A Nomogram Model Based on Preoperative Coagulation-Fibrinolysis Biomarkers for Predicting Postoperative Hemorrhage After Pancreaticoduodenectomy: A Retrospective Study With Internal Validation.

Abstract

Aim: Post-pancreaticoduodenectomy hemorrhage (PPH) remains a life-threatening complication. However, conventional coagulation assays fail to capture the dynamic hemostatic disturbances preceding overt bleeding. This study aimed to evaluate whether preoperative coagulation-fibrinolysis biomarkers could improve PPH risk stratification and support the development of an interpretable predictive model.

Methods: We retrospectively analyzed 315 adults who underwent elective pancreaticoduodenectomy, randomly allocated to a training cohort (n = 220) and an internal validation cohort (n = 95). Preoperative variables included routine coagulation indices and molecular markers, including thrombin-antithrombin complex (TAT), plasmin-α2-plasmin inhibitor complex (PIC), and D-dimer. Multivariable logistic regression was used to identify independent predictors and construct a predictive nomogram. Model performance was assessed in terms of discrimination (area under the curve [AUC]), calibration (Brier score and Hosmer-Lemeshow goodness-of-fit test), and decision curve analysis (DCA).

Results: In the training cohort, patients who developed PPH exhibited significantly higher levels of TAT, PIC, and D-dimer, and lower fibrinogen (FIB). Multivariable analysis identified four independent predictors: FIB (odds ratio (OR) = 0.46), PIC (OR = 1.79), D-dimer (OR = 1.59), and TAT (OR = 1.15). The nomogram demonstrated good discrimination, with an AUC of 0.81 in the training cohort and 0.75 in the validation cohort. Calibration performance was acceptable, with Brier scores of 0.11 (training) and 0.15 (validation). Although the Hosmer-Lemeshow test indicated no statistically significant lack of fit (p > 0.05 in both cohorts), visual inspection of the validation calibration curve suggested moderate deviations in the intermediate-risk range, likely reflecting the limited sample size. Shapley Additive exPlanations‌ (SHAP) analysis identified PIC as the most influential predictor in the model output.

Conclusions: Preoperative coagulation-fibrinolysis "process markers" may provide additional information for assessing PPH risk. The four-marker nomogram (PIC, TAT, D-dimer, and FIB) offers an interpretable framework for perioperative risk stratification and demonstrates satisfactory overall performance. However, external validation in larger, independent cohorts is warranted to confirm model generalizability and calibration stability.