Liver Biomarker Improves AHA/ACC 10-year ASCVD Risk Prediction in US and China Cohorts with ML

Introduction: Accurate stratification of hard atherosclerotic cardiovascular disease (ASCVD) risk remains challenging despite advances in prevention. Liver function biomarkers (LFBs), particularly gamma - glutamyl transferase (GGT), have been linked to cardiovascular outcomes, yet their contribution to hard ASCVD risk prediction is not well defined. Methods: This study analyzed data from the National Health and Nutrition Examination Survey (NHANES, 2005 - 2018) to assess cross - sectional associations between LFBs and 10 - year hard ASCVD risk estimated by the ACC/AHA Pooled Cohort Equations. Multivariable regression, restricted cubic splines, and mediation analyses were applied to examine independent and dose - response relationships. External validation was performed in the China Health and Retirement Longitudinal Study (CHARLS) and NHANES using machine learning models (CoxBoost, Naive Bayes and Random Forest). Results: Among 5,731 NHANES participants, GGT showed an independent linear association with hard ASCVD risk (P - trend = 0.003), partly mediated by systolic blood pressure (44.8%), HbA1c (19.0%), and high density lipoprotein cholesterol (13.4%). Machine learning (ML) models incorporating GGT, alkaline phosphatase (ALP), and globulin alongside traditional risk factors improved predictive accuracy, with Naive Bayes achieving an AUC of 0.751 in NHANES validation. Conclusions: GGT is an independent and biologically plausible biomarker of hard ASCVD risk, acting through cardiometabolic pathways. Incorporating LFBs into risk prediction models, particularly with machine learning, enhances risk stratification and may facilitate early identification of high - risk individuals.