Risk Prediction Models for Post-Endoscopic Retrograde Cholangiopancreatography Pancreatitis: A Systematic Review and...

Area of Science:

• Gastroenterology and Endoscopy
• Medical Informatics and Predictive Modeling
• Clinical Risk Stratification

Background:

• Endoscopic retrograde cholangiopancreatography (ERCP) is vital for pancreaticobiliary disorders but commonly leads to post-ERCP pancreatitis (PEP).
• PEP is a significant complication with potential for severe morbidity and mortality, necessitating effective risk identification.
• Accurate prediction of PEP is crucial for implementing timely and appropriate preventive measures.

Purpose of the Study:

• To systematically review and evaluate existing risk prediction models for post-ERCP pancreatitis (PEP).
• To establish an evidence-based framework for clinical decision-making regarding PEP risk.
• To guide future development of more robust and validated PEP prediction models.

Main Methods:

• Systematic review and meta-analysis adhering to PRISMA 2020 guidelines.
• Comprehensive literature search of multiple databases up to February 2025 for studies developing or validating PEP predictive models.
• Independent data extraction and quality assessment using the PROBAST tool by two reviewers.

Main Results:

• Eleven studies with 12 predictive models were analyzed, showing a pooled AUC of 0.798.
• Key independent predictors identified include endoscopic sphincterotomy (EST), prior pancreatitis, age, operative time, and intubation difficulty.
• Significant heterogeneity, potential bias, limited external validation (3/12 studies), and predominant use of logistic regression were noted.

Conclusions:

• Current PEP prediction models offer good discrimination but suffer from methodological limitations and insufficient external validation.
• Validated models can aid in risk stratification to guide preventive strategies such as rectal NSAIDs and hydration.
• Future research should focus on multicenter prospective validation, calibration, decision-curve analysis, and explainable AI models.