Application of a Machine Learning Predictive Model for Recurrent Acute Pancreatitis

Area of Science:

• Gastroenterology
• Medical Informatics
• Predictive Analytics

Background:

• Acute pancreatitis is a leading cause of hospitalization for pancreatic disease.
• A significant subset of patients experience recurrent acute pancreatitis (RAP) after an initial episode.

Purpose of the Study:

• To construct and validate predictive models for recurrent acute pancreatitis (RAP).
• To identify key clinical factors associated with RAP development.

Main Methods:

• Utilized data from 531 first-episode acute pancreatitis patients.
• Employed machine learning, specifically XGBoost, for model development.
• Assessed model performance using AUC, accuracy, sensitivity, specificity, and decision curves.

Main Results:

• The XGBoost model demonstrated strong predictive performance (AUC 0.779) in the test set.
• Key predictors for RAP included drinking, smoking, elevated triglycerides, and ANC occurrence.
• The model achieved an accuracy of 0.763 and a negative predictive value of 0.922.

Conclusions:

• The XGBoost model is a promising tool for predicting RAP.
• This predictive capability can assist in identifying and managing high-risk individuals.