A novel biomarker combination of adenosine deaminase and total bile acids for predicting 3-year coronary artery disease risk in middle-aged and older hospitalized patients

Coronary artery disease (CAD) onset is a leading cause of mortality and morbidity. Short-term risk assessment, which generates greater urgency than long-term estimates, can prompt immediate preventive actions. However, validated biomarkers and tools for near-term prediction are scarce. This study aimed to identify and validate novel biomarkers for predicting 3-year incident CAD to facilitate timely and targeted risk management. This retrospective cohort study included 353 patients without history of CAD in the index hospitalization. Based on follow-up coronary angiography results, patients were classed into CAD (incident CAD within 3 years following their index hospitalization, n = 130) and control groups ( n = 223). Cox regression, restricted cubic splines (RCS), and eXtreme Gradient Boosting (XGBoost) machine learning (ML) modeling were used to identify and validate biomarkers and construct a risk predictive ML model. Elevated baseline plasma adenosine deaminase (ADA) and total bile acids (TBA) levels were significantly associated with a higher risk of CAD onset within 3 years. Multivariable Cox regression confirmed baseline ADA ( HR = 1.525, 95% CI 1.246–1.867, P < 0.0001 ) and TBA ( HR = 1.253, 95% CI 1.055–1.489, P = 0.01 ) as independent predictors of 3-year CAD risk. RCS analysis revealed linear dose-dependent relationships between both biomarkers and CAD risk. Combining ADA and TBA (cutoff > 12.25) enhanced risk stratification, with the high-risk group exhibiting a 2.795-fold increased relative risk and an elevated hazard ratio of 3.11 (95% CI 1.985–4.878) in the fully adjusted Cox model. Subgroup analyses confirmed consistent associations across all strata. An XGBoost model incorporating seven biomarkers, including ADA and TBA, achieved excellent predictive performance with an AUC of 0.846 (95% CI 0.803–0.888) for the full dataset. SHAP analysis identified ADA and TBA as the most influential predictors. Survival analysis demonstrated that patients identified as high risk by the XGBoost model faced a nearly eightfold higher CAD incidence within 3 years compared to low-risk individuals (HR = 7.97). ADA and TBA serve as potential biomarkers for predicting 3-year CAD risk in middle-aged and older adults, and their incorporation into an XGBoost model yields superior predictive performance.