The establishment and validation of a clinical prediction model for postoperative biliary fistula after pancreaticoduodenectomy.

Background: At present, pancreaticoduodenectomy (PD) is a classic surgical treatment for benign and malignant tumors around ampulla. The operation is complicated and postoperative complications are frequent. Biliary fistula is the most common anastomotic fistula after pancreatic fistula. Our objective is to analyze the risk factors for biliary fistula after PD and to construct a nomogram to predict biliary fistula after PD. Methods: The clinical data of a total of 196 patients who underwent PD from March 2014 to March 2024 in Yueyang Hospital Affiliated to Hunan Normal University and The Third Xiangya Hospital of Central South University were retrospectively analyzed. The number of included patients was divided in the ratio of 7:3 using the random split method, with 130 patients in the training set and 66 patients in the validation set. Predictors were screened and a nomogram prediction model was constructed by one-way logistic regression analysis, Lasso regression analysis and multifactorial logistic regression analysis. The discriminative ability, consistency and clinical effectiveness of the models were assessed by area under the curve (AUC) of the working characteristics of the subjects, calibration curve and decision curve analysis (DCA). Results: The results of multifactorial logistic regression analysis showed that diabetes, low albumin, postoperative gastroparesis, abdominal bleeding, abdominal infection, and postoperative pancreatic fistula were the independent risk factors for biliary fistula after PD (P<0.05). The AUC of the column-line graph prediction model constructed from the above factors was 0.807 [95% confidence interval (CI): 0.652-0.962] in the training set and 0.782 (95% CI: 0.517-1.000) in the validation set, suggesting that the diagnostic efficacy of the model was better, and the calibration curves plotted in the training and validation sets were closer to the standard curves, suggesting that the model consistency was better. The plotted DCA curves also indicated a significant positive net gain. Conclusions: The nomogram prediction model constructed by diabetes, albumin, postoperative gastroparesis, abdominal bleeding, abdominal infection, and postoperative pancreatic fistula can well identify high-risk patients with postoperative biliary fistula (POBF) in PD, which has a good clinical application value.