Improved outcome prediction in acute pancreatitis with generated data and advanced machine learning algorithms

Area of Science:

• Medical Informatics
• Computational Medicine
• Clinical Decision Support

Background:

• Traditional scoring systems for acute pancreatitis (AP) severity have limitations in predictive accuracy.
• Accurate prediction of AP severity is crucial for timely and appropriate patient management.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) algorithms in improving the prediction of AP severity.
• To compare the performance of ML models against established clinical scoring systems.

Main Methods:

• A retrospective analysis of 101 AP patients was performed, with data augmented to 250 cases using ADASYN.
• Supervised ML models, including random forest (RF) and XGBoost (XGB), were trained and validated.
• Model performance was assessed against Ranson's, modified Glasgow, and BISAP scores using AUC, F1 score, and recall.

Main Results:

• The random forest (RF) model achieved the highest performance with an AUC of 0.89, F1 score of 0.82, and recall of 0.82.
• BISAP and Glasgow criteria showed moderate performance, while Ranson's criteria were least effective.
• RF significantly outperformed all traditional clinical scores in predicting adverse outcomes.

Conclusions:

• Machine learning models, especially RF, demonstrate superior predictive accuracy for AP severity compared to traditional methods.
• The integration of ML into clinical workflows holds promise for improving prognostic accuracy and patient care in AP.