A w-ACT model for sarcopenia among community-dwelling older adults based on National Basic Public Health Services:...

Sarcopenia leads to substantial health and well-being impairments in older adults, underscoring the need for early detection to facilitate intervention. Despite its importance, community settings face challenges with data accessibility, model interpretability, and predictive accuracy.

To develop a local, data-driven, machine learning-based predictive model aimed at identifying high-risk sarcopenia populations among community-dwelling older adults.

The study encompassed 910 participants over 60 years old from the National Basic Public Health Services (NBPHS) program. Sarcopenia was ascertained by the Asian Working Group for Sarcopenia (AWGS) criteria. We leveraged Logistic Regression and seven additional machine learning models for risk prediction, employing the LASSO method for feature selection, employing LASSO regression with 10-fold cross-validation for feature selection. The optimal lambda.1se threshold identified four key predictors forming the w-ACT model (weight, Age, Calf circumference, Triglycerides). A comprehensive set of 10 diagnostic indicators was utilized to assess model performance.

The Random Forest-based w-ACT model demonstrated superior performance, with an AUC of 0.872 (95%CI: 0.793,0.950) (validation set) and MCC of 0.566, 0.841 (95%CI: 0.777,0.904) (test set) and MCC of 0.511. Key predictors included weight, age, calf circumference, and triglycerides. SHAP analysis confirmed clinical interpretability.

The w-ACT model offers a reliable, interpretable tool for community-based sarcopenia screening, leveraging accessible variables to guide preventive care.