Cardiac and systemic biological age capture complementary aging signals associated with cardiovascular disease.

BACKGROUND: Biological ages capture aging signatures associated with aging-related outcomes, but most models use systemic and unimodal indices. The added value of organ-specific and multimodal information for assessing organ-specific disease associations remains unclear. METHODS: We enrolled 1,830 participants (77.81% male; age 55.87 ± 11.07 years) undergoing blood chemistry testing and echocardiography. Cardiac, systemic, and multimodal biological age models were trained using supervised machine learning with 20-fold cross-validation to predict chronological age from LASSO-selected variables among 15 echocardiographic indices, 73 blood-based indices, or their combination. Biological age acceleration (BAA) was derived using modality-specific age-bias correction functions fitted in the apparently healthy reference cohort and standardized before analysis. Associations between BAA and prevalent cardiovascular disease (CVD; ICD-10 I00-I99) were evaluated using multivariable logistic regression. Models were developed in 937 apparently healthy participants; CVD analyses compared 810 cases with 1,020 CVD-negative participants. RESULTS: Cardiac biological age showed moderate accuracy (mean absolute error [MAE] = 6.08 years; R 2 = 0.34), with A1 (late diastolic transmitral flow velocity) and septal e' (early diastolic mitral annular velocity) as the leading SHAP contributors. Systemic biological age performed better (MAE = 4.60 years; R 2 = 0.65), with estimated glomerular filtration rate and creatinine as the leading contributors. Cardiac BAA showed a numerically larger association with prevalent CVD than systemic BAA (odds ratio [OR] = 1.36, 95% confidence interval [CI] 1.23-1.52 vs. OR = 1.31, 95% CI 1.20-1.43). Cardiac and systemic biological ages showed partial overlap (R 2 = 0.25). The multimodal model achieved the best age-prediction performance (MAE = 4.29 years; R 2 = 0.70) and was associated with prevalent CVD (OR = 1.40, 95% CI 1.28-1.53). Subtype-specific associations were broadly directionally consistent; significance was reached for hypertension, whereas estimates for smaller subtypes remained imprecise. CONCLUSIONS: Cardiac and systemic biological age models captured complementary aging information and showed distinct strengths in chronological-age estimation and cross-sectional CVD assessment. The multimodal model achieved the lowest MAE and highest R 2, supporting the complementary value of multimodal integration for biological aging assessment.