Why Biological Age Clocks Disagree, and How to Use Them Anyway

Send one blood sample to three biological age tests and you will get three answers. This surprises patients and it should not surprise clinicians, though it frequently does, because the marketing around these tests presents a single number with a confidence the underlying measurement does not support.

The disagreement is real, well documented, and more informative than the numbers themselves.

What the discordance looks like

The pattern above is illustrative rather than drawn from a single published case, and it reflects what the 2026 literature describes when different methods are applied to the same person. A DNA methylation clock returns one figure, a blood biomarker algorithm returns a substantially lower one, and telomere length places the patient at roughly their chronological age. Three numbers, no obvious way to reconcile them.

There are at least five fundamentally different approaches in circulation, each measuring a different aspect of the ageing process. Their disagreement reflects differing questions rather than error, which is a more interesting statement than it first appears.

What each family is actually measuring

Method	What it measures	What it is reasonably good at	Principal limitation
First-generation methylation clocks	DNA methylation patterns trained to predict chronological age	Tracking chronological age closely	Trained to predict age, so deviation from age is partly noise
Second-generation clocks (PhenoAge, GrimAge)	Methylation patterns trained against mortality and morbidity outcomes	Population-level risk stratification	Individual-level precision is weaker than population performance suggests
Pace-of-ageing measures (DunedinPACE)	Rate of change rather than accumulated age	Detecting a change in trajectory	Requires repeated measurement to say anything
Blood biomarker algorithms	Composite of standard clinical markers	Interpretable, inexpensive, actionable inputs	Reflects current physiology more than accumulated ageing
Telomere length	Average leukocyte telomere length	Historically important in ageing research	High measurement variability; weak individual predictive value

Reading down that table, the divergence starts to look less like a defect. A pace-of-ageing measure and an accumulated-age measure are answering different questions, and expecting them to agree is a little like expecting a speedometer and an odometer to display the same number.

The reliability question

Reproducibility is the harder problem, and the field has been candid about it. Test-retest reliability of the older clocks remains imperfect even after principal-component-based refinements to DunedinPACE and revisions to GrimAge. Short retesting intervals magnify the issue considerably, and tracking month to month invites measurement noise to present itself as progress or decline.

Sex-specific calibration is another unresolved variable, and clinicians applying clock data to individual patients should treat it as such pending dedicated analyses.

A further point deserves emphasis because it runs against how these tests are marketed. The clocks detect large deviations from expected ageing reasonably well. Their sensitivity to lifestyle intervention over the timescales patients care about is considerably weaker. A patient who improves their diet, starts training and sleeps better for four months may see a clock reading that moves in the wrong direction, and that movement is more likely to be noise than an indictment of their effort.

Using them without overselling them

Three practices make this workable.

Pick one method and stay with it. Comparing a methylation result from one provider against a biomarker algorithm from another produces a difference that reflects the methods rather than the patient.

Extend the interval. Annual is defensible. Quarterly is measuring variability. The temptation to retest quickly is strongest exactly when a patient has made changes and wants confirmation, which is when noise is most likely to be misread as signal.

Position the result as one input among many. A clock reading placed alongside apolipoprotein B, fasting insulin, blood pressure, grip strength and cardiorespiratory fitness sits in proportion. Presented as a headline figure, it acquires an authority that the measurement does not carry, and it becomes very difficult to walk back.

Explaining a discordant result

Patients who receive three different numbers often conclude that the whole exercise is meaningless. The more accurate explanation is available and lands well when it is given directly: the tests are measuring different things, the differences between them are informative, and the reading that matters is what happens to a single measure over years rather than what any one measure says today.

The comparison that seems to help most is a familiar one. Two bathroom scales rarely agree to the pound, which is why people are told to use the same one. Nobody concludes from this that weight is unmeasurable.

Where a synthesis layer earns its place

A biological age result becomes useful in context and misleading in isolation, and assembling that context by hand across several years and several providers is slow work.

This is the specific problem Longevitix was built for. Clock results, laboratory panels, wearable-derived trends, medications and clinical notes are held as one longitudinal record, so a reading can be interpreted against the trajectory of everything else rather than on its own. The platform separates guideline-level evidence from emerging signals, and biological age measurement currently sits firmly in the second category. Every figure it presents traces back to its source, which means a clinician can see the method, the date and the provider behind a number before deciding how much weight to give it.

That traceability is not a technical nicety. In an area where the measurement is genuinely uncertain, a system that presents a confident number without showing its provenance is not helping. It is just adding a fourth answer to the three the patient already has.