Construction and validation of a machine learning model to predict the sarcopenic obesity population

Sarcopenic obesity (SO) imposes a heavy medical burden on both individuals and society, highlighting the importance of early identification and timely intervention. This study aims to construct a clinical prediction model based on machine learning (ML) methods to assist in the early screening of SO. Data from the National Health and Nutrition Survey from 2011 to 2018 were used for model development. The included data were randomly divided into a training set and an internal test set with a ratio of 7:3. Eight ML methods - logistic regression, decision tree, neural network, K-nearest neighbor, Naive Bayes, support vector machine, extreme gradient boosting (XGBoost), and light gradient boosting machine - were used to develop the model. The area under the receiver operating characteristic, calibration curves, and decision curve analysis were employed to evaluate the predictive performance of the constructed model. The SHapley Additive exPlanations method was used to illustrate the importance of variables in the model. After feature screening, 7 indicators were used as ML features to construct the model, including hepatic steatosis index, bone mineral content, skeletal muscle mass to visceral area ratio, history of heart failure, atherogenic index of plasma, neutrophil-to-lymphocytes ratio, and serum phosphorus level. The area under the receiver operating characteristic values of logistic regression, decision tree, neural network, K-nearest neighbor, Naive Bayes, support vector machine, XGBoost, and light gradient boosting machine were 0.83 (95% confidence interval [CI] = 0.81-0.85), 0.82 (95% CI = 0.80-0.85), 0.86 (95% CI = 0.84-0.88), 0.86 (95% CI = 0.85-0.88), 0.83 (95% CI = 0.81-0.85), 0.85 (95% CI = 0.83-0.87), 0.88 (95% CI = 0.86-0.90), and 0.87 (95% CI = 0.85-0.89), respectively, in the training set, and 0.86 (95% CI = 0.83-0.89), 0.79 (95% CI = 0.76-0.83), 0.83 (95% CI = 0.80-0.87), 0.82 (95% CI = 0.79-0.86), 0.83 (95% CI = 0.80-0.86), 0.85 (95% CI = 0.82-0.88), 0.84 (95% CI = 0.81-0.88), and 0.85 (95% CI = 0.82-0.88), respectively, in the internal test set. Among these models, XGBoost showed the best prediction performance in the training set. The calibration and decision curve analysis curves showed good predictive performance. The XGBoost-based model in this study performed well and combined 7 clinical predictors to help diagnose SO in adults.