Prognostic nomogram for acute pancreatitis after percutaneous biliary stent insertion in patients with malignant obstruction

This retrospective study was approved by the institutional review board of our hospital. Data collection consent was obtained. Data of 377 consecutive patients with MBO who underwent PTBS in our center between March 2016 and July 2021 were collected. For this study, inclusion criteria included: (1) complete clinical data including laboratory indexes and imaging information; (2) MBO confirmed based on radiological and/or pathological findings. Exclusion criteria included: (1) patients with a history of pancreatitis in the recent 3 months at admission or (2) a history of pancreatectomy. After that, 314 patients were included in this study. Among them, 159 were men and 155 were women. The median age of the patients was 65.0 years (range, 28—92 years). The patient inclusion process is illustrated in Fig. 1. The diagnosis of the primary tumor was established based on laboratory and radiology findings of 145 patients and pathological results of 169 patients.

All patients were required to fast at least 8 h prior to the procedure. The stent insertion was performed under local anesthesia. The intrahepatic bile duct was punctured using a 21-gauge Chiba needle (Cook, Bloomington, IN, USA). In case of a successful puncture, a 0.018-inch guidewire was inserted, and thereafter a 4F introducer sheath (Neff Percutaneous Access Set, Cook, Bloomington, IN, USA) was introduced. We then performed cholangiography to evaluate the obstruction site. Following this, a 0.035-inch guidewire was advanced to the duodenum across the obstruction site with a 4F catheter. After measurement of the length of stricture, the stent was introduced over the guidewire and then deployed across the stricture to cover the bile duct approximately 1.5—2 cm distal and proximal to the obstruction to prevent tumor infiltration. Stent graft patency was confirmed with repeat cholangiography. Moreover, the external drainage tube (8F, Cook, Bloomington, IN, USA) was inserted in patients with infection, and the iodine-125 seeds (0.8 mci, Xinke, Shanghai, China) strand was inserted for intraluminal radiotherapy in some patients with their permission. The puncture approach was occluded with gel foam pledgets through a sheath. Three types of uncovered SEMS (Self-Expanding Metallic Stent) with a diameter of 8 mm and lengths from 60 to 100 mm were used in the current study (E-Luminexx [Bard Peripheral Vascular, Tempe, AZ], S.M.A.R.T [Cordis, Milpitas, CA], and Zilver [Cook, Bloomington, IN]). All procedures were carried out by two interventional radiologists with more than 10 years of experience.

Acute pancreatitis was diagnosed based on the Atlanta classification [13], which requires the presence of two or more of the following criteria: persistent abdominal pain accompanied by vomiting and nausea; the level of serum amylase of at least three times over the limit of normal; and radiology features including CT or ultrasonography.

During the period of hospitalization, all patients were followed daily monitoring the levels of serum amylase and clinical condition. The first postoperative serum amylase levels were determined at 3 h after the procedure. Radiology examinations were used to confirm pancreatitis when the serum amylase levels were over three times upper the limit of normal without clinical symptoms. Patients with pancreatitis received somatostatin and fasting therapy.

Data of each patient was extracted from individual medical records and image systems. The characteristics of patients consisted of age and gender, primary tumor, underlying disease, previous biliary drainage, preoperative infection, laboratory indices, hs-CRP, and location of stricture. The procedure-related characteristics included operation time, external drainage tube insertion, Iodine-125 seed strand insertion, stent length, number of stents, stent insertion across the papilla, and visualization of the pancreatic duct.

For subgroup analysis, enrolled patients were classified into two groups (non-pancreatitis and pancreatitis) according to clinical outcomes. Missing data were excluded from the analyses. Continuous variables were evaluated using the Mann–Whitney U-test and depicted as medians and IQR (Inter-Quartile Range). Categorical data were compared using chi-square or Fisher's exact tests as appropriate and presented as frequencies.

We developed a nomogram for pancreatitis after PTBS in three steps:(1) a univariate analysis with one variable was used at a time to identify potential risk factors(ie, young age, stent insertion across the papilla, and visualization of the pancreatic duct); (2) significant variables above were subjected to the backward multivariate logistic regression analysis method to establish the independent predictor for post-procedural pancreatitis; (3) a nomogram was developed by entering the results of regression into the “rms” and “shiny” package of R software. For developing a nomogram, each factor was scored on basis of estimated logistic regression coefficients. The biggest impact factor was determined and sequentially other factors were scored in proportion to the points assigned to the biggest impact factor. This translated complex mathematical models into a simple graph of scaled variables facilitating a quick approximation of event probability [14].

We further evaluated the performance of the nomogram in terms of discrimination and calibration. The ability of the nomogram to distinguish non-pancreatitis from pancreatitis was assessed by calculating the area under the receiver operator characteristic curve (AUC). Moreover, the predictive performance was validated with bootstrap resampling repeated 1000 times and then compared the predicted and observed probabilities of pancreatitis in patients with MBO. The decision curve analysis (DCA) was used to confirm the clinical utility of this predictive scoring system. A P-value of < 0.05 was considered statistically significant. All data analyses were implemented using R software V.4.0.2 (Beijing Foreign Studies University, Beijing, China; www.​r-project.​org).

Among the 314 patients, 41 (13.1%) developed pancreatitis. All patients recovered at a mean of 3.2 days (range, 1–7 days) under somatostatin and fasting therapy during the period of hospitalization. No severe pancreatitis occurred and none needed further surgical therapy or intensive care. The baseline of patients was summarized in Table 1. Patients with stent insertion across the papilla represented the larger proportion of pancreatitis (80.5%), with only 19.5% of pancreatitis patients with stent above the papilla. Visualization of the pancreatic duct was more common among pancreatitis than among non-pancreatitis (46.3% vs 5.1%), and young patients (< 60 years) were more common as well (56.1% vs 34.1%). There were no significant differences in other factors between the two groups (Table 1). Subsequently, the variables of young age (p = 0.003), stent insertion across the papilla (p < 0.001), and visualization of the pancreatic duct (p < 0.001) in the univariate analysis were considered for the multivariate logistic regression analysis. Finally, the results of backward stepwise regression analysis revealed that the following factors were markedly associated with increased risk of pancreatitis after PTBS: young age (OR = 2.57, 95% CI 1.16 to 5.69), stent insertion across the papilla (OR = 6.47, 95% CI 2.66 to 15.70), and visualization of the pancreatic duct (OR = 15.40, 95% CI 6.07 to 39.03) (Table 2). Besides, fifty-one (16.2%) patients with other complications were observed in this study, including 29 (9.2%) of puncture site pain, 11 (3.5%) of infection, 8 cases (2.5%) of bile leak, and 3 (1.0%) of bleeding. All patients recovered under conservative treatment before discharge.

Based on the multivariate logistic regression analysis, we constructed a nomogram for predicting the probability of post-procedural pancreatitis. In this nomogram, we used the age of patients as a continuous variable, which increases the precision of the model and allows for more individualized risk prediction. Besides, each value of a variable corresponds to a score, and the corresponding scores for the three variables included in the model were summed to achieve a total score for an individual. The total score was then projected onto a total point scale to obtain the probability of post-procedural pancreatitis for each patient (Fig. 2).

The nomogram exhibited great discrimination, with an AUC of 0.807 (95% CI 0.730 to 0.883, Fig. 3) and good sensitivity (78.0%) and specificity (65.6%). Second, the optimism-corrected AUC obtained from bootstrap resampling (1000 times) was 0.803, suggesting excellent internal validation. The calibration curve showed that the predicted probabilities of pancreatitis risk agreed well with the observed probabilities (Fig. 4). The DCA curve indicated that the net benefit per patient increases as the model curve is extended, confirming the potential clinical utility of the nomogram (Fig. 5).