Machine learning model has been proved to be superior to existing prediction scores for mortality prediction of Acute Pancreatitis (AP).

The use of most of previous ML models is limited in clinical practice, mainly due to the lack of explainability in the clinical setting.

Our study used XGBoost model with SHAP method in a critically ill AP cohort to predict mortality, and identified variables that facilitated the model to make a reasonable prediction, thus making it more transparent and reliable.

In the future prospective data are warranted to refine the model, thus to integrate it into clinical decision support systems.

A gradient-boosting ML (XGBoost) model was developed and externally validated based on two public databases: Medical Information Mart for Intensive Care (MIMIC, training cohort) and the eICU Collaborative Research Database (eICU-CRD, validation cohort).

Read More