Biological age tests: what an epigenetic clock measures

Reviewed 10 sourcesPublished by PepVault

Biological age tests compared: what Horvath, PhenoAge, GrimAge and DunedinPACE measure, what predicts death best, what moved a clock in trials, and the noise.

A white-haired woman in a green tank top and white skirt playing pickleball with a wooden paddle
Last reviewed
Sources
10
Questions
6
Compounds
3

The short answer

An epigenetic clock reads chemical tags called methylation at hundreds of spots on your DNA, usually from blood, and turns them into an age or a pace of aging. The newer clocks predict best: each extra year of GrimAge carried 10% higher risk of death in large cohorts. Pick one clock, retest in the same lab after a year, and treat a shift of a year or two as noise.

Every number checked at its primary source · · 10 sources · How we research

Evidence: Clock definitions and mortality links come from large cohort studies; the intervention results come from post hoc analyses of two randomized trials (CALERIE, DO-HEALTH) and one uncontrolled 9-person pilot (TRIIM).

A white-haired woman in a green tank top and white skirt playing pickleball with a wooden paddle

What does an epigenetic clock actually measure?

In a New Zealand birth cohort tracked through four checkups between ages 26 and 45, the slowest-aging member aged 0.40 biological years per calendar year and the fastest aged 2.446. Same birth year, roughly six times the speed. A biological age test tries to read that speed from one blood sample.

It does it through methylation: small chemical tags that sit on DNA at sites called CpGs and help switch genes on or off. The genes themselves do not change with age, but the pattern of tags does, in a way regular enough to read like a clock1. A clock is a formula, built with machine learning, that weighs a few hundred of those sites and outputs a number.

What the number means depends on what the formula was trained to predict. The first clocks were trained on calendar age. The newer ones were trained on health: blood chemistry, mortality, or how fast organs decline over decades. That one choice decides what a result can tell you.

A smiling woman in a blue top leaning on her kitchen counter with her phone and a notebook

PepVault for iPhone

Keep your protocol in PepVault

PepVault's bloodwork log keeps each panel by date; markers it does not list, such as albumin, can be added by name, so the same nine sit side by side at every draw.

Download on the App Store

7 days free on the yearly plan for new subscribers.

Horvath, Hannum, PhenoAge, GrimAge, DunedinPACE: which clock is which?

The mortality column uses one comparison where the authors could run it: in Lu's 2019 validation, each clock's age acceleration (how far it reads above calendar age) was tested against time to death in the same cohorts5. The blood-chemistry PhenoAge's figure comes from its own check in US national survey data3, and DunedinPACE was tested later, per standard deviation of pace rather than per year6.

ClockTrained onReportsDeath risk in the papers
Horvath (2013)Calendar age, 8,000 samples from 51 tissues and cell types; 353 CpGs1An age+2% per year of acceleration5
Hannum (2013)Calendar age, blood of 656 people aged 19 to 101; 71 CpGs2An age+4% per year5
PhenoAge, blood chemistry (2018)Mortality risk in 9,926 US adults, from 9 routine lab markers plus age3An age+9% per year3
DNAm PhenoAge (2018)The blood-chemistry PhenoAge above, rebuilt from 513 CpGs3An age+5% per year5
GrimAge (2019)Time to death, through DNA estimates of 7 blood proteins and smoking pack-years; 1,030 CpGs5An age+10% per year5
DunedinPACE (2022)The 20-year rate of decline in 19 organ markers in one birth cohort; 173 CpGs6A pace (1.0 = one year per year)+26% per standard deviation faster, in 771 older men6

Each clock's years are on its own scale, so Lu's team compared the clocks by P value (how sure the link is), and the winner is clear: GrimAge's P value for time to death was 2.0E-75, against 8.9E-5 for Horvath's clock in the same data5. GrimAge also tracked coronary heart disease, cancer and fatty liver on CT scans5. DunedinPACE matched GrimAge's effect sizes and added prediction of disease, disability and death beyond it6.

Horvath's clock matches calendar age within 3.6 years for half of test samples1, and that is its limit: a clock built to match your birthday says little about your health. Hannum's blood clock had an error of 3.9 years in the people it was built on and 4.9 years in a second group2.

Can a routine blood test give you a biological age?

Yes. The blood-chemistry PhenoAge needs no DNA test at all. It combines your calendar age with nine markers most standard panels already include3:

  • Albumin and alkaline phosphatase (liver)
  • Creatinine (kidney)
  • Glucose (metabolism)
  • C-reactive protein, or CRP (inflammation)
  • White blood cell count, lymphocyte percent, mean cell volume and red cell distribution width (immune and blood count)

PepVault's bloodwork log keeps each panel by date; markers it does not list, such as albumin, can be added by name, so the same nine sit side by side at every draw. Log it in PepVault

It held up outside the data it was built on. In 11,432 US adults followed for 12.6 years, an older PhenoAge predicted death after adjusting for calendar age and sex, including in people with no reported disease and a normal BMI4. People with more diseases had older PhenoAges4.

When to get help If any of the nine comes back flagged outside the lab's range, take that marker to your doctor on its own merits, whatever the age number says.

Has anything slowed a clock in a trial?

Three studies stand out. Two are randomized trials (one group gets the treatment, a control group does not, by chance), analysed after the fact in what is called a post hoc analysis; one is a small pilot with no control group.

StudyWho and whatResult
CALERIE220 adults without obesity; 25% calorie cut prescribed (11.9% achieved on average) or usual diet, for 2 years8DunedinPACE 2–3% slower at 12 and 24 months; PhenoAge and GrimAge did not change significantly8
DO-HEALTH777 adults aged 70 and over; vitamin D 2,000 IU, omega-3 1 g a day and/or home exercise, for 3 years9Against placebo, omega-3 slowed PhenoAge, GrimAge2 and DunedinPACE; all three treatments added up on PhenoAge; PhenoAge and GrimAge2 2.9–3.8 months lower at 3 years9
TRIIM9 men aged 51 to 65; growth hormone with DHEA and metformin, for 1 year, no control group10Epigenetic age 1.5 years below baseline; GrimAge 2 years lower, still lower 6 months after stopping10

CALERIE is the strongest result8. The effect grew with the dose: people who cut more than 10% of calories saw larger changes on DunedinPACE than those who cut less8. From earlier cohort data, the authors estimate that a 2–3% slower pace could mean up to 10–15% lower mortality risk, similar to quitting smoking8. That is an estimate; the trial did not count deaths.

DO-HEALTH gives omega-3 the clearest single-supplement result on these clocks so far: PhenoAge and GrimAge2 a few months younger after three years9. Vitamin D and exercise alone moved none of the clocks9. Its effect on DunedinPACE, about 1%, was smaller than calorie restriction's9.

TRIIM made the biggest claim, a reversal rather than a slowing10. It was built to regrow the thymus, the gland that trains immune T cells, and the authors themselves call for a properly powered follow-up10. Starting growth hormone or metformin is a doctor's decision.

A clock moving is a change in a biomarker, not proof of a longer life. None of these trials measured lifespan or disease as its clock outcome.

How much do results vary from one test to the next?

More than a single number suggests. When the same blood samples were run twice, Horvath's clock differed by a median of 1.8 years between the two runs and by up to 4.8 years7. Across the other clocks tested the median gap was 0.9 to 2.4 years, and the largest gaps reached 4.5 to 8.6 years7. The DNA version of PhenoAge was the noisiest: a median of 2.4 years and a worst pair 8.6 years apart7. That is lab noise alone, before any real change.

Two clocks are far steadier. GrimAge scored an intraclass correlation of 0.989 on those replicates (1.0 would mean identical runs), and its worst gap was about half the spread between people, where Horvath, Hannum and PhenoAge went past the whole spread7. DunedinPACE scored 0.96, partly because its builders left out the least reliable CpGs6. Retrained versions of the older clocks, built from principal components (a way of pooling many CpGs so single-site noise cancels), brought most replicates within 1.5 years7.

Commonly reported: two kits from different companies giving ages years apart, and a repeat test moving by a year or more with no change in habits. Both fit the replicate numbers above.

How do you get a result worth comparing?

  1. Choose a clock trained on health and steady on repeat runs, GrimAge or DunedinPACE, or a principal-component version of PhenoAge, rather than one built only to match calendar age67.
  2. Stay on that one clock; a GrimAge result and a Horvath result are different measures and cannot be compared.
  3. Use the same lab and the same sample type each time, since the clocks' error differs from one tissue to another1.
  4. Wait at least a year before retesting; CALERIE's change was there at its first check, 12 months8, and DO-HEALTH measured at 3 years9.
  5. Read any change against the noise: on the older clocks a shift under about two years can be the lab, not you7.
  6. For the blood-chemistry PhenoAge, get all nine markers from one draw, so the inputs match.

Used this way, a clock is a slow-moving dashboard number. In the trials it moved by a few months, or about two years in one 9-person pilot8910, so a promise of a ten-year drop goes beyond anything a trial has shown.

Track it in the app

Sources

  1. Horvath S. DNA methylation age of human tissues and cell types. Genome Biol. 2013. PubMed 24138928
  2. Hannum G et al. Genome-wide methylation profiles reveal quantitative views of human aging rates. Mol Cell. 2013. PubMed 23177740
  3. Levine ME et al. An epigenetic biomarker of aging for lifespan and healthspan. Aging (Albany NY). 2018. PubMed 29676998
  4. Liu Z et al. A new aging measure captures morbidity and mortality risk across diverse subpopulations from NHANES IV. PLoS Med. 2018. PubMed 30596641
  5. Lu AT et al. DNA methylation GrimAge strongly predicts lifespan and healthspan. Aging (Albany NY). 2019. PubMed 30669119
  6. Belsky DW et al. DunedinPACE, a DNA methylation biomarker of the pace of aging. eLife. 2022. PubMed 35029144
  7. Higgins-Chen AT et al. A computational solution for bolstering reliability of epigenetic clocks. Nat Aging. 2022. PubMed 36277076
  8. Waziry R 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. PubMed 37118425
  9. Bischoff-Ferrari HA et al. Individual and additive effects of vitamin D, omega-3 and exercise on DNA methylation clocks of biological aging in older adults from the DO-HEALTH trial. Nat Aging. 2025. PubMed 39900648
  10. Fahy GM et al. Reversal of epigenetic aging and immunosenescent trends in humans. Aging Cell. 2019. PubMed 31496122

Newsletter · Weekly

The weekly peptide briefing.

New guides, the week’s peptide news, one tool. Free, one email a week, unsubscribe anytime.

We’ll email you a link to confirm. Your address is used for the briefing only. Privacy

PepVault for iPhone

Keep your protocol in PepVault

Log every dose, track every vial, and see your protocol come together.

Download on the App Store

7 days free on the yearly plan for new subscribers.

Explore the live demo