The patient population is the same as that described in a previous study: 590 patients underwent ERCP with biliary stent implantation between June 2021 and December 2023 (Fig. 1). Bleeding from stenting is characterized by intraoperative bleeding that necessitates electrocoagulation, hemostatic clips, or epinephrine spray to control bleeding; bleeding occurring after stenting; or postoperative symptoms, such as vomiting, black stools, drainage of bloody material, or a postoperative decrease in hemoglobin levels exceeding 20 g/L[10]. Bleeding can occur either during the operation or in a delayed fashion, typically approximately 10–14 days after surgery [11]. Ethical approval for the study was obtained from the hospital and patient information was kept anonymous.
Inclusion and exclusion criteria:
The inclusion criteria were: 1) hospitalized patients, 2) aged over 18 years; 3) patient gave consent for stent placement, 4) ceased antiplatelet and anticoagulant medications 5-7 days before stent placement and resumed medication 3 days post-procedure, 5) procedure performed by a qualified and experienced physician, and 6) complete case information and data
The exclusion criteria were: 1) incomplete stenting, 2) incomplete medical records, 3) lack of follow-up, 4) preoperative gastrointestinal bleeding and bleeding from non-stenting procedures, 5) post-stenting perforation, 6) patients not admitted for initial stenting or stent replacement, 7) mortality during admission, 8) patients who underwent biliary-intestinal anastomotic stenting.
The clinical characteristics analyzed in this study were: age, sex, smoking, drinking. Clinical indications were: benign, duodenal cancer, pancreatic head cancer, ampullary cancer, gallbladder cancer, biliary duct cancer, liver cancer. Background diseases were:hypertension, diabetes, hyperlipemia, CHD, papillary diverticulum, liver cirrhosis, calculus of bile duct, diameter>1cm, renal-failure. Essential operational factors: EST, balloon dilatation, basket extraction, lithotripsy, spyglass examination, ENBD, gallbladder wall thickening, stent type, and amount>1. Patient's medical background includes cholecystectomy, subtotal gastrectomy, and use of anticoagulant medicine;Laboratory test results and pertinent data:p-WBC, p-PLT, p-HB, p-LYMPH, p-MONO, p-NEUT, p-ALT, p-AST, p-TBIL, p-DBIL, p-IBIL, p-γ-GT, p-ALP, p-ALB, p-BUN, p-SCR, a-WBC, a-PLT, a-HB, a-LYMPH a-MONO,a-NEUT, a-ALT, a-AST, a-TBIL, a-DBIL, a-IBIL, a-γ-GT, a-ALP, a-ALB, a-BUN, a-SCR, FIB, PT-INR, PT, APTT, and TT.
Ethical issues
Ethical clearance was obtained from Tianjin Third Central Hospital and Medicine Ethical Review board (IRB).Due to the retrospective nature of the study, the Ethics Committee of Tianjin Third Central Hospital waived the need to obtain informed consent.The retrospective study complied with the ethical principles of the Declaration of Helsinki and the Measures for Ethical Review of Biomedical Research Involving Human Beings. Informed immunity was obtained as this study only utilized pre-existing medical records or data, did not involve the identity or privacy of the subjects, did not pose any risk or harm to the subjects, and the data were anonymized when analyzed.
Statistical analysis
Categorical data are represented by frequencies and percentages, whereas continuous variables are represented by means of a normal distribution and median quartiles for a non-normal distribution. ROC curves were generated for all variables (Fig. 2) and a significance test for correlation (level=0.95) was conducted to calculate AUC values (Fig. 3). A heat map was created using the correlation coefficients (Fig. 4) to visualize variables with significant covariance or correlation and to identify the most appropriate variables for logistic regression. Data dimensions and predictors were selected using the LASSO regression technique (Fig. 5) to eliminate "suspected risk factors.” The dataset was randomly divided into a training set (n=354) and a test set (n=236) in a 6:4 ratio. A multifactorial logistic regression analysis was conducted on the identified "suspected risk factors" to create a nomogram of bleeding risk factors (Fig. 6).
During model validation, we employed a bootstrap approach for internal validation using 1,000 samples of the raw data. The consistency index (CI) and 95% confidence intervals were used to assess the discriminatory ability of the predictive model. The predictive power of the model was assessed using the area under the curve (AUC) of the participants’ characteristics (ROC). An AUC value greater than 0.75 suggests that the model has sufficient discriminatory ability. The accuracy of the model was confirmed using the calibration curves. A decision curve analysis (DCA) was performed to examine net patient benefit. The proposed nomogram would yield a net gain in forecasting stent insertion bleeding if the threshold likelihood of bleeding fell between 0 and 0.8, as demonstrated by the well-calibrated predictive model and DCA(Fig. 8).
R software (version 4.3.2; R Foundation for Statistical Computing, Vienna, Austria) was used, and P < 0.05 was considered statistically significant. The following R packages were used:(corrplot), (glmnet), (caret), (CBCgrps), (nortest), (tidyverse), (ggpubr), (rms), and (pROC). SPSS software (version 27.0) was also used in the statistical analysis.