Skip to main content

Blood-panel and wearable “biological age” versus lab aging clocks

Blood-panel memberships and wearable apps now show a biological age. What each number is built from, what it was checked against, and what it cannot tell you, compared with a lab aging clock.

10 min readEdited by Illes Muller
Contents

Three kinds of age number at a glance

The same words, “biological age”, are used for three different kinds of number. This page is organized by method, not by company, and does not rank products.

Blood-panel age

e.g. Phenotypic Age, Klemera–Doubal

Built from
Routine lab markers from a standard blood draw (nine in Phenotypic Age; other versions use other sets) plus your calendar age
Trained or checked against
Phenotypic Age: markers chosen because together they predicted death in a large US health survey, then checked in a later round. Klemera–Doubal: a statistical method that uses calendar age as one input
What it can’t tell you
Whether lowering the number lowers your risk; how much it moves between two draws days or weeks apart (no study found); how well a company’s own version works (not reported on the pages we checked)

Wearable age or pace

from a watch, ring or band

Built from
Watch or ring sensor data (heart rate, heart rate variability, estimated VO2max, sleep, activity), sometimes with age, BMI or lab values
Trained or checked against
Research models: death risk or calendar age in large studies, mostly tested by their own developers. Consumer features: methods only partly disclosed, with no published test against health outcomes found
What it can’t tell you
Whether your number predicts your health; it carries the error of the sensor estimates, and we found no published comparison with a DNA methylation clock

DNA methylation clock

e.g. Horvath, GrimAge, DunedinPACE

Built from
Chemical tags on DNA, usually from a blood sample
Trained or checked against
Varies by clock: calendar age, a blood-panel score, time to death, or 20-year decline in one birth cohort
What it can’t tell you
Your individual risk; a steady reading after emergency surgery, in late pregnancy or in severe illness, when some clocks rise for a while; which clock is “right” when they disagree

Blood-panel ages

A blood-panel age turns routine lab results into a single number in years. The best-known recipe is Phenotypic Age, often called PhenoAge. It combines your calendar age with nine blood measures. They are albumin, creatinine, glucose, C-reactive protein (a marker of inflammation), the share of lymphocytes, red cell size, the variation in red cell size, alkaline phosphatase and white cell count. Researchers chose these markers, and how much each one counts, because together they predicted who died during follow-up in a large US health survey. The same team then checked the score in a later, separate round of that survey. Among 11,432 adults, a higher score went with a higher risk of death, even at the same calendar age. An older recipe, the Klemera–Doubal method, combines several markers and uses calendar age itself as one of its inputs.

Calendar age is one of the inputs, and in the developers’ validation data Phenotypic Age tracked calendar age closely (a correlation of 0.94, where 1 would be a perfect match). The extra information Phenotypic Age adds beyond calendar age is real but small. In the developers’ check, the score predicted death within ten years only slightly better than calendar age alone. In an independent study of about 115,000 UK Biobank participants, almost all of them men, adding the score to standard cardiovascular risk factors barely improved prediction of heart attack, stroke and related disease, and the standard Framingham risk score did better. Even so, a 2024 review by the Biomarkers of Aging Consortium lists Phenotypic Age among blood-based measures “validated in cross-population studies”: its link with death was confirmed in a separate group of people followed for at least ten years. So the link with death holds up in large groups, but the score adds little to what calendar age and routine risk factors already tell you.

Some of the inputs move a lot from week to week. In healthy people, C-reactive protein typically varies by about a third around a person’s own usual level, while albumin varies by about 2.5%, according to the EFLM Biological Variation Database, a reference kept by European laboratory-medicine specialists. So a single blood draw may catch some of these markers above or below your usual level. In our searches of PubMed in November 2026, we found no published study that measured how much a blood-panel age itself changes between two draws a few days or weeks apart.

Several companies offer a blood-panel age. Function Health says its biological age “calculates mortality risk and phenotypic age based on various lab test results” (checked November 2026). InsideTracker says InnerAge 2.0 builds on the Klemera–Doubal method “in combination with InsideTracker’s own blood biomarker dataset” (checked November 2026). Its developer documentation says the score also uses body mass index. Superpower says its clinical team turns its results “into a PhenoAge estimate” (checked November 2026). The studies above tested the published recipes, not any company’s own version. On the pages we checked, none of the three companies reported a study of its own number against health outcomes, or a figure for how much that number varies between repeat tests.

Wearable ages and paces

A sleeping person’s wrist on a white duvet, wearing a fabric band with a faint green sensor light.
Wearable ages are built from watch or ring sensor data, such as heart rate, sleep and activity.

Some wearable “ages” come from research labs. Scientists have built age scores from motion sensors and from the light-based pulse sensor used in smartwatches.

Each model was tested mainly by the team that built it. As of 3 November 2026, we found no independent group that had re-tested any of them.

Fitness itself is strongly linked with survival. In 122,007 patients sent for treadmill tests at one US hospital, the least fit quarter had about five times the adjusted risk of death of the top 2 to 3 percent. In that study, fitness was estimated from treadmill performance, not measured from breathing gases. A Norwegian study of about 37,000 healthy adults followed for 24 years found a similar link, using a fitness estimate that needed no exercise test. The link held in people under 60 at the start. These are observational studies: they link fitness with survival, but they do not prove cause and effect. They also do not test any company’s age formula or a watch’s fitness estimate.

The inputs have error. A review of 14 studies concluded that, for watch estimates of VO2max, “the estimation error at the individual level is large”. In a study of 13 adults, overnight heart rate and heart rate variability agreed with an ECG better on some devices than others. The same researchers later noted that companies do not fully publish how they calculate these numbers, which makes like-for-like checks harder. In a one-night study of 62 adults wearing six wrist devices, sleep-stage tracking agreed with a lab sleep study only fairly to moderately.

As of 3 November 2026, we found no peer-reviewed study showing that any company’s wearable age or pace predicts death or disease. One company published a paper on its own wearable fitness age. All the authors worked for the company, and the paper did not compare it with lab-measured fitness or health outcomes. We also found no study comparing a wearable age with a DNA methylation clock in the same people. The closest compared individual smartwatch measures, such as daily steps and resting heart rate, with DunedinPACE in 621 adults aged 50 to 79. The strongest correlation was about 0.3, and no combined wearable age was calculated.

What the wearable companies say

Checked November 2026. What each company says below is its own statement, not a finding we confirmed. The other sentences are our notes on what those pages do and do not show.

None of these pages, as we read them, reports how much the number varies on repeat measurement, or tests it against health outcomes. That leaves aside Whoop’s white paper, which we could not open.

Lab DNA methylation clocks

Gloved fingers hold a glass microarray slide with four shimmering sample areas in front of an open lab scanner.
Epigenetic tests read DNA methylation, chemical tags on DNA, often with a microarray slide like this one.

DNA methylation clocks read chemical tags on DNA, usually from blood, that shift with age. Each clock was trained on a different target. The original Horvath clock was trained to predict calendar age. DNAm PhenoAge was trained to predict the blood-panel Phenotypic Age score. GrimAge was built to predict time to death. DunedinPACE was trained on how fast 19 measures of body-system health declined over 20 years in one New Zealand birth cohort.

Lab clocks are not perfectly steady either. For six widely used clocks, measuring the same sample twice gave answers up to 9 years apart; retrained versions of the same clocks mostly agreed within 1.5 years. In its developers’ test, DunedinPACE gave highly consistent results when the same samples were run again. In small studies where fresh samples were taken from the same people hours or days apart, around a meal, a stress test or a stay at high altitude, most clocks were much less stable than when one sample was simply rerun. In small human studies, readings from newer clocks rose after emergency surgery for a broken hip, toward the end of pregnancy and in severe COVID-19, then fell back, at least partly, with recovery; planned operations did not show the same rise.

No biomarker of aging had been approved by US regulators for clinical use when the Biomarkers of Aging Consortium reviewed the field in 2023. In 2024 the consortium wrote that there was still no agreement on how such measures should be validated before clinical use. Whether using them helps people feel, function or live longer still has to be tested in prospective trials.

How blood-panel ages and DNA clocks relate

In the same people, blood-panel ages and DNA methylation clocks often disagree. In 964 New Zealanders born in the same year, so that calendar age could not explain the differences, different aging measures showed “low agreement.” In over 9,000 older Americans, DNA methylation and blood-chemistry measures were only moderately correlated (correlations of 0.1 to 0.4, where 1 would be perfect agreement). Yet GrimAge, DunedinPoAm (an earlier version of DunedinPACE) and all three blood measures were each linked to survival. Different studies have found different measures to be the strongest. One team that compared several measures in a US health survey concluded that studies “should not treat proposed measures of aging as interchangeable.” When the two disagree, that alone does not show which one is wrong: they are built from different measurements and were trained on different targets.

What regulators have said

Whoop’s app listing says its products are for general fitness and wellness and are not medical devices. The same listing marks the app as a regulated medical device for its ECG feature. Oura says its ring is not a medical device.

In July 2025, the FDA sent Whoop a warning letter about its blood-pressure estimates. The letter does not mention the age feature. In a June 2026 close-out letter, the FDA said it does not intend to enforce device requirements for the modified blood-pressure feature. It said this applies to that feature and not to any other product.

The FDA’s updated wellness guidance, issued in January 2026, is not binding. It says the FDA may treat some sensor-based estimates, such as blood pressure or heart rate variability, as wellness features if they meet several conditions. One condition is that values resembling clinical ones are validated. It does not mention age scores.

The FTC’s health-products guidance for businesses (December 2022) says that, as a general matter, advertised health benefits need support from randomized controlled trials in people. It is guidance, not law, and it does not mention wearables or biological age. We found no regulator document that assesses a wearable age score.

Questions to ask of any age number

Whichever number you are looking at, these questions help place it.

  1. What was it trained or checked against: calendar age, death or disease, or nothing published?
  2. Who tested it: the company that sells it, or an independent group?
  3. How much does it move between two readings when nothing has changed?
  4. How accurate are the device readings it is built from, for people like you?
  5. Does it tell you more than the readings it is built from, such as your fitness or resting heart rate?

If a report flags a blood value, or you have symptoms or health worries, see when to talk to a doctor. To compare biological age tests side by side, see Compare.

Sources

Show all 47 sources, with our notes on authors' ties to companies where relevant

The pictures are AI-generated images of typical scenes, not photographs of the named companies’ products or staff. Sources checked 3 November 2026. See how we make and check content or report a correction.