Development and Validation of a Machine Learning-Based Risk Prediction Model for Sarcopenia in Community Hospital Patients: A Retrospective Cohort Study

Xue Zhao¹Wang Yao²Jiawei Shen¹Xinyu Tang³Jue Zheng¹Chang Guo¹Sun Ye¹Miqiong Li^1*Chao Wang^2,4*Peihao Yin^5,6,7*

• ¹Department of General Practice, Changshou Community Health Service Center of Putuo District, Shanghai, China
• ²Department of Oncology, Shanghai Jiaotong University School of Medicine Affiliated Ruijin Hospital, Shanghai, China
• ³Department of General Practice, Changshou Community Health Service Center of Putuo District,, Shanghai, China
• ⁴Shanghai Jiao Tong University, Shanghai, China
• ⁵Department of General Surgery, Putuo People’s Hospital, School of Medicine, Tongji University, Shanghai, China
• ⁶Department of Clinical Nutrition, Putuo People’s Hospital, School of Medicine, Tongji University, Shanghai, China
• ⁷Department of Hospital Infection Control and Prevention, Putuo People’s Hospital, School of Medicine, Tongji University, Shanghai, China

Sarcopenia, a progressive age-related loss of skeletal muscle mass and strength, represents a growing public health challenge amid global population aging. Early detection remains difficult with conventional diagnostic approaches. This study aimed to develop and validate reliable machine learning (ML) models to identify key risk factors for sarcopenia in community hospital settings. Using retrospective data from 1,650 patients at a community health center, we collected comprehensive demographic, clinical, and lifestyle variables. Twelve ML models—including Random Forest, Support Vector Machine, XGBoost, and Logistic Regression — were constructed and evaluated using 5-fold cross-validation. The CatBoost, LightGBM, and Gradient Boosting Decision Tree models demonstrated superior predictive performance, with area under the receiver operating characteristic curve (AUROC) values of 0.999, 0.996, and 0.995, respectively. SHapley Additive exPlanations (SHAP) analysis revealed that SARC_Cal_score, body mass index (BMI), and age belong to the most influential predictors, while a greater chronic disease burden was positively associated with sarcopenia risk. In conclusion, ML models show substantial potential for clinical application in identifying sarcopenia risk, thereby supporting early intervention strategies. This approach enhances detection capabilities and provides a practical tool for individualized treatment planning in community-based elderly care. Future research should integrate additional biomarkers and environmental factors to further improve model accuracy and facilitate integration into clinical workflows.