What this number is, and what it is not

Biological age has a precise meaning here, and it is narrower than the phrase suggests. It is the chronological age at which your blood chemistry would be unremarkable. If your markers look like those of the average 58-year-old and your passport says 65, the calculator returns 58. That is the entire claim.

It is not a measurement taken from your cells. It is not a forecast of how long you will live. Nothing here is read off a clock inside you — it is arithmetic on fourteen laboratory values, compared against tens of thousands of other people's. I think the arithmetic is worth doing. That is why I built the page. I also think the phrase “biological age” does more work than it has earned, and the honest version of it is duller and more useful.

Two algorithms, because they disagree

The page shows two measures side by side, and they will not agree. That is deliberate.

Klemera-Doubal Biological Age [1] asks a single question: what age best explains all twelve markers at once? Each marker is regressed against age in a reference population, and a marker earns its weight from how tightly it tracks age there and how little scatter it has around that line. Systolic blood pressure earns the most. White cell count earns almost nothing.

PhenoAge [3] was built to answer a different question. Levine and colleagues fitted nine markers and age against ten-year mortality in NHANES III, then rescaled the resulting risk back into years. So PhenoAge is a mortality risk wearing the clothes of an age. It correlates more tightly with chronological age than Klemera-Doubal does, and it predicts death better.

When the two diverge by more than a few years, it usually means the inflammatory and metabolic markers are pulling in one direction and the structural ones in another. A gap of eight years between them is worth taking to your own doctor. An average of the two would hide exactly the thing you wanted to know.

How the calculation actually works

Klemera-Doubal starts by fitting, in the reference population, a straight line for each marker against age: a slope k, an intercept q, and the scatter around the line s. Your biological age is then the weighted combination

BA = ( Σ (xm − qm) ⋅ km/sm2 + CA/sBA2 ) / ( Σ (km/sm)2 + 1/sBA2 )

The second term in each half is the correction Klemera and Doubal added to their own first formula. Without it the estimate is unbiased but noisy, and the noise gets worse the weaker your markers are. With it, your chronological age is folded back in as one more piece of evidence, weighted by how much true biological age varies around chronological age in the reference population. On this biomarker set that term carries about two thirds of the answer for men and about half for women.

Rearrange that formula and something useful falls out. The gap between your biological and chronological age is exactly

BA − CA = Σ (xm − expectedm) ⋅ wm

where expected is what the regression predicts for someone your age and w is a fixed weight per marker. Each marker contributes an exact number of years, and those years sum to the gap with nothing left over. That is the chart the calculator draws. It is a decomposition, not an approximation of one.

PhenoAge takes a different route. The nine markers and your age go into a linear predictor, that becomes a ten-year mortality probability through a Gompertz survival function, and the probability is inverted back onto the age scale. I use the coefficients as Levine published them in 2018, unchanged.

Where the reference data comes from, and why that matters to your result

The Klemera-Doubal algorithm here is trained on NHANES III, the American national survey run between 1988 and 1994, in non-pregnant adults aged 30 to 75, fitted separately for men and women. That is 5,368 men and 5,995 women. I trained it against the biomarker set the BioAge toolkit calls V2 [2] and checked the result against the toolkit's own published supplement: projected into NHANES IV it correlates with chronological age at 0.935 across 35,955 people, which is the number the supplement reports, on the sample size the supplement reports.

Here is the part that changes how you should read your own result. The reference is a population measured more than thirty years ago. Adults measured between 1999 and 2018 score about two years younger than that reference on this algorithm — the median is −2.4 years for men and −2.0 for women. So “two years younger than my age” is not a good result. It is the ordinary one.

That is why the calculator puts a percentile next to every number, and why I would look at the percentile first. It compares you against 17,404 men or 18,551 women measured in the modern era, which is the comparison you actually wanted.

On the four-week diet projection

The toggle is not a simulation. It reports what happened in one randomised controlled trial, to the same quantity this page displays.

The Nutrition for Healthy Living study randomised 104 adults aged 65 to 75 to one of four diets in a two-by-two design: omnivorous or semi-vegetarian, crossed with higher fat or higher complex carbohydrate. Protein was held at 14% of energy in all four arms. Every meal was delivered for four weeks, and people ate as much of it as they wanted. Andrews and colleagues then calculated the Klemera-Doubal gap before and after [4].

The omnivorous higher-fat arm — closest to what everyone had been eating already — did not move. Measured against it, the omnivorous lower-fat arm came out 3.51 years lower (95% CI 0.35 to 6.68), and the semi-vegetarian higher-fat arm 3.52 years lower (95% CI 0.36 to 6.68). The semi-vegetarian lower-fat arm moved 3.14 years in the same direction but crossed zero (95% CI −0.04 to 6.32). Those confidence intervals are wide enough to drive a bus through, and the calculator shows them rather than the point estimate alone.

Hold three things at once here. This was a real randomised trial, with delivered food and weighed records, not a questionnaire study. But the authors themselves warn against reading it as reversed ageing: four weeks is long enough to change the physiology the markers read, and nowhere near long enough to change an ageing trajectory. And the effect is a group mean — your own response is not the group mean, and the calculator is showing you a plausible shift, not a prediction about you.

The carbohydrate in those diets was whole and minimally processed, and the arms that lowered the gap were also the arms highest in fibre and lowest in fat. The finding does not transfer to a diet built on refined carbohydrate, and the authors say so directly.

Who this is wrong for

  • Anyone under 20, or over about 85. The algorithm is trained on adults aged 30 to 75 and the modern reference thins out at the top. Outside that range the calculator gives you a number and a warning; treat the warning as the more reliable of the two.
  • Anyone acutely unwell. CRP and white cell count respond to an infection within hours. Blood taken during or just after one will return a biological age that describes the infection, not you.
  • Anyone with kidney or liver disease. Creatinine, urea and alkaline phosphatase are doing something specific in that setting, and the algorithm reads them as ageing because it has no way to know otherwise.
  • Pregnancy. Pregnant women were excluded from the training data, and almost every marker here shifts in pregnancy.
  • Anyone wanting a single verdict. Two draws a few weeks apart will not agree. The number is worth repeating over a year; it is not worth much once.

What I would do with the result

Look at the percentile before the age. Then look at the bar chart, because the two or three markers carrying most of the gap are the only part of this that points anywhere. In practice it is usually blood pressure, HbA1c and CRP doing the work, and all three are things you and your doctor already have levers for.

Then repeat it in a year with blood drawn the same way, fasting, when you are well. The trajectory across three of these is worth more than any single one of them, which is the same thing I say about growth charts and for the same reason.

Questions people actually ask about this

What is biological age, as this calculator means it?

It is the chronological age at which your blood chemistry would be unremarkable. If your markers look like those of the average 58-year-old and your passport says 65, the calculator returns 58. That is the whole of the claim. It is a statement about where your laboratory values sit relative to a reference population, not a measurement of how much life you have left and not a reading taken from your cells.

Which algorithms does it use?

Two, side by side. Klemera-Doubal Biological Age, which asks what age best explains twelve markers at once, and PhenoAge, which was fitted to predict death within ten years and then rescaled into years. They answer different questions and they disagree, often by several years. That disagreement is information, not a fault, and the page shows both rather than averaging them into a single tidier number.

Where do the numbers behind it come from?

The Klemera-Doubal algorithm is trained here on NHANES III, the American survey run between 1988 and 1994, in non-pregnant adults aged 30 to 75, fitted separately for men and women. PhenoAge uses the coefficients Levine and colleagues published in 2018, unchanged. I reproduced both against the published BioAge toolkit and its supplement: the correlation with chronological age comes out at 0.935 on 35,955 people, which is the figure the supplement reports.

Why does it say I am younger than I am?

Partly because you may be, and partly because of the reference population. The training data is American adults measured between 1988 and 1994, and the median adult measured between 1999 and 2018 scores about two years younger than that reference. So a result of minus two is average, not good. The percentile shown beside each result is the honest comparison, because it places you against people measured in the modern era rather than against the training set.

How much does one blood draw really tell me?

Less than the decimal places suggest. CRP can double with a cold, glucose moves with what you ate, and a single result carries all of that. Two draws a few weeks apart will not give the same answer. The number is worth something as a starting point and as a thing to repeat; it is not worth anything as a one-off verdict, and I would not act on a single reading.

Does the four-week diet projection mean I can reverse my ageing?

No, and the authors of the trial it comes from say so explicitly. In a fully fed randomised trial in 104 adults aged 65 to 75, swapping an omnivorous higher-fat diet for an omnivorous lower-fat, higher-fibre one moved the Klemera-Doubal gap by about three and a half years in four weeks. Four weeks is not long enough to change how you are ageing. What it is long enough to change is the physiology the markers are reading, which is a different and more modest claim.

Which markers move the number most?

For the Klemera-Doubal measure, systolic blood pressure carries the most weight, followed by blood urea nitrogen, albumin and HbA1c; uric acid and white cell count carry almost none. This is not a ranking of clinical importance. A marker earns weight in this algorithm by tracking chronological age tightly in the reference population, which is a different property from mattering to your health.

Can I use this instead of seeing a doctor about my results?

No. The calculator reads your numbers as a population statistician would and is blind to everything that makes them mean something — a kidney problem behind a high creatinine, an infection behind a high CRP, a medicine you take every morning. An abnormal individual result needs a doctor who can see the rest of you. This tool is for the pattern across the panel, once somebody competent has looked at the parts.

References

The methods, the coefficients, the training data and the trial behind the diet toggle.

  1. Klemera P, Doubal S. A new approach to the concept and computation of biological age. Mech Ageing Dev. 2006;127(3):240–248. doi:10.1016/j.mad.2005.10.004 The estimator this calculator uses, including the correction that folds chronological age back in as evidence.
  2. Kwon D, Belsky DW. A toolkit for quantification of biological age from blood chemistry and organ function test data: BioAge. GeroScience. 2021;43(6):2795–2808. doi:10.1007/s11357-021-00480-5 The reference implementation and the cleaned NHANES III and IV datasets. The V2 biomarker set used here is the one defined in its supplement, and the correlation of 0.935 on 35,955 people is the figure I reproduced against it.
  3. Levine ME, Lu AT, Quach A, et al. An epigenetic biomarker of aging for lifespan and healthspan. Aging (Albany NY). 2018;10(4):573–591. doi:10.18632/aging.101414 The PhenoAge coefficients, used here exactly as published.
  4. Andrews CJ, Ribeiro RV, Gosby A, et al. Short-term dietary intervention alters physiological profiles relevant to ageing. Aging Cell. 2026;25(5):e70507. doi:10.1111/acel.70507 The four-week randomised trial behind the diet projection, and the source of every coefficient and confidence interval it shows.
  5. Levine ME. Modeling the rate of senescence: can estimated biological age predict mortality more accurately than chronological age? J Gerontol A Biol Sci Med Sci. 2013;68(6):667–674. doi:10.1093/gerona/gls233 The first application of Klemera-Doubal to NHANES, and the source of the biomarker selection approach the V2 set follows.