Risk prediction of biliary infection after endoscopic drainage for malignant perihilar biliary obstruction: A 10-year multicenter retrospective study

Abstract

BACKGROUND

As a prominent drainage method, endoscopic retrograde cholangiopancreatography (ERCP) with stenting has been universally employed to treat malignant perihilar biliary obstruction (MPHBO). Nonetheless, postoperative biliary infection (PBI) constitutes a remarkable complication associated with this procedure, which can result in fatal outcomes under some circumstances.

AIM

To investigate the risk factors and predict the occurrence of PBI following ERCP drainage in patients suffering from MPHBO.

METHODS

This retrospective study analyzed data from patients who underwent ERCP drainage at three different centers. Independent risk factors for PBI were identified by adopting multivariate analyses. Logistic regression model and artificial neural network (ANN) models were developed and validated to predict PBI.

RESULTS

A total of 288 patients who underwent 403 ERCP procedures were included in the study. The incidence of PBI was 39% (158/403). As evidently demonstrated by multivariate analysis, the Bismuth-Corlett classification (odds ratio [OR] = 1.412; 95% confidence interval [CI]: 1.144-1.743; P = 0.001), hypokalemia (OR = 4.080; 95%CI: 1.958-8.505; P < 0.001), and aspartate transaminase (AST) (OR = 1.003; 95%CI: 1.000-1.006; P = 0.021) were independent risk factors for PBI. Simultaneously, both a logistic regression model (area under the curve [AUC] = 0.734) and an ANN model (AUC = 0.867) were developed by adopting these factors. As suggested by a validation through 45 additional cases, the ANN model demonstrated an AUC of 0.940, surpassing the logistic regression model’s AUC of 0.791.

CONCLUSION

The Bismuth-Corlett classification, hypokalemia, and AST levels were identified as independent risk factors for PBI following ERCP drainage. The ANN model was proven to be an effective approach for the anticipation of the PBI occurrence.

Keywords: Malignant perihilar biliary obstruction; Endoscopic retrograde cholangiopancreatography; Postoperative biliary infection; Risk factors; Artificial neural network

Core Tip: This study established logistic regression and artificial neural network (ANN) models to preoperatively predict postoperative biliary infection—a serious complication worsening surgical outcomes and short-term prognosis in patients with malignant perihilar biliary obstruction undergoing endoscopic retrograde cholangiopancreatography drainage. Multivariate logistic regression identified key preoperative risk factors, notably hypokalemia, Bismuth-Corlett classification, and elevated aspartate transaminase levels. The ANN model demonstrated markedly superior predictive performance compared with logistic regression. These models offer clinicians a practical tool for early identification of high-risk patients, enabling timely, targeted interventions to mitigate infection-related complications and improve postoperative outcomes.