A drop in epigenetic age observed during a trial does not, on current evidence, show that participants' health has improved. A systematic review from October 2026 gathers 26 human studies measuring, before and after an intervention, at least one epigenetic clock and one health measure (PubMed ). No study with a comparison group saw the two change in a uniformly concordant way.
An epigenetic clock is an algorithm that estimates age from DNA methylation. Methylation refers to small chemical marks placed on specific sites of the genome, which shift over the years. We described how these clocks work in a dedicated article.
The review's authors draw a clear conclusion. These clocks remain exploratory or supporting biomarkers for now. They are not validated as surrogate endpoints, the measures that stand in, within a trial, for the health event one is trying to prevent (PubMed ).
Twenty-six studies, no concordant comparative trial
Researchers in Zhengzhou asked a question that fits in one sentence. When an intervention moves an epigenetic clock, do health measures (the clinical outcomes) move in the same direction?
The 26 included studies split into two halves. Thirteen are comparative: one group receives the intervention, another serves as a reference. The other thirteen compare participants with themselves, before and after, without a control group (PubMed ). This second design cannot separate the effect of the intervention from what would have happened anyway.
Their breakdown leaves little room for concordance. Fifteen studies are mixed: some clock-outcome pairs move together, others do not. For six, concordance remains indeterminate; four allow no conclusion on either the clock or health (PubMed ).
A single study shows uniform concordance. It has no control group, and that concordance is limited to intermediate outcomes, meaning biological or body measurements. No study of what matters directly to participants (how they feel, what they can do, whether they survive) showed uniform concordance (PubMed ).
None of the thirteen studies with a comparison group showed a uniformly concordant change between the epigenetic clock and clinical outcomes.
The analysis also goes down to the individual level. Are the participants whose clock falls most the ones whose health improves most? Eleven studies asked this question. Their clear findings remain isolated and were not replicated across independent studies (PubMed ).
No combination of intervention, clock and outcome met the criteria for a meta-analysis, a statistical pooling of results across studies. The certainty of evidence is rated very low for every question examined (PubMed ).
The precedent of surrogate endpoints
A marker that predicts risk does not, by that fact alone, become a target whose improvement protects. Cardiology learned this from the CAST trial, published in 1991.
After a heart attack, ventricular premature beats (early heartbeats arising in the heart's lower chambers) were associated with a higher risk of sudden death. Suppressing them should logically have reduced that risk. The trial therefore selected patients in whom encainide or flecainide, two heart-rhythm drugs, suppressed these beats. It randomly assigned 1,498 of them to the drug or to a placebo (PubMed ).
Over an average follow-up of ten months, 63 of the 755 patients on treatment died, against 26 of the 743 on placebo (PubMed ). The targeted marker was corrected; mortality had more than doubled.
Fleming and DeMets drew a general rule from such failures in 1996. For a marker to stand in for the health event, the intervention's effect on that marker must reliably predict its effect on that event (PubMed ). That condition often fails: a disease can cause harm through pathways that bypass the marker, and a treatment can act through unforeseen mechanisms. A marker can thus predict prognosis without forecasting a treatment's effect.
Epigenetic clocks stand exactly at this stage. GrimAge, a clock trained on mortality data, predicts lifespan in cohorts, the large populations followed over many years (PubMed ). Yet in 2024, the international consortium on biomarkers of aging noted that no consensus governed their validation as surrogate endpoints (PubMed ). The October 2026 review found no such validation at the trial level (PubMed ).
Why the clock moves while health stays put
Measurement noise
In 2022, a team at Yale University compared repeated analyses of the same samples. For six well-known clocks, the gap between two analyses of the same sample reached up to 9 years (PubMed ). Recalculated versions of these clocks, built to resist noise, brought most gaps below 1.5 years.
Four years later, the same team assessed 18 clocks. Most proved highly reproducible at the technical level. Their biological stability remains low to moderate: across closely spaced samples taken under varying conditions (meals, stress, environment), their values shift between measurements (PubMed ). A fall of a few years between two blood draws may therefore reflect, in part, the conditions of sampling.
Clocks that contradict each other
The CALERIE trial provides an example. Two hundred and twenty adults without obesity were randomly assigned to 25% calorie restriction or an unrestricted diet for two years (PubMed ).
Their blood samples underwent a post hoc analysis, one that was not planned in the trial's original protocol. DunedinPACE, a clock that estimates the pace of aging, indicated a slowdown under restriction. PhenoAge and GrimAge, two clocks that estimate an age, did not move beyond what chance explains. The measured effects were small, although the authors argue that modest slowing can weigh heavily on population health (PubMed ).
The Yale team broadened this finding in 2026. It applied eleven additional clocks to two studies published on the basis of a single one: none showed an effect distinguishable from chance (PubMed ). Its authors propose three confidence criteria. A result gains credibility when the recalculated versions respond, as do clocks trained on mortality or on the pace of aging. It gains more when several clocks move in the same direction.
A result built on a single clock
Published in 2021, the pilot trial often cited for its “3.23 years” of rejuvenation illustrates this case. Its eight-week program combined guidance on diet, sleep, exercise and relaxation. It also included probiotics and phytonutrients, compounds of plant origin. Forty-three healthy men aged 50 to 72 were randomly assigned to this program or to a control group with no intervention (PubMed ).
Calculated from saliva with the 2013 Horvath clock alone, epigenetic age fell by 3.23 years compared with controls. The p-value, that is, the probability of observing a gap at least as large if the program had no effect, was 0.018 (PubMed ). It falls below the conventional threshold of 0.05.
The per-group figures, reported in 2021 and corrected by the authors in 2024, shed light on this gap. The 18 analyzed participants in the program group lost 2.04 years of epigenetic age (p = 0.043). Over the same period, the epigenetic age of the 20 controls rose by 1.10 years (p = 0.191) (PubMed ). Part of the between-group gap therefore comes from this rise among the controls, itself indistinguishable from chance.
We have updated our article on the epigenetic clock, which presented this result as a pilot signal awaiting confirmation. The clock used, a first-generation one, was designed to estimate chronological age (PubMed ). Used alone, it could not test any of the three confidence criteria proposed at Yale. The same team reanalyzed the trial's data: none of the eleven added clocks showed a decrease distinguishable from chance (PubMed ). In that reanalysis, the decline measured by the original clock among program participants did not itself reach the 0.05 threshold.
Reading a claim of “years regained”
Four checks place a result expressed as biological age.
- Control group. Without one, a drop may reflect measurement noise, the conditions of sampling or regression to the mean. This term describes the tendency of an extreme value to move closer to normal at the next measurement.
- Outcome set in advance. An analysis decided after the fact, or picked from among several clocks, multiplies the chances of finding a result by luck.
- Agreement across several reliable clocks. This combines two of the three confidence criteria proposed by the Yale team (PubMed ).
- Health measure tracked alongside. Physical function, blood pressure, blood sugar, lipids or clinical events: without one, the clock speaks alone.
What would turn a clock into proof
Epigenetic clocks do not leave this review disqualified. Some, such as GrimAge, predict mortality in cohorts (PubMed ). New computational methods are also making them more reliable (PubMed ). What they lack are long trials in which a fall in the clock would forecast the intervention's effect on chronic disease and survival. The CALERIE authors also call for such trials, with follow-up long enough to measure these events (PubMed ).
Until then, a clock that falls in a trial points to a lead worth following. The benefit is still counted in preserved function, in diseases avoided and in years lived.
Frequently asked questions
References
- Yi M, Li M, Li H, Gui Y, Wang Z. Concordance between changes in epigenetic clocks and changes in clinical outcomes in geroscience intervention trials: A systematic review. Exp Gerontol. 2026;226:113367 (PubMed ).
- Echt DS, Liebson PR, Mitchell LB, et al. Mortality and morbidity in patients receiving encainide, flecainide, or placebo. The Cardiac Arrhythmia Suppression Trial. N Engl J Med. 1991;324(12):781-788 (PubMed ).
- Fleming TR, DeMets DL. Surrogate end points in clinical trials: are we being misled? Ann Intern Med. 1996;125(7):605-613 (PubMed ).
- Lu AT, Quach A, Wilson JG, et al. DNA methylation GrimAge strongly predicts lifespan and healthspan. Aging (Albany NY). 2019;11(2):303-327 (PubMed ).
- Moqri M, Herzog C, Poganik JR, et al. Validation of biomarkers of aging. Nat Med. 2024;30(2):360-372 (PubMed ).
- Higgins-Chen AT, Thrush KL, Wang Y, et al. A computational solution for bolstering reliability of epigenetic clocks: Implications for clinical trials and longitudinal tracking. Nat Aging. 2022;2(7):644-661 (PubMed ).
- Sehgal R, Borrus DS, Gonzalez J, et al. Biological Versus Technical Reliability of Epigenetic Clocks and Implications for Disease Prognosis and Intervention Response. Aging Cell. 2026;25(8):e70635 (PubMed ).
- Waziry R, Ryan CP, Corcoran DL, et al. Effect of long-term caloric restriction on DNA methylation measures of biological aging in healthy adults from the CALERIE trial. Nat Aging. 2023;3(3):248-257 (PubMed ).
- Borrus DS, Sehgal R, Armstrong JF, et al. When to Trust Epigenetic Clocks: Avoiding False Positives in Aging Interventions. Aging Cell. 2026;25(10):e70756 (PubMed ).
- Fitzgerald KN, Hodges R, Hanes D, et al. Potential reversal of epigenetic age using a diet and lifestyle intervention: a pilot randomized clinical trial. Aging (Albany NY). 2021;13(7):9419-9432 (PubMed ).
- Fitzgerald KN, Hodges R, Hanes D, et al. Correction: Potential reversal of epigenetic age using a diet and lifestyle intervention: a pilot randomized clinical trial. Aging (Albany NY). 2024;16(5):4943-4945 (PubMed ).
- Horvath S. DNA methylation age of human tissues and cell types. Genome Biol. 2013;14(10):R115 (PubMed ).



