A machine learning-based biological aging prediction and its associations with healthy lifestyles: the...

This study aims to establish a biological age (BA) predictor and to investigate the roles of lifestyles on biological aging. The 14,848 participants with the available information of multisystem measurements from the Dongfeng-Tongji cohort were used to estimate BA. We developed a composite BA predictor showing a high correlation with chronological age (CA) (r = 0.82) by using an extreme gradient boosting (XGBoost) algorithm. The average frequency hearing threshold, forced expiratory volume in 1 second (FEV₁ ), gender, systolic blood pressure, and homocysteine ranked as the top five important features for the BA predictor. Two aging indexes, recorded as the AgingAccel (the residual from regressing predicted age on CA) and aging rate (the ratio of predicted age to CA), showed positive associations with the risks of all-cause (HR (95% CI) = 1.12 (1.10-1.14) and 1.08 (1.07-1.10), respectively) and cause-specific (HRs ranged from 1.06 to ∼1.15) mortality. Each 1-point increase in healthy lifestyle score (including normal body mass index, never smoking, moderate alcohol drinking, physically active, and sleep 7-9 h/night) was associated with a 0.21-year decrease in the AgingAccel (95% CI: -0.27 to -0.15) and a 0.4% decrease in the aging rate (95% CI: -0.5% to -0.3%). This study developed a machine learning-based BA predictor in a prospective Chinese cohort. Adherence to healthy lifestyles showed associations with delayed biological aging, which highlights potential preventive interventions.