Blood Biochemistry Age Clock - An interpretable estimate of biological age and predictor of all-cause-mortality derived from routinely ordered blood biomarkers.
Key claim: A Blood Biochemistry Age Clock (BBAC) built from 11 routinely ordered blood biomarkers yields an interpretable biological-age estimate that predicts all-cause mortality and chronic-disease incidence comparably to or, when optimized, better than PhenoAge in UK Biobank and NHANES cohorts.
Abstract
“Omic” aging clocks(ACs) represent an established predictor of biological-age(BA)/all-cause mortality(ACM); however, they are expensive and lack interpretability/explainability. We developed a Blood-Biochemistry Age Clock (BBAC) composed of 11 routine blood biomarkers. The AC functions by transforming the individual’s levels of blood biomarkers to years added/subtracted according to “biomarker-ACM risk-profile”. The BBAC and PhenoAge were calculated for individuals of the UKB cohort and the results were associated with ACM and disease incidence (eight common chronic diseases) using univariant/multivariate Cox mortality analysis. The BBAC calculation procedure was further optimised. Mortality analysis was likewise performed on NHANES dataset with additional ACs (PCAge/LinAge) serving as predictors. Comparisons between ACs were done using Akaike Information Criterion (AIC). In univariate ACM prediction on UKB cohort BBAC outperformed PhenoAge (AIC - 910749.6 vs. 912914.9). Similarly, BBAC was a better univariate disease incidence predictor(lower AIC). In multivariate ACM prediction PhenoAge exceeded BBAC (AIC - 890855.6 vs. 893294.4). Nevertheless, the optimised version of BBAC was better in predicting ACM in uni (BBAC/PhenoAge AIC - 154165.7/155217.6) as well as in multivariant setting (BBAC/PhenoAge AIC - 151006.9/151031.9). For multivariant disease incidence prediction the results were mixed. On the NHANES cohort BBAC and PhenoAge performed similarly, while PCAge and LinAge achieved the best mortality prediction power. BBAC is an explainable/interpretable AC with good ACM prediction ability.
Why this matters
Most phenotypic-age clocks (PhenoAge included) were trained once and then propagated through the field without many challengers using the same inputs. BBAC demonstrates that an interpretable 11-marker model built on routinely ordered bloods can match or, when optimized, beat PhenoAge on all-cause mortality and chronic-disease incidence in UK Biobank and NHANES — a rare direct head-to-head in a measurement space dominated by legacy constructs. It does not test intervention responsiveness, and matching PhenoAge on prediction is a predictive-validity claim, not evidence that moving BBAC moves aging.