Development and validation of a dynamic prediction tool for post-endoscopic retrograde cholangiopancreatography early biliary tract infection in patients with choledocholithiasis

Abstract:

Objective To develop a prediction tool for post-endoscopic retrograde cholangiopancreatography (ERCP) early biliary tract infection (PEEBI) in patients with choledocholithiasis, and assist clinical decision-making before ERCP and early personalized intervention after ERCP. Methods An observational bidirectional cohort study was adopted to select inpatients with choledocholithiasis who underwent ERCP in a hospital. Directed acyclic graph (DAGs) and the least absolute shrinkage and selection operator (LASSO) were used to predict PEEBI based on logistic regression, and the models were compared and validated internally and externally. Results From January 1, 2020 to September 30, 2023, a total of 2 121 patients with choledocholithiasis underwent ERCP were enrolled, of whom 77 (3.6%) developed PEEBI, mostly in the first 2 days after surgery (66.2%). The major influencing factors for PEEBI were non-iatrogenic patient-related factors, namely diabetes mellitus(OR=2.43, 95%CI: 1.14-4.85), bile duct malignancy (OR=3.95, 95%CI: 1.74-8.31) and duodenal papillary diverticulum (OR=4.39, 95%CI: 1.86-9.52). Compared with the LASSO model, the DAGs model showed higher ability (3.0%) in comprehensive discrimination (P=0.007), as well as good differentiation performance (D=0.133, P=0.894) and calibration performance (χ²=5.499, P=0.703) in external validation. Conclusion The DAGs model constructed in this study has good predictive performance. With the help of this tool, targeted early preventive measures in clinical practice can be taken to reduce the occurrence of PEEBI.