When your biological-age number moves, what moved?
Repeating a laboratory result is not the same as measuring a stable trait
A biological-age test can give the same answer twice from the same blood sample and still be a shaky guide to the person who provided it. An analysis of 18 epigenetic clocks makes that distinction central. Before celebrating a younger number, it helps to ask what kind of change the test can reliably detect.
Sehgal and colleagues used the TranslAGE platform to examine DNA methylation measurements across existing datasets. Published in Aging Cell, the work was a benchmarking study, not a trial of an anti-aging treatment. It assessed both technical and biological reliability.
Two different kinds of consistency
Technical reliability asks whether a laboratory obtains similar results when it runs the same sample again. Most clocks performed well on standard replicate assays. Some were sensitive to factors such as position on a testing slide or the method used to extract DNA. Certain principal-component-based clocks remained more robust across technical conditions.
Biological reliability asks a different question: do repeated samples from the same person, taken within a short interval, produce consistent readings? The researchers examined measurements around meals, acute stress and environmental exposures, as well as samples collected over days.
Here, many clocks were less stable. A test could be technically reproducible while varying substantially within the person. The distinction matters because a technically sound laboratory measurement can still be an uncertain guide to a treatment's effect.
The intervention problem
Imagine testing before and after a new regimen and seeing an apparently younger biological age. If the samples were collected under different conditions, part of the difference might reflect those conditions rather than a durable change attributable to the regimen. That example illustrates the measurement problem; it does not establish that every favorable result is noise.
In the analysis, more reliable clocks tended to produce more stable associations with cognitive outcomes and more consistent responses in intervention datasets. Less reliable measures could yield variable or misleading effects. Technical reproducibility did not predict biological reliability.
How far the finding goes
The datasets varied in size and setting, and several biological-reliability experiments were small and short. The study did not evaluate every consumer service or identify a universal commercial winner. Nor does it show that all epigenetic research lacks value.
Its contribution is more useful than that: it identifies a condition that a meaningful progress measure must satisfy. For an individual, a small change between two tests should not automatically become evidence of rejuvenation. For researchers, stable measurement conditions and biological reliability need to be part of the design.
The aspiration remains compelling. A trustworthy measure could help reveal which interventions matter. But the ability to produce a number comes before the ability to interpret that number with confidence.
Read the research: Primary study / PubMed
Study publication: August 2026. Source edition: October 22, 2026.