Multivariable prediction model for both 90-day mortality and long-term survival for individual patients with perihilar cholangiocarcinoma: does the predicted survival justify the surgical risk?

Abstract Background The risk of death after surgery for perihilar cholangiocarcinoma is high; nearly one in every five patients dies within 90 days after surgery. When the oncological benefit is limited, a high-risk resection may not be justified. This retrospective cohort study aimed to create two preoperative prognostic models to predict 90-day mortality and overall survival (OS) after major liver resection for perihilar cholangiocarcinoma. Methods Separate models were built with factors known before surgery using multivariable regression analysis for 90-day mortality and OS. Patients were categorized in three groups: favourable profile for surgical resection (90-day mortality rate below 10 per cent and predicted OS more than 3 years), unfavourable profile (90-day mortality rate above 25 per cent and/or predicted OS below 1.5 years), and an intermediate group. Results A total of 1673 patients were included. Independent risk factors for both 90-day mortality and OS included ASA grade III–IV, large tumour diameter, and right-sided hepatectomy. Additional risk factors for 90-day mortality were advanced age and preoperative cholangitis; those for long-term OS were high BMI, preoperative jaundice, Bismuth IV, and hepatic artery involvement. In total, 294 patients (17.6 per cent) had a favourable risk profile for surgery (90-day mortality rate 5.8 per cent and median OS 42 months), 271 patients (16.2 per cent) an unfavourable risk profile (90-day mortality rate 26.8 per cent and median OS 16 months), and 1108 patients (66.2 per cent) an intermediate risk profile (90-day mortality rate 12.5 per cent and median OS 27 months). Conclusion Preoperative risk models for 90-day mortality and OS can help identify patients with resectable perihilar cholangiocarcinoma who are unlikely to benefit from surgical resection. Tailored shared decision-making is particularly essential for the large intermediate group.


Introduction
Perihilar cholangiocarcinoma (pCCA) is the most common malignancy of the biliary tree 1 . In Western countries, the incidence of pCCA is about 1-2 patients per 100 000 2,3 . The aim of surgery is resection with negative surgical margins 4,5 . Surgical resection is challenging because the tumour arises at or near the biliary confluence in proximity to the vascular inflow structures of the liver. Therefore, surgical resection typically requires major liver resection with extrahepatic bile duct resection, and often reconstruction of the portal vein and/or hepatic artery. The overall 90-day mortality rate after resection of pCCA is about 12 per cent in large nationwide Western studies 6 . This postoperative mortality ranks as one of the highest in surgical oncology. The most common cause of postoperative death is liver failure, typically owing to a small liver remnant aggravated by infectious complications 7 .
Many studies of surgery for pCCA referred to resection as the only potentially curative treatment for patients with pCCA. The chance of cure after resection, however, was only 15 per cent for patients with extrahepatic biliary tract cancer 8 . The 10-year recurrence-free survival rate after resection of pCCA was only 5 per cent 9 . Patients with lymph node metastasis rarely experience long-term survival 10 . The patient and multidisciplinary team should determine whether the predicted long-term overall survival (OS) after resection justifies the short-term surgical risk. Weighing these two outcomes requires consideration of factors for surgical risk known before operation (for example ASA fitness grade) and for OS (for example hepatic artery involvement). Several studies [11][12][13] developed prognostic models for either short-or long-term postoperative outcomes. These models, however, are mostly unsuitable for preoperative decision-making because they include factors such as margin status that become available only after surgical resection.
The aim of this study was to develop two prognostic models with factors known before surgery for 90-day mortality and long-term OS after resection of pCCA. The individually predicted 90-day mortality and long-term OS can together guide the decision whether the predicted long-term OS justifies the predicted 90-day mortality.

Methods
Patients from the collaborative multicentre international database managed by the International Hepato-Pancreato Biliary Association (IHPBA) Perihilar Cholangiocarcinoma Collaboration Group, undergoing a major liver resection for pathologically confirmed pCCA between 2000 and 2019, were included in this retrospective cohort study. A total of 25 participating centres worldwide included a median of 79 (i.q.r. 42-100) consecutive patients. Each centre collected data using a standardized and deidentified data file. Patient and tumour characteristics, clinical parameters, and laboratory results were collected retrospectively from medical archives. Patients were excluded from the present study if they had undergone minor liver resection or extrahepatic bile duct resection only, associating liver partition and portal vein ligation for staged hepatectomy, had unresectable disease at surgical exploration, R2 resection margins, or had undergone liver transplantation. Ethical approval and individual informed consent were waived by the Institutional Medical Ethics Committee of the Amsterdam University Medical Centre.

Patient work-up and management
Variations in surgical expertise, hospital volume, and standard of care across hospitals and over time were accepted owing to the retrospective nature of the study and long inclusion period. Most patients underwent preoperative endoscopic or percutaneous transhepatic biliary drainage of the anticipated future liver remnant (FLR). When the anticipated FLR was considered inadequate, patients underwent portal vein embolization (PVE).

Definitions and outcomes
A major liver resection was defined as resection of at least three Couinaud segments. Resection of fewer than three Couinaud segments or an extrahepatic bile duct resection only were considered minor resections. Preoperative cholangitis was defined by the presence of fever, abdominal pain, or leucocytosis requiring biliary drainage, as defined previously in the DRAINAGE trial 14 . Postoperative mortality was defined as death within 90 days after resection. OS was defined as the interval between surgery and the date of death or last follow-up. Tumour margins were considered free (R0 resection) when all resection and circumferential margins were free from tumour on pathological examination. Hepatic artery or portal vein invasion was defined by the presence of abutment of at least 180° on radiological imaging (CT and/or MRI).

Statistical analysis
Statistical analyses were conducted with R 3.5.1 (https:// cran.r-project.org). Continuous data are reported as median (i.q.r.), and were compared using Mann-Whitney U tests. Categorical parameters are presented as counts and percentages, with analysis using χ 2 tests. The TRIPOD recommendations were followed for the development of this prediction model (Supplementary material) 15 . Multivariable models for predicting 90-day mortality were constructed using logistic regression analyses, with outcomes reported as linear predictors for comparing magnitude of correlation and ORs with corresponding 95 per cent confidence intervals. Univariable and multivariable Cox proportional hazard models were constructed for predicting OS, with outcomes reported as linear predictors for comparing magnitude of correlation and HRs with corresponding 95 per cent confidence intervals. Only factors that are available before surgery were considered for both models. Factors were included in multivariable models based on backwards selection with a cut-off of P < 0.050. Factors predictive of 90-day mortality were also included in the long-term survival model and vice versa. Model discrimination was presented as Harrell's C-index. Missing data were imputed using the mice package for R 3.5.1. A sensitivity analysis was undertaken for the last 6 years of the inclusion period (2014-2019). P values were two-tailed and P < 0.050 considered to be statistically significant. Individual patients were categorized in three groups based on arbitrary chosen cut-offs following expert consensus on the predicted outcomes of both models: favourable profile for surgical resection (90-day mortality rate below 10 per cent and predicted OS above 3 years), unfavourable profile (90-day mortality risk above 25 per cent and/or predicted OS below 1.5 years), and the intermediate group.

Patient characteristics
Baseline characteristics of the cohort are shown in Table 1. In total, 2136 patients were identified in the registry; 463 patients were excluded for the following reasons: no major liver resection (227), diagnosis other than pCCA on pathological examination of the resected specimen (201), surgical resection took place before the year 2000 (21), and macroscopically unresectable disease (R2 resection, 14). This led to 1673 eligible patients in the present study. A total of 1260 patients (79.5 per cent) presented with jaundice at first presentation.
The majority of patients underwent preoperative biliary drainage (1389, 83.0 per cent). Preoperative cholangitis occurred in 338 patients (

Ninety-day mortality
The 90-day mortality rate for the entire cohort was 13.6 per cent (228 of 1673). In the group aged less than 70 years, the 90-day mortality rate was 11.2 per cent (108 of 964), compared with 18.  In the sensitivity analysis, HRs and ORs were comparable to those for the complete cohort, with slightly better C-index values. A calculator for predicting both long-term OS and 90-day mortality for the individual patient is available at https:// dhoppener.shinyapps.io/risk_versus_harm_app/.

Discussion
In this study, two preoperative prognostic models were developed for predicting 90-day mortality and long-term OS after resection of pCCA. Patients were categorized in three groups based on predicted risk for both outcomes. A favourable risk profile (90-day mortality rate below 10 per cent and predicted OS above 3 years) was observed in 17.6 per cent of patients, reflecting those who were likely to benefit from surgical resection. An unfavourable risk profile (90-day mortality rate above 25 per cent and/or predicted OS below 1.5 years) was found in 16.2 per cent of patients, who were unlikely to benefit from surgery. An intermediate risk profile, observed in 66.2 per cent of patients, would require a more tailored approach balancing patient preferences regarding surgical risk and long-term OS.
A recent systematic review 6 reported a pooled 90-day mortality rate of 12 per cent after major liver resection for pCCA in Western countries. This is comparable to the 90-day mortality rate of 13.6 per cent in the present study, which excluded patients who underwent extrahepatic bile duct resection only. Postoperative mortality after resection for pCCA is mostly due to posthepatectomy liver failure 16,17 . Reduction in liver failure and mortality can be achieved by adequate biliary drainage and preoperative augmentation of the FLR with PVE 18 . PVE was performed in about 60 per cent of patients in a large Asian study 19 , with a low postoperative mortality rate of 2 per cent. Only 16 per cent of patients in the present study underwent PVE. More liberal use of PVE may increase the FLR volume, and decrease postoperative liver failure and mortality 20 .
Surgical resection of pCCA offers the best chance of long-term survival. The 5-year OS rate of 17 per cent and median OS of 27 months in this study is comparable to values in other series. Published median OS after resection of pCCA ranged from 20 to 40 months 21 , with 5-year OS rates ranging from 10 to 30 per cent 1 . Although long-term survival after resection of pCCA remains poor, the median OS with palliative systemic chemotherapy for advanced biliary tract cancer was only 11 months, with no survivors beyond 3 years 22 . However, these patients had advanced disease. OS in patients with resectable pCCA receiving palliative chemotherapy would probably be better. Nevertheless, the potential survival benefit of surgical resection should be carefully balanced against the risk of 90-day mortality. In addition to postoperative mortality, surgical resection is associated with considerable morbidity, with prolonged hospital stay and recovery after discharge.
The influence of (neo)adjuvant chemotherapy was not included in this preoperative risk evaluation. Only a few small retrospective studies of neoadjuvant chemotherapy for biliary tract cancer are available 23 . These studies mainly included patients with locally advanced or borderline resectable biliary tract cancer, for whom neoadjuvant chemotherapy might offer downstaging. A currently ongoing phase II trial, the NACRAC study 24 [25][26][27] .
Patients (and surgeons) might differ regarding the 90-day mortality risk that they believe would justify resection. This study assumed an arbitrary predicted 90-day mortality risk of 25 per cent above which a resection seems rarely justified. Similarly, patients (and surgeons) might also differ regarding the long-term OS that they believe would justify the surgical risks. This study assumed a predicted OS of 18 months below which a resection seems rarely justified. In surgical oncology, unfavourable patient factors and more advanced disease are often associated with both poor short-and long-term outcomes. In the present study, ASA grade III-IV, tumour diameter, and right-sided resection were associated with both worse 90-day mortality and worse long-term OS. Future research should investigate how patients and surgeons both balance surgical risk and long-term OS. Moreover, patients rely primarily on the information from the surgeon. Adequate use of shared decision-making can increase patient satisfaction and decrease decisional regret 28 .
Several staging systems for pCCA are used to guide treatment including the AJCC, Bismuth-Corlette, and Blumgart systems [29][30][31] . However, these staging systems consider only anatomical aspects of the tumour. To address these limitations, several prognostic models have been developed for survival after surgical resection of pCCA 12,32-35 . These models typically include margin status, nodal status, and tumour differentiation. Although these variables are strong independent prognostic factors, they are available only after surgical resection and cannot therefore be used for preoperative decision-making.
This study should be viewed in the light of several limitations. First, inherent to all retrospective studies, the diagnosis and treatment of patients differed between the 25 participating centres and changed over time during the two-decade inclusion period. However, to achieve its aim, this study required a very large cohort that was otherwise not available. Second, about one-third of patients with resectable pCCA on imaging do not undergo resection because of occult metastatic disease or more advanced disease. These patients were not included in this multicentre cohort. Therefore, the predicted OS and 90-day mortality apply only to patients who do undergo resection. Most studies presented outcomes only for patients with resected pCCA. Studies including all patients with resectable pCCA (for example the DRAINAGE and INTERCPT trials) reported a much higher perioperative mortality rate 14,36 . The proportion of patients with primary sclerosing cholangitis (PSC) was small (4 per cent), as most patients with PSC underwent liver transplantation and were therefore not included in this cohort of resections. Finally, patients with a suspected diagnosis of pCCA but a different diagnosis after resection (5-10 per cent) were not included in this multicentre cohort 37 . In the preoperative setting, the final diagnosis is not known. Consequently, observed OS will be slightly longer than predicted by the model because of patients with non-malignant disease.