What biological age tests can and cannot tell you

A number that sounds like a fact

"Your biological age is 42." It is a compelling sentence, it fits on a dashboard, and it is the most commercially successful idea in longevity medicine.

It is also an estimate produced by a statistical model, and what the model was trained to predict determines entirely what the number means.

What the main approaches actually are

• Epigenetic clocks. The best‑developed. They measure DNA methylation — chemical marks on DNA — at selected sites, and combine them into an age estimate. The first generation (Horvath, Hannum) were trained to predict chronological age, which means a perfect score would tell you only what your passport says. Later ones (PhenoAge, GrimAge, DunedinPACE) were trained on mortality, on clinical measures, or on the rate of change itself, and those predict health outcomes better.
• Telomere length. The oldest idea and among the weakest in practice. Measurement is highly variable between laboratories, and individual telomere length is a poor predictor of anything at the individual level.
• Composite clinical scores. Built from ordinary blood markers — albumin, creatinine, glucose, CRP, white cell counts, and so on. Unglamorous, cheap, and in several comparisons they hold their own against far more expensive tests.

The three questions worth asking

What was it trained to predict? A clock trained on chronological age is trained to agree with your birthday. A clock trained on mortality is at least aimed at something that matters. Vendors rarely volunteer which.

How reproducible is it? This is the weakest point. Studies that ran the same sample twice have found test‑retest variation of a magnitude that can swallow the differences people are being sold — several years, in some cases. If splitting one blood tube gives two answers years apart, a change of two years after six months of effort means nothing.

Does it respond to anything? Some clocks shift with interventions in small trials. Whether that shift corresponds to any real change in ageing, or simply to the clock's inputs moving, is unresolved.

What the field's own researchers say

This is the part that rarely appears in marketing. Several of the scientists who built these clocks have publicly cautioned against direct‑to‑consumer use, on the grounds that the tests are validated for comparing groups in research, not for telling an individual how fast they are ageing.

A measure can be genuinely useful across a thousand people and nearly useless for one of them. That is not a flaw in the science; it is a misuse of it.

What the number can do to you

Two failure modes, both real:

• A reassuring result in someone with a genuinely bad metabolic panel. The clock says 38, the HbA1c says pre‑diabetes, and the clock is the one that gets believed.
• An alarming result in someone healthy, which produces anxiety and a receptiveness to whatever is sold next. That sequence is not accidental in some business models.

Where we stand

We do not offer biological age testing, and the reason is the same question that governs every line of our panel: would the result change what we do next?

It would not. If your HbA1c, lipids, liver and kidney function are good, a clock saying 48 changes nothing. If they are bad, a clock saying 38 should not reassure anybody. The actionable information is in the ordinary markers, and those cost less.

That is a position, not a proof. The field is moving, the second‑generation clocks are better than the first, and this may look unduly conservative in five years. It may also look prudent.

If you are buying one anyway

1. Ask which clock, and what it was trained on.
2. Ask for the test‑retest reproducibility figure. A vendor who cannot supply it has not asked.
3. Do not use it to decide anything your blood panel disagrees with.
4. Do not retest at short intervals. The change will be smaller than the noise.

The boundary

None of this says the underlying science is bad — it is among the more interesting work in biology. It says that the distance between a good research tool and a consumer product with a number on it is substantial, and that the number is not yet something to make decisions on.