Introduction
Endoscopic papillectomy (EP) for duodenal ampullary tumors is primarily performed for lesions confined to the papilla, particularly adenomatous changes. However, early complications such as bleeding (0-21.6%), pancreatitis (0-23.1%), and perforation (0-8.3%) occur at relatively high rates. Post-EP bleeding, due to anatomical characteristics of the ampulla of Vater, can be challenging to manage and may worsen, making it critical to identify factors associated with this complication to reduce its frequency.
Aims & Methods
The study aimed to investigate the factors associated with post-endoscopic papillectomy bleeding in patients with ampullary tumors. This study included 39 patients (39 lesions), excluding residual and accessory ampullary lesions, who underwent EP at our institution between March 2011 and October 2024. We evaluated univariate analysis and SHAP values based on tree-based models for several factors related to post-EP bleeding. Hyperparameters of each model were optimized by optuna, and those models were trained with the Python package scikit-learn (version 1.3.0).
Results
The cohort included 1 adenocarcinoma, 32 adenomas, 1 neuroendocrine neoplasm, and 5 cases of ampullary inflammation, with a mean age of 66.7 years (23 males, 16 females). Patient factors included: The American Society of Anesthesiologists (ASA) Physical Status Classification I/II/III in 5/21/13 participants, median Charlson Comorbidity Index (CCI) of 1 (range 0-7), 15 cases with a history of alcohol consumption, median Brinkman Index of 100 (range 0-1680), median total bilirubin of 0.7 mg/dL (range 0.2-2.1), median serum amylase of 80 IU/L (range 36-296), median platelet count of 198,000/μL (range 113,000-494,000), median PT-INR of 0.98 (range 0.86-1.14), median APTT of 28.7 seconds (range 23.0-39.4), median CEA of 1.9 ng/mL (range 0.3-12.8), and median CA19-9 of 7.7 U/mL (range 0.8-76.4). Antithrombotic agents were used in 4 participants (1 on low-dose aspirin, 1 on clopidogrel, and 2 on direct oral anticoagulants), and periampullary diverticula were observed in 3 cases. Median tumor size was 10.3 mm (range 2.0-45.0). Procedural factors included a median total procedure time of 56 minutes (range 23-230), with 36 cases receiving endoscopic biliary stenting and 38 cases receiving endoscopic pancreatic duct stenting. The electrosurgical units ICC200/VIO300D/VIO3 were used in 9/19/11 cases, respectively. Protease inhibitors were used in 9 participants, NSAID suppositories (25mg/50mg) in 5/24 cases, and local hemostatic measures included epinephrine spray in 8 cases, thrombin spray in 8 cases, and hemostatic peptide solutions in 5 cases. Clipping for ulcer after resection was performed in 33 cases. Post-EP bleeding within 30 days occurred in 12 cases (30.8%), including 9 moderate and 3 severe cases. Univariate analysis revealed that a history of alcohol consumption was significantly associated with post-EP bleeding (Odds ratio [OR] 2.33, 1.14-4.79, P = 0.02). Feature importance was evaluated using the mean absolute SHAP values across all participants. In the Random Forest model, the top contributing features were CA19-9 (mean |SHAP| = 0.026), CCI (mean |SHAP| = 0.024), and a history of alcohol consumption (mean |SHAP| = 0.023). In contrast, the XGBoost model identified tumor size (mean |SHAP| = 0.315), a history of alcohol consumption (mean |SHAP| = 0.244), and platelet count (mean |SHAP| = 0.198) as the most important features.
Conclusion
In our cohort, a history of alcohol consumption is a robust factor associated with post-endoscopic papillectomy bleeding.
