Predicting the Risk of Clinically Significant Pancreatic Fistula Following Pancreatoduodenectomy Based on Preoperative CT Parameters

Abstract

Objective. To develop and perform internal validation of a scoring scale based on preoperative computed tomography parameters for predicting the risk of clinically significant pancreatic fistula after pancreatoduodenectomy, and to compare the prognostic characteristics of this scale with those of established models: Fistula Risk Score, Alternative Fistula Risk Score, and Preoperative Fistula Risk Score.Materials and Methods. A retrospective observational study was conducted at the O. O. Shalimov National Scientific Center for Surgery and Transplantology from January 2022 to November 2023. The main cohort (n = 234) was used to develop and internally validate a scoring scale based on preoperative computed tomography parameters in accordance with the recommendations of the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis guidelines. A predefined control cohort (n = 106), in which risk-adapted preventive interventions were not applied, was used for a direct comparison of the developed scale with existing scales. The endpoint was clinically significant postoperative pancreatic fistula (2016 International Study Group on Pancreatic Surgery classification, grades B and C). Candidate predictors were identified exclusively during the preoperative phase: pancreatic parenchymal density on native computed tomography, diameter of the main pancreatic duct, the “duct-to-parenchyma” ratio in two orthogonal planes, and body mass index. Discriminatory ability was assessed using ROC analysis and calculation of the area under the ROC curve with 95% confidence intervals, as well as by comparing the areas under the ROC curves (DeLong’s test); risk stratification—based on the observed incidence of clinically significant postoperative pancreatic fistula in risk categories and the trend test; calibration—based on the intercept, slope, and Brier’s measure; clinical utility—based on the results of clinical decision curve analysis; internal stability—using the bootstrap resampling method (2,000 iterations).Results. In the control cohort, clinically significant pancreatic fistula developed in 23 (21.7%) patients. The developed scale (0–6 points) demonstrated high discriminatory power: an area under the ROC curve of 0.850 (95% confidence interval 0.765–0.936). Compared with the Fistula Risk Score (area under the ROC curve 0.769; 95% confidence interval 0.665–0.873), the proposed scale demonstrated statistically significant (p = 0.028) superior discrimination, whereas no differences were found in this regard compared with the Alternative Fistula Risk Score (area under the ROC curve 0.851; p = 0.972) and the Preoperative Fistula Risk Score (area under the ROC curve 0.820; p = 0.570). Risk stratification was clinically interpreted: the incidence of clinically significant postoperative fistula increased from 3.8% in the low-risk group to 53.3% in the very high-risk group (p for trend < 0.001). Calibration of the proposed model was characterized by the intercept approaching 0 and the slope approaching 1, indicating good agreement between predicted and actual risks. The developed scale demonstrated the greatest net benefit within the range of clinically relevant thresholds of approximately 10–45%. Bootstrap validation confirmed the stability of the model’s discriminatory ability (mean area under the ROC curve of 0.850).Conclusions. We propose an objective and reproducible preoperative tool for predicting clinically significant pancreatic fistula following pancreatoduodenectomy, characterized by high discriminatory power, clinically interpretable risk stratification, proper calibration, and potential clinical utility for risk-adapted planning of perioperative management. External validation using independent multicenter cohorts and a prospective evaluation of the effectiveness of risk-oriented preventive strategies are needed.