Why aging clocks disagree

In a two-year trial, two groups of healthy adults were compared: those who cut their calories and a group who ate as usual, and everyone's blood was tested with three DNA-based aging clocks. On one clock, DunedinPACE, the diet group's pace of aging slowed slightly. On the other two, PhenoAge and GrimAge, the groups did not differ.^[1]

All three clocks read the same blood samples, and none of the three was broken. Each clock was taught to predict something different, and that choice decides what its number means. The same applies when your own report lists several clocks: the numbers are not rival guesses at one true age, so neither the highest nor the most flattering one is the “real” result. Once you know what a clock was taught, you can tell which one answers your own question. If you are new to the idea of a biological age, start with what is biological age?

What does an aging clock learn?

The clocks on this page start from DNA methylation: chemical tags at hundreds of thousands of places on your DNA, some of which shift with age. Researchers choose a target, a number they want to predict. A statistical method then selects places on the DNA, and how much each counts, that best predict the target in a training group of people.^[2][3] The finished clock is that formula. Applied to your sample, it returns its estimate of the target it learned.

The target is the most important fact about a clock. The three families of clocks use three kinds of target, and those targets sit in different places in time.^[2][3][4][5]

What does an age-like clock tell you?

The first widely used clocks were taught calendar age. Steve Horvath's 2013 clock was built from about 8,000 samples of 51 healthy tissues and cell types.^[2] Gregory Hannum's clock, from the same year, was built on blood from 656 people aged 19 to 101.^[6] Their output is an age in years, and the interesting part is the gap: a result of 47 for someone who is 50.

The design has a catch. A clock that predicted calendar age perfectly would carry no information about aging beyond the calendar age you already know.^[7] The gap is the part of the pattern that calendar age does not explain, and a clock built to match the calendar was not designed to capture health. In the PhenoAge developers' comparisons, age-like clocks predicted later death less well than PhenoAge did.^[3] Researchers call the gap “age acceleration”, but it is a difference at one moment, not a speed.

What does a risk-like clock tell you?

Risk-like clocks were taught a target tied to health. PhenoAge was built in two steps. First, researchers used a large US health survey to combine calendar age and nine routine blood measures, such as glucose, albumin and C-reactive protein, into a score weighted by how well they predicted death. Then they taught a DNA clock to predict that score from blood taken at the same visit, in an Italian study.^[3]

GrimAge also used two steps. It first built DNA-based estimates of seven blood proteins and of lifetime smoking. It then combined them with age and sex into a score trained on how soon people died in a long-running US heart study, and rescaled that score into years. In the developers' own tests, it predicted time to death better than earlier DNA clocks.^[4] A second version adds estimates of two more blood measures, C-reactive protein and the blood-sugar marker HbA1c, so “GrimAge” on two reports may not be the same formula.^[8]

Both report years, but those years are a risk scale. A higher number means a higher estimated risk in the groups the clock was built on. It is not a count of years left.^[3][4]

What does a pace clock tell you?

DunedinPACE comes from a study that has followed 1,037 children born in Dunedin, New Zealand, in 1972 and 1973. At ages 26, 32, 38 and 45, researchers measured 19 markers of the heart, metabolism, kidneys, liver, immune system, teeth and lungs, and worked out how fast each person's markers had changed. They then taught a DNA clock to predict that speed from one blood sample taken at 45.^[5]

The result is a pace, not an age. A pace of 1.0 means one year of biological change per calendar year, the average in that New Zealand group, so 1.1 would be about 10% faster than that average.^[5] Because the clock learned the speed of the previous two decades, it estimates a recent pace from a single sample. It cannot say how fast you aged at 20, or how fast you will age at 70.

A pace learned between 26 and 45 may not describe later decades. One study that compared people of different ages reported bursts of molecular change around 44 and 60, but in 2026 the journal questioned the reliability of its peak analyses.^[9]

Why does one person get different numbers?

Put the three targets side by side and disagreement is expected. Someone whose DNA tags look typical for their age but who has smoked for years may get an age-like result near their calendar age and a higher GrimAge, because GrimAge includes an estimate of smoking.^[4] Each number answers its own question, and the units differ too: an age in years cannot be set beside a pace such as 1.08.

The trial in the opening, called CALERIE, shows the same thing across a group. Its 220 healthy adults aged 21 to 50 were randomly assigned to a diet group, asked to eat a quarter fewer calories (they managed about 12% on average), or to eat as usual. In this one trial, DunedinPACE was slower in the diet group by about 2 to 3%. PhenoAge and GrimAge, in versions built to give steadier repeat readings, did not differ between the groups.^[1]

A clock that stays still is not proof that nothing happened. The trial ended when the diet did, so it could not show whether the slower pace meant less illness later. The average shift was also small next to the differences between people: the researchers estimated that the two groups' DunedinPACE changes overlapped by close to 90%.^[1]

Can two tests disagree for other reasons?

Yes. Different targets are only one cause. When the same blood sample was measured twice, the original versions of widely used clocks, including Horvath, Hannum and PhenoAge, differed noticeably, sometimes by several years; versions built to reduce noise agreed much more closely.^[10] Several clocks built on blood gave much higher ages from saliva or cheek swabs of the same people.^[11] Blood samples taken hours or days apart varied more than one sample measured twice.^[12] Running the same DNA on an older and a newer version of the lab chip that reads the tags shifted some clocks by several years.^[13]

Most of this research used small groups in research labs; comparing commercial tests on the same people is the next step. Until then, results are easiest to compare when they come from the same test and the same type of sample. See how to compare two results and the limitations of aging clocks.

Which kind answers your question?

Start from what you want to know. A report's fine print or the provider's documentation should name the clock behind a number.

To see which clocks each consumer test reports, use Compare; the short quiz suggests a kind of test to start with.

Sources

Peer-reviewed sources for the scientific statements above, with the commercial ties their authors declare.

1. Waziry et al. (2023). Effect of long-term caloric restriction on DNA methylation measures of biological aging in healthy adults from the CALERIE trial. Nature Aging. Two authors are named DunedinPACE inventors; the declaration also reports commercial licensing. A later correction changed only an author affiliation.
2. Horvath (2013). DNA methylation age of human tissues and cell types. Genome Biology. The author is a named inventor on a university patent application.
3. Levine et al. (2018). An epigenetic biomarker of aging for lifespan and healthspan. Aging. Two authors are named inventors on a university patent application.
4. Lu et al. (2019). DNA methylation GrimAge strongly predicts lifespan and healthspan. Aging. Written by the clock's developers; two authors are named inventors on a university patent application.
5. Belsky et al. (2022). DunedinPACE, a DNA methylation biomarker of the pace of aging. eLife. Several authors are named inventors of DunedinPACE, which the paper says is licensed to a commercial company.
6. Hannum et al. (2013). Genome-wide methylation profiles reveal quantitative views of human aging rates. Molecular Cell. Blood samples from 656 people aged 19 to 101.
7. Sluiskes et al. (2024). Clarifying the biological and statistical assumptions of cross-sectional biological age predictors: an elaborate illustration using synthetic and real data. BMC Medical Research Methodology. Statistical analysis and simulation by an academic group; no competing interests declared.
8. Lu et al. (2022). DNA methylation GrimAge version 2. Aging. Written by the clock's developers; two are named inventors on a university patent application, and one co-founded the non-profit foundation that licenses it.
9. Shen et al. (2024). Nonlinear dynamics of multi-omics profiles during human aging. Nature Aging. 108 people aged 25 to 75 followed for a median of 1.7 years; the age patterns compare people of different ages. In July 2026 the journal added an Editor's Note that the reliability of the analyses used to detect age-related peaks is under review.
10. Higgins-Chen et al. (2022). A computational solution for bolstering reliability of epigenetic clocks. Nature Aging. Two authors built clock metrics licensed to a test company.
11. Apsley et al. (2025). Cross-tissue comparison of epigenetic aging clocks in humans. Aging Cell. Two co-authors are DunedinPACE inventors (licensed to a test company), although the paper declares no conflicts.
12. Sehgal et al. (2026). Biological versus technical reliability of epigenetic clocks and implications for disease prognosis and intervention response. Aging Cell. Two authors are co-inventors of another clock and have consulted for test companies.
13. Tay et al. (2025). DNA methylation age differences between an older and a newer version of the lab chip. Communications Biology. Two authors co-founded a non-profit foundation that licenses clock patents; another advised a longevity company.