Construct prediction models for low muscle mass with metabolic syndrome using machine learning.

Background: Metabolic syndrome (MetS) and sarcopenia are major global public health problems, and their coexistence significantly increases the risk of death. In recent years, this trend has become increasingly prominent in younger populations, posing a major public health challenge. Numerous studies have regarded reduced muscle mass as a reliable indicator for identifying pre-sarcopenia. Nevertheless, there are currently no well-developed methods for identifying low muscle mass in individuals with MetS.

Methods: A total of 2,467 MetS patients (aged 18-59 years) with low muscle mass assessed by dual-energy X-ray absorptiometry (DXA) were included using data from the 2011-2018 National Health and Nutrition Examination Survey (NHANES). Least Absolute Shrinkage and Selection Operator (LASSO) regression was then used to screen for important features. A total of nine Machine learning (ML) models were constructed in this study. Area under the curve (AUC), F1 Score, Recall, Precision, Accuracy, Specificity, PPV, and NPV were used to evaluate the model's performance and explain important predictors using the Shapley Additive Explain (SHAP) values.

Results: The Logistic Regression (LR) model performed the best overall, with an AUC of 0.925 (95% CI: 0.9043, 0.9443), alongside strong F1-score (0.87) and specificity (0.89). Five important predictors are displayed in the summary plot of SHAP values: height, gender, waist circumference, thigh length, and alkaline phosphatase (ALP).

Conclusion: This study developed an interpretable ML model based on SHAP methodology to identify risk factors for low muscle mass in a young population of MetS patients. Additionally, a web-based tool was implemented to facilitate sarcopenia screening.