Online calculators for predicting the risk of anastomotic stricture after hepaticojejunostomy for bile duct injury after cholecystectomy: a multicenter retrospective study

Background: Anastomotic stricture is a common underlying cause of long-term morbidity after hepaticojejunostomy (HJ) for bile duct injury (BDI) following cholecystectomy. However, there are no methods for predicting stricture risk. This study was aimed at establishing two online calculators for predicting anastomotic stricture occurrence (ASO) and stricture-free survival (SFS) in this patient population. Methods: The clinicopathological characteristics and follow-up information of patients who underwent HJ for BDI after cholecystectomy from a multi-institutional database were reviewed. Univariate and multivariate analyses of the risk factors of ASO and SFS were performed in the training cohort. Two nomogram-based online calculators were developed and validated by internal bootstrapping resamples (n=1000) and an external cohort. Results: Among 220 screened patients, 41 (18.64%) experienced anastomotic strictures after a median follow-up of 110.7 months. Using multivariate analysis, four variables, including previous repair, sepsis, HJ phase, and bile duct fistula, were identified as independent risk factors associated with both ASO and SFS. Two nomogram models and their corresponding online calculators were subsequently developed. In the training cohort, the novel calculators achieved concordance indices (C-indices) of 0.841 and 0.763 in predicting ASO and SFS, respectively, much higher than those of the above variables. The predictive accuracy of the resulting models was also good in the internal (C-indices: 0.867 and 0.821) and external (C-indices: 0.852 and 0.823) validation cohorts. Conclusions: The two easy-to-use online calculators demonstrated optimal predictive performance for identifying patients at high risk for ASO and with dismal SFS. The estimation of individual risks will help guide decision-making and long-term personalized surveillance.


Introduction
BDI remains a potentially devastating complication following cholecystectomy, with a possible reduced life quality and a high rate of subsequent litigation. The treatment of BDI may vary from simple stenting of minor injuries to complicated surgical reconstruction in the case of major injuries. Rouxen-Y HJ is the preferred method for severe BDI, and a series of studies have shown good outcomes after this procedure [1][2][3][4][5] .
HIGHLIGHTS • The first model for predicting the stricture risk of patients who underwent hepaticojejunostomy (HJ) for bile duct injury (BDI) after cholecystectomy. • Our model demonstrated optimal predictive performance for identifying patients at high risk. • This easy-to-use online calculator can be used by both clinicians and patients. • Estimating individual risks will help guide decision-making and long-term personalized surveillance.
Although advances in surgical techniques and perioperative management have made HJ a safe procedure for BDI over the past decades, anastomotic stricture, a major complication after HJ and the leading cause of long-term major morbidity, remains a major concern. The incidence of anastomotic stricture reportedly ranges from 10 to 30% [6,7] . Patients who develop anastomotic strictures also experience inevitable readmission, increased medical costs, and prolonged psychological pressure. Accordingly, the accurate prediction of stricture can help surgeons identify high-risk patients and perform timely interventions. To the best of our knowledge, however, limited number of studies have reported the factors associated with long-term prognosis after HJ, and no predictive models have been proposed for either anastomotic stricture occurrence (ASO) or stricture-free survival (SFS). As such, it is desirable to develop an accurate and pragmatic model for predicting ASO and SFS. In this study, we sought to identify the parameters independently influencing ASO and SFS and incorporated them into two nomogram-based calculators to assess the risk for anastomotic stricture in patients undergoing HJ for BDI after cholecystectomy to facilitate the calculation of the risk/benefit ratio of individual patients to guide decision-making.

Study population
Using a multicenter database, with approval from the institutional review boards of each participating institution, data on consecutive patients who underwent HJ for IBDI after cholecystectomy were identified between 2002 and 2022 in the eight hepato-pancreatic-biliary centers of China. The exclusion criteria were as follows: (1) perioperative death; (2) missing data on important clinical or follow-up variables. All HJ procedures of patients enrolled in this study were performed in an open approach. BDI was defined as any damage to the bile duct or biliary tree including bile leaks [8] . All the patients in this study were with major BDI. The type of BDI was defined by Strasberg classification, in which type E1 is a circumferential injury to the common duct more than 2 cm from the bifurcation; type E2 is a circumferential injury to the common duct less than 2 cm from the bifurcation; type E3 is a circumferential injury to the common duct at the bifurcation; type E4 is injury proximal to the bifurcation involving both main right and left hepatic ducts; and type E5 is a combined injury to the common duct and a major aberrant right hepatic duct [9,10] . This retrospective study was conducted in accordance with the Declaration of Helsinki and the Ethical Guidelines for Clinical Studies (No. Quick-PJ 2022-08-29) and was considered exempt from requiring informed consent. Besides, this study was registered with ResearchRegistry.com (Unique Identification Number: researchregistry8389). Data have been reported as per STROCSS 2021 criteria [11] .
All postcholecystectomy BDI repair was performed via Rouxen-Y HJ by specialized hepatobiliary surgeons from the participating centers. Dissection and adhesiolysis were started towards the liver hilum to identify and lengthen the healthy bile duct stump. Division of hilar plate and limited hepatectomy were performed when needed. The common hepatic, left and right hepatic ducts were further dissected based on BDI level until the healthy and vascularized biliary stump was reached. The segmental ducts were sutured together, if possible, to enable the construction in one, rather than two jejunal anastomoses. A 40-60 cm long Roux-en-Y jejunal loop was prepared and then transferred to the right abdominal cavity through the transverse mesocolon on the right edge of the median colonic pedicle. HJ was performed as wide as possible by end-to-side, mucosamucosa, single-layer of Vicryl, or polydioxanone synthetic (4/0 or 5/0) sutures (interrupted, continuous, or both). At least one intraabdominal drain was placed behind the anastomosis. The diameter of the proximal bile duct used for HJ after BDI was generally unchanged or dilated compared with that before cholecystectomy. Therefore, a stent was not routinely used during HJ, and no such patients were enrolled in this study. No technical modifications were made during the study period.

Follow-up and data collection
Patients were followed up after 1 month, 3 months, 6 months, 1 year, and then annually thereafter. Each visit included a physical examination, liver function test, and abdominal ultrasound to determine the status of the liver. Magnetic resonance cholangiopancreatography was performed in patients presenting with recurrent cholangitis to assess the patency of HJ. Telephone interviews were used as a complementary follow-up method. Baseline data, clinical features, intraoperative variables, hospital morbidity and mortality, and long-term prognosis were recorded. Previous repair in this study was defined as any attempt at biliary repair between BDI and specialist HJ after referral [12] , such as primary suture or clip, repair over T-tube, HJ, and end-to-end anastomosis. The HJ phase was defined as early (< 14 days), intermediate (14-90 days), or late ( > 90 days after BDI). Sepsis was assessed in the presence of at least one of the following features: leukocytosis more than 15 000/ml, fever > 38.5°C, angiocholitis episodes, peritonitis, or intra-abdominal abscess [13] . Bile fistula was defined as bilirubin concentration in the drains exceeding serum bilirubin with a consecutive change of clinical management or occurrence of a bilioma necessitating drainage [14] . All data were screened and collected from the computerized BDI database by a specialized research assistant.

Predictive endpoints
As the present study focused on post-HJ anastomotic strictures, the predictive endpoints were ASO and SFS. Anastomotic stricture was defined as the presence of abdominal symptoms, cholangitis, and abnormal liver function tests in conjunction with a stricture at the HJ, diagnosed with percutaneous transhepatic cholangiography, computed tomography, or MRI, requiring intervention [15] . SFS was calculated from the date of HJ to either the date of stricture occurrence or the date of the last follow-up.

Statistical analysis
Categorical variables were summarized as numbers (n) and proportions (%) and compared using χ 2 test or Fisher's exact test, as appropriate. The survival curve of SFS was manifested using Kaplan-Meier method and compared using the log-rank test. Logistic and Cox proportional hazards regression models were used to identify independent factors for predicting ASO and SFS, respectively. Variables with significance (P < 0.10 in the univariate logistic and Cox regression analyses) were included in the subsequent multivariate regression analysis. Two calculators for ASO and SFS were generated independently to predict individual risks based on the significant factors originating from the multivariate analysis performed by a backward step-down selection process applying a threshold of P-value less than 0.05. The performance of the models was evaluated visually using Harrell's C-index and calibration curves by comparing the nomogram predictions with the actual observed endpoints [16] .  For validation, the discrimination and calibration capabilities of the calculators from the training cohort were verified using the same methods as in the external validation cohort. The bootstrapping technique (1000 repetitions), which is based on random sampling with replacement, was used for internal validation. In addition, the predictive performances of the two models and any of the four screened factors were compared using receiver operating characteristic (ROC) curves and decision curve analysis (DCA) [17,18] . To facilitate their incorporation into clinical practice, the model formulas were also coded into web-compatible versions. Statistical analyses were conducted using IBM SPSS Statistic, version 23.0 (IBM Corp.), and R software version 4.1.3 (http://www.r-project.org/). A P-value less than 0.05 was considered to indicate a significant difference in a twotailed test.

Patient cohorts and clinicopathologic features
During the study period, 283 patients who underwent HJ for BDI after cholecystectomy at eight hepato-pancreatic-biliary centers. Among them, 63 (22.26%) patients who met the exclusion criteria were excluded: 2 (0.71%) died in the perioperative period and 61 (21.55%) lacked complete clinical or follow-up data. A total of 220 patients were enrolled in the whole cohort, including 51 (23.18%) males and 169 (76.82%) females. Specifically, 118 (46.09%) patients from the First, Second, Third, and Fourth affiliated hospitals of Anhui Medical University were identified and set as the training cohort. Based on the same screening criteria, an independent group consisting of 102 (36.04%) patients from the remaining four centers during the same period were included in the present study and served as an external validation cohort (Fig. 1) 64%)]. There were no significant differences (P < 0.05) in baseline characteristics between the two cohorts as shown in Table 1.

Development of online calculators
The results of univariate and multivariate analyses of ASO and SFS based on accessible variables in the training cohort are listed in Tables 3 and 4, respectively. Factors that significantly affected ASO and SFS in the univariate analysis were subjected to multivariate analysis. Finally, four variables, including previous repair, sepsis, HJ phase, and bile duct fistula, which were identified as independent risk factors associated with both ASO and SFS from the multivariate analysis (P < 0.05), were used to construct the nomograms (Fig. 3). Both models demonstrated satisfactory precision in predicting stricture risk.  (Figs. 4 and 5). In addition, ROC curves for the two calculators were plotted in the training cohort (Fig. 6).

Comparisons of the models with four independent predictors
Comparisons of the discriminatory performance and predictive power between the calculators and the four independent predictors of ASO and SFS using ROC and DCA curves are shown in Figure 7. The areas under the curve of the proposed models in the training cohort were superior to those of any of the four variables in predicting ASO (area under the curve: 0.841 vs. 0.649-0.681) and SFS (time-dependent ROC). Furthermore, Figure 7 also demonstrates that the proposed models had the highest DCAs over most ranges of the probability thresholds, suggesting that the two models were more beneficial than independent factors and the best ability to predict the two study endpoints.

Discussion
Despite the popularity of cholecystectomy, BDI remains a severe complication related to cholecystectomy and may result in both immediate and long-term morbidity and mortality. Patients with severe BDI may require surgical intervention. A retrospective study of a nationwide dataset of patients who underwent cholecystectomy between 2001 and 2011 demonstrated that the incidence of BDI requiring surgical repair was 0.1% [20] . The long-term prognosis of repair for BDI can be dismal even in high-volume hepatobiliary centers with experienced surgeons. The stricture rate after HJ ranged from 10 to 30% [7] . The stricture rate in the present study was 18.64%, with most of the strictures occurring in the second year, and the median interval between HJ and stricture was 34.34 months, which is consistent with the findings of Stilling et al. [7] . Therefore, the high incidence of stricture after HJ is a major risk factor associated with the longterm life quality of patients with BDI. Accurate prediction of ASO and SFS can potentially help implement individualized therapy for patients undergoing HJ for BDI after cholecystectomy. Although multiple efforts have been made to explore the association between clinical variables and long-term stricture situations, none of these studies have further developed a prediction model [12,15,21] . In this scenario, the present study, therefore, developed and externally validated two nomogram-based calculators that could accurately predict ASO and SFS in this population. By considering a wide variety of prognostic factors and complex mathematical relationships, the current calculators individualized the risk for ASO and SFS for each patient and demonstrated a greater prognostic significance than the four independent risk factors, including previous repair, sepsis, HJ phase, and bile fistula. These four predictors can be easily accessed, making the models feasible in clinical practice. Furthermore, consistent with our results, previous studies have determined these factors as independent factors for ASO and SFS [12,15,21,22] . Almost all relevant studies have shown that the timing of sound biliary repair by experienced surgeons is a critical factor in improving the prognosis of BDI patients [5,[22][23][24] . Previous studies revealed that the success rate of repair performed by specialists can reach 79%; however, it is lower than 30% in the case of nonspecialists or beginners [25,26] . We concluded that nonspecialist repair may exert adverse effects on the outcomes of BDI in both psychological and technical aspects. First, nonspecialists tend to minimize the subsequent influence of intraoperative adverse incidents and are timid in the treatment of BDI. Second, limited by experience, nonspecialists are not likely to make a comprehensive assessment of the degree of BDI and thus, an appropriate treatment would not be chosen. Third, nonspecialists are inferior to specialists in the selection of anastomotic techniques, anastomotic materials, and management of postoperative complications.
Repair timing was classified into early, intermediate, and late phases. The traditional view is that repair surgery should be performed 3 months after biliary duct injury to facilitate HJ when the biliary wall and surrounding inflammation completely subside, the proximal bile duct dilates, and the range of bile duct ischemia is clear. With the development of surgical skills and equipment, emerging evidence supports the safety of early repair at experienced centers. Particularly, proven skills for thin bile duct anastomosis used in liver transplantation also bring about more probabilities for early repair. Perera et al. compared the outcomes of two groups of patients with BDI and found that immediate and early repair after BDI results in comparable, if not better, longterm outcomes compared with late repair when performed by specialists, with stricture rates of 18, 5, and 29%, respectively [27] . This study also suggested that repair attempts in low-volume hospitals before referral and late repair were associated with a higher risk for ASO and SFS. Besides, the suture method was not an independent risk factor associated with both ASO and SFS, which suggested that it was the quality rather than the method of anastomosis that independently affected the prognosis of patients with BDI. As such, a direct repair by experienced surgeons in the early phase in high-volume hospitals was highly recommended.
A large cohort study of 529 BDI patients who underwent HJ revealed that sepsis control before the repair was a protective factor against anastomotic failure, suggesting that adequate antibiotic and drainage treatment should be considered for patients with sepsis [28] . Bile duct fistula after HJ was also shown to increase the rate of anastomotic stricture after HJ, and bile fistula may induce a pre-anastomotic inflammatory response that resulted in fibrosis with stricture formation. In addition, based on our institutional experience, the placement of two drainages above and below the anastomotic stoma respectively during HJ for adequate drainage and removing them successively are beneficial for the management of bile fistula. More than 50% of the ASOs were diagnosed within 24 months and 75% within 36 months. These earlier ASOs were associated more with perioperative factors, including the four predictors and surgical techniques, while latter ASOs were associated more with the anatomical changes of the digestive system and nonfunctioning of Oddi sphincter after HJ [29,30] . The trend revealed by the Kaplan-Meier curve was also consistent with that in previous studies [7,15,[31][32][33] . Therefore, timely interventions regarding these four risk factors can reduce the incidence of ASO and improve the overall prognosis of patients after HJ.
In the present study, we also programmed two online calculators that can be accessed using computers, smartphones, or other mobile devices on the basis of nomograms that significantly improved the approachability of such predictive models. Our predictive model had reasonable discriminatory power with C-indices of 0.841 and 0.763 for ASO and SFS, respectively. In addition, the calibration plot demonstrated the satisfactory accuracy of our model in both the training and validation cohorts. In addition, both DCAs and ROCs revealed that the proposed models were the highest over most ranges of probability thresholds, indicating that the models were quite robust. These models provide easy-to-use tools for comprehensive prognosis evaluation and decision-making. For those patients with a high risk for ASO or dismal SFS according to the calculators, perioperative management should be enforced, and follow-up during the first 3 years after HJ should be performed more closely. The present study has several limitations. First, the retrospective nature of the methodology, institutional variation in surgical practices, and a high proportion of patients lost to follow-up may lead to inherent biases. Second, the models were established based on perioperative data, which made preoperative prediction impossible. In addition, prospective studies are required to validate the reliability of these models.
Irrespective of these limitations, our study has a number of strengths, such as the large sample size, the multi-institutional nature of data, and the use of an external validation dataset. The greatest strength may lie in the fact that this pioneering study provides a comprehensive insight into anastomotic stricture and makes it possible to create a 'risk flag' in the electronic medical record by identifying individuals who are at high risk for ASO and poor SFS and then provide individualized management. The potential for the seamless integration of high-value indicators of risk and clinical practice has a broad application prospect, which is also a promising direction for future research.

Conclusions
The two calculators developed for the first time for predicting the risk for ASO and SFS in patients who underwent HJ for BDI after cholecystectomy showed superior performance and discriminative power compared with those of a single independent factor. Both clinicians and patients can use them to stratify patients according to the underlying risk, allowing the development of personalized surveillance programs and a better distribution of health resources.

Ethical approval
The study procedures were approved by the Institutional Ethics Committee of the First Affiliated Hospital of Anhui Medical University (No. Quick-PJ 2022-08-29).