Biological Age Tests: What Epigenetic Clocks Can and Can't Tell You
Epigenetic clocks are a genuine scientific advance and a useful research tool — but the single "biological age" number a consumer test hands you comes with heavy caveats. Here's what the evidence actually supports.
By Wellness Wire Editorial
Send off a saliva swab or a tube of blood, and a growing number of companies will mail back a single striking number: your "biological age." Sometimes it lands below the candles on your birthday cake, sometimes above. The science underneath is real and genuinely interesting — but the tidy figure on the report hides a lot of uncertainty. Here is what epigenetic clocks can and cannot tell you, based on the published evidence.
What the test actually measures
Epigenetic clocks do not sequence your genes. They read DNA methylation — chemical tags that attach to specific spots on DNA, called CpG sites, and help switch genes on or off. Some of these tags change in fairly predictable ways as we age. The first widely used clock, published by Steve Horvath in 2013, tracked 353 CpG sites across roughly 8,000 samples spanning 51 tissue and cell types. It predicted chronological age remarkably well, correlating at r=0.97, with a median error of about 2.9 years in its training data and 3.6 years in independent test data.
But note what that clock was built to do: predict how old you are, not how healthy you are. This kind of "first-generation" clock nails your calendar age — which tells you little you didn't already know from your birth certificate.
Four clocks, four questions
Researchers have since built clocks that ask different questions, and the differences matter.
- First generation — Horvath (2013)trained to estimate chronological age from 353 CpG sites. It measures how old you are.
- Second generation — PhenoAge (Levine et al., 2018)its 513 CpG sites were trained not on birthdays but on a "phenotypic age" derived from nine clinical biomarkers plus chronological age. A one-year rise in DNAm PhenoAge was associated with a 4.5% higher risk of death from any cause, and it predicted cancers, healthspan, physical functioning, and Alzheimer's disease — outperforming the older chronological clocks.
- GrimAge (Lu et al., 2019)a composite built from seven methylation-based stand-ins for blood proteins plus a methylation estimate of smoking history. In its original study it was among the strongest methylation predictors of time to death, coronary heart disease, and cancer, with higher GrimAge robustly associated with earlier mortality. It was designed for population risk, not individual diagnosis.
- Third generation — DunedinPACE (Belsky et al., 2022)rather than a static age, it estimates the pace of aging, distilled from 19 organ-system measures tracked over about 20 years in a New Zealand birth cohort. It was deliberately engineered for consistency, reaching high test-retest reliability (ICC=0.96).
There is no single "best" clock. They answer different questions — chronological age, mortality and healthspan, or the rate of aging — and they frequently disagree. That is why one confident "your biological age is X" number oversimplifies what is really a family of research tools.
Why individual results wobble
The clocks were built and validated on large populations, and that is where they perform best. Applied to one person on one day, they are shakier. A 2025 review found that measuring the same biological sample several times can produce age estimates that differ by up to about nine years, because many of the individual CpG sites they rely on are technically noisy. The authors concluded that these clocks "fail to meet common standards for clinical utility compared with established biomarkers," and warned that "applying epigenetic clocks in individual-level decision making can be uninformative and potentially harmful," risking treating normal biological variation as if it were disease.
Applying epigenetic clocks in individual-level decision making can be uninformative and potentially harmful.
The clocks also disagree with one another. A 2026 editorial in eBioMedicine noted that today's clocks "capture broad trends but often diverge in their estimates," and cited a 2025 analysis of 14 clocks across 174 disease outcomes in which no single model emerged as uniformly superior. The editorial also flagged that some trials already use clocks as endpoints without proof they can track intervention-driven change.
Tracking your own number over time is especially unreliable: a change in lab, assay platform, or processing batch can shift the result on its own, unless all your samples are re-run together. And a deeper question stays open. These clocks are statistical associations; none of the studies shows that methylation changes cause aging or disease — the shifts may drive it, merely accompany it, or lag behind it. Whether a clock can validly show that an anti-aging intervention "worked" remains unproven.
Regulation and who was studied
Regulatory status matters here, and it is easy to misread. Commercial biological-age tests are largely laboratory-developed tests, which have generally not gone through FDA premarket review of their analytical and clinical validity. Under the long-standing CLIA framework, a lab had to show a test measures what it claims (analytical validity) but not that the result is clinically meaningful (clinical validity). The FDA notes that such direct-to-consumer tests "typically do not have the FDA's independent assurance of analytical validity, clinical validity, or clear communication of test results." No epigenetic clock is an FDA-approved diagnostic. The FDA issued a 2024 final rule that would have phased in tighter oversight of these tests, but a federal court vacated that rule in 2025, the agency did not appeal and reverted to the prior framework — so laboratory-developed tests remain outside routine FDA premarket review. Separately, at least one epigenetic-age test has reported an FDA "registration" — but registration is an administrative listing, not approval or clearance of a test's accuracy or clinical validity. Bottom line: the number on the report carries no regulatory guarantee, and the human evidence behind it is predictive at the population level, not proof of cause and effect in any one person.
Who the clocks were built on also shapes how far the results travel. A 2023 review found the underlying methylation data come "predominantly from individuals of European heritage," and warned that "prediction models may underperform in social groups that were poorly represented in the training data." Smaller effect sizes have been documented for Black American than white American participants, and the social and demographic makeup of training samples is often poorly reported. Because the tests mostly use blood or saliva, a reading may also not reflect what is happening in other organs.
What a result can and can't tell you
None of this makes epigenetic clocks worthless. The U.S. National Institute on Aging says epigenetic-age estimates "can aid prediction of age-related health outcomes, including multiple chronic diseases, impaired cognitive function, functional limitations, and mortality in older adults." That is a meaningful capability for research and population health.
But the same agency adds a caveat worth taping to your mirror: "other factors, such as demographics, socioeconomic status, mental health, and health behaviors, are comparable — and often more robust — predictors of late-life health outcomes." In plain terms, cheaper and simpler information often predicts your health at least as well as a methylation test. And an individual reading can be thrown off by a recent illness, stress, or ordinary day-to-day fluctuation.
So treat a biological-age result as a conversation starter and a population-level research metric — not a diagnosis, a verdict, or a scoreboard to optimize. If a number worries you, the more reliable predictors the research keeps returning to are things like your health behaviors, mental health, and social and economic circumstances — and the most useful next step is usually a conversation with a professional who knows your history, not another test.
- 01 Horvath S., Genome Biology, 2013 pubmed.ncbi.nlm.nih.gov ↗Original 353-CpG multi-tissue clock trained to predict chronological age (r=0.97).
- 02 Levine M.E. et al., Aging (Albany NY), 2018 pubmed.ncbi.nlm.nih.gov ↗DNAm PhenoAge, a 513-CpG second-generation clock tied to mortality, cancer, and Alzheimer's.
- 03 Lu A.T. et al., Aging (Albany NY), 2019 aging-us.com ↗GrimAge composite predictor; among the strongest DNAm predictors of time to death, coronary heart disease, and cancer.
- 04 Belsky D.W. et al., eLife, 2022 pubmed.ncbi.nlm.nih.gov ↗DunedinPACE third-generation pace-of-aging clock, built for reliability (ICC=0.96).
- 05 Apsley A.T., Etzel L., Ye Q., Shalev I., Epigenomics, 2025 pmc.ncbi.nlm.nih.gov ↗Review finding up to ~9-year replicate noise; warns clocks lack clinical utility for individuals.
- 06 eBioMedicine (The Lancet Discovery Science) editorial, 2026 pmc.ncbi.nlm.nih.gov ↗Editorial: clocks diverge; cites a 14-clock, 174-outcome analysis with no uniformly superior model.
- 07 National Institute on Aging (NIH), 2022 nia.nih.gov ↗Clocks aid prediction of health outcomes, but demographics and behaviors are comparable or more robust predictors.
- 08 Watkins et al., Environmental Epigenetics, 2023 academic.oup.com ↗Training data skew toward European ancestry; models may underperform in underrepresented groups.
- 09 DTC tests are largely LDTs lacking FDA assurance of analytical/clinical validity; no clock is FDA-approved.
- 10 A federal court vacated the FDA's May 2024 LDT final rule in 2025; the agency reverted to the prior framework, leaving LDTs outside routine premarket review.
Wellness Wire Editorial
The Wellness Wire editorial team. Our explainers are researched from primary sources — clinical-trial reports and FDA labeling — and written for general education. They are not individually reviewed by a clinician and are not a substitute for medical advice.
