Combining biomarkers such as blood-test measurements into an age-related score can summarize many values in one number. Expressing that number in years makes comparisons look straightforward. Whether it carries the same meaning for people surveyed in a later period is a separate question.
I examined how an ageing clock fits its development data and how it behaves in later data. A good fit during development should not stand in for evidence of later stability was the starting point of this study.
What to decide before looking at later data
I used biomarkers from a public US health survey to develop models in an earlier period and apply them to a later one. If the calculation is continually adjusted after seeing the later data, it becomes difficult to ask whether the original model carries across unchanged. I therefore separated model development from the out-of-time evaluation.
That separation concerns more than the final equation. Decisions about measurements, units and the handling of values can also affect the result. I organized the analysis so those conditions were set in the development data before being applied to later data, and prepared the manuscript alongside the analytical material.
What a moving score can mean
Temporal drift here concerns changes in the pattern of scores across survey periods. The sampled population and measurement context can change together, so a shift cannot automatically be read as a change in people’s rate of ageing. The comparison across survey periods uses different respondents in each period, so it cannot be read as tracking one person’s ageing over time.
I examined chronological-age prediction, stability of the score’s calibration across periods, and association with health outcomes separately. A useful result on one does not settle the others. Keeping the comparisons separate makes it easier to explain what a score might be used for, and what the analysis has not established.
I revised the manuscript and submitted it to GeroScience, which acknowledged receipt on 13 September 2026. It has not been accepted for publication. Predictive performance and medical interpretation remain unpublished. I do not present this work as a new clock proven superior to existing tools or as a way to judge an individual’s health. Once there is a live version of record and an active DOI, I plan to publish the conditions of the out-of-time evaluation alongside the results.