External Validation of the Estimation of Physiologic Ability and Surgical Stress (E-PASS) Risk Model to Predict Operative Risk in Perihilar Cholangiocarcinoma | Clinical Decision Support | JAMA Surgery | JAMA Network
[Skip to Navigation]
Figure 1.  Area Under the Curve (AUC) Analysis for Estimation of Physiologic Ability and Surgical Stress Models (Comprehensive Risk Score [CRS] and Fixed CRSf [CRSf]) as a Discriminant of Postoperative In-Hospital Mortality
Area Under the Curve (AUC) Analysis for Estimation of Physiologic Ability and Surgical Stress Models (Comprehensive Risk Score [CRS] and Fixed CRSf [CRSf]) as a Discriminant of Postoperative In-Hospital Mortality
Figure 2.  Morbidity and Mortality Rates Among Risk Groups of Estimation of Physiologic Ability and Surgical Stress Models
Morbidity and Mortality Rates Among Risk Groups of Estimation of Physiologic Ability and Surgical Stress Models

CCI indicates Comprehensive Complication Index; CRS, Comprehensive Risk Score; and CRSf, fixed CRS.

Table 1.  Demographics and Operative Characteristics of the Study Cohort
Demographics and Operative Characteristics of the Study Cohort
Table 2.  Postoperative Outcomes and Eventsa
Postoperative Outcomes and Eventsa
Table 3.  In-Hospital Mortality Risk, as Predicted by E-PASS Models and Actual Number of Deaths
In-Hospital Mortality Risk, as Predicted by E-PASS Models and Actual Number of Deaths
1.
Nuzzo  G, Giuliante  F, Ardito  F,  et al; Italian Chapter of the International Hepato-Pancreato-Biliary Association.  Improvement in perioperative and long-term outcome after surgical treatment of hilar cholangiocarcinoma: results of an Italian multicenter analysis of 440 patients.  Arch Surg. 2012;147(1):26-34.PubMedGoogle ScholarCrossref
2.
van Gulik  TM, Kloek  JJ, Ruys  AT,  et al.  Multidisciplinary management of hilar cholangiocarcinoma (Klatskin tumor): extended resection is associated with improved survival.  Eur J Surg Oncol. 2011;37(1):65-71.PubMedGoogle ScholarCrossref
3.
Nagino  M, Ebata  T, Yokoyama  Y,  et al.  Evolution of surgical treatment for perihilar cholangiocarcinoma: a single-center 34-year review of 574 consecutive resections.  Ann Surg. 2013;258(1):129-140.PubMedGoogle ScholarCrossref
4.
Farges  O, Regimbeau  JM, Fuks  D,  et al.  Multicentre European study of preoperative biliary drainage for hilar cholangiocarcinoma.  Br J Surg. 2013;100(2):274-283.PubMedGoogle ScholarCrossref
5.
de Jong  MC, Marques  H, Clary  BM,  et al.  The impact of portal vein resection on outcomes for hilar cholangiocarcinoma: a multi-institutional analysis of 305 cases.  Cancer. 2012;118(19):4737-4747.PubMedGoogle ScholarCrossref
6.
Coelen  RJ, Wiggers  JK, Nio  CY,  et al.  Preoperative computed tomography assessment of skeletal muscle mass is valuable in predicting outcomes following hepatectomy for perihilar cholangiocarcinoma.  HPB (Oxford). 2015;17(6):520-528.PubMedGoogle ScholarCrossref
7.
Kennedy  TJ, Yopp  A, Qin  Y,  et al.  Role of preoperative biliary drainage of liver remnant prior to extended liver resection for hilar cholangiocarcinoma.  HPB (Oxford). 2009;11(5):445-451.PubMedGoogle ScholarCrossref
8.
Haga  Y, Ikei  S, Ogawa  M.  Estimation of Physiologic Ability and Surgical Stress (E-PASS) as a new prediction scoring system for postoperative morbidity and mortality following elective gastrointestinal surgery.  Surg Today. 1999;29(3):219-225.PubMedGoogle ScholarCrossref
9.
Haga  Y, Ikei  S, Wada  Y,  et al.  Evaluation of an Estimation of Physiologic Ability and Surgical Stress (E-PASS) scoring system to predict postoperative risk: a multicenter prospective study.  Surg Today. 2001;31(7):569-574.PubMedGoogle ScholarCrossref
10.
Haga  Y, Wada  Y, Takeuchi  H,  et al.  Estimation of Physiologic Ability and Surgical Stress (E-PASS) for a surgical audit in elective digestive surgery.  Surgery. 2004;135(6):586-594.PubMedGoogle ScholarCrossref
11.
Haga  Y, Ikejiri  K, Wada  Y,  et al.  A multicenter prospective study of surgical audit systems.  Ann Surg. 2011;253(1):194-201.PubMedGoogle ScholarCrossref
12.
Wang  H, Wang  H, Chen  T, Liang  X, Song  Y, Wang  J.  Evaluation of the POSSUM, P-POSSUM and E-PASS scores in the surgical treatment of hilar cholangiocarcinoma.  World J Surg Oncol. 2014;12:191.PubMedGoogle ScholarCrossref
13.
Haga  Y, Miyamoto  A, Wada  Y, Takami  Y, Takeuchi  H.  Value of E-PASS models for predicting postoperative morbidity and mortality in resection of perihilar cholangiocarcinoma and gallbladder carcinoma.  HPB (Oxford). 2016;18(3):271-278.PubMedGoogle ScholarCrossref
14.
Edge  SBBD, Compton  CC, Fritz  AG, Greene  FL, Trotti  A, eds; American Joint Committee on Cancer.  AJCC Cancer Staging Manual. 7th ed. New York, NY: Springer; 2010.
15.
de Graaf  W, van Lienden  KP, Dinant  S,  et al.  Assessment of future remnant liver function using hepatobiliary scintigraphy in patients undergoing major liver resection.  J Gastrointest Surg. 2010;14(2):369-378.PubMedGoogle ScholarCrossref
16.
Dindo  D, Demartines  N, Clavien  PA.  Classification of surgical complications: a new proposal with evaluation in a cohort of 6336 patients and results of a survey.  Ann Surg. 2004;240(2):205-213.PubMedGoogle ScholarCrossref
17.
Slankamenac  K, Graf  R, Barkun  J, Puhan  MA, Clavien  PA.  The Comprehensive Complication Index: a novel continuous scale to measure surgical morbidity.  Ann Surg. 2013;258(1):1-7.PubMedGoogle ScholarCrossref
18.
Rahbari  NN, Garden  OJ, Padbury  R,  et al.  Posthepatectomy liver failure: a definition and grading by the International Study Group of Liver Surgery (ISGLS).  Surgery. 2011;149(5):713-724.PubMedGoogle ScholarCrossref
19.
Koch  M, Garden  OJ, Padbury  R,  et al.  Bile leakage after hepatobiliary and pancreatic surgery: a definition and grading of severity by the International Study Group of Liver Surgery.  Surgery. 2011;149(5):680-688.PubMedGoogle ScholarCrossref
20.
Rahbari  NN, Garden  OJ, Padbury  R,  et al.  Post-hepatectomy haemorrhage: a definition and grading by the International Study Group of Liver Surgery (ISGLS).  HPB (Oxford). 2011;13(8):528-535.PubMedGoogle ScholarCrossref
21.
Vibert  E, Boleslawski  E, Cosse  C,  et al.  Arterial lactate concentration at the end of an elective hepatectomy is an early predictor of the postoperative course and a potential surrogate of intraoperative events.  Ann Surg. 2015;262(5):787-792.PubMedGoogle ScholarCrossref
22.
Janssen  KJ, Donders  AR, Harrell  FE  Jr,  et al.  Missing covariate data in medical research: to impute is better than to ignore.  J Clin Epidemiol. 2010;63(7):721-727.PubMedGoogle ScholarCrossref
23.
Rubin  DB.  Multiple Imputation for Nonresponse in Surveys. New York, NY: John Wiley & Sons; 1987.
24.
Sakata  J, Shirai  Y, Tsuchiya  Y, Wakai  T, Nomura  T, Hatakeyama  K.  Preoperative cholangitis independently increases in-hospital mortality after combined major hepatic and bile duct resection for hilar cholangiocarcinoma.  Langenbecks Arch Surg. 2009;394(6):1065-1072.PubMedGoogle ScholarCrossref
Original Investigation
December 2016

External Validation of the Estimation of Physiologic Ability and Surgical Stress (E-PASS) Risk Model to Predict Operative Risk in Perihilar Cholangiocarcinoma

Author Affiliations
  • 1Department of Surgery, Academic Medical Center, Amsterdam, the Netherlands
JAMA Surg. 2016;151(12):1132-1138. doi:10.1001/jamasurg.2016.2305
Key Points

Question  What is the value of the Estimation of Physiologic Ability and Surgical Stress (E-PASS) risk model and its modified preoperative version (mE-PASS) in predicting in-hospital mortality after resection for perihilar cholangiocarcinoma?

Findings  In this retrospective study that included 156 patients, both models had adequate discriminative performance despite poor mE-PASS calibration. Both models were able to distinguish groups with low (0.0%-3.6%), intermediate (8.3%-9.0%), and high (25.0%-28.3%) mortality risk.

Meaning  The E-PASS models accurately identify patients at high risk of in-hospital mortality after resection for perihilar cholangiocarcinoma, thereby allowing risk assessment and shared decision making.

Abstract

Importance  Resection of perihilar cholangiocarcinoma (PHC) is high-risk surgery, with reported operative mortality up to 17%. Therefore, preoperative risk assessment is needed to identify high-risk patients and anticipate postoperative adverse outcomes.

Objective  To provide external validation of the Estimation of Physiologic Ability and Surgical Stress (E-PASS) risk model in a Western PHC cohort.

Design, Setting, and Participants  The E-PASS variables were obtained from a database that included 156 consecutive patients who underwent resection for suspected PHC between January 1, 2000, and December 31, 2015, at the Academic Medical Center, Amsterdam, the Netherlands. The accuracy of E-PASS using intraoperative variables and its modified form that can be used before surgery (mE-PASS) in predicting mortality was assessed by area under the curve analysis (discrimination) and by the Hosmer-Lemeshow goodness-of-fit test (calibration).

Main Outcomes and Measures  In-hospital mortality, severe morbidity (Clavien-Dindo grade≥III), and a high Comprehensive Complication Index.

Results  Among 156 patients included in the study, the median age was 63 years, and 62.8% (n = 98) were male. Of them, 85.3% (n = 133) underwent major liver resection. Severe morbidity occurred in 51.3% (n = 80), and in-hospital mortality was 13.5% (n = 21). Both E-PASS and mE-PASS had adequate discriminative performance, with areas under the curve of 0.78 (95% CI, 0.67-0.88) and 0.79 (95% CI, 0.70-0.89), respectively, while E-PASS showed better calibration (P = .33 vs P = .02, Hosmer-Lemeshow goodness-of-fit test). The ratios of observed to expected mortality were 1.31 for E-PASS and 1.24 for mE-PASS. Both models were able to distinguish groups with low risk, intermediate risk, and high risk, with observed mortality rates of 0.0% to 3.6%, 8.3% to 9.0%, and 25.0% to 28.3%, respectively. Severe morbidity and a high Comprehensive Complication Index were more frequently observed among high-risk patients.

Conclusions and Relevance  Both E-PASS models accurately identify patients at high risk of postoperative in-hospital mortality after resection for PHC. The mE-PASS model can be used before surgery in outpatient settings and allows for risk assessment and shared decision making.

Introduction

Resection of perihilar cholangiocarcinoma (PHC) offers the only chance for long-term survival. The procedure typically consists of a combined extrahepatic bile duct and liver resection and often requires an extended hemihepatectomy or vascular reconstruction to obtain a radical resection.1,2 This technically challenging and aggressive approach in mostly postcholestatic livers contributes to a high postoperative mortality rate, ranging from 5% to 17% even in experienced centers.1,3-6

Both patient-related factors and surgical variables are important contributors to substantial operative risk, resulting in high morbidity and mortality. Factors that have been associated with adverse postoperative outcomes in PHC include advanced age,6 preoperative cholangitis,3 small future liver remnant (FLR) volume,7 portal vein reconstruction,5 and intraoperative blood loss.4 However, several of these factors can only be determined at the time of surgery. Ideally, a PHC-specific risk model would aid the clinician in the phase of preoperative risk assessment and shared decision making by identifying high-risk patients. Although much needed, no such model specifically addressing PHC is available to date.

More than a decade ago, Japanese colleagues developed a scoring system to predict postoperative outcome after elective gastrointestinal surgery.8,9 That model is based on the hypothesis that postoperative complications result from a disruption of homeostasis due to overwhelming surgical stress exceeding a patient’s reserve capacity, thus addressing both preoperative and surgical variables. The Estimation of Physiologic Ability and Surgical Stress (E-PASS) model has proved effective in predicting postoperative morbidity and mortality after various gastrointestinal surgical procedures, although it has mainly been studied in Asian populations.10 This model was modified (mE-PASS) by reducing the number of surgical variables and allocating fixed stress scores (median values) to specific procedures.11 The risk score thereby is clinically valuable at the time of surgical planning.

Both E-PASS models were recently shown to accurately predict postoperative outcome in PHC at 2 Asian institutions.12,13 Therefore, the present study aimed to provide external validation of these models in a Western PHC cohort.

Methods
Study Population and Preoperative Workup

A waiver was granted from the Institutional Review Board of the Academic Medical Center, Amsterdam, the Netherlands, for approval of this retrospective study. The need for written or oral informed consent was waived. Data were retrospectively obtained from a prospective database that included all consecutive patients who underwent resection for suspected PHC between January 1, 2000, and December 31, 2015, at the Academic Medical Center. Perihilar cholangiocarcinoma was defined as a tumor mass or seemingly malignant stricture at or near the biliary confluence, arising between the origin of the cystic duct and the segmental bile ducts.14

Preoperative optimization included endoscopic or percutaneous biliary drainage of at least the FLR in the presence of obstructive cholestasis with jaundice. Portal vein embolization was performed for a small FLR volume (computed tomography volumetry <35%) or when hepatobiliary scintigraphy indicated poor FLR function.15 Any episode of preoperative cholangitis induced by biliary decompression was treated with antibiotics and, when indicated, additional drainage or drain revision. Patients underwent surgery not earlier than they had fully recovered from drainage-related complications and cholestasis.

Surgical Procedures

Patients underwent radical resection of the tumor encompassing hilar resection with en bloc (extended) hemihepatectomy, including the caudate lobe in most cases. Excision and reconstruction of the left or right portal vein or its bifurcation were performed when involved by the tumor. Complete lymphadenectomy of the hepatoduodenal ligament was routinely performed. For biliary reconstruction, end-to-side anastomoses of the segmental ducts and a Roux-en-Y jejunal loop were constructed.2 In selected patients with Bismuth-Corlette type I or II tumors, only an extrahepatic bile duct resection without liver resection was performed. Frozen section of the proximal and distal bile duct margins was routinely performed to ensure tumor-free margins. All resections were performed by staff surgeons with extensive hepatobiliary expertise.

E-PASS Models

Both the E-PASS and mE-PASS scoring systems are regression models that have previously been described in detail by Haga et al.11,13 Briefly, a Comprehensive Risk Score (CRS) is calculated by combining a Preoperative Risk Score (PRS) consisting of 6 preoperative variables and a Surgical Stress Score (SSS) consisting of 3 surgical variables.

The PRS is calculated using the following formula: PRS = −0.0686 + 0.00345X1 + 0.323X2 + 0.205X3 + 0.153X4 + 0.148X5 + 0.0666X6, where X1 is age, X2 is the presence (1 point) or absence (0 points) of severe heart disease (New York Heart Association class III-IV or severe arrhythmia requiring mechanical support), X3 is the presence (1 point) or absence (0 points) of severe pulmonary disease (vital capacity <60% or forced expiratory volume in 1 second <50%), X4 is the presence (1 point) or absence (0 points) of type 1 or 2 diabetes (World Health Organization [WHO] criteria), X5 is the WHO performance status index (range, 0-4 points), and X6 is the American Society of Anesthesiologists physiological status classification (range, 1-5 points). The WHO status was checked at first presentation in our center.

The SSS is calculated using the following formula: SSS = −0.342 + 0.0139X1 + 0.0392X2 + 0.352X3, where X1 is blood loss (in grams) divided by body weight (in kilograms), X2 is the operative time (in hours), and X3 is the extent of the skin incision (0 points for a minor incision or laparoscopic or thoracoscopic surgical procedure, 1 point for laparotomy or thoracotomy alone, and 2 points for laparotomy and thoracotomy). Because all patients in the study cohort underwent laparotomy alone, the value for the extent of the skin incision consistently was 1.

The CRS is calculated using the following formula: CRS =  −0.328 + (0.936 × PRS) + (0.976 × SSS). For the modified model (mE-PASS), a fixed CRS (CRSf) was computed by combining the PRS and a fixed SSS (SSSf). Predefined values for different resection types were 0.401 for extrahepatic bile duct resection only, 0.453 for liver segmentectomy, 0.663 for hemihepatectomy, and 1.025 for extended hemihepatectomy.11

The CRSf is calculated using the following formula: CRSf = 0.052 + (0.58 × PRS) + (0.83 × SSSf). The predicted in-hospital mortality rate (Y) for both E-PASS models was then calculated with the following predefined equations: CRS less than 0.159 (Y ≈ 0), CRS between 0.159 and 2.98 (Y = −0.465 + 1.192[CRS] + 10.91[CRS]2), CRS at least 2.98 (Y = 100), CRSf less than 0.326 (Y ≈ 0), and CRSf at least 0.326 (Y = −0.0541[CRSf] + 0.197[CRSf]2 − 0.00328).

All variables needed for calculating the CRS and CRSf were obtained from our PHC database. Total scores and predicted in-hospital mortality were then calculated for each individual patient.

End Points

Study end points were in-hospital mortality and postoperative morbidity. All complications occurring within 30 days after surgery or in the hospital were scored according to the Clavien-Dindo grading system (range, I-V).16 Severe morbidity was defined as Clavien-Dindo grade III-V. Also, the Comprehensive Complication Index (CCI) (range, 0-100), which summarizes all postoperative complications, was calculated for each patient according to the incidence and consequence of postoperative events.17 Complications, such as posthepatectomy liver failure, biliary leakage, and hemorrhage, were scored and graded according to the International Study Group of Liver Surgery criteria, with grades B and C considered as clinically relevant and severe morbidity.18-20

Statistical Analysis

Continuous variables are presented as the mean (SD) or median (range) for nonnormal distributed data. Categorical variables are expressed as counts and percentages. The CCI was transformed into a categorical variable using a CCI greater than the third quartile as the cutoff for low and high CCI.21

The predictive performance of the E-PASS models was analyzed in terms of discrimination and calibration. Discrimination, the ability to distinguish patients who died after surgery from those who did not, was assessed by area under the curve analysis with 95% CIs. Areas under the curve exceeding 0.70 and 0.80 were considered acceptable and excellent discrimination, respectively. Calibration, the agreement between predicted mortality and observed mortality, was assessed by the Hosmer-Lemeshow goodness-of-fit test, with a significant outcome (P < .05) indicating poor calibration.

Observed in-hospital mortality rates were then compared between 3 risk categories (low, intermediate, and high) based on the total CRS and CRSf scores. Furthermore, the incidence of severe complications among these groups was observed and compared using the Pearson χ2 test. The association between CRS and severe morbidity (CCI above the third quartile) was also assessed in multivariable analysis by logistic regression, which was adjusted for sex, preoperative biliary drainage, preoperative cholangitis, preoperative bilirubin levels, FLR volume, and vascular reconstruction.

Two variables needed for CRS calculation had missing data, namely, intraoperative blood loss (10.9% [17 of 156] missing data) and body weight (1.9% [3 of 156] missing data). To avoid bias, multiple imputation with 10 imputed data sets was performed for these missing data before calculating the CRS.22 A regression model was used with baseline values, E-PASS, and outcome variables. Data were subsequently pooled using the rule by Rubin.23 There were no missing data for CRSf calculation.

Analyses were performed using statistical software (SPSS Statistics for Windows, version 23.0; IBM and R, version 3.1.2; R Foundation for Statistical Computing). Two-tailed P < .05 was considered statistically significant.

Results
Patient Demographics and Operative Characteristics

A total of 156 consecutive patients (98 [62.8%] male) underwent resection for presumed PHC during the study period. Demographics and operative characteristics are listed in Table 1. Most patients (139 [89.1%]) had undergone preoperative biliary drainage, and a high incidence (59 [37.8%]) of preoperative cholangitis was noted. A combined extrahepatic bile duct resection and liver resection was performed in 137 (87.8%) patients, with 133 (85.3%) undergoing a major hepatectomy (ie, ≥3 Couinaud segments). Portal vein reconstruction was performed in 36 (23.1%) patients. Observed median SSS values for patients undergoing formal hemihepatectomy and extended hemihepatectomy were 0.790 and 0.995, respectively.

Pathological examination confirmed PHC in 129 patients (82.7%) and revealed other disease in 27 patients. Intraductal papillary neoplasm of the bile duct without invasive component was found in 7 patients (4.5%), other malignant neoplasms in 4 patients (2.6%), and benign lesions in 16 patients (10.3%).

Postoperative Events

Twenty-one patients (13.5%) died in the hospital after surgery. All of these patients had undergone a major hepatectomy. There was a suggestion of higher mortality after extended hemihepatectomy compared with standard hemihepatectomy (24.0% [12 of 50] vs 11.3% [9 of 80], P = .06). Causes of death were liver failure (13 [61.9%] patients), sepsis with multiorgan failure (7 [33.3%] patients), or myocardial infarction (1 [4.8%] patient). Liver remnant volume below 30% was significantly associated with either sepsis or liver failure–related death (odds ratio, 3.18; 95% CI, 1.15-8.80; P = .03).

A total of 122 patients (78.2%) developed 1 or more postoperative complications of any grade. The total number of complications among all patients was 341 (median, 2 per patient), and the median CCI was 29.6. The incidence of complications with Clavien-Dindo grade III or higher was 80 of 156 (51.3%). An overview of postoperative complications is summarized in Table 2. Biliary leakage was the most frequent complication, occurring in 53 patients (34.0%), and 47 (30.1%) of these events were classified as International Study Group of Liver Surgery grade B or C. The median hospital stay was 13 days (range, 4-95 days), including the day of admission.

E-PASS Model Performance

Discriminative power of the E-PASS models to predict in-hospital mortality is shown in Figure 1. Areas under the curve for CRS and CRSf were 0.78 (95% CI, 0.67-0.88) and 0.79 (95% CI, 0.70-0.89), respectively, indicating acceptable discrimination. Agreement between predicted mortality and observed mortality was best for E-PASS, as shown by a pooled P = .33 for the Hosmer-Lemeshow goodness-of-fit test for 10 imputed data sets. The mE-PASS model showed poor fit between predicted mortality and observed mortality (P = .02). Subgroup analysis revealed that mE-PASS calibration was poor for patients undergoing formal hemihepatectomy (P = .002) but was acceptable for those undergoing extended hemihepatectomy (P = .14). The ratios of observed to predicted mortality for the present cohort were 1.31 for E-PASS and 1.24 for mE-PASS (Table 3). When the risk of formal hemihepatectomy was reweighted using the observed median SSS of 0.790 for this procedure in the study cohort, mE-PASS calibration for patients undergoing this type of liver resection remained poor (P = .02).

Observed morbidity and mortality among the CRS and CRSf risk groups are shown in Figure 2. Observed mortality rates ranged from 3.6% (2 of 55) in the low-risk group to 8.3% (4 of 48) in the intermediate-risk group and to 28.3% (15 of 53) in the high-risk group (P < .001) for CRS. For the mE-PASS model, observed mortality rates were 0.0% (0 of 22) in the low-risk group, 9.0% (7 of 78) in the intermediate-risk group, and 25.0% (14 of 56) in the high-risk group (P = .004) for CRSf. A significant higher incidence of severe complications (Clavien-Dindo grade≥III) was observed in the high-risk group compared with the low-risk and intermediate-risk groups (75.5% [40 of 53] vs 36.4% [20 of 55] vs 41.7% [20 of 48], respectively; P < .001) for CRS. Even so, high CCI was comparable in the low-risk and intermediate-risk groups for CRS but was more frequent in the high-risk group (14.5% [8 of 55] vs 10.4% [5 of 48] vs 49.1% [26 of 53], respectively; P < .001). For the mE-PASS model, a higher complication rate was observed in the highest-risk group for CRSf, but it was only significantly higher compared with the lowest-risk group (60.7% [34 of 56] vs 31.8% [7 of 22], respectively; P = .02). High CCI was significantly increased in both the intermediate-risk and high-risk CRSf groups (0 vs 21.8% [17 of 78] and 40.0% [22 of 55], respectively; P = .001).

Multivariable logistic regression analysis revealed that CRS (adjusted odds ratio, 13.57; 95% CI, 3.34-55.22) and preoperative cholangitis (adjusted odds ratio, 10.57; 95% CI, 3.15-35.40) were independent risk factors for severe postoperative morbidity. These results are summarized in the eTable in the Supplement.

Discussion

The present study confirms the value of the E-PASS models for predicting postoperative in-hospital mortality after resection for PHC. The risk scores showed adequate predictive performance in this external Western validation cohort. These models provide a valuable tool for preoperative risk assessment and shared decision making as early as at the initial outpatient consultation. The mE-PASS model can be used in the preoperative setting, which has obvious benefits to both the clinician and patient.

Liver surgery has improved during the past decades, and preoperative optimization techniques, such as portal vein embolization and biliary drainage, have reduced the risks in subgroups of patients. However, morbidity and mortality after resection of PHC remain high even in specialized, high-volume centers. Mortality rates in the literature range from 5% to 17% in these centers, and operative risk has been reported to be even higher in selected patients who require right hemihepatectomies or extended liver resections.1,3-6 Furthermore, preoperative cholangitis remains a major concern because it increases the mortality risk more than 3-fold.3,24 Such high-risk procedures warrant adequate patient counseling because these high rates might not be acceptable for some patients. Morbidity after PHC is also substantial and is reported in up to 68%.2 However, reporting severe morbidity as the presence of a Clavien-Dindo grade III or higher complication inadequately addresses the overall postoperative events in patients. Most patients develop several minor and major complications; therefore, the CCI might provide more accurate insight into the burden of postoperative adverse events.

To date, there is no operative risk model specifically addressing PHC. The E-PASS models were originally developed in a large set of gastrointestinal surgical procedures and showed better predictive performance than well-known scores like the Physiological and Operative Severity Score for the Enumeration of Mortality and Morbidity (POSSUM).9-11 Two studies12,13 from China and Japan, respectively, found that the models were valuable to predict mortality after surgery for PHC. However, these studies were flawed by heterogeneous populations and small sample sizes. In the study by Wang et al,12 only 62 patients with PHC underwent resection (23 of whom had hepatectomies), and their analyses also included 38 patients who had unresectable tumors. The study by Haga et al13 also included patients with gallbladder carcinoma and included only 56 major liver resections. Therefore, the present study comprising 156 patients from a well-defined cohort adds substantial weight to the value of the E-PASS models in PHC.

However, there are some limitations to the E-PASS models. First, because E-PASS contains several operative variables, such as blood loss and operative time, this model might be suboptimal for preoperative risk assessment, with preoperative estimation of these variables possibly resulting in substantial underestimation or overestimation of operative outcomes. The modified model (mE-PASS) uses SSSf, making it suitable for preoperative use and shared decision making. However, calibration analysis showed a poorer fit for the mE-PASS model, which could be explained by underestimation of the surgical stress of formal hemihepatectomy in our cohort, although reweighting of operative risk did not lead to significant improvement in model calibration. Despite imbalance between the predicted mortality risk and the observed mortalities, the mE-PASS model could accurately identify patients at higher risk of dying after surgery. Overall, both models somewhat underpredicted in-hospital mortality. Furthermore, predictors of outcomes in PHC, such as biliary drainage status of the FLR, preoperative cholangitis, and portal vein reconstruction, are not included, rendering the model most likely not the ideal predictive one. Preoperative cholangitis remained an independent risk factor for severe morbidity in our analysis after adjustment for CRS, which includes the WHO performance score. Last, severe heart and pulmonary disease were almost nonexistent in our cohort, which could be owing to patient selection because patients with New York Heart Association class III or IV, severe arrhythmia, or decreased vital capacity are usually not considered adequate candidates for major liver surgery.

Despite the lack of PHC-specific factors in the E-PASS models, these scores showed acceptable discrimination for mortality in our cohort. Based on their individual CRS and CRSf scores, patients could be categorized into low-risk, intermediate-risk, and high-risk groups. Severe morbidity and mortality varied substantially, with a low (0.0%-3.6%) mortality rate in the low-risk group, an intermediate (8.3%-9.0%) mortality rate in the intermediate-risk group, and a high (25.0%-28.3%) mortality rate in the high-risk group. Severe morbidity and high CCI were more frequently observed among high-risk patients. We chose to also calculate the CCI for individual patients because it provides more accurate information about the overall postoperative events compared with the Clavien-Dindo classification.17 Because the CRS and CRSf equations are more difficult to compute compared with a simple point-adding risk score, a CRS-derived total risk points score may be more convenient to use in clinical practice. The total risk points score has previously been described by the E-PASS investigators.10 To aid the clinician in predicting operative risk in patients with PHC, the eFigure in the Supplement has been included to allow automatic calculation of E-PASS scores and in-hospital mortality based on the equations by Haga et al.11

There are several limitations to our study. Although this study includes one of the largest single-center series of patients with PHC, to our knowledge, the small sample size resulted in few events. The consequent statistical uncertainty in the analyses is reflected by the wide 95% CIs. A second limitation is the considerable amount (up to 17%) of missing data (mainly intraoperative blood loss) for calculating the CRS values, which was overcome by multiple imputation with 10 imputed data sets. Although this limitation may have introduced some bias, imputation of missing variables has been shown to decrease the risk of bias by not excluding those patients with missing variables.22

Conclusions

The present study shows that the E-PASS models accurately identify patients at higher risk of postoperative in-hospital mortality after resection for PHC. In the absence of models based on specific risk factors in PHC, the E-PASS models allow risk assessment and patient counseling as early as at the initial outpatient visit and can thus support shared decision making.

Back to top
Article Information

Correction: This article was corrected on October 12, 2016, to fix an x-axis label in Figure 2.

Accepted for Publication: May 19, 2016.

Corresponding Author: Robert J. S. Coelen, MD, Department of Surgery, Academic Medical Center, Meibergdreef 9, 1105 AZ Amsterdam, the Netherlands (r.j.coelen@amc.nl).

Published Online: August 31, 2016. doi:10.1001/jamasurg.2016.2305

Author Contributions: Drs Coelen and van Gulik had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Drs Coelen and Olthof contributed equally to this work.

Study concept and design: Coelen, Olthof, van Gulik.

Acquisition, analysis, or interpretation of data: Coelen, Olthof, van Dieren, Besselink, Busch.

Drafting of the manuscript: Coelen, Olthof.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Coelen, Olthof, van Dieren.

Administrative, technical, or material support: Coelen, Olthof, Besselink.

Study supervision: van Dieren, Busch, van Gulik.

Conflict of Interest Disclosures: None reported.

References
1.
Nuzzo  G, Giuliante  F, Ardito  F,  et al; Italian Chapter of the International Hepato-Pancreato-Biliary Association.  Improvement in perioperative and long-term outcome after surgical treatment of hilar cholangiocarcinoma: results of an Italian multicenter analysis of 440 patients.  Arch Surg. 2012;147(1):26-34.PubMedGoogle ScholarCrossref
2.
van Gulik  TM, Kloek  JJ, Ruys  AT,  et al.  Multidisciplinary management of hilar cholangiocarcinoma (Klatskin tumor): extended resection is associated with improved survival.  Eur J Surg Oncol. 2011;37(1):65-71.PubMedGoogle ScholarCrossref
3.
Nagino  M, Ebata  T, Yokoyama  Y,  et al.  Evolution of surgical treatment for perihilar cholangiocarcinoma: a single-center 34-year review of 574 consecutive resections.  Ann Surg. 2013;258(1):129-140.PubMedGoogle ScholarCrossref
4.
Farges  O, Regimbeau  JM, Fuks  D,  et al.  Multicentre European study of preoperative biliary drainage for hilar cholangiocarcinoma.  Br J Surg. 2013;100(2):274-283.PubMedGoogle ScholarCrossref
5.
de Jong  MC, Marques  H, Clary  BM,  et al.  The impact of portal vein resection on outcomes for hilar cholangiocarcinoma: a multi-institutional analysis of 305 cases.  Cancer. 2012;118(19):4737-4747.PubMedGoogle ScholarCrossref
6.
Coelen  RJ, Wiggers  JK, Nio  CY,  et al.  Preoperative computed tomography assessment of skeletal muscle mass is valuable in predicting outcomes following hepatectomy for perihilar cholangiocarcinoma.  HPB (Oxford). 2015;17(6):520-528.PubMedGoogle ScholarCrossref
7.
Kennedy  TJ, Yopp  A, Qin  Y,  et al.  Role of preoperative biliary drainage of liver remnant prior to extended liver resection for hilar cholangiocarcinoma.  HPB (Oxford). 2009;11(5):445-451.PubMedGoogle ScholarCrossref
8.
Haga  Y, Ikei  S, Ogawa  M.  Estimation of Physiologic Ability and Surgical Stress (E-PASS) as a new prediction scoring system for postoperative morbidity and mortality following elective gastrointestinal surgery.  Surg Today. 1999;29(3):219-225.PubMedGoogle ScholarCrossref
9.
Haga  Y, Ikei  S, Wada  Y,  et al.  Evaluation of an Estimation of Physiologic Ability and Surgical Stress (E-PASS) scoring system to predict postoperative risk: a multicenter prospective study.  Surg Today. 2001;31(7):569-574.PubMedGoogle ScholarCrossref
10.
Haga  Y, Wada  Y, Takeuchi  H,  et al.  Estimation of Physiologic Ability and Surgical Stress (E-PASS) for a surgical audit in elective digestive surgery.  Surgery. 2004;135(6):586-594.PubMedGoogle ScholarCrossref
11.
Haga  Y, Ikejiri  K, Wada  Y,  et al.  A multicenter prospective study of surgical audit systems.  Ann Surg. 2011;253(1):194-201.PubMedGoogle ScholarCrossref
12.
Wang  H, Wang  H, Chen  T, Liang  X, Song  Y, Wang  J.  Evaluation of the POSSUM, P-POSSUM and E-PASS scores in the surgical treatment of hilar cholangiocarcinoma.  World J Surg Oncol. 2014;12:191.PubMedGoogle ScholarCrossref
13.
Haga  Y, Miyamoto  A, Wada  Y, Takami  Y, Takeuchi  H.  Value of E-PASS models for predicting postoperative morbidity and mortality in resection of perihilar cholangiocarcinoma and gallbladder carcinoma.  HPB (Oxford). 2016;18(3):271-278.PubMedGoogle ScholarCrossref
14.
Edge  SBBD, Compton  CC, Fritz  AG, Greene  FL, Trotti  A, eds; American Joint Committee on Cancer.  AJCC Cancer Staging Manual. 7th ed. New York, NY: Springer; 2010.
15.
de Graaf  W, van Lienden  KP, Dinant  S,  et al.  Assessment of future remnant liver function using hepatobiliary scintigraphy in patients undergoing major liver resection.  J Gastrointest Surg. 2010;14(2):369-378.PubMedGoogle ScholarCrossref
16.
Dindo  D, Demartines  N, Clavien  PA.  Classification of surgical complications: a new proposal with evaluation in a cohort of 6336 patients and results of a survey.  Ann Surg. 2004;240(2):205-213.PubMedGoogle ScholarCrossref
17.
Slankamenac  K, Graf  R, Barkun  J, Puhan  MA, Clavien  PA.  The Comprehensive Complication Index: a novel continuous scale to measure surgical morbidity.  Ann Surg. 2013;258(1):1-7.PubMedGoogle ScholarCrossref
18.
Rahbari  NN, Garden  OJ, Padbury  R,  et al.  Posthepatectomy liver failure: a definition and grading by the International Study Group of Liver Surgery (ISGLS).  Surgery. 2011;149(5):713-724.PubMedGoogle ScholarCrossref
19.
Koch  M, Garden  OJ, Padbury  R,  et al.  Bile leakage after hepatobiliary and pancreatic surgery: a definition and grading of severity by the International Study Group of Liver Surgery.  Surgery. 2011;149(5):680-688.PubMedGoogle ScholarCrossref
20.
Rahbari  NN, Garden  OJ, Padbury  R,  et al.  Post-hepatectomy haemorrhage: a definition and grading by the International Study Group of Liver Surgery (ISGLS).  HPB (Oxford). 2011;13(8):528-535.PubMedGoogle ScholarCrossref
21.
Vibert  E, Boleslawski  E, Cosse  C,  et al.  Arterial lactate concentration at the end of an elective hepatectomy is an early predictor of the postoperative course and a potential surrogate of intraoperative events.  Ann Surg. 2015;262(5):787-792.PubMedGoogle ScholarCrossref
22.
Janssen  KJ, Donders  AR, Harrell  FE  Jr,  et al.  Missing covariate data in medical research: to impute is better than to ignore.  J Clin Epidemiol. 2010;63(7):721-727.PubMedGoogle ScholarCrossref
23.
Rubin  DB.  Multiple Imputation for Nonresponse in Surveys. New York, NY: John Wiley & Sons; 1987.
24.
Sakata  J, Shirai  Y, Tsuchiya  Y, Wakai  T, Nomura  T, Hatakeyama  K.  Preoperative cholangitis independently increases in-hospital mortality after combined major hepatic and bile duct resection for hilar cholangiocarcinoma.  Langenbecks Arch Surg. 2009;394(6):1065-1072.PubMedGoogle ScholarCrossref
×