Inferential boundaries of age prediction: why prediction does not establish biological age measurement

Age prediction through chronological clocks derives a biological age from various measurements, but this output is often treated as the biological age itself. This leads to misinterpreting ageing acceleration and considering intervention-induced reductions as rejuvenation. By analyzing the age supervision, we identify an age-task statistic that is not a biological-age construct.

This distinction impacts biomarker construction and validation. Even at optimal population performance, the observable distribution and age-prediction capabilities can still have incompatible biological-age meanings. To resolve this ambiguity, one needs assumptions or evidence beyond the age task. The squared-error age loss focuses on within-age output dispersion but does not define its biological direction.

When given age and background, a gap merely re-expresses the compressed score; complete age recovery, despite measurement heterogeneity, eliminates it. Biological covariance allows for genuine prognostic value without establishing a unique construct. In allogeneic hematopoietic stem-cell transplantation, recipient-blood scores display donor-lineage affiliation excess under a score-pairing null.

NHANES data show that age-trained scores enhance the prediction of five-year mortality beyond age and background. Nonetheless, modeling the source measurements and mortality supervision at a matched scalar capacity yields additional improvements. Hence, the true biological object must dictate study design, measurement choices, and representation.

Validation should establish the claimed measurement relation rather than depend solely on age-prediction success.