Understanding your biological-age estimate

What biological-age models measure, how uncertainty affects the result, and how to interpret changes over time.

Where the clocks come from

PhenoAge (Levine et al., 2018) was built on NHANES: nine routine blood markers plus age, fitted to predict mortality in a US population sample and validated on later cohorts. A ten-year gap between PhenoAge and chronological age carried roughly a doubling of mortality risk in the derivation data. Epigenetic clocks (Horvath, GrimAge) do something similar with DNA methylation. All of them are better predictors of what happens to a population than chronological age alone.

What that means for one person

A population instrument read on one person carries two kinds of uncertainty. The first is the model's own error. The second, and larger in practice, is biological variation: re-draw the same person next week and albumin, CRP and white-cell count move, and the clock moves with them. CRP alone varies about 40% within a person from draw to draw.

That is why a clock reported as a single number is misleading. Propagate the within-person variation through the model and PhenoAge on a healthy adult spans several years at 95%. A change of two years between two draws is usually inside that range.

What moves it, and what doesn't

The nine PhenoAge inputs are dominated by inflammation and metabolic markers. In practice that means the clock responds to things a practice can change: glucose control, inflammation, body composition, sleep and training load. It does not see lipids, blood pressure or fitness directly, which is why a second, broader engine is useful beside it.

It also responds to things that are not ageing. A vaccine ten days before a draw raises CRP. A blood donation lowers hematocrit and can move the red-cell indices. A hard training block moves white-cell count. Each of those can shift the clock by a year or more for one draw.

How to read one responsibly

• Ask for the range, not just the point.
• Ask which draw fed it, and what was happening in the ten days before that draw.
• Treat a change smaller than the range as no change.
• Look at the inputs. A clock that moved because CRP moved is telling you about CRP.
• Use two methods that share few inputs. Where they agree, the signal is stronger; where they disagree, the disagreement is the finding.

What CHAI does with this

CHAI shows the range, the Centurion Clock and PhenoAge figures and the factors that shaped the estimate on the home screen, flags "no clear trend yet" when the swing exceeds the move, keeps input notes for provisional assumptions, and dates every protocol event, vaccine and donation on the same timeline as the draws, so what changed around a move is on the chart beside it.

References: Levine ME et al., "An epigenetic biomarker of aging for lifespan and healthspan," Aging 2018;10(4):573–591. Correction, PLoS Medicine 2019. EFLM Biological Variation Database for within-person coefficients of variation.