Comparison between XGboost model and logistic regression model for predicting sepsis after extremely severe burns

Area of Science:

• Medical informatics
• Computational biology
• Burn critical care

Background:

• Sepsis is a life-threatening complication following severe burns.
• Accurate prediction of sepsis is crucial for timely intervention and improved patient outcomes.
• Existing predictive models may not fully capture the complexity of sepsis development in burn patients.

Purpose of the Study:

• To evaluate the predictive performance of an Extreme Gradient Boosting (XGboost) model compared to a multivariable logistic regression (LR) model for sepsis prediction in patients with extremely severe burns.
• To identify key clinical and demographic factors associated with sepsis development post-burn.

Main Methods:

• An observational study utilizing patient demographic and clinical data from medical records.
• Development and evaluation of two predictive models: XGboost and LR.
• Model performance was assessed using the area under the curve (AUC) of the receiver operating characteristic (ROC) curve.

Main Results:

• The study included 103 patients with extremely severe burns; 19% developed sepsis.
• The XGboost model achieved a higher predictive performance (AUC = 0.91) compared to the LR model (AUC = 0.88).
• SHAP analysis identified fibrinogen, neutrophil-to-lymphocyte ratio (NLR), body index (BI), and age as significant predictors of sepsis.

Conclusions:

• The XGboost model demonstrates superior predictive efficacy for sepsis in patients with extremely severe burns.
• Fibrinogen, NLR, BI, and age are significantly correlated with sepsis development after severe burns.
• These findings suggest XGboost as a promising tool for early sepsis detection in this vulnerable patient population.