What epigenetic clocks actually measure (DunedinPACE vs GrimAge vs PhenoAge)

By Marcus Renshaw ·

Biological age is the number every longevity company wants to sell you, and epigenetic clocks are how they estimate it. Send in a blood or saliva sample, and a few weeks later a report says you are aging faster or slower than the calendar. The problem is that “biological age” is not one measurement. PhenoAge, GrimAge, and DunedinPACE are the three names you will see most, they will give you three different answers, and that is not a mistake. They were trained to predict different things. If you are going to track this, it helps to know which question each clock is actually answering, and how much to trust a single test.

An epigenetic clock reads DNA methylation to estimate biological age. First-generation clocks were trained to predict chronological age. Second-generation clocks like PhenoAge and GrimAge were trained on health and mortality, so they estimate risk. DunedinPACE measures your rate of aging per year rather than a fixed age. A single reading is noisy, so trends beat one number.

What is an epigenetic clock reading?

Your DNA does not change much over your life, but the chemical tags sitting on top of it do. Methylation is the main one: small marks that switch genes louder or quieter, and their pattern shifts in predictable ways as you age. An epigenetic clock is a statistical model that reads the methylation state at a few hundred or few thousand specific sites and outputs a number.

The key thing to understand is that the number is only as meaningful as what the model was trained to predict. A clock trained to guess your birth certificate age is answering a different question from one trained to predict who dies sooner. That training target is the whole story behind why the three popular clocks disagree.

Why do PhenoAge, GrimAge, and DunedinPACE give different answers?

Because they belong to different generations, each built on a better target than the last.

First-generation clocks (Horvath, Hannum) were trained to predict chronological age. They are accurate at that and were a genuine breakthrough, but predicting the calendar is not the same as predicting health.

Second-generation clocks were trained on health outcomes instead. PhenoAge was built on nine clinical biomarkers tied to mortality, things like CRP, glucose, and white cell counts, plus age. GrimAge went further and was trained directly on time to death, incorporating methylation proxies for smoking and blood proteins, which makes it the strongest single-clock predictor of mortality risk we have.

DunedinPACE is different again. Instead of estimating a fixed age, it was trained on how fast a group of people physically declined over years of follow-up, so it reports a rate. A score of 1.0 means you are aging one biological year per calendar year; 1.2 means twenty percent faster. That makes it the natural instrument for the measure, intervene, retest loop, because rate is what an intervention is supposed to change.

Which clock should you actually look at?

It depends on the question. If you want the best single estimate of mortality risk, GrimAge is the one with the strongest track record. If you want to track whether your protocol is bending your trajectory year over year, DunedinPACE is built for exactly that, because it measures pace rather than a level. PhenoAge sits in between and has the advantage that you can partly reconstruct it from an ordinary blood panel, which makes it cheap to sanity-check.

What you should not do is treat any of these as a target to be gamed. The number is a proxy for a process, the same way a reference range is not the same as an optimal value. Driving one clock reading down without the underlying health improving is measuring the shadow, not the thing.

How reliable is a single test?

Less than the crisp printout suggests, and this is the part the reports rarely lead with. A widely cited 2022 analysis found that ordinary technical noise could produce deviations of up to nine years between two replicates of the same sample for several prominent clocks. You could look biologically 50 on one run and 59 on the next, with nothing about you having changed.

The same group offered a fix: rebuilding the clocks from principal components, which pulls the stable aging signal out of the noise and gets most repeat measurements to agree within about a year. Some testing companies now use these more reliable versions, so it is worth asking which one you are getting. The practical rule is the same one that governs how often it is worth retesting any marker: do not react to a single reading, and space retests far enough apart that a real change can outrun the noise. One clock result is a data point, not a verdict.

Can you actually move the number?

Sometimes, and the honest answer is that the evidence is still young. Epigenetic clocks are now being used to screen aging interventions because they read out faster than waiting decades for mortality data, and a large 2024 effort harmonized dozens of human intervention studies across sixteen clocks to see what actually shifts them. Some interventions moved the clocks; some did not; and, as the rapamycin reversal showed, a few favorites came up short precisely when the community started grading them this way. That is the instrument doing its job. A clock that only ever confirmed your hopes would be useless.

So the reasonable stance is to use these clocks as a slow-moving trend line, not a scoreboard. If a change in your protocol lines up with a genuine shift in pace across a couple of well-spaced, reliable tests, that is a signal worth weighing alongside your bloodwork and your training. A single dramatic before-and-after is almost always noise.

How I use it

I have tracked PhenoAge, which is a static snapshot: useful as a periodic sanity check against my clinical markers, but it tells me where I am, not which direction I am moving. The reason I am eyeing DunedinPACE is that rate is the better instrument for evaluating a protocol year over year, and year over year is the timescale this whole approach runs on. I treat any single result as one noisy reading, retest on a long cadence, and only care about the trend. The clock is a supporting marker, the same way a resting heart rate or an ApoB is, not the thing I am optimizing in isolation.

If you want to start tracking biological aging without paying for a clock first, begin with the cheap markers you can measure at home and build the logging habit. One number, recorded properly over time, teaches you more about your own trajectory than a single expensive test does, and the free BP tracker is a no-cost way to start that discipline.

FAQ

What does an epigenetic clock actually measure?

It reads DNA methylation, the chemical tags that sit on your DNA and shift in predictable patterns as you age, at a set of specific sites, and runs them through a statistical model to output a number. What that number means depends entirely on what the model was trained to predict, whether chronological age, mortality risk, or the rate of aging.

What is the difference between PhenoAge, GrimAge, and DunedinPACE?

They were trained on different targets. PhenoAge was built on clinical biomarkers linked to mortality. GrimAge was trained directly on time to death and is the strongest single predictor of mortality risk. DunedinPACE was trained on how fast people physically declined over years, so it reports a rate of aging rather than a fixed age. Different jobs, different numbers.

Which epigenetic clock is the most accurate?

There is no single most accurate clock, because accuracy depends on the question. GrimAge is best validated for predicting mortality risk. DunedinPACE is best for tracking whether an intervention is changing your rate of aging over time. First-generation clocks like Horvath predict chronological age well but say little about health.

How reliable is a single epigenetic age test?

Not very, on its own. Technical noise can produce differences of up to nine years between two runs of the same sample for several clocks. Newer principal-component versions of the clocks cut that noise so most repeats agree within about a year. Either way, you should track trends across well-spaced tests rather than reacting to one reading.

Can you lower your epigenetic age?

Possibly, but the evidence is still early. Clocks are now used to screen aging interventions, and studies across many clocks show some interventions shift them and others do not. Treat a clock as a slow trend line: a real change should show up across a couple of reliable, well-spaced tests, not as a single dramatic before-and-after, which is usually noise.

Is DunedinPACE better than PhenoAge for tracking a protocol?

For tracking, generally yes, because DunedinPACE measures your rate of aging per year, which is exactly what an intervention is meant to change. PhenoAge gives a static snapshot of where you are, which is useful as a periodic check but harder to read as a trajectory. Many people use PhenoAge as a cheap sanity check and DunedinPACE for year-over-year tracking.

Sources