Randomized trial demonstrates effective risk stratification for cardiovascular-kidney-metabolic syndrome using machine learning in diverse populations, indicating clinical potential.
Key Points
This research aims to develop and validate a machine learning classifier for predicting stages of cardiovascular-kidney-metabolic syndrome using standard indices.
Analyzed data from 12,106 participants in NHANES survey.
Selected 10 biomarkers through feature selection after addressing multicollinearity.
Evaluated several machine learning algorithms, with LightGBM showing the highest performance (ROC AUC: 0.88).
LightGBM model achieved a ROC AUC of 0.88 for predicting CKM syndrome stages.
External validation with CHARLS dataset yielded a ROC AUC of 0.84.
SHAP analysis identified metabolic markers such as eGDR and TyG as key predictors for CKM stages.