Overall ranking
Mean rank Lower is better
Arithmetic mean of the selected domain ranks, weighted equally.
Age prediction accuracy
R-squared Higher is better
R-squared describes how closely clock estimates track chronological age across datasets.
Disease discrimination
Area Under the Curve Higher is better
Area under the ROC curve for distinguishing disease status across datasets; 0.50 is chance-level discrimination.
Mortality association
Hazard ratio Higher is better
Sample-count-weighted median dataset-level Cox hazard ratio per 1 SD adjusted age acceleration.
Cell-composition influence
Absolute residual shift (years) Lower is better
Mean absolute residual shift after adjusting for seven cell fractions, averaged equally across datasets.
Within-person reliability
Mean absolute deviation Lower is better
Mean absolute difference between predicted-age change and elapsed years, averaged equally across datasets.
Geographic generalizability
Country-balanced MAE (years) Lower is better
Median across countries of within-country median dataset MAE, with equal country weights.
Responsiveness to interventions
Direction-aligned age difference (years) Higher is better
Mean direction-aligned adjusted age difference across interventions and exposures, with equal comparison weights and median reference directions.