How Often Should You Test Biological Age?

One vendor recommends retesting in three months. Larger trials that detected group-level changes measured participants two or three years apart. Neither interval has been validated as a routine schedule for tracking one person. People retest because they want to know whether what they’re doing to live longer and better is showing up in the number. It’s a little like trying to spot a glacier moving from the kitchen window: the real question is whether the movement is larger than the measurement error. Humanic Blue is examining that gap between a changed number and a meaningful change.

What we found starts with a blank space: no medical body sets a schedule for this. The US Preventive Services Task Force’s A and B recommendations contain no entry for biological age, epigenetic testing or aging clocks, and the NHS Health Check doesn’t include one.

Measurement error helps explain why repeating a test may not answer the question. A small difference can reflect the test’s variation rather than a lasting change in the body. But error alone cannot tell us how long to wait: that also requires evidence about real changes over time and whether tracking them improves care.

Published error on standard epigenetic clocks runs up to 9 years between replicates, and principal-component versions brought most replicate differences within about 1.5 years. That is technical variation in the test, not a recommended waiting period.

Commercial suggestions range from months to a year. The larger intervention trials measured group effects after two to three years. Those are different kinds of evidence, and neither supplies a clinically validated personal retest interval.

Is there an official retest schedule for biological age tests?

No. We looked for one where schedules usually live.

The USPSTF, which sets US preventive-screening intervals, lists hypertension screening (Grade A, 2021) and statin use (Grade B, 2022) but nothing on biological age or DNA methylation.

In England, the NHS says “you should be invited to an NHS Health Check by your GP or local council every 5 years” if you’re 40 to 74; the check “usually includes: measuring your height and weight, measuring your waist, a blood pressure test, a cholesterol test, and possibly a blood sugar level test.” No clock.

That absence matters most if you’re new to this. An epigenetic clock is a statistical model derived from chemical marks on DNA; it is not included in the preventive-screening schedules compared here.

Idan Shalev, Associate Professor of Biobehavioral Health at Penn State, and Abner Apsley, a postdoctoral researcher at the University of Illinois Urbana-Champaign, put it plainly in The Conversation in May 2026: these tests “do not meet the standards required of common medical tests.” Two researchers who study these clocks are saying the clocks aren’t held to the bar a medical test has to clear.

A screening interval is a promise that a test is worth repeating on a schedule. Nobody has made that promise here.

How big is the test-retest error on an epigenetic clock?

This is the number the whole decision hangs on, and vendors describe it in incompatible ways.

A 2022 Yale-led study ran the same blood through six clocks twice. Older clocks differed by as much as nine years; redesigned principal-component clocks brought most repeat results within about 1.5 years. DunedinPACE also repeated tightly when the same sample used the same laboratory platform, but agreement weakened when the testing chip changed. Same blood, different machinery, slightly different answer—the laboratory version of moving the goalposts while insisting they are calibrated.

Commercial relationships sit beside those results. Yale licenses the redesigned metrics to Elysium, which supplied some study data, and Morgan Levine formerly advised the company. DunedinPACE inventor Daniel Belsky occasionally advises TruDiagnostic, its exclusive licensee.

TruDiagnostic recommends testing every three months and advertises 98% reproducibility. Elysium reports agreement within zero to 1.5 years; Levine once suggested annual or twice-yearly testing. A percentage and a number of years cannot be compared directly, which is precisely the information a personal schedule would need. The exact replicate data, platforms, disclosures and vendor claims

How long did the trials that actually detected change wait?

The four often-cited trials used four intervals and four measures. An eight-week pilot in about 40 men, in which participants knew their group assignment, reported a 3.23-year difference on an older saliva clock—but that change fits inside the clock’s published nine-year replicate error, the work has not been replicated, and two authors use or earn from the program.

A three-month fasting-mimicking-diet analysis reported a change in a blood-chemistry score, not an epigenetic clock; founder Valter Longo holds an interest in the company supplying the diet. The larger methylation studies waited much longer. CALERIE measured adults over two years and moved DunedinPACE by 2–3%, while two other clocks stayed put. DO-HEALTH measured about 780 older adults at baseline and three years and found differences of only three to four months on selected clocks.

Those studies establish that group-level changes can be researched over years. They do not turn two or three years into a personal testing schedule—or eight weeks into one because the headline has better cheekbones. The full trial designs, numbers and conflicts

How much can a research trial's measurement interval tell us about consumer retesting?

What can the trial intervals tell us?

Set the numbers side by side. Most technical replicate differences on the tested principal-component clocks were within about 1.5 years. CALERIE reported a 2–3% group effect on pace of aging after two years, while DO-HEALTH reported group differences of several months on selected clocks after three years. These quantities are not directly interchangeable: one is replicate error, one is a rate and one is an age estimate.

The intervals come from different sources:

• 3–6 months — a schedule promoted by TruDiagnostic. A vendor suggestion does not establish that this interval reliably tracks individual aging.
• 6–12 months — Levine’s “annually or bi-annually” for Elysium. The comparison studies here do not validate this interval for individuals. A year passing on the calendar cannot be treated as a known one-year change on every biological-age clock.
• 2–3 years — the interval at which CALERIE and DO-HEALTH detected group effects in hundreds of people. That does not establish two to three years as a clinical schedule for an individual.

Christopher Hine, PhD, is a principal investigator at Cleveland Clinic Research.

He gave NBC News the plain version in April 2026: “Only doing it once isn’t so telling, because you don’t know if you’re [on] an upward trajectory or downward trajectory in terms of the rate of your aging.” His point is about direction: a single dot on a chart isn’t a line. That observation does not establish which interval can reliably detect a meaningful change in one person.

Matt Kaeberlein, PhD, CEO of Optispan and Affiliate Professor at the University of Washington (disclosure: Optispan sells health-optimisation programs), wrote in March 2026 that a diagnostic must show its precision, its accuracy and its use in guiding care, and “Epigenetic age tests fail on all three counts.” The precision point is the one this article is about.

What information is missing before a clock's error range can support a retest schedule?

What can shift a single reading without you aging?

Several things, many of them absent from the kit card.

• Recent illness: small COVID-19 studies found clock changes during or after serious infection and declines during recovery. The evidence is preliminary, but it shows that recovery can look like age reversal.
• Sample type: saliva and blood contain different cells. A saliva result and a blood-spot result are not interchangeable just because both arrive as a number with “years” beside it.
• Cell mix and smoking: blood clocks read immune cells, and results can shift as that mix changes. Some clocks also build smoking history into the model.
• Laboratory platform: agreement fell when identical samples were run on different methylation chips. Technical reliability and the stability of a living person across two days are separate problems.

We did not find controlled evidence that ordinary short sleep, vaccination or alcohol changes one clock reading, so those common claims are not being smuggled in through the side door. The studies and platform results

How does this compare with the bloodwork your doctor actually schedules?

For perspective, it helps to look at tests that do come with a schedule, the unglamorous ones nobody sells in a nice box. US guidance suggests annual blood-pressure screening from age 40, with longer intervals for younger adults at lower risk. England’s NHS Health Check runs every five years from 40 to 74. Lipids are checked sooner after starting or adjusting a statin because the result guides treatment. Epigenetic biological age has no guideline interval. The source comparison is below.

These schedules have a defined purpose: detecting a treatable condition or checking a response to treatment. The epigenetic-clock studies discussed here do not establish an equivalent clinical use or retest schedule.

What’s the case for testing rarely, or never?

Repeated commercial testing means repeated expense without a validated interval. Cardiologist Eric Topol has advised consumers not to buy current clocks because standardization and reliability remain unsettled. Northwestern longevity director Douglas Vaughan called epigenetic age testing not ready for general consumers, while DunedinPACE developer Belsky described the clocks as research tools that may interest curious people.

The deeper problem survives even if a clock moves: no trial has shown that a person who lowers one lives longer or stays healthier because of it. A changing biomarker is not automatically a changed outcome. Sometimes the most scientifically responsible testing schedule is “not until the test can answer a decision you actually need to make.”

How do you make two tests comparable?

Several differences can complicate a comparison. Knowing them can explain why two results differ; it does not create a validated personal testing protocol.

• Same vendor, same product. Different companies run different algorithms on different samples; a TruAge number and an Index number aren’t on the same scale.
• Same sample type. Blood spot to blood spot, saliva to saliva. Horvath 2013 and the DunedinPACE methods above are why.
• Same array and clock version. Belsky’s 450k-to-EPIC comparison lost reliability; DO-HEALTH used PC versions of clocks that most consumer reports don’t name. Ask which version you got and whether it changed between orders.
• Illness and recovery can affect the score. Calzari found acceleration six months after infection. That does not establish a waiting period after an acute illness or tell an individual when a clock result becomes interpretable.
• Ask for the error. TruDiagnostic states “98% reproducibility,” Elysium states “zero to 1.5 years.” The error needs units that match the reported result: years for an age estimate, or rate units for a pace estimate.
• Pace and age have different units. Levine’s advice was to watch “your cumulative rate of aging”; Belsky’s DunedinPACE is designed as a rate. The trials measured group-level clock changes. They did not validate either a clock choice or a retest schedule for monitoring one person.

The protocols and vendor pages reviewed here do not establish a standard time of day or fasting procedure for consumer retesting. The collecting laboratory can explain its own preparation instructions; these studies do not supply a universal protocol.