Early prediction of severe acute pancreatitis using machine learning

Area of Science:

• Gastroenterology
• Medical Informatics
• Computational Biology

Background:

• Acute pancreatitis (AP) is a frequent cause of hospitalization.
• Severe AP (SAP) carries a high mortality risk (nearly 30%).
• Accurate risk stratification is crucial for timely SAP management.

Purpose of the Study:

• To develop and evaluate machine learning algorithms for predicting severe AP (SAP).
• To compare the performance of ML models against established risk scores (HAPS, BISAP).

Main Methods:

• Development of three ML models: logistic regression, neural networks, and XGBoost.
• Model performance assessed using Area Under the Receiver Operating Characteristic (AUROC) curves.
• Comparison of ML models against Harmless Acute Pancreatitis Score (HAPS) and Bedside Index for Severity in Acute Pancreatitis (BISAP).

Main Results:

• Trained and tested on 61,894 patients.
• XGBoost model achieved an AUROC of 0.921, outperforming logistic regression and neural networks.
• XGBoost model demonstrated superior AUROC compared to HAPS and BISAP for SAP identification.

Conclusions:

• Machine learning shows promise in enhancing AP risk stratification accuracy.
• Improved accuracy can lead to earlier treatment and intervention for SAP patients.