Preoperative predictive model based on computed tomography imaging features for pancreatic fistula risk after pancreaticoduodenectomy

Abstract

Abstract Purpose: This study aimed to establish a predictive nomogram model to anticipate the risk of clinically relevant postoperative pancreatic fistula (CR-POPF) after pancreaticoduodenectomy (PD) at an earlier stage. Methods: Data were retrospectively collected from patients who underwent PD at the First Affiliated Hospital of Wannan Medical College. Subsequently, univariate and multivariate logistic regression analyses were performed on relevant factors to identify independent risk factors for CR-POPF. This led to the development of a risk prediction nomogram model based on imaging data. The model's predictive performance and calibration were assessed using ROC curve analysis and calibration curves, then combined with DCA to evaluate the model's clinical utility, and compared with existing models. Results: Multivariable regression analysis showed that BMI (OR = 1.365, P < 0.001), extracellular volume fraction (ECVF) (OR = 0.884, P < 0.001), main pancreatic duct diameter (MPD) (OR = 0.263, P < 0.001), and the short axis of the pancreatic neck cross-section (OR = 1.374, P = 0.014) are independent risk factors for CR-POPF. There is a significant positive correlation between ECVF and pancreatic fibrosis; Compared with existing risk scoring systems, the model developed in this study showed a superior fit and had the smallest misclassification error. Conclusion: The results of this study indicate that the nomogram model provides a good predictive effect for the preoperative assessment of the risk of CR-POPF, and that ECVF is a readily obtainable predictor of CR-POPF, with a high correlation between ECVF and pancreatic fibrosis, and the pancreatic texture is classified based on ECVF.