Yale-led analysis harmonized 51 studies under one standardized pipeline, and senolytics came out looking shaky.
Send in a vial of blood, get back a number, skip the decades of waiting it used to take to know if something is actually slowing your aging. That’s the pitch behind epigenetic aging clocks, and it’s a big part of why direct-to-consumer biological age testing has grown into its own industry over the past few years. A new study just put that pitch under harsher scrutiny than it usually gets.
Researchers at Yale University, publishing in Nature Medicine on 21 August, pooled 51 published and private human intervention studies (3,128 blood samples in total) and ran each through the same standardized pipeline, calculating 16 widely used epigenetic clocks plus 94 additional DNA methylation biomarkers for each dataset [1]. Different studies have historically processed data differently, making honest comparisons across the field nearly impossible. This is one of the first attempts to fix that.
The results: 19 of the 51 interventions produced a statistically significant decrease in epigenetic age. After correcting for the false-positive risk that comes with testing so many biomarkers at once, that number dropped to 13. Five interventions actually increased epigenetic age. The remaining 26, a bit more than half the studies included, showed no significant movement either way.
Pick one of these already-published interventions at random, in other words, and the odds don’t favor it doing anything measurable at all.
Senolytics didn’t hold up
The sharpest finding concerns senolytics, drugs designed to clear out senescent cells, the aged cells that stop dividing but keep pumping out inflammatory signals that damage the tissue around them. Senolytics also happen to be one of the most heavily marketed ideas in consumer longevity right now.
Five senolytic studies made it into the analysis, and the results didn’t line up. Different biomarkers moved in opposite directions within the same study. The same biomarker sometimes moved in opposite directions across separate studies of the same intervention. To separate real signal from noise, the researchers set two bars: a genuine effect should move biomarkers of the same generation, in the same direction and magnitude, within a single study, and a repeat study of the same intervention should reproduce that movement. Senolytics cleared neither one.
Only two interventions in the entire dataset cleared both bars. Anti-TNF therapies, anti-inflammatory drugs used for rheumatoid arthritis and inflammatory bowel disease, shifted nearly all second-generation biomarkers by comparable amounts across multiple independent studies.
Two separate Mediterranean diet protocols, tested in different healthy cohorts, decreased overlapping sets of biomarkers in ways that matched each other. As a category, drugs produced significantly bigger effects than either lifestyle changes or supplements.
Not every clock measures the same thing
There’s a finding here that matters just as much for anyone shopping for a consumer test: the clocks aren’t interchangeable. Older, first-generation clocks, including the original Horvath and Hannum measures, moved sporadically with no consistent pattern. The researchers found that a meaningful chunk of what the Horvath clock registers may just be random biological noise, not something an intervention could plausibly move.
Newer, second-generation clocks built for reliability did better. DunedinPACE, designed to track the pace of aging rather than accumulated age, decreased significantly in 16 interventions and increased in just one, the largest effect size anywhere in the analysis.
A related measure called PCGrimAge produced the strongest statistical evidence overall. Different clocks also suited different interventions better than others: DunedinPACE stood out for lifestyle changes, GrimAgeV2 for drugs.
What the researchers won’t claim
The paper is unusually direct about its own limits. A biomarker responding to an intervention, the authors write, is a prerequisite for eventually using it as a shortcut for real health outcomes, not proof that it already works as one. A clock moving isn’t the same thing as a person becoming biologically younger, and the study can’t say whether the short-term shifts it measured actually translate into a longer or healthier life down the road.
One disclosure worth flagging: several authors are affiliated with TruDiagnostic, a commercial epigenetic testing company that supplied some of the private datasets and controls the algorithms behind two of the clocks analyzed. The paper discloses the relationship directly.
For a longevity field that has leaned hard on the promise of a fast blood-test answer, this study is a useful gut check. It doesn’t say biological age testing is worthless, and it doesn’t say every anti-aging product on the market is snake oil. It says the honest answer to “is this working,” for most products people are already paying for, is that nobody has rigorously checked yet. Until someone does, a lower number on a test report isn’t the same thing as more years or better years, actually gained.
