Related Experiment Videos
Development and Validation of a Predictive Model for ICU-acquired Weakness in Sepsis Patients: An Interpretable
Yuan Du1, Yu Hong Guo1, Hao Ran Ye1
1Beijing Hospital of Traditional Chinese Medicine, Capital Medical University, Beijing 100010, China.
Biomedical and Environmental Sciences : BES
|August 6, 2026
Summary
An interpretable random forest model accurately predicts intensive-care-unit-acquired weakness (ICU-AW) in sepsis patients using early ICU data. Key predictors include early antibiotic use, mechanical ventilation, SOFA scores, and age, aiding clinical decision-making for prevention.
Area of Science:
- Critical care medicine
- Neurology
- Machine learning applications in healthcare
Background:
- Intensive-care-unit-acquired weakness (ICU-AW), encompassing critical illness polyneuropathy (CIP) and myopathy (CIM), is a frequent neuromuscular complication in sepsis patients.
- Early identification of ICU-AW is crucial for timely intervention and improved patient outcomes.
Purpose of the Study:
- To develop and validate an interpretable machine learning model for the early prediction of ICU-AW in sepsis patients.
- To identify key predictive features for ICU-AW using advanced feature selection techniques.
Main Methods:
- Utilized the MIMIC-IV database (3,842 sepsis patients) and local hospital data for model development and validation.
- Employed LASSO regression and the Boruta algorithm for feature selection from baseline demographics, vital signs, severity scores, and laboratory data.
- Constructed and evaluated seven machine learning models, including random forest, assessing performance via AUC, accuracy, sensitivity, specificity, and calibration.
Main Results:
- The random forest model demonstrated superior performance with an AUC of 0.772 in the validation set and 0.753 in the external cohort.
- SHapley Additive explanations (SHAP) analysis identified early antibiotic use, early mechanical ventilation, SOFA scores, and age as significant predictors of ICU-AW.
- The optimal model exhibited good calibration and the highest net benefit, indicating clinical utility.
Conclusions:
- An interpretable random forest model effectively predicts ICU-AW risk in sepsis patients using early ICU data.
- The model, enhanced by SHAP analysis, can serve as a valuable clinical decision-making tool for early risk identification and optimized prevention strategies.