Predicting Post-ERCP Pancreatitis Using Machine Learning: Risk Stratification and Feature Importance Analysis

Despite advances in understanding the risk factors of post-endoscopic retrograde cholangiopancreatography (ERCP) pancreatitis (PEP), this adverse event remains frequently unpredictable. This study aims to develop a machine learning (ML) model to predict PEP risk.

Data were collected from a prospective ERCP registry on patients with naïve papilla who underwent ERCP between 2022 and 2024. CatBoost and eXtreme Gradient Boosting algorithms were trained to estimate PEP risk and the performance of the resulting models was assessed using the area under the receiver operating characteristic (AUC) with 10-fold cross-validation.

Of 1330 screened patients, 1190 met the inclusion criteria, and 170 (14.3%) developed PEP. The best-performing algorithm was CatBoost, which consisted of eight features: age, sex, normal papilla morphology, pancreatic duct cannulation, difficult cannulation, abnormal bilirubin levels, common bile duct diameter, and successful stone extraction. This model achieved an AUC of 68.8% (70.4% sensitivity, 67.2% specificity, 26.5% positive predictive value, and 92.0% negative predictive value). The CatBoost model effectively stratified patients into low-, intermediate-, and high-risk groups, with corresponding PEP incidences of 5.7%, 21.0%, and 40.0%, respectively.

ML is highly promising for prediction of PEP. Future studies should focus on multicenter data, inclusion of multimodal data, severity risk-stratification, and real-time application.