Developing a sarcopenia identification model with eight machine learning algorithms

Sarcopenia is characterized by progressive loss of skeletal muscle mass and function. It affects approximately 10–16% of older adults globally and up to 20.7% in China. Sarcopenia is significantly associated with an increased risk of falls, fractures, hospitalization, and all-cause mortality. Despite a growing number of machine-learning models for sarcopenia identification, fewer than 10% of these models have been externally validated. Modeling data were obtained from a community-based screening program of 919 older adults in Fenghua District, Ningbo in 2021; external validation was performed in an independent study of 1,037 older adults in Changshan County, Quzhou during 2023. Machine learning models were developed and compared to predict the risk of sarcopenia. Eight machine learning algorithms were used: logistic regression (LR), support vector machine (SVM), neural network (NN), random forest (RF), extreme gradient boosting (XGBoost), light gradient boosting machine (LGBM), K-nearest neighbors (KNN), and decision tree (DT). Model performance and interpretability were assessed, with interpretability elucidated using Shapley additive explanations (SHAP). Seven key variables were identified including body mass index (BMI), fat-free mass (FFM), age, sex, waist-to-hip ratio (WHR), alcohol use, and smoking status through using univariate LR, the Boruta algorithm, and least absolute shrinkage and selection operator regression. According to the model performance evaluation, the LR model achieved excellent discrimination ability for both internal (area under the curve [AUC]: 0.937, 95% confidence interval [CI]: 0.906–0.969) and external validation (AUC: 0.918, 95% CI: 0.898–0.938), with high sensitivity and specificity. Interpretability analysis revealed FFM and sex as the most important predictors. The LR model based on seven variables demonstrated acceptable performance in predicting sarcopenia in older adults through both internal and external validation. It may be used as a low-cost initial screening tool.