How biological age is calculated: four methods and why their numbers disagree

“What is my biological age” sounds like a question with one answer. It has several, and they disagree: run the same blood sample through different methods and you get different numbers. That is not a flaw in the measurement. The methods are counting different things, and understanding what each one counts is more useful than collecting another number.

What is being measured

Chronological age counts time. Biological age is an attempt to reflect accumulated change in cells and tissues — how far the body has actually travelled along the path of ageing. The general framing is covered separately in what biological age is and how it is measured. This piece is about something else: how that number is produced, and why the answers diverge.

Method one: epigenetic clocks

The best-known class. DNA methylation — chemical marks at particular sites in the genome — changes with age. A model is trained to predict chronological age from those marks, and the gap between prediction and passport age is read as “biological”.

The first such models were published in 2013: Horvath's clock, trained across multiple tissues, and Hannum's clock, trained on blood. Hannum's clock was trained on blood from adults aged 19–101. Horvath's is multi-tissue, built on a sample spanning childhood to old age; both predict calendar age.

The next generation was trained differently — not on age, but on outcomes. PhenoAge (Levine et al., 2018) was trained on a composite clinical “phenotypic age” itself derived from mortality in NHANES III; GrimAge (Lu et al., 2019) on mortality and associated biomarkers. DunedinPACE (Belsky et al., 2022) stands apart: it estimates not age but the pace of ageing — how many years of biological ageing accrue per calendar year — and was built on a birth cohort: the pace was modelled from repeat assessments between ages 26 and 45.

Why this matters to a reader. First-generation clocks answer “how old do your molecular marks look”. Second-generation clocks answer “how well does your molecular profile predict the risk of adverse outcomes in the cohorts studied”. Those are different questions, and there is no reason for them to give the same answer.

Method two: composite scores from clinical labs

No epigenetics here. Ordinary laboratory values — inflammation, metabolism, kidney and liver function, blood counts — are combined into a single number. The classic approach of this class is the Klemera–Doubal method (2006) — a mathematical way of estimating biological age from a set of age-correlated biomarkers. The specific marker panels are later applications of it, for example on NHANES data.

The practical advantage: the inputs come from a routine panel a person already has. The drawback: the number depends entirely on which markers are in the set, and different authors use different sets.

Method three: functional tests

Gait speed, grip strength, VO2max — measurable properties of how the body works right now. They are rarely labelled “biological age”, yet in observational cohorts they are consistently associated with age-related outcomes — gait speed studied in people aged 65 and over, grip strength in middle-aged and older adults, cardiorespiratory fitness in adult cohorts. We covered one of them separately: grip strength in the clinic.

Functional tests measure not molecular state but its result: how accumulated change shows up in how the body performs.

Method four: consumer tests

A separate category worth naming plainly. Commercial “find out your biological age” kits rest on the same classes of model, but the model's composition, training set and reproducibility are often undisclosed. A number whose origin cannot be checked cannot be interpreted.

Why the numbers disagree

Three reasons, all of them expected:

1. The models were trained on different targets. One predicts calendar age, another outcome risk, a third the pace of change. Agreement was never part of the design.
2. Different tissues, different cohorts. A model trained on adult blood does not transfer automatically to another tissue or another age group.
3. Technical reproducibility. Re-measuring the same sample does not return an identical result; the size of that spread differs between models and is not always published.

The practical consequence: comparing your number from one method with someone else's from another is meaningless. Only measurements made by the same method — ideally in the same laboratory — are comparable.

What these methods do not say

• They do not diagnose and do not prescribe. That is a clinician's work.
• An observational association between a marker and an outcome does not mean that changing the marker changes the outcome. Those are different claims, and the second does not follow from the first.
• Epigenetic clocks and composite biological-age scores are not regulator-approved and are not recommended by professional societies as instruments for individual clinical decisions: they were developed and validated on groups, not for one person. Gait speed and grip strength are used in geriatrics and in sarcopenia assessment — but on their own, outside the “biological age” frame.

What to do with this

More useful than a single figure is the trend within one method: how your value behaves over time. That is the argument for keeping measurements in one place and in comparable form — in the Lonevi record you can see how your values move from test to test, rather than one number detached from its history.

And it is worth discussing with a clinician in the frame it was produced in: which method, which laboratory, what spread.

---

This material is informational and does not replace consultation with a doctor.