Related Experiment Video
Updated: Aug 15, 2026
Identification of Disease-related Spatial Covariance Patterns using Neuroimaging Data
Published on: June 26, 2013
Development and external validation of a composite biomarker-based machine learning model for sarcopenia risk
Pengcheng Mei1, Tao Ying1, Jing Wu1
1Department of Geriatrics, The Third People's Hospital of Chengdu (Affiliated Hospital of Southwest Jiaotong University), College of Medicine, Southwest Jiaotong University, Chengdu, Sichuan, China.
Background:
Sarcopenia is common in patients with cardiovascular disease (CVD) and is associated with functional decline and adverse clinical outcomes. However, practical tools for early risk stratification in this population remain limited, particularly those incorporating composite metabolic biomarkers.
Objective:
To identify the most informative composite biomarker and the optimal machine learning (ML) algorithm for the development and validation of a sarcopenia risk stratification system in patients with CVD.
Methods:
We analyzed data from the China Health and Retirement Longitudinal Study (CHARLS, 2011-2015), the English Longitudinal Study of Ageing (ELSA, 2004-2013), and a hospital-based clinical cohort enrolled between 2023 and 2024. Cross-sectional analyses in CHARLS (n = 1,343) were used to examine the associations between candidate composite biomarkers and sarcopenia. Nine ML models were developed in a longitudinal CHARLS sub-cohort free of sarcopenia at baseline (n = 887), with hyperparameters optimized by 10-fold cross-validation. The best-performing CatBoost model incorporating TyG-BMI was further interpreted using Shapley Additive Explanations (SHAP) and externally validated in ELSA (n = 823) and the clinical cohort (n = 2,497), on the basis of which the Cardiovascular Disease-Sarcopenia Risk Score (CVD-SRS) was derived.
Results:
Among the evaluated composite biomarkers, TyG-BMI showed the highest discriminative ability for sarcopenia (AUC = 0.938). The CatBoost model showed good discrimination across cohorts, with AUCs of 0.982 in the training set, 0.907 in the internal validation set, 0.891 in ELSA, and 0.900 in the clinical cohort. The CVD-SRS provided consistent risk stratification across the internal and external validation cohorts.
Conclusions:
A CatBoost model integrating TyG-BMI showed good performance for sarcopenia risk stratification in patients with CVD. The CVD-SRS may facilitate early screening and help identify individuals who warrant further sarcopenia assessment in clinical practice.
