Measuring How Old You Really Are: The Next Generation of Biological-Age Clocks

The first generation of these tools — Horvath's original epigenetic clock, then PhenoAge, then GrimAge — read patterns of chemical tags on DNA called methylation marks. Feed in a blood sample, and the algorithm returns an estimate of biological age, or a pace-of-aging score like DunedinPoAm that tries to capture how fast the clock is ticking right now. They are not crystal balls. They are statistical models, trained on large populations, and they carry the usual caveats about averages and outliers. But they have proved sturdy enough to attract serious researchers, and sturdy enough to spawn successors.

One of the more interesting successors arrived this year. A team led by Mboning and colleagues published BayesAge 2.0, a maximum-likelihood algorithm that predicts transcriptomic age from RNA-sequencing data — that is, from the body's gene-expression patterns rather than its methylation marks. Where most older clocks lean on linear regression, BayesAge 2.0 uses a Poisson model suited to count-based gene-expression data and a smoothing technique called LOWESS to handle the fact that genes do not change in a straight line over a lifetime. The authors report that the new method reduces a stubborn problem in the field — age bias, the tendency of these models to overestimate the young and underestimate the old — and that it runs faster than the elastic-net regressions that have dominated the space.

A better speedometer is useful only if the car can be steered. The more provocative question is whether everyday medical decisions register on these clocks at all. A new analysis from the Baltimore Longitudinal Study of Aging — one of the longest-running aging cohorts in the world — took a careful swing at it. Researchers examined 27 common drug categories and their association with phenotypic aging markers across four domains: body composition, energetics, homeostatic mechanisms, and neuroplasticity. By comparing each participant to themselves over time, the team tried to filter out the genetic and early-life noise that muddies most observational work.

Five drug categories tracked with measurable reductions in phenotypic-aging markers. Vitamin D was associated with a roughly three-quarter-year decrease in the body-composition marker. Bisphosphonates, the bone-density drugs, lined up with about a two-year reduction in the energetics marker. Proton pump inhibitors, thyroid hormones, and thiazide diuretics also showed reductions in one or more domains. The confidence intervals are wide and the effects modest, but the direction is consistent and, for a field that has spent years arguing whether any of this is movable, that is news.

A word of caution is in order, and the study's authors would be the first to offer it. These are associations, not proofs of cause. A man on a thiazide for his blood pressure differs in many ways from a man who is not, and statistical adjustment can only do so much. Nor does a small numerical shift in a biological-age marker translate cleanly into more birthdays. What it does suggest is that the clocks are sensitive enough to pick up signals from ordinary clinical care — which is itself a meaningful step.

The clocks themselves are not interchangeable, and a third study this year drove that point home. Researchers analyzing data from the National Health and Nutrition Examination Survey looked at reproductive profiles, epigenetic aging, and mortality in 770 post-menopausal women aged 50 to 85. Using a clustering technique called latent profile analysis, they sorted the women into four reproductive patterns, including one defined by premature menopause.

The women in the premature-menopause group showed a faster pace of aging on DunedinPoAm and higher mortality — a hazard ratio of 1.40, with about 36 percent of that excess risk statistically mediated by the accelerated pace-of-aging score. But here is the wrinkle: PhenoAge and GrimAge, the two most widely cited biological-age clocks, did not flag the same group as older. Different clocks, in other words, are listening for different things. Pace-of-aging measures may be more sensitive to certain life-history signals than the snapshot clocks that estimate a single biological age. For readers of this column, the lesson is not about reproductive history specifically; it is about humility. When you read that someone's biological age is X, ask which clock said so.

Direct-to-consumer biological-age tests have multiplied in the last few years, and they are not all created equal. Some report a methylation-based age. Some report a pace-of-aging score. A few, riding the wave that produced BayesAge 2.0, will soon offer transcriptomic readouts. The science behind them is real, but it is also young. The numbers shift between labs, between blood draws, and between clocks. Treat them the way you would treat a single blood-pressure reading taken at a health fair: interesting, possibly useful, not a verdict.

The more durable takeaway from this year's research is structural. The field is moving from cataloging biological age toward asking what moves it. Some of what moves it, the Baltimore data suggest, may already sit in your medicine cabinet for reasons that have nothing to do with longevity. None of this is a license to start or stop a prescription on your own. It is a reason to have a more interesting conversation with the doctor you already see — about why you are on what you are on, and whether the regimen still fits the man you are now, not the one you were ten years ago.