Why biological age tests disagree

Epigenetic clocks are technically excellent and biologically fragile, and those two things turn out to be almost unrelated. A 2026 study in Aging Cell found that clocks reproducing the same sample with intraclass correlations above 0.90 could still shift measurably after a meal. No clock reached excellent biological reliability.

Are epigenetic clocks accurate?

It depends entirely on what you mean by accurate, and Sehgal and colleagues, published in Aging Cell in 2026, separated the two meanings.

Technical reliability asks whether running the same blood sample twice gives the same answer. Most clocks do well: intraclass correlation coefficients above 0.90 for SystemsAge, GrimAge and the principal-component-based clocks. Hannum and PhenoAge sat around 0.7 to 0.8.

Biological reliability asks whether the answer holds steady across short-term states that should not change how old you are. Here pooled ICCs ranged from 0.4 to 0.7, and no clock reached the 0.90 threshold conventionally described as excellent. Eating a meal, acute stress and pollution exposure all moved results, sometimes into the poor range.

The finding that matters most: technical and biological reproducibility were essentially uncorrelated (r = 0.0168). A clock that reproduces a sample perfectly can still be the one most disturbed by whether you had breakfast.

What about pace-of-ageing measures?

They answer a different and arguably better question. DunedinPACE, published in eLife in 2022, was built from the Dunedin birth cohort of 1,037 people tracked from age 26 to 45 across 19 biomarkers spanning cardiovascular, metabolic, renal, hepatic, immune, dental and pulmonary systems. Rather than estimating an age, it estimates a rate: how fast organ systems are deteriorating now.

It predicts cognitive and physical decline in midlife, and cardiovascular disease, stroke and mortality in older adults. Its technical reliability is high, ICC 0.96.

But it was among the weaker performers on biological reliability in the 2026 analysis, scoring poorly under acute stress. Conceptually stronger, practically still sensitive to the state you were in when the blood was drawn. Both of those things are true at once.

This is the same problem that shows up in the gap between a single fitness measurement and the slope you are on: one reading describes a day, and ageing is not a thing that happens in a day.

The bottom line

• Most epigenetic clocks show excellent technical reliability (ICC above 0.90) but only moderate biological reliability (pooled ICC 0.4 to 0.7).
• The two measures are nearly uncorrelated (r = 0.0168), so a technically precise clock is not necessarily a stable one.
• DunedinPACE measures rate of ageing rather than an age, with ICC 0.96 technically, though it scored poorly for stability under acute stress.