Research Square · Sep 25, 2026
Identification of Core Biomarkers and Temporal Mechanism Investigation in Cardiovascular Disease: A Machine Learning and Cross-Lagged Path Model Approach
By Zhang M, Huang Y, Liang Z et al.
Objective Utilizing the China Health and Retirement Longitudinal Study (CHARLS) database, this research aims to develop a risk prediction model for cardiovascular diseases (CVD) in Chinese residents, identify core predictive indicators, establish their causal relationships with CVD, and provide screening tools for CVD prevention and management. Methods National survey data from 2015 CHARLS database were analyzed. Univariate analysis first identified CVD-related factors. Subsequently, four machine learning algorithms - regularized logistic regression, random forest, support vector machine, and XGBoost - were optimized using 5-fold cross-validation. Data were partitioned into training-test sets (8:2 ratio). Optimized algorithms constructed CVD risk models while evaluating predictive performance. Participants with complete 2011 baseline data were selected for cross-lagged path analysis to establish causal relationships between core predictors and CVD. Results The study enrolled 10,666 participants (mean age 61.58 ± 9.43 years; 53.93% female; 63.20% rural residents) with CVD prevalence of 44.13% (4,707/10,666). Univariate analysis revealed significant differences (all P < 0.05) between CVD and non-CVD groups in gender, residence, education level, age, white blood cell count, mean corpuscular volume, blood urea nitrogen, fasting glucose, total cholesterol, triglycerides, HDL-C, LDL-C, C-reactive protein, HbA1c, cystatin C, TyG index, TyG-BMI index, creatinine, urea, hematocrit, and hemoglobin.The XGBoost model demonstrated optimal performance. Both cystatin C and TyG-BMI index emerged as top predictors across all models. Cross-lagged path analysis revealed bidirectional causality between CVD and TyG-BMI index, while unidirectional causality existed between CVD and cystatin C. Conclusion The high CVD prevalence in Chinese residents was effectively predicted by XGBoost model. TyG-BMI index and cystatin C serve as core predictive markers. The bidirectional CVD-TyG-BMI relationship and unidirectional CVD-cystatin C association suggest comorbidity risks between cardiovascular, metabolic, and renal diseases.