Objective:

To identify the most informative composite biomarker and the optimal machine learning (ML) algorithm for developing and validating a sarcopenia risk stratification system in patients with cardiovascular disease (CVD).

Approach:

• Data Analysis: Analyzed data from CHARLS, ELSA, and a clinical cohort to examine associations between composite biomarkers and sarcopenia.
• Machine Learning Model Development: Developed nine ML models using a longitudinal CHARLS sub-cohort, optimizing hyperparameters via 10-fold cross-validation.
• Model Validation: Validated the best-performing CatBoost model incorporating TyG-BMI in ELSA and the clinical cohort, deriving the Cardiovascular Disease–Sarcopenia Risk Score (CVD-SRS).

Key Findings:

• TyG-BMI showed the highest discriminative ability for sarcopenia (AUC = 0.938).
• The CatBoost model demonstrated good discrimination with AUCs of 0.982 (training set), 0.907 (internal validation), 0.891 (ELSA), and 0.900 (clinical cohort).
• The CVD-SRS provided consistent risk stratification across internal and external validation cohorts.

Interpretation:

A CatBoost model integrating TyG-BMI demonstrated good performance for sarcopenia risk stratification in patients with CVD.

Limitations:

• The study may be limited by the generalizability of findings across diverse populations.
• Potential biases in data collection and analysis methods could affect results.

Conclusion:

The CVD-SRS may assist in early screening and identifying individuals who require further sarcopenia assessment.