Biological aging measures for all-cause mortality prediction in two population-based cohorts

Abstract Biological aging measures may improve mortality prediction beyond chronological age and conventional risk factors. We evaluated PhenoAge, the Klemera–Doubal method (KDM), homeostatic dysregulation (HD), frailty index (FI), and leukocyte telomere length (TL) in TwinGene (n = 9,617) and UK Biobank (UKB; n = 314,064). Separate multivariable Cox models evaluated each biological aging measure, adjusting for age, sex, body mass index, smoking, and education, with twin-pair-clustered robust variance estimation in TwinGene. Per 1-SD increase, mortality HRs for age-adjusted PhenoAge and KDM residuals, log HD, and FI ranged from 1.11–1.27 in TwinGene and 1.24–1.40 in UKB; TL HRs were 0.96 (95% CI: 0.93–0.99) and 0.94 (0.92–0.95), respectively. In 10-fold cross-validation, Cox prediction models adding PhenoAge residual, FI, and TL to the conventional model increased Harrell’s C-index from 0.780 to 0.787 in TwinGene and from 0.720 to 0.746 in UKB. Incremental prognostic gains were larger in UKB.