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

BACKGROUND

Endoscopic retrograde cholangiopancreatography (ERCP) serves as an essential procedure for diagnosing and treating pancreaticobiliary disorders, however it frequently results in post-ERCP pancreatitis (PEP), its most common complication. Identifying risk factors for PEP is crucial given its potential for significant morbidity and mortality.

OBJECTIVE

To systematically review and evaluate existing PEP risk prediction models, thereby establishing an evidence-based framework to inform clinical decision-making and guide future model development.

DESIGN

Systematic review and meta-analysis conducted in accordance with PRISMA 2020 guidelines.

MATERIALS AND

METHODS

We performed a comprehensive literature search across multiple databases (PubMed, Scopus, Embase, etc.) from their inception till February 2025. The inclusion criteria focused on studies that developed or validated predictive models for PEP in adult patients undergoing ERCP. Two reviewers independently extracted data and assessed methodological quality using the PROBAST tool.

RESULTS

From 466 initially identified records, we analyzed 11 studies comprising 12 predictive models. The pooled area under the receiver operating characteristic curve (AUC) was 0.798 (95% confidence interval: 0.727-0.876). Key independent predictors of PEP included endoscopic sphincterotomy (EST), previous pancreatitis, age, operative time, and intubation difficulty. However, significant methodological heterogeneity and potential bias were evident across studies. Most models were developed using logistic regression, with only one study employing machine learning techniques. Notably, only three studies conducted external validation, underscoring substantial limitations in current predictive modeling approaches.

CONCLUSION

Existing PEP prediction models demonstrate overall good discrimination but are limited by methodological heterogeneity and scarce external validation. Clinically validated models can support risk stratification to guide preventive strategies (e.g., rectal NSAIDs, aggressive hydration, selective prophylactic pancreatic stenting) alongside clinical judgment. Future research should prioritize multicenter prospective validation, robust calibration and decision-curve analyses, impact assessments, and explainable machine learning models integrated with electronic health record systems.