Construction and validation of a risk prediction model for pancreatic fistula after pancreaticoduodenectomy / 中华内分泌外科杂志

ABSTRACT

Objective:To investigate the risk factors of pancreatic fistula after pancreaticoduodenectomy and establish the risk graph model of pancreatic fistula.Methods:The clinical data of 147 patients undergoing pancreaticoduodenectomy from Jan. 2018 to Jan. 2021 in Department of Hepatobiliary Surgery of Northern Theater Command General Hospital were retrospectively analyzed. The independent risk factors for postoperative pancreatic fistula were determined by univariate and multivariate analysis, and the linear graph model for predicting individual pancreatic fistula was drawn. The area under the subject operating characteristic curve was used to evaluate the model differentiation, the calibration curve was used to evaluate the model calibration, and finally the clinical application value of the model was evaluated by the clinical decision curve (DCA) .Results:The incidence of pancreatic fistula was 38.1%, including grade B pancreatic fistula in 49 cases and Grade C pancreatic fistula in 7 cases. Univariate analysis showed that operation method, body mass index (BMI), pancreatic texture, pancreatic duct diameter and lesion location were the related factors for postoperative pancreatic fistula. Multivariate analysis showed that BMI>25 kg/m 2, pancreatic soft texture, pancreatic duct diameter ≤3 mm and non-pancreatic diseases were independent risk factors for postoperative pancreatic fistula. According to the results of multiple factors, a prediction model of the nomogram was drawn, and the area under the subject operating characteristic curve of the model was calculated as AUC=0.792 (95% CI: 0.718-0.867). The calibration curve was drawn through internal verification of re-sampling, and the fitting curve swung around the 45° reference line, showing a high calibration degree; Clinical decision curve (DCA) analysis showed that the threshold probability was between 15% and 75% for maximum net benefit. It had good clinical application value. Conclusions:BMI>25 kg/m 2, soft pancreas, pancreatic duct diameter ≤3 mm and non-pancreatic diseases are independent risk factors for pancreatic fistula after pancreaticoduodenectomy. The established line graph model has good predictive efficiency and can effectively predict the occurrence of postoperative pancreatic fistula.