Your Biological Age Test Might Be Wrong. Here's Why.

A patient walked into my clinic last month holding two biological age reports. One, from a saliva-based test kit he ordered online, told him his biological age was 52. The other, from a blood-based epigenetic panel run through a longevity clinic overseas, told him he was 41. He is 46. "So which one is right?" he asked, placing both reports on the desk between us like competing witness statements. "Am I ageing well or badly?"

It is a question I am hearing more often. Biological age testing has gone from a niche research curiosity to a consumer product you can order on your phone. Instagram and TikTok are full of longevity influencers showing off their results, celebrating being "biologically 32 at 45" or panicking about being "biologically 60 at 38". Brands are marketing these kits like wellness horoscopes. And the honest answer to my patient's question is more complicated, and more important, than most of those influencers are letting on.

Because a major study published in Nature Communications in late 2025, comparing 14 of the most widely used epigenetic clocks head to head across 174 disease outcomes in nearly 19,000 people, found something that should give everyone selling these tests pause: no single clock was best at everything. Not one. The clocks that predicted cardiovascular disease well often missed liver disease. The ones that tracked mortality missed neuropsychiatric conditions. And some of the most marketed consumer tests were not even in the comparison.

What a Biological Age Test Is Actually Measuring

To understand why two tests can give the same person an eleven-year spread, you need to know what sits under the bonnet.

Most biological age tests on the market today are epigenetic clocks. They work by measuring DNA methylation, a chemical modification where methyl groups (tiny clusters of one carbon and three hydrogen atoms) attach to specific spots on your DNA. These spots are called CpG sites, short for cytosine-phosphate-guanine dinucleotides. As you age, methylation patterns at these sites shift in predictable ways. Some sites gain methyl groups. Others lose them. An epigenetic clock is a mathematical algorithm that reads the methylation pattern at a carefully selected set of CpG sites and calculates an "age" from the pattern.

The first-generation clocks, built by Steve Horvath at UCLA in 2013 and Gregory Hannum at UC San Diego in the same year, were trained to match chronological age as closely as possible. They were brilliant pieces of science. But predicting how old someone is and predicting how fast they are ageing are not the same problem. A 50-year-old who smokes, drinks heavily, sleeps four hours a night and has uncontrolled diabetes might still have a Horvath clock age close to 50, because the original Horvath clock was trained to match birthdays, not health trajectories.

That limitation drove the development of second-generation clocks. In 2019, Ake Lu, working with Horvath and colleagues, published GrimAge. Instead of training the algorithm on chronological age, they trained it on mortality risk. GrimAge incorporates surrogate markers for smoking history, inflammatory proteins (like adrenomedullin and growth differentiation factor 15), and metabolic health. It asks a fundamentally different question: not "how old does your DNA look?" but "how close to dying does your DNA look?" That is a grimmer question, which is where the name comes from.

Then in 2022, Daniel Belsky and colleagues at Columbia University published DunedinPACE (Pace of Aging Calculated from the Epigenome). DunedinPACE does something different again. It was built from longitudinal data on the Dunedin cohort, a famous group of 1,037 New Zealanders tracked from birth since 1972. Rather than estimating a static biological age ("you are biologically 53"), DunedinPACE estimates a rate of ageing ("you are ageing at 1.15 years per calendar year"). That distinction matters clinically, because if you make lifestyle changes and retest in a year, a rate-based clock can tell you whether your pace of ageing has slowed. A static clock might not pick that up.

The Study That Sorted the Clocks

All of these clocks are published, peer-reviewed, and scientifically credible. So which one should a patient trust? That is the question a team at the University of Exeter and the UK Biobank set out to answer.

Their study, published in Nature Communications, took 18,859 participants from the Generation Scotland cohort, measured all 14 major epigenetic clocks in the same blood samples, and then followed the participants for over ten years, tracking 174 incident disease outcomes and all-cause mortality. The design is important. Rather than each clock being tested on its own favourite dataset, every clock was tested on the same people, using the same samples, against the same outcomes. A fair race.

Three findings stood out.

First, no single clock was uniformly the best. Different clocks excelled at different things. GrimAge and GrimAge2 showed the strongest associations with respiratory disease, liver disease, and overall mortality. DunedinPACE performed particularly well for cognitive decline, metabolic conditions, and tracking the pace of ageing in response to lifestyle change. The first-generation Horvath and Hannum clocks were the weakest performers overall, though they still had some predictive value for selected conditions.

Second, the second-generation mortality-trained clocks consistently outperformed the first-generation chronological-age-trained clocks. This makes sense. If you train an algorithm to match birthdays, it gets good at matching birthdays. If you train it to predict who is going to get sick and die, it gets good at predicting who is going to get sick and die. The second generation clocks capture something real about biological decline that the first generation missed.

Third, adding epigenetic age data to traditional clinical risk models improved disease prediction beyond standard factors. In other words, these clocks are not just research toys. Even after accounting for age, sex, smoking status, BMI, and other standard risk factors, accelerated epigenetic ageing added independent predictive power for cardiovascular disease, respiratory disease, and mortality. The clocks are measuring something that blood pressure and cholesterol alone do not capture.

Why Two Tests Give You Two Different Ages

Back to my patient with the eleven-year gap. Here is why his two reports disagreed so dramatically.

Different sample types. His online test used saliva. His clinic test used blood. Almost all validated epigenetic clocks were trained on blood-derived DNA. Saliva is a mixture of about 65 per cent immune cells and 35 per cent buccal (cheek) epithelial cells, and those two cell types have fundamentally different methylation signatures. When you apply a blood-trained clock to saliva without proper correction, the result can be off by years. Some commercial tests apply correction factors, but the accuracy of those corrections is variable and not always transparently disclosed. As researchers interviewed in The Scientist have noted, the gap between a rigorously validated clock and a commercially optimised wellness test can be enormous, sometimes spanning decades of apparent age for the same sample.

Different platforms. His saliva test used a lower-resolution methylation array. His blood test used the Illumina EPIC array, which reads roughly 850,000 CpG sites. These are different instruments measuring different sets of DNA markers, and comparing results across them is like comparing a photograph taken on a phone camera with one taken on a medical-grade scanner. Both capture reality. But one captures a lot more of it.

Different algorithms. Even if you use the same sample type and the same platform, different clocks will give different numbers. A study published in an October 2024 issue of Aging titled "When to Trust Epigenetic Clocks" demonstrated that short-term fluctuations in diet, illness, environmental exposures, and even the time of day when the sample was taken can shift estimated age substantially. Epigenetic changes are dynamic, not static. A single snapshot can be noisy.

The bottom line for my patient was this: neither test was "right" in the way his blood pressure reading is right. Both were imprecise estimates from different instruments using different methods, and the gap between them told us more about the tests than about him.

The GrimAge Validation That Kept Clinicians' Attention

Despite the inconsistency problem, there is genuine clinical signal in these clocks, and one study published in Epigenetics in 2025 showed it cleanly. Zhu, He, Wang, Zhao and colleagues ran a retrospective cohort analysis of 1,942 participants in the US National Health and Nutrition Examination Survey (NHANES), median age 65, and tested whether GrimAge and GrimAge2 age acceleration predicted who actually died over the follow-up period.

They found that GrimAge and GrimAge2 were the only two clocks, among all those tested, that showed approximately linear and positive associations with all three mortality outcomes: all-cause, cancer-specific, and cardiac. Each year of GrimAge acceleration (meaning your GrimAge is one year older than your chronological age) was associated with a hazard ratio of roughly 1.5 for all-cause mortality. In practical terms, a person whose GrimAge was five years older than their birthday had meaningfully higher odds of dying in the follow-up period than someone whose GrimAge matched their chronological age, even after adjusting for the usual suspects (smoking, BMI, blood pressure, glucose).

This is not a small thing. It suggests the clock is picking up biological deterioration that traditional risk scores miss.

DunedinPACE and the Brain

On the other side of the evidence, DunedinPACE has carved out its own niche. A March 2026 medRxiv preprint found that among all tested epigenetic clocks, DunedinPACE showed the strongest and most consistent associations with cognitive performance across multiple domains. The other clocks, including GrimAge, showed no consistent pattern with cognition.

That finding matters in Singapore specifically. Our population is ageing fast. The Department of Statistics projects that by 2030, roughly one in four residents will be aged 65 or older. Dementia prevalence in Singapore was estimated at around 10 per cent in those aged 60 and above using the 10/66 diagnostic criteria in the 2015 WiSE (Well-being of the Singapore Elderly) study by Subramaniam and colleagues. Cognitive decline is the thing many Singaporeans fear more than cancer, and a tool that reliably tracks the pace of brain ageing could, in principle, help identify people who would benefit most from early intervention.

The caveat: the March 2026 data is a preprint, not yet peer-reviewed. And DunedinPACE, like all clocks, has not yet been validated as a formal clinical endpoint by any regulatory body. It is useful. It is not gospel.

Can You Actually Change Your Biological Age?

This is the question every patient really wants answered. And the honest read is: probably, somewhat, in some people, with caveats the size of a textbook.

The most cited trial is the 2021 pilot RCT by Fitzgerald, Baer and colleagues, published in Aging. They randomised 43 healthy men aged 50 to 72 into an eight-week lifestyle programme (diet rich in methyl-donor nutrients, moderate exercise, stress reduction, sleep guidance, supplemental probiotics and phytonutrients) or a control group. The treatment group showed a 3.23-year decrease in Horvath DNAmAge compared with controls (p = 0.018). Blood biomarkers shifted too: serum 5-methyltetrahydrofolate rose by 15 per cent and triglycerides fell by 25 per cent.

It is a genuinely interesting result, and it has been cited thousands of times. But the limitations are real. The sample was small (18 in the treatment arm). Only men were included. The intervention was multimodal (diet, exercise, sleep, stress, supplements all at once), so we do not know which component did the heavy lifting. And it used the first-generation Horvath clock, which, as we just discussed, is the weakest predictor of actual health outcomes. A 3.23-year shift on the Horvath clock is not necessarily the same as a 3.23-year shift on GrimAge or DunedinPACE. The team themselves have called it a pilot study requiring larger replication.

What we can say with more confidence is that the lifestyle factors most consistently associated with slower epigenetic ageing across multiple large cohort studies are not surprising. A 2026 review in eBioMedicine (part of The Lancet family) by Seale, Horvath, Teschendorff, Eynon and Voisin, summarising the field's clinical translation progress, highlighted that higher BMI, smoking, elevated fasting glucose, and poor blood pressure control are the factors most reliably associated with accelerated ageing across multiple cohorts and multiple clock types, including DunedinPACE. Reversing those factors is associated with slower measured ageing. No supplement, peptide injection or hyperbaric chamber session has evidence of comparable magnitude.

The Singapore Context

Singapore is one of the longest-lived nations on earth. Life expectancy at birth here is now 84.1 years. But lifespan and healthspan are not the same thing. The Singapore Burden of Disease study has consistently shown that while we live long, the last decade of many Singaporeans' lives is marked by chronic disease, disability and diminished independence. The gap between life expectancy and healthy life expectancy is roughly eight to ten years, which means the average Singaporean spends a substantial portion of their final years in poor health.

That is the gap biological age testing is trying to address. If we can identify people who are ageing faster than their birthday suggests, we can intervene earlier, with the boring, proven tools (blood pressure control, glucose management, smoking cessation, physical activity, sleep, diet) rather than waiting for the heart attack or the dementia diagnosis.

Singapore's Ministry of Health released an updated Screening Test Review Committee Report in February 2026, refreshing recommendations for cancer screening, diabetes, hypertension, hyperlipidaemia and osteoporosis. That report uses a sensible three-tier framework: Category 1 tests are suitable for population-wide screening; Category 2 tests are appropriate for individual-level discussion with your doctor based on your risk profile; and Category 3 tests lack sufficient evidence. Epigenetic age testing is not in the STRC framework yet, but if it were, it would almost certainly sit in Category 2: a tool for discussion with your doctor, not something to screen every Singaporean at a polyclinic.

That is the right framing. A biological age test is most useful when it is interpreted by someone who knows your medical history, your medications, your family history, and your lifestyle context. A number on a consumer report, interpreted alone at midnight on your phone, can cause anxiety or false reassurance in roughly equal measure.

What My Patient Went Home With

We went through his two reports together. I explained the sample-type issue, the platform difference, and the fact that neither number was precise enough to take literally. We looked at his actual clinical markers: his fasting glucose, his lipid panel, his blood pressure, his VO2max from a recent cardiopulmonary exercise test, his liver stiffness from a FibroScan, his body composition from a DXA scan. Those individual measurements, each validated and reproducible, told a much more granular and actionable story than a single "biological age" number ever could.

Where biological age testing added value was in the conversation it started. He wanted to know if his lifestyle was working. He wanted a sense of trajectory. We agreed that if he wanted to track his rate of ageing over time, a blood-based test using a validated second-generation clock (like DunedinPACE or GrimAge) repeated at consistent intervals, ideally annually, interpreted in the context of his full clinical picture, could be a useful data point. Not the data point. A data point. One instrument in an orchestra, not a solo performance.

He left clinic with a sensible plan: optimise the things with the largest evidence base (his sleep was poor, his blood pressure was borderline, his fibre intake was low), retest in twelve months with the same platform and the same clock, and compare the trajectory. Not a number. A direction.

That, for now, is the honest read on biological age testing. The science is real and progressing fast. The clocks are capturing something meaningful about how we age. But the consumer market has outrun the evidence, and a single number from a single test, interpreted without clinical context, is about as useful as weighing yourself on two different scales in two different countries and panicking about the difference.

If this is a conversation you want to have, have it with your own doctor, who knows your history, your medications, and what the number means for you, not for a population average. That is where the value is.

Dr Samuel Choudhury, MBBS (NUS) · MPH (Johns Hopkins) · GDFM · GDFP Derm

References

1. McCartney DL, Hillary RF, Conole ELS, et al. An unbiased comparison of 14 epigenetic clocks in relation to 174 incident disease outcomes. Nature Communications. 2025;16:9006. doi:10.1038/s41467-025-66106-y
2. Seale K, Horvath S, Teschendorff AE, Eynon N, Voisin S. Epigenetic clocks: advancing biological age measures towards meaningful clinical use. eBioMedicine (The Lancet). 2026;115:105663. doi:10.1016/S2352-3964(26)00056-3
3. Zhu T, He Y, Wang Y, Zhao L, et al. GrimAge and GrimAge2 Age Acceleration effectively predict mortality risk: a retrospective cohort study. Epigenetics. 2025;20(1):2530618. doi:10.1080/15592294.2025.2530618
4. Belsky DW, Caspi A, Corcoran DL, et al. DunedinPACE, a DNA methylation biomarker of the pace of aging. eLife. 2022;11:e73420. doi:10.7554/eLife.73420
5. Horvath S. DNA methylation age of human tissues and cell types. Genome Biology. 2013;14(10):R115. doi:10.1186/gb-2013-14-10-r115
6. Lu AT, Quach A, Wilson JG, et al. DNA methylation GrimAge strongly predicts lifespan and healthspan. Aging. 2019;11(2):303-327. doi:10.18632/aging.101684
7. Fitzgerald KN, Hodges R, Hanes D, et al. Potential reversal of epigenetic age using a diet and lifestyle intervention: a pilot randomized clinical trial. Aging. 2021;13(7):9419-9432. doi:10.18632/aging.202913
8. Ministry of Health Singapore. Report of the Screening Test Review Committee 2026, Volume 1. February 2026. Available at: moh.gov.sg
9. Subramaniam M, Chong SA, Vaingankar JA, et al. Prevalence of dementia in people aged 60 years and above: results from the WiSE study. Journal of Alzheimer's Disease. 2015;45(4):1127-1138. doi:10.3233/JAD-142769
10. Singapore Department of Statistics. Death and life expectancy: latest data. Available at: singstat.gov.sg