you.
Astra Editorial, reviewing the epigenetic clock literature
It measures DNA methylation levels at specific genomic sites and combines them, using weights learned from a training dataset, into a single number. Different clocks were trained to predict different targets: chronological age, clinical phenotype and mortality, or pace of change over time.12345
No. CALERIE's methylation analysis found a split result: some DNA methylation measures of biological aging shifted with long-term calorie restriction, others did not move in the same direction.6
It is an interesting pilot finding, not confirmed proof. TRIIM enrolled nine men with no control group, which is a hypothesis-generating design, not a confirmatory one.8
Treat it cautiously. Epigenetic clocks have a documented test-retest reliability problem serious enough that researchers built a dedicated computational method to improve it for clinical trial and longitudinal use.7
No. Astra does not sell anything shown to reverse biological age. This article exists to explain what the underlying science measures, not to sell a biological age intervention.
Astra does not sell anything that reverses biological age, and no product on the market has been shown to do that in a large, well-controlled human trial. What this article can offer instead is a clearer read of what a biological age number is actually measuring, so a report claiming you are years younger reads as a research instrument's output rather than a verdict. Keep reading in the Astra Learn library, and sign up below if you want an update when the evidence changes.
This guide is educational and is not medical advice. Compounded medications are not FDA-approved. Speak with a licensed physician about your own care.