For a decade, the promise of epigenetic aging clocks has been that a blood test could tell you whether an anti-aging intervention is working, without waiting decades to see how many people die. A new analysis suggests most of the interventions people are already paying for do not move the needle at all.
Researchers at Yale University and collaborators pooled 51 longitudinal human intervention studies covering 3,128 blood samples, then ran each dataset through a single standardized pipeline that calculated 16 prominent epigenetic clocks and 94 additional DNA methylation biomarkers. The results were published in Nature Medicine on August 21.
Nineteen of the 51 interventions significantly decreased epigenetic age across the biomarker panel, with the number dropping to 13 after correction for multiple testing. Five significantly increased it. The remaining 26, slightly more than half, showed no significant effect in either direction.
The Buzziest Anti-Aging Category Came Out Looking Shaky
The most pointed finding involves senolytics, the class of compounds meant to clear senescent cells and among the most heavily promoted ideas in consumer longevity.
Five senolytic studies were included. Within individual studies, different biomarkers moved in opposite directions. Across studies, the same biomarker sometimes moved in opposite directions. The authors concluded that the effects of senolytics on epigenetic aging may be inconsistent.
To sort signal from noise, the team proposed two rules. A genuine intervention should move biomarkers of the same generation in the same direction and magnitude within a study, and a second study of the same intervention should move the same biomarkers. Senolytics satisfied neither.
Two interventions cleared both bars. Anti-TNF therapies, used for rheumatoid arthritis and inflammatory bowel disease, shifted nearly all second-generation biomarkers by similar magnitudes across multiple studies. Two different Mediterranean diet protocols in healthy cohorts decreased overlapping sets of biomarkers. Pharmacological interventions as a category produced significantly larger effects than lifestyle changes or supplements.
Two supplement studies using the same product changed the same biomarkers but in inconsistent directions, which the authors describe as generating uncertainty rather than evidence.
Not All Clocks Are Measuring the Same Thing
The analysis also sorts the clocks themselves, which matters because direct-to-consumer testing companies sell results from very different algorithms under the same biological age label.
First-generation clocks trained to predict chronological age, including the original Horvath and Hannum measures, moved sporadically and without a clear pattern. Horvath1 correlated with its own stochastic component, meaning a substantial share of what it registers is random epigenetic drift rather than biology an intervention could plausibly change. That pattern had been proposed from work quantifying the stochastic component of aging using cross-sectional data, and this analysis tested it longitudinally.
Second-generation, reliability-optimized measures performed far better. DunedinPACE, designed to capture the rate of aging rather than accumulated age, decreased significantly in 16 interventions and increased in only one, the largest effect size in the analysis. PCGrimAge produced the strongest statistical evidence. GrimAgeV2, PCPhenoAge, and SystemsAge also showed consistent decreases.
Different clocks responded to different things. DunedinPACE was the standout for lifestyle interventions, registering decreases in 8 of 15. GrimAgeV2 was the standout for drugs, with 8 significant decreases across 14 pharmacological studies, but showed movement in only 1 lifestyle study.
Health status mattered too. Biomarkers responded more strongly in participants with diagnosed disease than in healthy volunteers, which the authors attribute to greater baseline dysregulation and more room to improve. DunedinPACE was the exception, responding similarly in both groups.
What the Numbers Do Not Prove
The paper is explicit about a gap that consumer marketing tends to skip. Responsiveness is a prerequisite for using a biomarker as a surrogate endpoint, not proof that it is one. A clock that moves is not the same as a person who has become biologically younger.
The authors write that it remains unclear whether short-term biomarker changes correspond to long-term improvements in disease, healthspan, or lifespan. They list the absence of an empirically defined minimal clinically important difference as a key barrier and caution that second-generation clocks trained on composite risk factors, such as C-reactive protein in GrimAgeV2, may register shifts in inflammatory state rather than aging biology itself.
Other limitations are structural. The 51 studies differed substantially in age range, duration, design and quality. Preprocessing and batch correction were not harmonized across datasets. The analysis pooled effect sizes descriptively rather than through a formal meta-analysis, and the authors say that interpretation should remain within intervention classes.
Standardized effect sizes were modest. For widely used clocks, where epigenetic age deviation typically has a standard deviation of about 4 to 6 years, the authors calculate that an effect size near 0.2 corresponds to roughly a 1-year shift.
Where the Signal Actually Lives
The most useful practical result may be that global clocks average away information.
The team also tested 78 explainable biomarkers derived from component scores, of which 39 showed significant changes. Smoking cessation produced its largest decrease in a lung-specific score. Hyperbaric oxygen therapy also hit lung. Metformin moved inflammation, brain, and metabolic scores. Gastric bypass did not alter the general second-generation clock but decreased a metabolic score, with an effect size of 0.43.
Across seven different diets, the musculoskeletal score was the only system measure that decreased in all of them. An epigenetic proxy for triglycerides moved in six of seven.
Several authors are affiliated with TruDiagnostic, a commercial epigenetic testing company that supplied private datasets and controls the algorithms for two of the clocks analyzed. The company's role is disclosed in the paper. Anyone considering a consumer biological age test should recognize that these measures remain research tools rather than validated clinical endpoints, and should make health decisions based on a clinician rather than a test report.
Related MedicalDaily Coverage
Key Questions Answered
What is an epigenetic clock?
An algorithm that estimates biological age or pace of aging from chemical tags on DNA, usually measured in a blood sample.
What did the researchers do?
They harmonized 51 published and private intervention studies, totaling 3,128 samples, and recalculated the same 16 clocks and 94 other biomarkers for each dataset.
Which clocks performed best?
DunedinPACE showed the largest effect sizes and PCGrimAge the strongest statistical significance. Both are second-generation, reliability-optimized measures.
What happened with senolytics?
Five studies produced inconsistent results within and across studies, which the authors say raises doubt about the consistency of senolytic effects on epigenetic aging.
Does a lower clock reading mean someone got younger?
No. The study tests whether these biomarkers respond to interventions, not whether a response translates into longer or healthier life.
Should this change what anyone takes?
No. The paper is designed to guide clinical trial design, not personal supplement decisions.
