Biological age tests, ranked by how much you should trust them

The going rate for finding out how old you really are is about €300, a saliva tube and a three-week wait. What comes back is a number, a colour-coded chart, and — if the marketing is any good — the faint implication that you have been handed a reading off some internal odometer. It is a genuinely impressive piece of science. It is also, for the specific purpose most people buy it for, worse than a stopwatch and a wall.

That is the argument of this post, stated plainly: rank biological age tests by the evidence behind them, how repeatable they are, and whether the number responds to anything you do, and the cheap function tests beat the expensive molecular ones. Not because the molecular ones are junk — they aren’t — but because “predicts death across ten thousand people” and “can tell whether you had a good year” are different jobs, and only one of them is what you’re paying for.

We already ranked the underlying longevity biomarkers by how well they predict mortality. This is the shopping version: the actual tests you can take, in the order I’d trust them.

How I ranked them: three filters, all of which have to pass

Most “best biological age test” lists rank by price, packaging, or how confident the founder sounds on a podcast. I used three filters, and a test has to clear all three to rank high.

1. Evidence. Does the output predict hard outcomes — death, disease, disability — in large prospective cohorts, independent of chronological age? This is the one everyone quotes.

2. Reliability. If you took the test twice this week, how different would the two answers be? This is the one nobody quotes, and it’s the one that decides whether a test can be a tracker. A year of genuinely good training might buy you a year or two of “biological age”. If the test’s own noise is bigger than that, you are watching static.

3. Responsiveness. If you do the work, does the number move — within a timeframe a human being will tolerate? A test that only changes over a decade is a fine research instrument and a terrible feedback loop.

Bit nerdy, but that third filter is why the ranking looks upside down compared to the price list. Right, the list.

1. A measured VO2 max (“fitness age”)

Evidence: exceptional. Reliability: good. Responsiveness: 8-12 weeks.

Cardiorespiratory fitness is the strongest single mortality predictor anyone has produced. In the 2018 JAMA Network Open analysis of 122,007 treadmill-tested patients (Mandsager et al.), the least-fit group carried roughly five times the mortality risk of the elite-fit group, with no upper limit of benefit. Convert your VO2 max to the age at which it would be an average score and you have “fitness age” — which is, unglamorously, the best-evidenced biological age number on the market.

It also passes filters two and three, which the molecular tests struggle with. A properly conducted test repeats tightly, and VO2 max responds to structured training inside a training block — our 12-week VO2 max plan exists because the number moves that fast. The catch is measurement: a true lab test needs a mask and a graded protocol to exhaustion. A watch estimate or a submaximal step test is a rougher instrument, but tracked consistently on the same device it still beats almost everything below. More on why VO2 max sits at the top of the evidence pile.

2. The free two-minute physical: grip, gait speed, one-leg stand

Evidence: strong. Reliability: high. Responsiveness: weeks to months.

Three tests, no equipment beyond a €20 hand dynamometer and a phone timer, and between them a stack of mortality data that shames most of what’s sold in a box.

Grip strength: in the PURE study (Leong et al., Lancet 2015, ~139,000 people across 17 countries), each 5 kg drop was associated with a 16% higher risk of all-cause death — outpredicting systolic blood pressure for cardiovascular death. Gait speed: a pooled analysis of 34,485 older adults (Studenski et al., JAMA 2011) found survival rose steadily with usual walking pace across every age band. And the 10-second one-leg stand: among 1,702 adults aged 51-75, those who couldn’t hold it had a 1.84× higher adjusted risk of all-cause death over a median seven years — 17.5% of the failures died versus 4.6% of those who passed (Araujo et al., British Journal of Sports Medicine 2022, DOI). We wrote up what the balance test actually measures separately.

None of these is as discriminating alone as VO2 max. Taken together, repeated monthly, at zero cost, and all three trainable — that’s the best trust-per-euro on this list by a distance.

3. Phenotypic age from a routine blood panel

Evidence: good. Reliability: moderate. Responsiveness: months.

The PhenoAge model takes nine ordinary blood markers — albumin, creatinine, glucose, CRP, lymphocyte percentage, mean cell volume, red cell distribution width, alkaline phosphatase, white cell count — plus your chronological age, and returns an age-equivalent that predicts mortality and healthspan better than your birthday does (Levine et al., Aging 2018, DOI). Your GP has probably already drawn most of it.

Two honest caveats. Some inputs wobble — CRP spikes with any infection, so one panel taken while you’re fighting a cold will libel you. And because it leans on markers that are abnormal when something is wrong, it’s better at catching something is going on than at scoring a healthy person’s incremental progress. Annual, not quarterly.

4. DunedinPACE — the pace-of-aging methylation test

Evidence: strong. Reliability: high, by design. Responsiveness: unclear.

This is the methylation test I’d actually take. Rather than estimating how old you are, DunedinPACE estimates how fast you’re currently aging, distilled from 19 organ-system indicators tracked across two decades in the Dunedin 1972-73 birth cohort, and deliberately built by excluding unreliable probes. It shows high test-retest reliability, associates with morbidity, disability and mortality, and adds prediction beyond GrimAge (Belsky et al., eLife 2022, DOI).

It sits at 4 rather than 1 because filter three is still open: we don’t have good evidence about how quickly, or how much, a real person’s DunedinPACE responds to training. It’s a research instrument that happens to be purchasable.

5. GrimAge and DNAm PhenoAge — the mortality-trained clocks

Evidence: excellent. Reliability: poor for individuals. Responsiveness: unproven.

GrimAge combines seven DNA-methylation surrogates for plasma proteins with a methylation estimate of smoking pack-years, and predicts time-to-death better than any clock before it (Lu et al., Aging 2019, DOI). Population-scale, this is superb science. Even the first mortality work on the original clocks held up: a 5-year gap between methylation age and real age carried a 21% higher mortality risk across four cohorts, 16% after adjusting for the usual suspects (Marioni et al., Genome Biology 2015, DOI).

Then filter two arrives. Technical noise alone produced deviations of up to nine years between replicate samples for six prominent clocks — the same blood, run twice (Higgins-Chen et al., Nature Aging 2022, DOI). Nine years. Whatever a year of disciplined training does to your epigenome, it is smaller than that. The same paper offers the fix — principal-component versions that agree within about 1.5 years — but you have to ask your provider whether they use one, and most consumer kits don’t say.

6. Body composition as “body age”

Evidence: moderate. Reliability: high (DEXA), poor (smart scales). Responsiveness: months.

Where fat sits matters more than what the scale says — visceral fat is the metabolically nasty kind, and a tape measure round the waist captures much of the signal for free. DEXA repeats well; consumer bioimpedance scales drift with hydration enough to make a two-day “gain” meaningless. Useful as one input, oversold as an age.

7. Questionnaire quizzes — including ours

Evidence: indirect. Reliability: perfect and meaningless. Responsiveness: only when you change an answer.

A quiz measures nothing. It maps your answers onto published hazard ratios and ranks you against population averages. That sounds damning until you notice that some of the strongest predictors in the literature are self-reported — usual walking pace and self-rated health both carry real signal.

So it’s an excellent first map: Sarvita’s free biological age test breaks the result into six system ages, which is genuinely useful for finding which part of you is aging fastest before you spend a cent. As a tracker it’s circular — the number moves when you tell it you started training, not when you actually got fitter. I work at Sarvita and I’d still rank it here.

8. First-generation clocks sold as “your biological age”

Evidence: weak for this purpose. Reliability: poor. Responsiveness: unproven.

The Horvath and Hannum clocks were trained to predict chronological age, and they do it beautifully — around 0.96 correlation with your passport. Which is the problem: a kit running a first-generation clock and reporting “your biological age is 41, and you’re 42” has mostly demonstrated that the assay worked. Any deviation is a mix of genuine aging signal and measurement noise, and you can’t tell which. If a provider won’t name its clock, assume this tier.

9. Telomere length tests

Evidence: weak at the individual level. Reliability: poor. Responsiveness: no.

Telomeres shorten with age; the biology is real and the Nobel is deserved. The individual test is still the weakest thing on this list. In an individual-participant meta-analysis of 12,199 adults, the shortest telomere quintile had a 23% higher all-cause mortality risk than the longest — modest next to a five-fold fitness spread — and the cause-specific estimates were too imprecise to call, at 1.29 (95% CI 0.83-2.00) for cardiovascular death and 1.10 (95% CI 0.88-1.37) for cancer (age-adjusted; Mons et al., American Journal of Epidemiology 2017, DOI). Layer laboratory variability on a small effect and a single result tells you approximately nothing about you.

The placement people will argue with

Putting GrimAge below a hand dynamometer looks like heresy, so let me be precise: on filter one, the clocks win. If I had one blood sample from ten thousand strangers and had to guess who’d die first, I’d take GrimAge over grip strength.

That’s not the question you’re asking. You’re asking whether your number went down. That’s a within-person question, and it’s decided by the ratio between the change you can produce and the noise the test carries. Fitness measures win that ratio by a mile — a real 3 ml/kg/min VO2 max gain dwarfs the test’s error, whereas a plausible year of epigenetic improvement sits comfortably inside a nine-year replicate spread.

The other honest caveat, which applies to everything above: these tests predict mortality partly because illness drags them down, not only because they cause anything. Being unwell makes you slow, weak and wobbly. Gaming a test buys you nothing; doing the training that moves it buys you the thing the test is a proxy for.

What would change this ranking

Three things, and I’d revise happily:

• Consumer kits shipping principal-component clocks and publishing their test-retest reliability. That fixes filter two directly and would push the methylation tier up several places.
• Intervention data on DunedinPACE. If a training or dietary trial reliably shifts pace-of-aging within a year, filter three opens and it has a strong case for the top.
• Cheap, repeatable proteomic organ-age panels. The organ-specific work is the most interesting thing in this field, and it isn’t reliable or affordable enough yet to rank.

How to actually use this

Budget €0. Time a 10-second one-leg stand, measure your usual walking pace over a marked distance, and note your resting heart rate. Repeat monthly. This is the highest-value thing on the page and it costs a stopwatch.

Budget ~€50. Add a hand dynamometer and a real VO2 max estimate — a submaximal step test, or a watch estimate tracked consistently on one device. Retest quarterly; that’s the interval at which training actually shows up.

Budget ~€300. Get a proper blood panel and score it as phenotypic age, annually. Spend the rest of the money on a gym membership. If you still want the methylation test, buy it once, for curiosity, and pick a provider that names its clock and quotes its reliability.

And whichever you use: track the trend, not the number. One reading is a data point with an error bar you can’t see. Four readings on the same test, same conditions, over a year, is information — which is the whole reason the cheap tests win. You’ll take them again. Nobody re-runs a €300 saliva kit monthly, and if you did, you’d mostly be measuring the lab.

If you want the concept behind all of this rather than the shopping list, what biological age actually means is the place to start.