Integrating sarcopenia into ICU-acquired weakness risk stratification: a machine learning-based prediction model for...

Sarcopenia is closely associated with weakness in elderly and various chronic disease populations. However, its specific role as a predisposing factor for ICU-acquired weakness (ICU-AW) in critically ill patients remains unclear. This study aimed to investigate whether sarcopenia is a predictor of ICU-AW and to develop a machine learning-based prediction model integrating sarcopenia for risk stratification.

A retrospective analysis was conducted on data from a prospectively maintained database. A total of 1,324 critically ill patients were enrolled and randomly divided into a training set (n = 927) and a validation set (n = 397) in a 7:3 ratio. Sarcopenia was assessed by measuring the skeletal muscle area at the third lumbar vertebra (L3) level using computed tomography. Least absolute shrinkage and selection operator (LASSO) regression and the Boruta algorithm were employed to identify key predictors. 10 machine learning models were developed and their predictive performance was compared. SHAP (SHapley Additive exPlanations) analysis was used to quantify the feature importance of each predictor in the optimal model.

Six predictors were identified by intersecting the results from LASSO regression and the Boruta algorithm in the training set: age, APACHE II score, sarcopenia, sepsis, mechanical ventilation, and lactic acid. Among the 10 machine learning models developed, the XGBoost model exhibited the best overall predictive performance in the training set, achieving an area under the curve (AUC) of 0.838 (95% confidence interval [CI]: 0.807-0.868) and the lowest Brier score of 0.137. Decision curve analysis and clinical impact curves confirmed its stable clinical predictive value. SHAP analysis revealed that sarcopenia ranked as the third most important predictor, following APACHE II and age. In the validation set, the XGBoost model maintained excellent discriminative ability and predictive performance, with an AUC of 0.843 (95% CI: 0.804-0.882) and good calibration, demonstrating a positive net benefit across a threshold probability range of 9%-92%.

Sarcopenia is an important predictor of ICU-AW in critically ill patients. Integrating sarcopenia into the XGBoost model effectively identifies high-risk patients, providing a valuable tool for early risk stratification in intensive care settings.