Sarcopenia prediction model based on machine learning and SHAP values for community-based older adults with cardiovascular disease in China.

Background: Sarcopenia (SP), is recognized as a complication of cardiovascular disease (CVD), but few relevant diagnostic models have been developed. This study aims to establish an interpretable diagnostic model for the occurrence of SP in older adult CVD patients living in Chinese community-dwelling (CD). Methods: We randomly selected participants with CVD recruited from CHARLS from 2011 to 2015 and divided them into a training set and a test set. In the training set, we processed and screened the predictor variables and addressed the data imbalance by the synthetic minority oversampling technique (SMOTE). Subsequently, we built four machine learning (ML) models to predict SP. After 100 iterations, we selected the best performing model for risk stratification by comparing model discrimination and calibration. Then, we analyzed the relationship between ML risk and SP using scatterplots and logistic regression (LR). Finally, the Shapley's Additive Explanatory Plot (SHAP) illustrates how each feature level affects the predicted probability of SP. Results: We ultimately included 1,088 CD older adults, 18.61% of whom reported SP. The optimal model, XGBoost, was selected for prediction and risk stratification. After both univariate (odds ratio [OR]: 12.45, Conclusion: This study develops a novel clinically-integrated tool that can be used to easily predict SP in the older adults population with CVD, providing a basis for the development of personalized therapeutic measures.