AJCC 7th edition of TNM staging accurately discriminates outcomes of patients with resectable intrahepatic cholangiocarcinoma

By the AFC-IHCC-2009 study group




This year, the 7th edition of the AJCC staging manual has for the first time attributed a unique pTNM staging to intrahepatic cholangiocarcinoma (IHCC) that is intended to replace the 2 Western and ideally also the 2 Eastern systems currently in use. This proposal, which has not yet been validated, was tested in the current study.


Among 522 patients operated on with curative intent for an IHCC between 1994 and 2008 in tertiary hepatobiliary centers, those with mass-forming-type IHCCs, an R0 resection, and accurate pathological node staging were retained for evaluation. The distribution of these patients and their actuarial survival in the new TNM stages (as well as in the 4 previous ones) were compared.


Only 163 patients fulfilled the inclusion criteria, mainly because of the lack of routine lymphadenectomy, but patients and tumors characteristics of this population were representative. These patients were evenly distributed between AJCC 7th edition stages (stage I, 28%; stage II, 32%; stage III, 35%), which was not the case for the other systems. With an average follow-up of 34 months in survivors, the AJCC 7th edition was more discriminating than the others in predicting survival (median for stage I not reached; for stage II, 53 months, P = .01; for stage III, 16 months, P < .0001). Survival of these patients according to the 2 Japanese classifications was identical to that anticipated.


The 7th edition is clinically relevant and may be applicable worldwide, provided routine lymphadenectomy at the time of surgery for IHCC becomes the standard of care. Cancer 2011. © 2010 American Cancer Society.

Intrahepatic cholangiocarcinoma (IHCC) is the second most frequent intrahepatic primary liver tumor. However, its diagnosis had been rather unreliable until the end of the 1980s, which explains the relative paucity of data. Current evidence is that its incidence is increasing,1, 2 that surgery is the only effective treatment,3-5 that resectability remains desperately low, at 10% to 20%,6, 7 and that the outcome after resection has hardly improved over the past decade,8, 9 at least not to the extent observed for hepatocellular carcinoma (HCC) or colorectal liver metastases.10, 11 The most recently reported 5-year survival rates range between 30% and 35%.4, 9, 12

The World Health Organization has given a discrete code to IHCC (ICD-0-3 C22.1) to differentiate it from HCC (C22.0).13 However, unlike in Japan, where 2 TNM staging systems have been specifically developed,14, 15 the same prognostic classification has historically been applied in the West for HCC and IHCC. The 6th edition of the AJCC/UICC classification, in particular, was based on the results of a retrospective analysis of 557 patients who had undergone curative partial liver resections of HCCs at 4 major hepatobiliary centers.16 Although largely validated worldwide for HCC,17 it has proven reliable when applied to patients with IHCC.9, 18

Therefore, the 7th edition of the AJCC staging manual, which takes effect this year, has for the first time proposed a novel staging specific to IHCC.19 It is largely based on an analysis of data extracted from the Surveillance, Epidemiology and End Results (SEER) program on 598 patients who had undergone cancer-directed surgery for IHCC.18 These patients were found to be inaccurately differentiated by the 6th edition of the AJCC as well as by both Japanese systems. In particular, a monotonicity of gradients was not observed.9, 18 Therefore, the data on these patients were used to construct a new model that is largely mirrored in this 7th edition.

Tumor staging is designed to allow the standardized exchange of information on disease extension20 that may subsequently be used to select the most appropriate therapeutic approach, identify indications for neoadjuvant or adjuvant chemotherapy, and allow comparison of cohorts. Considering that these goals are particularly relevant to IHCC, it requires that they be accurately validated, which has not been the case yet for this revised version. A concern is that the SEER database from which it is constructed contains a significant proportion of missing data on some parameters. Most notably, half the patients had not undergone lymphadenectomy, and the resection margin status was unknown.18

Single-center experience in the surgical management of IHCC is limited. The AFC-IHCC-2009 study group, which is composed of tertiary centers was therefore created in 2009 under the hospices of the French Association of Surgery (AFC) to produce a registry of operated-on IHCC patients. In the present study, we used some of these data to test the various staging systems. Care was taken to focus on patients who had undergone a curative R0 resection and for whom complete clinical and pathological data sets were available.


Patients Selection

Pathology databases of 12 tertiary hepatobiliary centers were comprehensively screened to identify patients who had undergone first resection of an IHCC between 1994 and 2008. Among them, 522 fulfilled the following criteria: 1) the biliary origin of the tumor was explicitly confirmed using strict pathological criteria, and in case of doubt, the diagnostic was retained only if an endoscopy had ruled out a colorectal cancer or if immunophenotyping (performed in 47%) had shown that tumor cells were CK7+ CK20-; 2) the tumor was a typical IHCC, and hepatocholangiocarcinomas, cystadenocarcinomas, and hilar cholangiocarcinomas were excluded; and 3) resection was a partial hepatectomy with or without an associated procedure and was macroscopically curative (R0 or R1).

These patients were further evaluated, but only those with mass-forming-type IHCCs, an accurate pT and pathological node (pN) status, a pathologically confirmed R0 resection, and comprehensive data sets including adequate follow-up were entered in the final analysis.

Recording of Data and Definitions

Following institutional approval, the complete medical files (including hospital charts, operation records, and pathology reports) of all patients were analyzed retrospectively. Variables of interest were recorded and marked as nonavailable when missing. In particular, these included those required to: 1) check inclusion criteria, 2) characterize the population, 3) stage patients in each TNM system (described below), and 4) measure survival from the time of hepatic resection. Follow-up was updated at the end of 2009. Features determining T classification, N status, and underlying liver disease were identified in the pathological records. Extension to adjacent organs excluded that to the gallbladder. The definition of an M1 status included peritoneal seeding discovered at the time of surgery as well as involvement of the celiac, periaortic, or caval lymph nodes.

Study Design and Statistical Analysis

Survival curves were constructed using the Kaplan-Meier product-limit method. A first set of analyses was performed in a N0M0 cohort to specifically analyze the T classification. The TNM stages were thereafter computed for the entire population. Survival curves within each staging system were compared with a log-rank test and with Cox regression analysis to determine the hazard ratio (HR). Statistical significance was defined as P < .05, and the Bonferroni method was applied when required.


Study Population

In total, 163 patients met the inclusion criteria. They represent 31% of the entire population, and the main reasons for patients' exclusion were the absence of routine lymphadenectomy and a non-R0 resection (Fig. 1). Characteristics of this cohort are summarized in Table 1, where they are compared with those used to create IHCC staging systems. In particular, the proportions of multiple and N+ tumors were identical to those in the Japanese series. The average follow-up for the 163 patients was 27 months (median, 18 months; range, 1-156 months), and 77 patients (47%) died during this period, all of their disease. The average follow-up for the patients who remained alive was 34 months (median, 26 months; range, 6-156 months).

Figure 1.

Patient selection is shown (pN, pathological node; PI, periductal infiltrating; IG, intraductal growth).

Table 1. Characteristics of Patients Operated on for an IHCC in the Present Study and in Those Used to Compute the Various Staging Systems
 Present StudyYamasaki14Okabayashi15Nathan18
  • Series used to construct the 5th and 6th editions of the AJCC/UICC classification were derived from patients with HCC and were therefore not included.

  • a

    Evaluated in 100% of the patients;

  • b

    data available in 54% of the patients;

  • c

    data available in 39% of the patients;

  • d

    among N0M0 patients;

  • e

    among M0 patients, and half the patients had no lymphadenectomy;

  • f

    including N2 extension;

  • g

    excluding N2 extension; na, not available; MF, mass-forming-type IHCC.

 SourceMulticenterMulticenterSingle centerSEER database
 Number of patients16313660598
 Study period1994-20081990-19961981-19991988-2004
Inclusion criteria    
 MF type IHCCYesYesYesNo
 R0 resectionYesNoNoNo
Patients characteristics    
 Age, y, mean63na63na
Tumor characteristics    
 Diameter, cm, mean7.6na5.1na
 Microvascular invasion38%ana48%a33%c
 Invasion of adjacent organ5%nana9%d
 Lymph node invasion36%ana37%17%e
 Median, mo36 monana21 mo
 5-year (actuarial)32%na29%18%

Distribution of Patients in the Various Classification Systems

Distribution of N0M0 patients between the various T classifications was unequal (Fig. 2A). T1 tumors, in particular, were present in only 4% of the patients in 2 classifications, compared with 45% in the 3 others, including the AJCC 7th edition. The same heterogeneity was observed for TNM stages (Fig. 2B). Stage I tumors were marginal in 2 systems (3% of the patients). Conversely, stage IV tumors were overrepresented (47%) in the Liver Cancer Study Group of Japan (LCSGJ) classification. Only 4 patients were stage IIIB in the AJCC 6th edition, suggesting that tumors extending to adjacent organs are either not resectable, not curable (R0 resection), or not N0. Overall, the AJCC 7th edition achieved the most uniform distribution of patients.

Figure 2.

Distribution of the study population in the 5 staging systems is shown. (A) T stages of 101 N0M0 patients are shown. (B) TNM stages of 163 patients are shown.

Survival According to T Classification

The median survival of N0M0 patients was 56 months (95% CI, 46-65 months), and their 1-, 3-, and 5-year actuarial survival rates were 87% (±3%), 70% (±5%), and 44% (±8%), respectively. These were overall correlated with T classification (Fig. 3), but the AJCC 7th edition was the only classification achieving a significant difference between all classifications (T1 vs 2, P = .01; T2 vs 3, P = .0001). Microvascular invasion had no impact on survival in patients with single tumors (T1 vs T2 of Okabayashi's classification). Patients with single tumors ≤2 cm without vascular extension (AJCC 5th edition and LCSGJ stage 1) had an excellent outcome after resection (median survival not reached), whereas all patients with extension to adjacent organs (T3/4 of AJCC 7th/6th edition) died within 18 months of surgery.

Figure 3.

Survival of N0M0 patients according to T stage in the 5 staging systems is shown (0, absent; +, present; −, either absent or present; VI, vascular invasion; UL, unilobar; BL, bilobar): (a) invasion of the main vascular branches; (b) invasion of adjacent organs (except gallbladder) or ruptured tumor; (c) serous membrane invasion is considered equivalent to vascular invasion. AJCC 5th edition16 and LCSGJ,14 difference was significant between stages 1-3 and stage 4; AJCC 6th edition16 and Okabayashi,15 all stages were significantly different except stages 1 and 2; AJCC 7th edition,18 the difference was significant between all stages.

Survival According to TNM Staging

The median survival of the entire cohort of 163 patients was 36 months (95% CI, 17-56 months), and their 1-, 3-, and 5-year actuarial survival rates were 79% (±3%), 50% (±5%), and 32% (±5%), respectively. These were correlated with TNM stage (Fig. 4), except in the AJCC 6th edition, where survival of stage IIIB (N0 tumors extending to adjacent organs) was worse than that of stage IIIC (N1 tumors irrespective of T classification). In Okabayashi's staging, there was some redundancy between stage IIIA and stage IIIB, suggesting that tumor multiplicity and lymph node invasion have comparable impacts on survival. A significant difference between all nonmetastatic stages was only observed in the AJCC 7th edition (stage I vs II, P = .01; stage II vs III, P < .0001). Median survival of patients with stage I was not reached, whereas that of patients with stage II was 53 months (95% CI, 48- 58 months; HR, 2.3; 95% CI, 1.1-5.1) and that of patients with stage III was 16 months (95% CI, 11-21 months; HR, 6.5; 95% CI, 3.2-13.4). Nine patients were M1, and none survived 24 months (median survival, 13 months; 95% CI, 9-16 months; HR, 12.5; 95% CI, 4.6-33.4).

Figure 4.

Survival according to TNM stage in the 5 staging systems is shown. AJCC 5th edition,16 the difference was significant between stages I and IV, between stages II and IIIB or more, and between stages IIIA and IIIB or more; LCSGJ,14 the difference was significant between stages I-III and stage IV; AJCC 6th edition,16 the difference was significant between stage I or II and stage IIIA or more, and between stage IIIA and IIIB or more; Okabayashi,15 the difference was significant between stage I or II and stage IIIA or more, and between stage IIIA and stage IV; AJCC 7th edition,18 the difference was significant between all stages.


This year in its 7th edition,19 the AJCC has implemented a unique staging for IHCC. It was designed following a retrospective analysis of the SEER database18 and was found superior to the 2 Western16 and 2 Japanese14, 15 pTNM classifications that had previously been used. The aim of the present study was to test this new system, which has not been validated yet, using a highly selected group of patients. Comparison with the 4 other systems was made in parallel for the distribution of patients among the various stages and for accuracy in predicting outcome.

Our study population mainly differs from that used to construct the 7th edition in 2 aspects. First, only patients with complete clinical and pathological datasets were retained for analysis. In particular those in whom N status was not clearly detailed in the pathological reports were excluded. One drawback of the SEER-based prognostic model is indeed the significant proportion of missing data.18 Most notably, pN status was available in only half the patients (as lymphadenectomy has historically not been performed routinely in the West), and those who had not undergone lymph node (LN) sampling were considered pN0 rather than pNx. This might have been a potential source of understaging, as preoperative imaging is particularly inaccurate in predicting lymph node extension.21 As a matter of fact, only 17% of M0 patients had LN invasion in the SEER-based study,18 which is less than the 35% to 45% incidence reported when lymphadenectomy is performed routinely.12, 21, 22 The second difference is that we only evaluated patients with a documented R0 resection. Although this has not been deemed mandatory in the past to define a TNM classification, the presence of microscopic (R1) or macroscopic (R2) residual tumor is now considered a necessary adjunct to TNM stage grouping.20 This is particularly relevant for IHCC because of the critical influence of resection status on outcome.4, 5, 9, 12 Resection status was not available in the SEER-based study,18 but the proportion of R1 or R2 resections was as high as 29% in another Western validation study.9

We first analyzed how our patients were distributed among the various TNM stages of the 5 TNM classifications. All are indeed designed to predict outcome following surgery, and they should therefore include a clinically relevant proportion of early-stage (not too low) and very advanced-stage (not too high) tumors. This was not the case for 2 of them. In the LCSGJ classification and in the 5th edition of the AJCC, the proportion of patients with stage I tumors (single tumors, less than 2 cm in diameter, and without vascular or LN invasion) was 3%. This is the same percentage as that reported in the description of the LCSGJ classification,14 and others have similarly reported figures ranging between 0% and 2%.9, 15 The reason is that unlike with HCC, it is exceptional to discover IHCC at an early stage because of the lack of a screening policy. In contrast, the proportion of stage IV patients in the LCSGJ classification was 47%. This figure, which is also the same as that previously observed,14 is particularly high, as this stage IV clusters advanced T stages, LN extension, and distant metastases. The 3 other classifications (Okabayashi and AJCC 6th and 7th editions) are mainly characterized by a progressive loss of influence of tumor size (Fig. 3) and a stricter definition of stage IV as the metastatic stage (Fig. 4). Consequently all 3 ensured a more coherent distribution of the patients but this was optimal in the new 7th edition of the AJCC, with an almost equal proportion of TNM stages I, II, and III patients (Fig. 2B). Of note, however, is that this revised version groups vascular invasion and multiple tumors into a common T2 stage (Fig. 3) and may benefit in the future20 from refined molecular or biological characterization.23

All 5 systems ensured coherent grouping with respect to survival with 3 exceptions that highlight the respective weight of pathologic variables. First, in Okabayashi's staging (as well as the 6th edition of AJCC), stages I and II were associated with comparable survival, suggesting that microvascular invasion has no influence in patients with single tumors. Second, in the 6th edition of AJCC, stage IIIB was associated with a poorer survival than was stage IIIC suggesting that the impact of tumor extension to adjacent organs is equivalent to that of metastases and is worse than that of LN extension. Third, the prognostics of stages IIIA and IIIB in Okabayashi's staging as well as of stages IIIB and IV in the 5th edition of AJCC were almost the same, suggesting that tumor multiplicity and LN extension have almost the same detrimental impact, as previously observed.15 Overall, the 7th edition of the AJCC was the only system in which all TNM stages were associated with significantly different outcomes.

According to this new edition, 28% of the patients had stage I tumors, and their 5-year survival was 62%. In contrast, 6% of the patients were stage IV, none of whom survived 2 years after surgery. Therefore, the rationale for operating on these latter patients is questionable. The same holds true for patients whose tumors extend to adjacent organs other than the gallbladder (see above). The remaining stages II and III patients, who comprised two-thirds of the population, had a median survival of 53 and 16 months, respectively, and their estimated 5-year survival was 27% and 14%, respectively. Patients who have stage III disease, characterized by LN extension, should be urgently entered into prospective controlled trials to assess the efficacy of adjuvant chemotherapy.24 Patients who have stage II disease, characterized by vascular invasion and/or multiplicity in the absence of LN extension, may also benefit in the future from this regimen. In the mean time, regular and prolonged follow-up should be required, as their survival curve appears to decline rapidly after the fourth postoperative year.

We acknowledge that the present study has some limitations. First, although the patients included were managed at tertiary university hospitals and had comprehensive pathological records, all pathological slides were not reviewed by a single pathologist. We therefore cannot ensure that the evaluation of microvascular invasion was uniform. However, although in previous studies, microvascular invasion has frequently been found to be a prognostic variable by univariate analysis, it is seldom an independent one by multivariate analysis.8, 14, 23, 25, 26 Second, serous invasion, which is a distinctive feature of the LCSGJ system,14 was rarely mentioned in the pathological reports. However, this feature has been shown to have no survival impact and that omitting it actually improved the accuracy of staging.27 Third, no attempt was made to further stratify N+ patients according to the number of invaded LNs28 or the LN ratio29 because of the lack of standardization of LN harvesting. Finally, our strict inclusion criteria (complete clinical and pathological data sets and proven R0 resection) resulted in 69% of our patients being excluded from analysis. Nevertheless, the present series is one of the largest reported so far. Furthermore, it is representative, as patient and tumor characteristics are comparable to those of previous studies,4, 8, 9, 15, 18, 23, 30 including a 36% incidence of N+ extension. In addition, the stage-stratified survival rates of our patients according to both Japanese classifications were identical to those reported in the description of these classifications.14, 15 These results do not support a difference between IHCC in the East and in the West and suggest worldwide applicability of our results.

In conclusion, we have shown that the 7th edition of the AJCC staging manual maximizes the clinical relevance and predictability of outcomes. Our stage-stratified survival figures appear to be even more discriminative than those provided in the original description of this TNM staging.18 The present study was performed in a Western population that behaved in exactly the same way as Japanese patients, suggesting worldwide applicability of this system. To ensure accuracy, HPB surgeons should get in the habit of sampling LNs routinely at the time of surgery, as LN extension of IHCC is much more prevalent than for other tumors that they are used to dealing with.21, 22


We are grateful to Sandrine Faivre, MD, PhD, for helpful discussion and to the members of the AFC-IHCC-2009 study group (in addition to the authors and by alphabetical order): Emmanuel Boleslawski (Hôpital Hurriez, Lille, France), Karim Boudjema (Hôpital Pontchaillou, Rennes, France), Denis Castaing (Hôpital Paul Brousse, Villejuif, France), Daniel Cherqui (Hôpital Henri Mondor, Créteil, France), Laurence Chiche (Hôpital Cote de Nacre, Caen, France), Christian Ducerf (Hôpital de la Croix Rousse, Lyon, France), Jean Hardwigsen (Hôpital de la Conception, Marseille, France), Jean Yves Mabrut (Hôpital de la Croix Rousse, Lyon, France), Gerard Pascal (Hôpital Paul Brousse, Villejuif, France), Patrick Pessaux (Hôpital Hautepierre, Strasbourg, France), and Michel Rivoire (Centre Léon Bérard, Lyon, France).


The authors made no disclosures.