A journalist takes four different biological age tests on the same day. Same blood. Same body. Same morning. One test tells her she’s 7 years younger than her chronological age. Another tells her she’s 3 years older. The third splits the difference. The fourth refuses to commit because it measures “pace of aging” rather than a number, but the pace is somehow both fast and slow depending on which version of the algorithm she runs.

This is not a hypothetical. Some version of this article has been published roughly once a year for the last five years, by Wired, The Atlantic, The Guardian, the Washington Post. The results never get better. They probably get worse, because the tests keep multiplying and the marketing keeps escalating.

Bryan Johnson built a personal brand on biological age numbers. Influencers post their results with the same energy they used to bring to deadlift PRs. Entire supplement protocols are justified by a single test showing some imagined “reversal” of a few years. And underneath all of it, the science says something uncomfortable: the tests aren’t reliable enough at the individual level to support any of the conclusions people are drawing from them.

This article walks through what biological age tests actually measure, why they disagree, and what to track instead while we wait for the field to catch up.

A little background

This is the fourth piece in the longevity series. The autophagy article covers cellular cleanup. The mTOR vs AMPK piece covers the molecular wiring everyone fights over. The resilience vs slowdown article covers the Attia/Longo/Sinclair disagreement and ends with a specific point: validated longevity biomarkers, if they ever arrive, would be the thing that could resolve the debate. This article is the answer to that question. They haven’t arrived yet. Here’s why.

What biological age even means

The concept goes back to a 1969 paper by Alex Comfort. The idea: two 50-year-olds are not equally aged. One runs marathons and has the cardiovascular system of a 35-year-old. The other has type 2 diabetes, sleep apnea, and chronic inflammation, and her body is functionally closer to 60. Chronological age is a count of birthdays. Biological age is supposed to capture how much wear and tear has accumulated.

That’s a real concept. People age at different rates. The question isn’t whether biological aging exists, it’s whether we can actually measure it.

The big modern advance came in 2013, when Steve Horvath at UCLA published a paper showing that patterns of DNA methylation (the chemical tags on your genome that change as you age) could be combined into an algorithm that predicts chronological age with remarkable accuracy across tissues. That paper kicked off a decade of follow-ups, refinements, and commercial products.

Today there are several major “epigenetic clocks” used in research and sold to consumers. They’re the heart of the biological age industry, and they’re worth understanding because they don’t all measure the same thing.

The major clocks, briefly

There are three generations of these clocks, each built with a different goal.

First generation: estimate chronological age. The Horvath clock (2013) and the Hannum clock were both trained to predict how old you actually are based on methylation patterns. They’re accurate at the population level (within a few years of true age) but turn out to be limited as health predictors. Knowing whether your methylation “looks older than your real age” is suggestive but doesn’t strongly predict whether you’ll get sick or die sooner.

Second generation: predict mortality and disease. PhenoAge (2018) and GrimAge (2019) were built differently. Instead of training the algorithm on chronological age, the researchers trained it on actual outcomes: clinical biomarkers that predict death (PhenoAge) or a composite that includes time-to-death and methylation surrogates of plasma proteins and smoking history (GrimAge). The result: clocks that correlate less perfectly with calendar age but are much better predictors of actual mortality. GrimAge in particular is currently the strongest methylation-based predictor of mortality, cardiovascular events, and cancer incidence we have.

Third generation: measure pace of aging. DunedinPACE (2022) is fundamentally different. Instead of asking “how old does your biology look?” it asks “how fast is your biology changing per calendar year?” It was trained on longitudinal data from the Dunedin Study, which has followed about a thousand New Zealanders since birth (now into their 50s) with measurements of physical and cognitive function. DunedinPACE outputs something like a speed: a value of 1.0 means you’re aging at one biological year per calendar year. 1.2 means you’re aging 20% faster than average. 0.8 means slower. It’s the only clock currently designed to be sensitive to recent lifestyle changes, which makes it the favorite for intervention trials.

These are very different instruments. A first-generation clock is essentially a guess-your-age party trick. A second-generation clock is a mortality predictor. A third-generation clock is a velocity measurement. Asking which one is the “real” biological age is like asking which is the real temperature: Celsius, Fahrenheit, or the rate at which a kettle is heating up. They’re measuring different things.

Why they disagree

So here’s the first problem with biological age testing: there isn’t one biological age. There are multiple, depending on what you decide to measure. And the clocks correlate with each other less than you’d expect.

A 2025 Nature Communications study that compared 14 different clocks across nearly 19,000 individuals found that first-generation clocks correlated strongly with each other (r > 0.90), which makes sense since they were all trying to predict the same thing. But DunedinPACE correlated only weakly with the Horvath clock (r = 0.13) and moderately with GrimAge (r = 0.58). In other words: even the leading clocks in the field don’t agree about who is aging fast and who is aging slow.

This isn’t a flaw in the tests. It’s a consequence of the fact that “aging” is not a single property. It’s a collection of related but distinct processes (epigenetic drift, mitochondrial dysfunction, cellular senescence, proteostasis loss, immune decline, and more). Different clocks pick up different signals. Asking which clock is correct is the wrong question. The honest answer is that they’re each capturing something real, and those somethings don’t fully overlap.

The real problem: noise

Here’s the second problem, and it’s the bigger one. Even within a single clock, the noise is enormous.

A 2022 study found that running the same blood sample through the same clock twice can give estimates that differ by up to 9 years. So a 40-year-old gets her blood drawn, the lab runs the analysis, and her result is 35. The same sample, processed again, returns 44. Same biology. Same lab. Different number.

Some of that noise comes from the wet-lab procedure (how the DNA is processed, the specific batch of reagents, the sequencing platform). Some comes from biological fluctuation: methylation patterns shift modestly with stress, sleep, recent meals, infection status, time of day. Saliva and blood from the same person on the same day give meaningfully different estimates, because the tissues themselves have different methylation profiles.

For research purposes, this noise can be averaged out across large populations. You can absolutely show that GrimAge predicts mortality in a 10,000-person cohort. That’s been done many times and the signal is real. But for an individual person, looking at a single test result, the signal is buried in the noise. The “I reversed my biological age by 4 years” social media post is almost certainly within the test-retest variance of the assay itself.

Daniel Belsky, the Duke epidemiologist who developed DunedinPACE, has been saying this publicly since 2017: “Based on these results, I’d say it’s premature to market aging tests to the public.” His position hasn’t changed. If anything, the recent press has firmed it up. The April 2026 Washington Post piece and the May 2026 Conversation piece (the latter co-authored by an epigenetics researcher) say the same thing in different words: useful at the population level for research, not reliable enough at the individual level for personal decisions.

Why this matters

A lot of the public longevity discourse rests on biological age numbers. Bryan Johnson’s entire Blueprint project is justified by his published epigenetic age scores. Influencer claims about supplements reversing aging are almost always backed by a single before-and-after test. The “rejuvenation Olympics” leaderboard ranks people by DunedinPACE.

If the underlying measurements have a margin of error larger than the effect sizes being claimed, none of this means what people think it means. A protocol that claims to have lowered your biological age by 3 years, measured by a test with a 9-year test-retest range, has not demonstrated anything.

This isn’t a critique of the science. The science is honest about the limitations. The problem is the commercial layer on top of the science, which sells precision the underlying assay can’t deliver.

There’s also a more subtle problem. Even if biological age tests were perfectly precise, they’re correlational, not causal. Showing that an intervention lowers your GrimAge by 2 years doesn’t prove the intervention is making you live longer. It proves the intervention changed something that correlates with mortality in large populations. Those aren’t the same statement. The intervention might be tweaking the surface readout without changing the underlying biology. Or it might be doing exactly what you want. We mostly don’t know yet.

What to track instead

If you actually want to monitor your health on a year-to-year basis right now, the answer isn’t trendy. It’s the same set of measurements your primary care doctor has been ordering for decades, plus a few additions from the longevity-medicine playbook. None of these are biological age tests. All of them are reliable, well-validated, and respond to known interventions on known timescales.

The annual blood panel: lipids (with ApoB as the better-than-LDL marker if you can get it), HbA1c, fasting glucose and insulin, inflammatory markers like high-sensitivity CRP, liver enzymes, thyroid panel, kidney function. Most of these are cheap. All of them have decades of outcome data.

Body composition: a DEXA scan once every year or two gives you visceral fat, muscle mass, and bone density. Visceral fat predicts metabolic disease. Muscle mass and bone density predict frailty. Both move on observable timescales.

Fitness: VO2max is one of the strongest predictors of all-cause mortality. You can get it measured properly at a sports medicine clinic, or estimate it well enough from a maximal running test or a good wearable. Grip strength is the second-best simple measurement. It correlates with mortality almost as well as VO2max and you can test it with a hundred-dollar dynamometer.

Functional measures: how fast can you walk a mile, how many push-ups can you do, how long can you hold a single-leg balance with your eyes closed. These are the actual things that decline in old age and the actual things that matter for your last decade.

Sleep: most decent wearables now give you usable sleep duration and stages. Track total sleep, deep sleep, and HRV. Within-person trends are more useful than absolute numbers.

This is the boring list. It’s also the list that every serious longevity-focused doctor I’m aware of (Attia, Lustig, Topol, Ramakrishnan if he wrote prescriptions) would put together. None of it requires an epigenetic clock. All of it tells you more about your actual trajectory than a methylation reading ever will at current precision.

What would change my mind

Biological age testing might get good. The trajectory isn’t hopeless. Second-generation clocks like GrimAge2 are meaningfully better than first-generation ones. Pace-of-aging measures are conceptually closer to what people actually want to know. The Nature Communications 2025 paper that found weak correlations between clocks also confirmed that second-generation clocks add real predictive value over chronological age for disease outcomes.

The things that would shift this article’s conclusion:

Lower test-retest variance. If new methods can get the same-sample variance down from years to months, individual-level inference becomes plausible.

Cross-clock convergence. If newer clocks start agreeing with each other at the individual level (not just population level), it becomes more plausible that they’re measuring something real and singular.

Intervention trials with hard outcomes. Right now most “this intervention lowers biological age” studies don’t follow people long enough to show whether they actually live longer. If the next decade of trials shows that interventions that move biological age also move mortality in the same direction at the same magnitude, the case for using these tests as surrogates gets much stronger.

Regulatory standardization. Right now any company can offer a “biological age” test with no requirement to validate their methodology. If the FDA or equivalent agencies start requiring real validation studies, the worst products will get pushed out.

None of these are unreachable. The field is moving. But none of them have happened yet, and the consumer market has gotten well ahead of the science.

A word of caution

If you’ve already taken one of these tests and gotten a “young” result, enjoy it but don’t update your behavior. If you’ve gotten an “old” result, don’t panic and don’t go on a supplement binge. The result you got is one draw from a distribution that’s wider than most of the effects people are chasing. A second draw next week could put you in a different category entirely.

What the result probably doesn’t tell you is whether you should be worried about your health. The boring measurements above will tell you that. They’ll also tell you what to do about it, which is more than any current biological age test can deliver.

The longevity industry has a habit of selling certainty that the underlying science doesn’t yet support. Biological age tests are the cleanest example of that gap right now. It’s worth being skeptical, not because the science is wrong, but because the science is more humble than the marketing.

---

References

1. Horvath, S. (2013). DNA methylation age of human tissues and cell types. Genome Biology, 14(R115). https://genomebiology.biomedcentral.com/articles/10.1186/gb-2013-14-10-r115
2. Levine, M.E., Lu, A.T., Quach, A., et al. (2018). An epigenetic biomarker of aging for lifespan and healthspan. Aging, 10(4), 573–591. https://www.aging-us.com/article/101414/text
3. Lu, A.T., Quach, A., Wilson, J.G., et al. (2019). DNA methylation GrimAge strongly predicts lifespan and healthspan. Aging, 11(2), 303–327. https://www.aging-us.com/article/101684/text
4. Belsky, D.W., Caspi, A., Corcoran, D.L., et al. (2022). DunedinPACE, a DNA methylation biomarker of the pace of aging. eLife, 11, e73420. https://elifesciences.org/articles/73420
5. Bernabeu, E., et al. (2025). An unbiased comparison of 14 epigenetic clocks in relation to 174 incident disease outcomes. Nature Communications. https://www.nature.com/articles/s41467-025-66106-y
6. Higgins-Chen, A.T., et al. (2022). A computational solution for bolstering reliability of epigenetic clocks: Implications for clinical trials and longitudinal tracking. Nature Aging. (Test-retest reliability study.)
7. Shalev, I., Apsley, A. (April 2026). These tests claim to tell your “biological age.” The science isn’t so simple. The Washington Post. https://www.washingtonpost.com/health/2026/04/29/biological-age-tests-at-home/
8. Shalev, I., Apsley, A. (May 2026). Biological age tests reveal what slows or hastens aging, but they’re useful only for researchers, not consumers. The Conversation. https://theconversation.com/biological-age-tests-reveal-what-slows-or-hastens-aging-but-theyre-useful-only-for-researchers-not-consumers-275974
9. Belsky, D.W., et al. (2017). Eleven Telomere, Epigenetic Clock, and Biomarker-Composite Quantifications of Biological Aging: Do They Measure the Same Thing? American Journal of Epidemiology.
10. Comfort, A. (1969). Test-battery to measure ageing-rate in man. The Lancet.