Do epigenetic age tests accurately measure how fast you are aging and show if your interventions work?
Claim attributed to Consumer testing companies (e.g. TruDiagnostic, Elysium Index) and longevity clinics/influencers who resell repeat testing. , Marketed as a personal "speedometer of aging" and proof a supplement, diet or protocol is "lowering your biological age." It repackages real population-level science as an individually precise, actionable diagnostic.
Across thousands of people these clocks really do predict death and disease; for one person at one timepoint the number carries years of technical noise, and no trial shows that acting on it helps. The science is real; the personal "speedometer" framing outruns it.
A blood marker that predicts death across thousands of people is not a speedometer for you, and a number that swings several years on a re-run cannot prove your supplement worked.
What it’s supposed to target
- DNA methylation clocks
- CpG methylation patterns
- Biological vs chronological age
- Intervention tracking (claimed)
These tests are a measurement, not a treatment, so the theory is interpretive. Patterns of DNA methylation (chemical tags on CpG sites) change with age in a predictable enough way that algorithms (Horvath, GrimAge, DunedinPACE and others) can estimate a biological age. The promise is that this number tracks how fast you are aging and can tell you whether an intervention is working.
The underlying science is real and these clocks do predict health outcomes at the population level. The over-reach is the individual, real-time use: a single person's reading carries meaningful noise, clocks disagree, and almost none are validated to detect whether your supplement changed your aging rate. A genuine research tool, sold as a personal speedometer it has not earned.
Mechanism is theory, not proof. A plausible pathway explains why something might work, not whether it does. The verdict rests on the evidence below, not the elegance of the theory.
What would have to be true
The clocks must track biological aging, not just chronological age: HOLDS for second/third-generation clocks (GrimAge, PhenoAge, DunedinPACE) at the population level.
A single test must be reproducible enough that one person's number is real: FAILS for uncorrected first-generation clocks (3 to 9 year replicate noise); holds better for PC-corrected clocks and DunedinPACE (ICC ~0.96).
A short-term change must reflect real biology, not noise or chance: LARGELY FAILS, most 'improvements' came from one clock and died under correction.
Acting on the result must improve an outcome: UNTESTED, no randomized trial exists.
What the evidence actually shows
The population science is genuinely strong
At the level of thousands of people, these clocks earn their reputation. GrimAge (Lu 2019), a third-generation clock trained on mortality-linked plasma proteins and smoking rather than on age itself, predicted time-to-death (P=2.0E-75), coronary heart disease and cancer better than chronological age and earlier clocks across the Framingham, Women's Health Initiative, Jackson Heart and InCHIANTI cohorts. DunedinPACE (Belsky 2022) measures a pace of aging built from 20 years of within-person decline across 19 organ systems. These are real, useful instruments. The catch is in the unit of analysis: predicting risk across a group is not the same as clocking one individual's speed.
For one person, the number is noisy and the 'improvement' is often illusory
Higgins-Chen et al. (2022) found that for six prominent clocks, technical noise alone shifted replicate measurements of the same blood sample by roughly 3 to 9 years; their principal-component (PC) versions restored agreement to within about a year, but many consumer tests still report uncorrected single-clock values. A follow-up (Borrus 2024) went straight at 'is my intervention working?': re-analyzing six intervention datasets, 5 of 6 'significant' effects appeared in only one first-generation clock, were not corroborated by PC variants, and 4 of 5 lost significance after multiple-testing correction; some even implied interventions increased biological age. As two aging researchers put it in The Conversation (2026), the clocks are 'highly effective research tools to study aging at the population level' but 'aren't designed to make claims about the health of individuals.'
Studies, graded, and who paid
GrimAge beat chronological age for time-to-death, CHD and cancer in multiple cohorts (Lu 2019).
Technical noise alone moved replicate measures of the same sample by ~3 to 9 years for six clocks (Higgins-Chen 2022).
5 of 6 'significant' effects came from one clock, vanished under PC variants, and 4 of 5 died after multiple-testing correction (Borrus 2024).
No RCT has tested this; actionability is unproven, not demonstrated.
| # | Study | Type | Size | Funding / COI | Key limitations |
|---|---|---|---|---|---|
| 7000 | Lu 2019, GrimAge predicts lifespan/healthspan | Biomarker development + prospective mortality prediction (multi-cohort) | 7,375 arrays / 6,935 individuals (Framingham, WHI, Jackson Heart, InCHIANTI) | Independent NIH-funded academic work; UC/UCLA holds a patent with Lu and Horvath as inventors. | Group-level prediction; not validated as a personal intervention-tracking tool. |
| 1000 | Belsky 2022, DunedinPACE pace-of-aging clock | Biomarker development from longitudinal birth cohort | Dunedin 1972-73 cohort, ~1,000 members, 4 timepoints over 20 yrs | Independent NIA/MRC funding; Duke/Univ. of Otago hold a licensed patent, authors as inventors. | High reliability (ICC ~0.96) but a research instrument; individual responsiveness to interventions not outcome-validated. |
| 6 | Higgins-Chen 2022, reliability of clocks (Nature Aging) | Methods/reliability re-analysis (technical replicates + longitudinal) | Six clocks on replicate and longitudinal methylation datasets | Independent NIA-funded; authors are patent inventors and disclose ties to TruDiagnostic/FOXO, yet conclusions are cautionary. (See source note: the abstract is verified; full disclosures are in the peer-reviewed paper.) | Reliability shown mainly for first-generation clocks; PC versions fix most of it. |
| 61 | Borrus 2024, false positives in intervention studies | Re-analysis of six intervention datasets (preprint) | Six longitudinal intervention datasets | Independent NIA-funded (R01AG060110, R01AG065403); authors disclose consulting ties to TruDiagnostic, FOXO, Cambrian BioPharma, LongevityTech.fund. | Preprint; re-analysis of existing data, not a new trial. |
| 8000 | Horvath 2013, foundational multi-tissue clock | Methods/biomarker development (cross-sectional, multi-cohort) | ~7,844 non-cancer samples, 51 tissues/cell types | Independent UCLA academic; no consumer-test sponsorship in the 2013 paper. | First-generation, trained on chronological age; later shown to have poor test-retest reliability. |
The damning reliability evidence comes from the clock developers' own NIA-funded lab, which strengthens it: they are flagging the limits of their own tools rather than overselling.
Clocks disagree on the same person, blood and saliva differ, and results shift with diet, fasting, illness and time of day; there is no standardized method across companies.
Unproven ≠ disproven
Whether acting on a personal epigenetic-age result improves any clinical outcome is unproven, not disproven: no randomized trial has tested it, partly because such a trial is long, costly and commercially unnecessary.
Where claim and evidence diverge
The claim is a composite: the population-prediction half is well supported, while the individual-accuracy and intervention-tracking half is weak or untested. The 'Misleading' verdict is the distance between the two.
The money trail
The consumer model is repeat testing: customers retest to 'track' whether interventions lower their age, so every reassuring result aligns with selling another test.
Several leading clock inventors are named patent holders and disclosed consultants/advisors to testing firms (TruDiagnostic, FOXO); the cautionary reliability research, by contrast, was independent NIA-funded work.
The honest read
Trust these clocks for what they were built for, group-level risk research, not as a personal speedometer. One result, or a single short-term 'improvement,' is mostly noise unless a reliability-corrected clock is used, and no trial shows that acting on the number helps you.
What would change this verdict
A randomized controlled trial showing that changing behavior in response to an epigenetic-age readout improves a hard clinical outcome (mortality or disease incidence).
Independent test-retest data showing the specific consumer clocks sold today are reproducible to within ~1 year per person and detect real intervention effects across multiple corrected clocks.
Sources
- Lu AT, Quach A, Wilson JG, et al. DNA methylation GrimAge strongly predicts lifespan and healthspan. Aging (Albany NY). 2019;11(2):303-327. PMC6366976.
- Belsky DW, Caspi A, Corcoran DL, et al. DunedinPACE, a DNA methylation biomarker of the pace of aging. eLife. 2022;11:e73420. PMC8853656.
- Higgins-Chen AT, Thrush KL, Wang Y, et al. A computational solution for bolstering reliability of epigenetic clocks. Nature Aging. 2022;2:644-661. (GSA conference abstract, Innovation in Aging 2021;5(Suppl 1):igab046.015, confirms the 3 to 9 year replicate-noise and PC fix.)
- Borrus T, Sehgal R, Higgins-Chen AT, et al. When to Trust Epigenetic Clocks: Avoiding False Positives in Aging Interventions. bioRxiv 2024.10.22.619720. PMC11526921.
- Shalev I, Apsley A. Biological age tests reveal what slows or hastens aging, but they're useful only for researchers, not consumers. The Conversation. 4 May 2026.
- Horvath S. DNA methylation age of human tissues and cell types. Genome Biology. 2013;14(10):R115.
- Epigenetic clocks: advancing biological age measures towards meaningful clinical use. (Editorial, eBioMedicine, 2026.) PMC12905613.
People also ask
- Are epigenetic age tests accurate for one person?
- Not precisely. Across large populations these clocks predict death and disease well, but for a single person at a single timepoint, technical noise alone moved repeat measurements of the same sample by roughly 3 to 9 years across six clocks.
- Can DNA methylation tests show if my supplements are working?
- No trial supports this. In one analysis, 5 of 6 significant intervention effects came from one clock, vanished under PC variants, and 4 of 5 disappeared after multiple-testing correction. No study shows acting on the result improves health.
- Do epigenetic clocks actually predict mortality?
- Yes, across populations. Mortality-trained clocks like GrimAge beat chronological age for time-to-death, coronary heart disease and cancer in multiple cohorts. They were built as group-level risk research tools, not as a personal aging speedometer.
- Why do my biological age test results keep changing?
- Largely technical noise. Replicate measures of the same sample shifted several years for six clocks, so a single short-term improvement is mostly noise unless a reliability-corrected clock is used. The consumer model relies on repeat testing to track interventions.
Compare head to head: Epigenetic age tests vs telomere tests: which is legit?
Caveat is journalism, not medical advice. We check public claims against published evidence; we don’t diagnose, treat, or tell you what to take.