Machine Learning-Based Prediction of Short-Term Mortality in Patients With Severe Acute Pancreatitis: A Multicenter...

Area of Science:

• Critical Care Medicine
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Severe acute pancreatitis (SAP) poses a significant risk of short-term mortality.
• Early identification of high-risk patients is crucial for timely intervention and improved outcomes.

Purpose of the Study:

• To pinpoint risk factors associated with short-term mortality in SAP.
• To develop and validate a predictive model for early identification of high-risk SAP patients.
• To provide a tool for enhanced clinical decision-making in SAP management.

Main Methods:

• A cohort of SAP patients was analyzed, divided into mortality and survival groups.
• Feature selection was performed using LASSO regression, Boruta algorithm, and RFE.
• Seven machine learning models were constructed and validated externally using MIMIC-IV data, with performance assessed via ROC curves, calibration, and decision curves.

Main Results:

• Ten critical features were identified, including mechanical ventilation, age, and blood urea nitrogen.
• The Gradient Boosting Machine (GBM) model exhibited superior predictive performance (AUC training: 0.964, testing: 0.927, validation: 0.811).
• SHAP analysis elucidated feature importance, and a web-based calculator demonstrated clinical utility.

Conclusions:

• The Gradient Boosting Machine (GBM) model is highly effective for predicting short-term mortality in severe acute pancreatitis patients.
• The developed model and associated calculator can assist clinicians in identifying at-risk individuals and guiding treatment strategies.