Biological Age: What the Science Says

Ask five different labs to measure your "biological age" and you may get five different numbers. That's not a sign the field is broken. It's a sign that "biological age" isn't one thing — it's closer to a panel of different instruments, each built to catch a different kind of trouble. A blood-chemistry formula, a DNA methylation sample, and a decades-long tracking of your organ function are all legitimate ways to estimate how much wear your body has accumulated. They just don't always agree, because they aren't measuring quite the same thing.

We think you should know that before you look at any number we give you — including ours. This page explains where biological-age testing came from, how the major approaches actually work, why we use a method called OMICmAge, and what a single test — ours included — can't tell you no matter how it's marketed.

Where biological-age testing came from

Biological-age measurement didn't start with DNA. For roughly two decades, telomere length — the protective caps on the ends of chromosomes, which shorten as cells divide — was the leading candidate biomarker, propelled by the 2009 Nobel Prize in Physiology or Medicine for telomerase research. A wave of commercial telomere tests followed in the 2010s. Telomere length is still scientifically relevant, but the field now generally treats it as a weaker individual-level predictor than what came next.

What came next arrived in two waves of DNA methylation research. In 2013, Steve Horvath and, separately, Gregory Hannum each published "first-generation" epigenetic clocks — models trained to predict chronological age itself from patterns of DNA methylation. They were remarkably accurate at guessing someone's actual age. But accuracy at guessing the calendar isn't the same as usefulness for predicting health, and that gap is exactly what the second wave set out to close.

Starting around 2018, researchers changed the training target. Morgan Levine's PhenoAge (2018) and Ake Lu's GrimAge (2019) were trained on clinical markers of physiological decline and on mortality itself, not on chronological age — and both got measurably better at telling apart two people of the same age who are aging differently. In 2022, Daniel Belsky's DunedinPACE took a different approach again: rather than a snapshot age, it estimates your current pace of aging, trained on two decades of repeated organ-system measurements from a single birth cohort followed since 1972.

OMICmAge, published in Nature Aging in February 2026 by a Harvard/Brigham and Women's Hospital research team working with the lab TruDiagnostic, is the newest entrant, and it takes a third approach: folding proteomic and metabolomic information into a methylation-only test, so a single sample carries information that used to require several separate, expensive lab panels.

Three families, not one test

It helps to think of biological-age tools as three families rather than one category:

• Phenotypic / clinical calculators — formulas built directly from routine blood work: inflammation markers, kidney and liver function, blood sugar. No DNA involved.
• Epigenetic (DNA methylation) clocks — read chemical tags on your DNA from blood, saliva, or a blood spot. Horvath, Hannum, PhenoAge's methylation version, GrimAge, and OMICmAge all sit in this family, each trained toward a different target.
• Pace-of-aging trackers — trained on repeated measurements of the same people over years, then distilled into a one-time test. DunedinPACE is the clearest example, and — notably — the one tool in this whole category with a real randomized-trial result behind it.

Why we use OMICmAge, specifically

The mechanism is easiest to picture as a relay. Researchers first identified 396 proteins, metabolites, and clinical values that correlated with a mortality-trained score called EMRAge, then trained small DNA-methylation-based "proxies" for each one — stand-ins that estimate things like your C-reactive protein or HbA1c level from methylation patterns alone. The final algorithm combines 990 methylation sites with 40 of the best proxies into a single score. The practical result: a test that only requires a methylation sample, but whose score is shaped by real information about your protein and metabolite biology, learned during training rather than measured fresh each time.

That's a genuine structural difference from the clocks that came before it. Older clocks share a meaningful share of their underlying DNA sites with each other — PhenoAge and the original Horvath clock share 50 sites, for instance. OMICmAge shares at most 3 sites with any prior clock, because it was built from a wider biological starting point distilled down into methylation, rather than methylation trained on methylation.

We want to be precise about what that buys, because it's tempting to round it up to "the most accurate clock available" and some marketing in this category does exactly that. The published data don't support an unqualified version of that claim. Across its validation cohorts, OMICmAge wasn't the single best predictor of everything: a related methylation-only model edged it out for mortality prediction in its own discovery cohort, and a competing clock (GrimAge's newer version) came out ahead for mortality prediction in an independent Scottish cohort. What OMICmAge does show consistently is strong, broad performance across several disease categories at once — rather than the single best score on any one metric, it's the broadest-reaching one, ranking in the top two clocks for type 2 diabetes and cardiovascular disease across every cohort tested, with the strongest depression association in two of them.

There's one more distinction we think is worth stating plainly rather than leaving out: DunedinPACE currently has something OMICmAge doesn't. In the CALERIE randomized controlled trial, caloric restriction was shown to actually slow DunedinPACE — real experimental evidence that an intervention can move that specific score, not just a correlation. No equivalent trial result exists yet for OMICmAge. If the question that matters most to you is "has anyone proven you can change this number on purpose," DunedinPACE currently answers it better than the method we use. We'd rather tell you that than let a confident-sounding page imply otherwise.

How we differ from other biological age providers

Most direct-to-consumer biological age tests return a single number from a single data type — often DNA methylation alone, sometimes a blood panel alone — with a generic list of "things that raise or lower biological age" attached, and no context specific to you. We differ in two ways, and we want to be precise about which is which.

The first is methodological: OMICmAge folds three data types (methylation, protein and metabolite proxies, and clinical variables) into a single score, rather than relying on one. That's the same structural distinction described earlier in this article — it's a difference in what goes into the number, not a claim that the number itself is more certain than a single-data-type score would be.

The second is interpretive: every result here is reviewed by a clinician against your personal history before it reaches you, instead of arriving as a raw automated output. That's a service difference, not a scientific one, and we treat it that way — see the next section for exactly what it does and doesn't add.

What we deliberately don't do here is name specific competitor products or companies. Every comparative claim on this page is a comparison to a category pattern (single data type, no clinical review layer), not to a named rival — the same standard we've applied to every other comparative claim in this article, including the ones about OMICmAge itself.

The clinical interpretation lens

"A clinician reviews your result" is a specific claim, and it's worth being specific about what it actually changes.

It does not change the strength of the underlying evidence. A clinician looking at your OMICmAge score cannot make the science behind it more validated than it is — it's still a single, non-independently-replicated study, and a review process doesn't add a second study. If you take one thing from this section, take that: clinical review is about what happens with an uncertain number, not a reason to treat the number as less uncertain.

What it does add is context that the algorithm itself can't supply. A score can be elevated for reasons that have nothing to do with long-run aging — a recent infection, a medication, an acute illness — the same kind of confound the underlying research flags at the population level (for instance, the OMICmAge paper's own authors note that an association between low body weight and a higher score in one cohort was probably explained by illness driving both, not by low weight itself being the risk factor). A clinician reviewing your specific history is positioned to ask "does anything about this person explain the number besides aging?" in a way a purely automated report isn't.

Concretely, that turns "your score is X" into "here's what might specifically be contributing to X for you, and here's what's worth raising with your own physician" — a starting point for a conversation, not a diagnosis, and not a promise that acting on the discussion will change a future test result.

Retesting: what it can — and can't yet — tell you

The reason anyone retests at all is to see whether a real change (more exercise, quitting smoking, a new health condition) shows up in the score. That only works if the test is reliable enough that measuring the same person twice, with nothing else changed, gives close to the same number. That property is called test-retest reliability, and it matters more here than it might first seem, because without it there's no way to tell whether a shift between two results reflects real biological change or ordinary assay noise.

This is where the two methods described in this article currently stand in very different places. DunedinPACE was explicitly built and validated for test-retest reliability — its developers restricted the underlying methylation data specifically to sites with low variation between repeat measurements on the same sample, and reported high test-retest reliability as a core design feature. That property is a direct reason DunedinPACE could be used as an endpoint in the CALERIE randomized trial: a measurement you can't trust to hold still on an unchanged person isn't usable to detect a real intervention effect.

No published test-retest reliability data exists yet for OMICmAge, the method we use. That's a real, open gap, not a technicality — and we'd rather say so than let the fact that some people retest every few months imply the question has been answered. Right now, if your OMICmAge score moves between two tests, there's no published basis for telling you how much of that movement is a genuine change in your biology versus normal measurement variation.

Our honest position, given that gap: we lean toward longer intervals between retests (annual, rather than monthly or quarterly) until reliability data for OMICmAge specifically is published, and we'd treat any more frequent retest recommendation as something to check against this gap rather than accept as evidence-based on its face — including if such a recommendation appears elsewhere in our own marketing.

What actually moves the number — and how confident to be about each part

This section deserves real care, because the strength of evidence varies enormously by claim.

Trial-level evidence (rare, and the strongest tier in this field):

• Caloric restriction slowed DunedinPACE in the CALERIE randomized trial — the one clear experimental result across this entire category of tests.

Observational evidence (consistent direction, not proof of cause):

• More weekly exercise and more years of education associate with a younger-reading score across essentially every clock studied, including OMICmAge.
• Obesity, current smoking, and heavier alcohol use associate with an older-reading score across the same range of tools.

Preliminary evidence (flagged, not headlined):

• Associations between certain supplements and a lower OMICmAge score have been reported in a single commercial-testing cohort. They haven't been replicated elsewhere, so we treat them as a lead worth watching, not a recommendation.

What a single test — ours included — can't tell you

Biological age tests vary a lot in validity from one company to the next, and there's no industry-wide standard that guarantees any two of them are measuring the same thing the same way. We'd rather say that plainly than assume you haven't already wondered about it.

What biological age reflects is functional and cellular signs of aging — not how many birthdays you've had. It's an estimate, drawn from measurable biological patterns, not a diagnosis and not a countdown. A single test cannot predict your exact lifespan, cannot guarantee a specific health outcome, and improving the number is gradual and incompletely understood, even by the researchers who built the underlying models. This is also, formally, a wellness and laboratory-developed test rather than an FDA-cleared diagnostic — a personal result is a meaningful research-grade data point, not clinical-grade evidence on its own.

Where the science is heading

A few open threads are worth watching. Researchers have flagged plans for a version of OMICmAge compatible with lower-cost methylation arrays, and for replacing statistically-estimated protein and metabolite proxies with direct, targeted lab assays to sharpen accuracy. More broadly, a December 2025 commentary in npj Aging ("Do we actually need aging clocks?") pointed out that across the field, different clocks still disagree with each other in post hoc analyses of finished trials — none yet behaves as a fully stable, dependable trial endpoint. Closing that gap, for OMICmAge and for the field generally, is the next real milestone, not the marketing superlatives already circulating ahead of it.

References

1. Horvath, S. "DNA methylation age of human tissues and cell types." Genome Biology 14 (2013): R115.
2. Hannum, G., et al. "Genome-wide methylation profiles reveal quantitative views of human aging rates." Molecular Cell 49, no. 2 (2013): 359–367.
3. Levine, M. E., et al. "An epigenetic biomarker of aging for lifespan and healthspan." Aging (Albany NY) 10, no. 4 (2018): 573–591.
4. Lu, A. T., et al. "DNA methylation GrimAge strongly predicts lifespan and healthspan." Aging (Albany NY) 11, no. 2 (2019): 303–327.
5. Belsky, D. W., et al. "DunedinPACE, a DNA methylation biomarker of the pace of aging." eLife 11 (2022): e73420. doi.org/10.7554/eLife.73420
6. Waziry, R., et al. "Effect of long-term caloric restriction on DNA methylation measures of biological aging in healthy adults from the CALERIE trial." Nature Aging (2023). nature.com/articles/s43587-022-00357-y
7. Chen, Q., Dwaraka, V. B., Carreras-Gallo, N., et al. "OMICmAge quantifies biological age by integrating multi-omics with electronic medical records." Nature Aging 6, no. 3 (2026): 722–737. doi.org/10.1038/s43587-026-01073-7
8. "Do we actually need aging clocks?" npj Aging (December 2025). nature.com/articles/s41514-025-00312-2
9. Unbiased comparison of 14 epigenetic clocks in relation to 174 incident disease outcomes, Generation Scotland cohort. Nature Communications. nature.com/articles/s41467-025-66106-y