In a hall of mirrors, a person walks up to a curved mirror and sees themselves now as a dwarf, now as a giant. Popular machine learning models behave in much the same way when asked to guess biological age from blood tests or brain scans: they stretch age at the extremes, underestimating it in older people and overestimating it in younger ones. This bias is not random error but a systematic distortion built into the very logic of training. The model learns from chronological age as a surrogate for true biological age, because true age is unobservable, and as a result predictions shrink toward the mean, like a rubber band pulled from both ends.
The study authors formalized this bias and proposed an unbiased regression method (UMLR), which at the training stage adds two constraints: the mean predictions in two age-divided subgroups must match the true means. It is a bit like calibrating a mirror at two points so that it does not distort proportions. In simulations, the method eliminated the bias, although the root mean square error became higher and the coefficient of determination lower, the price of honesty.
Validation on independent data confirmed: standard models had a bias slope of about 0.5 to 0.7, whereas UMLR was close to 1.0. This means that many published associations between accelerated aging and disease may be an artifact of bias rather than a biological signal. The method can be applied today when building any aging clock, to obtain correct estimates of biological age and associations in epidemiological and clinical research.
Let us return to the hall of mirrors. A curved mirror is amusing as long as we know it is curved. But if we took its reflection for reality, we would draw false conclusions about our own body. The same goes for aging clocks: before trusting them, we need to make sure they do not lie at the edges of the range. Otherwise we risk confusing accelerated aging with regression to the mean, and then medical recommendations will end up reversed, as in a mirror.
In the brain age case, biased models claimed that people with better cognitive function have brains that age faster, while unbiased ones said the opposite. The same dataset led to opposite medical recommendations depending on whether the bias was corrected.
In the series Altered Carbon, biological age can be separated from chronological age with technology, but even there estimates of cerebral cortex age suffer from systematic errors if the algorithm is not calibrated. The work reminds us that before trusting 'aging clocks', we need to make sure they do not lie at the edges of the range.
Not published on arXiv. Here are its page, the PDF and the formalization: you can check it yourself, not only read our account.