Your Biological Age Test Is Less Precise Than You Think

September 9, 2026; By rarebluemoon

A patient sits down with a printout. It says their biological age is 47. Their driver’s license says 54. They are pleased, and they want to know what we should do to keep that number moving.

I have had this conversation many times, and I have learned to slow it down. Not because the science behind these tests is nonsense. It is not. But because the number on that page carries far more uncertainty than the confident typography suggests, and how you interpret it determines whether the test was worth doing at all.

What these tests are actually measuring

The modern versions rest on DNA methylation. Methylation is a chemical modification of DNA that shifts in patterned ways across the lifespan, and those patterns can be read from a blood sample.

In 2013, Steve Horvath published the work that opened the field. He built a predictor using thousands of samples spanning dozens of human tissues and cell types, and it estimated chronological age with a correlation of 0.97 and a median error of about 2.9 years.

That is a remarkable technical achievement. It is also worth noticing what it was trained to do: predict how many birthdays you have had. The first-generation clocks were validated against the calendar, not against how well you are aging.

That distinction gets lost in marketing. When a clock says you are seven years younger than your chronological age, the honest reading is that your methylation pattern resembles that of people seven years younger. Whether that means your heart, your kidneys, and your brain are aging more slowly is a separate question the original clock was not built to answer.

The reliability problem almost nobody mentions

Here is the finding that changed how I discuss these tests.

In 2022, a team including researchers at Yale, the National Institute on Aging, and Columbia examined the test-retest reliability of six prominent epigenetic clocks. Technical noise alone produced deviations of up to nine years between replicates.

Read that again. Not nine years between two samples drawn a year apart. Nine years between replicate measurements of the same biological material, attributable to noise in the measurement itself.

The lead author put it plainly in interviews at the time: a clock could report that you are biologically 50 on one test and 59 on the next. The team went on to develop a principal component method that brought most replicate pairs within about a year and a half of each other, which is a genuine improvement and is not yet universally adopted across commercial offerings.

So when a patient shows me a single biological age number and asks what it means, my first question is which assay produced it and whether the provider has published reliability data. Many cannot answer.

Why a single number is the wrong unit

This is also why I am skeptical of any protocol built around chasing that number downward. If the measurement error is larger than the change you are trying to produce, you are not tracking your health. You are tracking noise, and you will interpret random fluctuation as evidence that whatever you happened to be doing worked.

What the better measures do differently

The field has not ignored this. Later work moved toward measuring the rate of aging rather than taking a single snapshot.

DunedinPACE is the clearest example. Rather than training against chronological age, its developers used two decades of data from a New Zealand birth cohort, tracking within-individual decline across 19 indicators of organ system integrity, then distilled that into a single blood test. Critically, they built it from a methylation dataset that deliberately excluded probes with low test-retest reliability, and the resulting measure showed high reliability and associations with morbidity, disability, and mortality.

That is a meaningfully different product. It is attempting to estimate how fast you are aging, using inputs chosen partly for their stability.

It is still a research instrument being applied in clinical settings, and it does not tell an individual patient what to do on a Tuesday. But it represents the direction of travel, and it illustrates the standard worth applying: was this measure validated against health outcomes, and is it stable enough to measure twice?

So should you get tested?

My answer is a qualified yes, with conditions that matter more than the test itself.

• Treat it as one input, never a verdict. A biological age result belongs alongside your lipids, your metabolic markers, your blood pressure, and your functional capacity. It does not outrank them.
• Do not compare across providers. Different assays and different algorithms produce different numbers. A result from one company and a result from another are not the same measurement.
• Retest on the same platform, and not too often. Given documented measurement noise, short-interval retesting mostly generates confusion. Annual, same-assay, same-lab comparison is far more informative.
• Understand the regulatory status. These are generally laboratory-developed tests, not FDA-cleared diagnostics. That is not a reason to dismiss them. It is a reason not to treat the output as a clinical diagnosis.
• Ask what would change based on the result. If the honest answer is nothing, the test is entertainment. Expensive entertainment.

What I actually use to guide decisions

When a patient comes to me wanting to age well, methylation testing is rarely where I start. It is where I sometimes finish, as a supplementary data point.

The measures I lean on hardest are unglamorous and heavily validated. ApoB and lipoprotein(a) for cardiovascular risk. Fasting insulin and hemoglobin A1c for metabolic trajectory. High-sensitivity CRP for inflammatory burden. Blood pressure, measured correctly and more than once. Grip strength, gait speed, and cardiorespiratory fitness, all of which track function rather than inferring it.

None of those produce a satisfying single number to put on a wellness report. All of them have decades of outcome data behind them, and all of them respond to interventions we understand.

That is the logic behind how we structure a longevity and wellness assessment and why the results feed into a longevity roadmap with scheduled reassessment rather than a one-time report. The value is in the trend line and in changing one variable at a time, not in the first data point.

Where this leaves you

Epigenetic aging research is among the more exciting areas in medicine, and I expect these measures to become genuinely useful clinical tools. Some are close now.

But the gap between the research literature and the consumer product is wide, and it is filled largely by confident marketing. The literature is candid about measurement noise and about what first-generation clocks were and were not trained to do. The marketing rarely mentions either.

If you want a number to be proud of, buy the test. If you want to age well, get the boring markers measured, measure them again in a year, and change one thing at a time. The second approach is less exciting. It is also the one supported by evidence that has been around long enough to trust.