Epigenetic Testing: What It Measures, and What It Does Not

What it measures

The first widely used clock was published in 2013. It estimated age from 353 methylation sites, built using 8,000 samples across 51 human tissues and cell types (Horvath, 2013). It was trained to do one thing: predict chronological age — the number on your ID.

The next generation changed the target. Rather than predicting the calendar, it was trained on clinical measures tied to lifespan and healthspan, and it predicts all-cause mortality, cancer risk, physical functioning and healthspan better than the earlier version (Levine et al., 2018).

A third approach does not estimate an age at all. It estimates how fast you are ageing, built from two decades of tracking nineteen indicators of organ-system function in the same people over time (Belsky et al., 2022). It answers “at what speed”, not “what age”.

Here is the practical consequence, and it is the most useful thing on this page. These tests were trained to predict different things, so they answer different questions and do not have to agree with each other. A report that hands you “your biological age” without naming the clock behind it is handing you an answer without the question.

What it does not measure

It is not a diagnosis. It does not detect disease, does not identify a cause, and does not replace any clinical test your doctor would order.

There is also a reliability problem worth knowing before you pay for one. Run the same sample twice and earlier clocks could return different numbers. The field responded by building versions specifically designed to reduce technical noise and improve test-retest reliability (Higgins-Chen et al., 2022), and the pace-of-ageing measure was built by deliberately excluding the methylation sites with poor reliability (Belsky et al., 2022).

If your plan is to repeat the test to track progress, that is the single most important question to ask: which version is being used, and how reproducible is it.

It does not tell you what to do either. These measures were built from associations across large populations, which is not the same as an instruction for one person.

And the point most often skipped: improving the number has not been shown to improve health outcomes. These clocks were validated as predictors, not as treatment targets. Moving a score is not the same as living longer or better, and no published work has closed that gap.

How to use the result

Results depend on the tissue sampled, the cell composition of that sample, the laboratory, and the version of the clock used. Comparing a result from one provider against a result from another is frequently meaningless.

Used sensibly, an epigenetic test is one measurement among several — repeated in the same laboratory, with the same method, and read by someone who knows which clock produced it. Used as a scoreboard, it will mostly tell you about the noise.