What does MAE (Mean Absolute Error) of 4.4 years mean? Is that clinically useful?

MAE is the average difference between the model’s predicted brain age and a person’s chronological age across the population. A common misconception is that lower MAE is always better — but in a biological age model, if you engineered MAE toward zero, the model would collapse into a chronological clock, and the gap between predicted biological age and chronological age — the signal you actually care about — would disappear. A portion of that 4.4-year MAE reflects real biological variance between people, which is what makes the score informative.

This is why the recent PLOS Biology analysis of brain-age models found that models with lower age-prediction accuracy actually showed higher sensitivity for detecting clinically meaningful deviations.

For context on the numbers: published EEG-based brain age models typically report MAE of 5–8 years. Even large-scale MRI-based models land in the 3–5 year range. BrainYears™ at r = 0.923 with MAE = 4.4 years is at the strong end of the published literature.

Critically, within-subject precision is far tighter than the cross-sectional MAE: our comparison group showed a mean change of just +0.07 years between assessments. In other words, the within-subject signal is far tighter than the 4.4-year cross-sectional MAE would suggest, and movement on an individual’s score can be interpreted with confidence.