Predictive modeling of ICU-AW inflammatory factors based on machine learning

Area of Science:

• Critical Care Medicine
• Biostatistics
• Biomarkers

Background:

• ICU-acquired weakness (ICU-AW) is a frequent complication in intensive care units.
• Early prediction and intervention are crucial to reduce ICU-AW incidence.
• Machine learning offers novel approaches for predicting complex clinical outcomes like ICU-AW.

Purpose of the Study:

• To develop and validate a machine learning model for predicting ICU-acquired weakness (ICU-AW).
• To identify key clinical and inflammatory factors associated with ICU-AW risk.
• To utilize easily accessible data for early clinical recognition of ICU-AW.

Main Methods:

• Least Absolute Shrinkage and Selection Operator (LASSO) regression for variable selection.
• Development of prediction models using logistic regression (LR), random forest (RF), and extreme gradient boosting (XGB).
• Validation of models using a training set (70%) and a test set (30%) of patient data, including inflammatory markers like IL-1β, IL-6, and IL-10.

Main Results:

• The logistic regression model incorporating inflammatory factors achieved an AUC of 82.1% on the test set.
• This model demonstrated superior performance and calibration compared to five other models.
• Nomograms were utilized for visual representation and prediction using the optimal model.

Conclusions:

• Accessible clinical characteristics and laboratory data, particularly inflammatory factors (IL-1β, IL-6, IL-10), are valuable for predicting ICU-AW.
• The developed model aids in the early identification of patients at risk for ICU-AW.
• Machine learning provides an effective tool for enhancing ICU patient care and outcomes.