AHPBA/AJCC consensus conference on staging of hepatocellular carcinoma: consensus statement

Background

Staging systems for cancer are developed with several objectives in mind ¹. First, staging subdivides patients with this naturally progressive disease into cohorts based on extent, and by inference severity of the disease, and predicts survival at each level of severity. This subdivision permits prognosis to be assigned to a cohort that matches particular staging criteria, which in turn allows reports from different centers and even different countries to be compared. Differences in stage of presentation and outcome in either treated or untreated patients from different centers can be assessed. Staging also allows investigators to decide whether particular cohorts are similar to or different from other cohorts. This objective gives each center a crude picture of how well or poorly it is doing with respect to a standard. Second, staging is used to select primary and adjuvant therapy, and to assess the outcome of therapy in a given cohort. Staging is used in the design of therapeutic trials to ensure uniformity of disease severity. It also assists in stratification of patients before randomization. Thus it is easier to be certain that differences in outcome are related to differences in therapy rather than differences in patient populations. Finally, staging is also a form of record-keeping for registry purposes.

Staging is not necessarily useful to establish prognosis in an individual patient, since each stage contains a spectrum of disease severity. An individual patient may or may not exactly match the criteria for a particular stage. Even if the patient can be classified into a particular stage, there may be other factors, not captured in the staging system that may affect survival.

The TNM system is the most widely accepted and used staging system. It is maintained by the American Joint Committee on Cancer (AJCC) and the Union Internationale Contre le Cancer (UICC). TNM was initially devised in 1954, and has been modified many times since then. There is an ongoing process to re-assess and revise the TNM. Cancer-specific site groups appointed by the AJCC review the literature and analyse data from large cohorts of patients. These committees decide whether the current staging systems are adequate, or whether they require updating on the basis of the literature and data that they have reviewed. Recommendations are submitted to AJCC, which in turn negotiates with the UICC in order to attain a global TNM system. The final result is published in the AJCC Manual of Cancer Staging.

The TNM system, as initially devised, used anatomical extent of disease as the sole factor to determine staging. More recently, attempts have been made to incorporate non-anatomical factors. For example, in breast cancer, estrogen and progesterone receptors are important prognostic factors, as is Her-2/Neu analysis. These factors may be more important than simple anatomical extent of disease.

Regarding hepatocellular carcinoma (HCC), many authors have noted that TNM does not accurately predict outcome, in part, because it does not consider liver function status and the presence or absence of cirrhosis and portal hypertension ²,³ or outcome following newer therapies such as orthotopic liver transplantation and ablation. As a result there have been numerous attempts to devise alternative staging systems for HCC, none of which have yet been universally accepted.

Development of a staging system

Classically, staging systems for cancer and other prognostic scores are developed by a process that identifies potential prognostic factors. This often begins with a literature search aimed at finding factors that other investigators have reported to be predictive of outcome. For HCC a large number of such variables have been identified. Tumor characteristics such as size, number, and distribution are obvious candidate risk factors in a prognostic score. Severity of liver disease is another. However, some workers have identified additional factors, such as AFP (alfafetoprotein) level, patient age, gender, race, and underlying liver disease (hepatitis B or hepatitis C). Various markers of liver function have been evaluated, including INR (international normalized ratio or prothrombin time), bilirubin, portal hypertension, and ascites. Also new tumor markers have been suggested as possible prognostic factors. These include AFP-L3 fraction, or PIVKA II.

Once a group of potentially useful prognostic factors has been identified the strength of association with survival in a sample population with the disease is assessed initially by univariate analysis, usually employing some form of regression analysis. Those factors most strongly associated with survival are then entered into a multivariate regression analysis, often using the Cox proportional hazard model. Some investigators include in the multivariate analysis only those factors statistically significantly associated with death, whereas others also include factors whose association with death does not quite reach statistical significance. Multivariate analyses can provide information regarding the strength of association between individual variables and survival and this information can be used to weight the variable in the staging system. The variables are then combined in a logical way to create the gradations or stages of the scoring system.

The next step is to construct Kaplan-Meier survival curves for each stage, and determine whether survivals in each stage are statistically different from each other, usually using the log-rank test. The staging system now has to undergo extensive validation. Most often the system is developed using retrospective data. Validation must include prospectively gathered data subject to the same analysis, by the same group and by others using different populations.

The development of staging systems for HCC was initiated in 1985 when Okuda published the first system to include variables of both tumor size and liver function. Renewed interest in the mid-1990s, largely stimulated by the increased incidence of HCC, has led to the development of several different staging systems from Japan ⁴, Italy ⁵,⁶,⁷, Spain ⁸,⁹,¹⁰, and China ¹¹ that have followed the classic pattern of development for clinical staging. At the same time the AJCC TNM staging for HCC has been updated and modified, and the scientific committee of the International Hepato-Pancreato-Biliary Association (IHPBA) has recommended an alternative pathologic staging system based on the Japanese system. However, this recommendation has not yet been accepted as an official staging system of the IHPBA by the Council of this organization.

In an attempt to standardize staging of HCC, the American Hepatico-Pancreatico-Biliary Association (AHPBA) organized a consensus conference that was co-sponsored by the AJCC at the Beth Israel Deaconess Medical Center in Boston in November 2002. The data were considered by a consensus development panel, the conclusions of which are presented below.

Methods

The goal was to have the Consensus Panel receive data on the current staging systems in use for HCC, consider these, and reach consensus regarding the current best staging system, and also define the limitations of the systems. The staging systems were presented by their proponents, with ample time for questions and discussion after each presentation. Priority for such questions was given to the panelists, but discussion included the 75 participants at the meeting.

The panel was provided with pertinent published material for review prior to the meeting. It had met by conference call to discuss a priori what evidence should be sought to reach a consensus as to the best staging system. It was not the goal of the panel to develop a new system, but rather to define the strengths and weaknesses of current systems.

The levels of evidence that the panel sought from the presenters and published data were analogous to those used by many specialty societies for the development of practice guidelines. The latter, whether they are intended to offer approaches to diagnosis, prognosis or treatment, are developed for use by physicians and have been subjected to rigorous scrutiny in terms of both their impact on outcomes and the quality of the evidence they employ. In the staging of HCC, the strength of a recommendation may be based on several criteria, each of which may convey its own impact on health care delivery (Table 1) ¹². These criteria were adopted by the American Association for the Study of Liver Diseases (AASLD) and the American Gastroenterological Association (AGA) for use in development of their practice guidelines. They are not mutually exclusive and may be additive. They are applicable to the development of a staging system for HCC.

Table 1. Categories reflecting the relevance of the evidence used in development of a staging system for HCC.

A	Survival benefit (by better prognostication and guidance to treatment)
B	Improved diagnosis (by non-invasive assessment prior to therapeutic intervention)
C	Improvement in quality of life
D	Relevant pathophysiological parameters improved (by evaluating impact of HCC on underlying liver disease and vice versa)
E	Impact on cost of health care

Modified from Gross et al.¹².

Second, in making a recommendation for adopting one system in preference over another, the quality of the evidence is paramount. These levels of evidence were adopted by the American College of Physicians and published in their Manual for Assessing Health Practices and Designing Practice Guidelines. These were also adopted by by the AASLD and AGA and published as a policy statement. They too are applicable with modification to the development of a staging system for HCC.

Grade 1 evidence (the highest level) would be derived from prospectively designed controlled studies, in which each factor used in the staging system is assessed initially by univariate analysis, and then all factors are subjected to multivariate analysis to derive the relative values in forecasting outcome (prognosis without therapeutic intervention or with randomized therapeutic interventions). The assessment of staging criteria derived by multivariate analysis would then need to be validated prospectively in a separate distinct group(s) of HCC patients. Furthermore, it would be necessary to exclude therapeutic intervention strategies as a variable in the analysis. For example, since the BCLC system makes a priori categorical therapeutic decisions in its algorithm (i.e. staging and treatment are locked together), it may be impossible to exclude therapeutic bias from the evaluation of this system, since therapy may influence prognosis (potentially, both negatively and positively).

Grade 2 evidence would be derived from retrospective analysis of large groups of HCC patients, or by meta-analysis of several such studies, and would not have been prospectively validated in large independent studies.

Grade 3 evidence may be based on clinical experience of a large referral center, or even expert committees offering such combined experience, but would not have been subjected to robust validation as required in grade 1 evidence.

Grade 4 is essentially empirical, albeit clinically ‘logical’ evidence that may have been adopted by individual centers and not subjected to any statistical analysis or prospective validation

The panel reconvened 1 month after the meeting for a full-day session to deliberate and come to consensus on the charge it had received:

• Define the strengths and weaknesses of the current HCC staging systems.

• Select the current most useful staging system.

• Indicate areas that need further definition and study.

Results

General comments

The current staging systems fall into two broad categories:

• Clinical staging systems: These characterize patients at initial presentation. The database required to complete the staging assessment includes serological and radiological features, such as radiological tumor extent or AFP concentration. The database does not include either the type of treatment to be offered, or any pathologic features that might influence outcome.

• Pathologic staging systems: These are applicable once tissue has been obtained. Tissue obtained by needle biopsy is usually insufficient for adequate staging, so that pathologic staging can only be applied to patients who have undergone liver resection or liver transplantation.

At the present time there is no single staging system that embraces all the needs of all the physicians caring for patients with HCC including accurate categorization of all patients, determination of appropriate treatment options, and establishment of prognosis. There are biases in the currently reported literature on staging, which depend on the interests and needs of the proponents of a system (hepatologist vs surgeon), and the population of patients seen at a given center (including underlying liver disease, whether detected by symptoms or by surveillance).

Surgeons and therefore systems that arise from surgical interest groups are focused on resectable lesions. Surgically treatable lesions form a small percentage of all HCCs, and since the surgical systems are largely based on this population they effectively categorize only this population. Hepatologists contact and manage a much broader spectrum of HCC. Consequently, staging systems based on this broader intake may be expected to categorize the overall population of HCCs more effectively, without however, providing the precision of categorization needed for surgical treatment.

Also, staging of HCC differs from staging of many other cancers because underlying liver disease may have a much greater impact on prognosis than the biology of the tumor. Another important point is that the natural history of HCC may vary according to its cause. Therefore staging systems developed from a population of hepatitis B-positive patients may not be reflective of cancer progression or prognosis in a population of HCC patients who are hepatitis C-positive. Thus in staging for HCC, appropriate consideration has to be given to all of these facts.

Clinical staging systems

Okuda. The first staging system, developed in the 1980s (Table 2) ⁴.

Table 2. The Okuda staging system for HCC ⁴.

Parameter	Value	Points
Tumor size	>50%	1
	<50%	0
Ascites	present	1
	absent	0
Serum albumin	>3	0
	<3	1
Serum bilirubin	<3	0
	>3	1
Stage 1–0 points
2–1–2 points
3–3–4 points

The strengths of the Okuda system are that it was in use for a decade, something no other system can claim. The biology of the tumor was recognized to the extent that tumor size was a staging criterion, and the contribution of liver disease was recognized by including the ascites, albumin, and bilirubin as criteria. Its weakness was that its ability to stratify patients by duration of survival was not good, particularly for patients with better prognosis, and that it has never been validated in a prospective manner.

CLIP (Cancer of the Liver Italian Program) (Table 3) ⁵,⁶,⁷. This staging system used classic techniques of analysis of variables. It only included patients with cirrhosis, and uses Child-Pugh score rather than its individual components.

Table 3. The CUP staging system for HCC.

Variables	0	1	2 points
i. Morphology	Single	Multiple	Massive
	<50%	<50%	or >50%
ii. Child's score	A	B	C
iii. AFPng/ml	<400	≥400
iv. Portal V thrombosis	No	Yes

Score = sum of points for 4 variables

CLIP, Cancer of the Liver Italian Program.

The strengths of this system are that it recognizes the contribution of tumor biology by including tumor morphology and size, AFP concentration and the presence or absence of portal vein thrombosis. The contribution of liver disease is assessed by the Child-Pugh score. The system has been prospectively validated by the investigators as well as by other groups. However, not all investigators have found the CLIP useful. The source of this variability is not clear, but may be related to differences in patient populations. For example, investigators at the Chinese University in Hong Kong found that CLIP did not adequately stratify early stage disease. However, their population included only a small proportion of patients with early stage disease, and it may be that if the population had better resembled the CLIP population, CLIP may have performed better.

However, this highlights a persistent problem in assessing the different reported staging systems, and that is the applicability to different populations as noted above. Can, for example, a staging system developed in cirrhotic hepatitis C patients, where many HCCs were identified by screening, be applied to a population of largely hepatitis B carriers in whom surveillance was uncommon?

BCLC (Barcelona Clinic Liver Cancer) (Table 4) ⁸,⁹,¹⁰. This group has a priori divided patients into those with early, intermediate, or late stage disease and constructed prognostic models for each stage. Early stage disease is suitable for radical therapies, such as resection. Intermediate stage disease may be suitable for resection or transplantation. Late stage disease is suitable only for experimental therapy and palliative or supportive therapy.

Table 4. BCLC staging system for HCC ⁸,⁹,¹⁰.

Stage	Performance status	Tumor stage	Liver function
Stage A (early HCC)
Al	0	Single <5 cm	No portal hypertension and normal bilirubin
A2	0	Single <5 cm	Portal hypertension and normal bilirubin
A3	0	Single <5 cm	Portal hypertension and elevated bilirubin
A4	0	Up to 3, <3 cm	Child-Pugh class A-B
Stage B (intermediate HCC)	0	Large multinodular	Child-Pugh class A-B
Stage C* (advanced HCC)	1–2	Vascular invasion or extrahepatic spread	Child-Pugh class A-B
Stage D* (end-stage HCC)	3–4	Any of the above	Child-Pugh class C

BCLC, Barcelona Clinic Liver Cancer.

*At least one of the conditions should be met.

The advantage of this system is that it is innovative and brings in new concepts. It recognizes the importance of tumor biology, by including the number and size of tumor nodules and the presence of vascular invasion. It also recognizes the importance of liver disease by including the Child-Pugh score and the presence of significant portal hypertension. The importance of the overall performance status is also recognized. It also provides a guide to therapy.

The weakness of this system is that it has not been prospectively validated. It requires an assessment of portal hypertension, which may not be precise, and it is somewhat complex to administer.

CUPI (Chinese University Prognostic Index) (Table 5) ¹¹. The Chinese University of Hong Kong have independently devised a staging system using classical methods ⁵. However, they used the standard TNM staging to define tumor extent, rather than looking at the individual parameters that make up the TNM staging. The staging also includes the presence of symptoms at presentation, as well as other widely used variables.

Table 5. CUPI staging system for HCC ¹¹.

		Variables	Points
i)	TNM	I & II	–3
		III	–1
		IV	0
ii)	Asymptomatic disease		–4
iii)	Ascites present		3
iv)	AFP ≥500 ng/ml		2
v)	Bilirubin	<2	0
		2–3	3
		>3	4
vi)	ALP ≥200 IU/L		3
Score = sum of points from 6 variables
Low risk	≤l
Intermediate risk	2–7
High risk	≥8
Mortality risks:
Low	3 month <30%
Intermediate	3 month 30–70%
High	3 month >70%

CUPI, Chinese University Prognostic Index.

The strength of this approach is that it is easily applicable. Like previous systems, it incorporates measures of tumor biology and liver function. The disadvantages are that it has not been prospectively validated and it requires consideration of as many as six variables. In addition, the CUPI has been developed in a different population than the other staging systems, namely a population of mainly hepatitis B carriers.

Pathologic staging systems

The pathologic staging systems that were considered included the latest version of the AJCC staging system and a Japanese staging system recommended by the scientific committee of the IHPBA and supported by the UICC.

AJCC (American Joint Committee on Cancer) (Table 6) ¹³,¹⁴. The latest edition of the AJCC Manual of Cancer Staging includes a new simplified TNM staging system for HCC, called the sT system. This system was developed using a large database of patients from two centers in the USA, and one each in Japan and France ¹⁰. The system was again developed using classical methods. However, the cohort studied included only patients who had a liver resection. This limits the application of the AJCC staging system to patients who have had either resection or transplantation. The analysis identified only five predictive factors: These were micro- or macro-vascular invasion (large branch of the portal vein), severe fibrosis or cirrhosis (Ishak score 5 or 6) ¹¹ and to a lesser extent, tumor size and tumor number. Nodal involvement and metastases were not independent predictors of outcome.

Table 6. AJCC staging system for HCC ¹³,¹⁴.

T classification:	Morphology		T classification
– Single tumor without vascular invasion			T1
– Single tumor with vascular invasion or multiple tumors, none >5 cm			T2
– Multiple tumors, any >5 cm or tumor involving major branch of portal or hepatic veins			T3
Staging:
Stage I	T1	N0	M0
Stage II	T2	N0	M0
Stage III A	T3	N0	M0
Stage III B	AnyT	N1	M0
Stage IV	Any T	Any N	M1

Also, fibrosis (severe or cirrhosis) downgrades any stage

AJCC, American Joint Committee of Cancer.

The AJCC system has the advantage that it has been validated. It was developed using a large cohort of patients, and conforms to the TNM system, the most widely used cancer staging system.

Note: When analysed by Kaplan-Meier survival curves, tumor size was only a predictor of outcome if vascular invasion was present. This is an unexpected finding, since virtually every other analysis as well as clinical experience indicates that larger tumors have a worse prognosis. The explanation may lie in the selection of the patient group, i.e. those with resectable lesions, in that it is likely that the only large tumors that were included were those that were slow growing and had a good prognosis. Rapidly growing tumors are most unlikely to be resectable.

UICC (United International Consensus Committee. A group established by the IHPBA, including authors from France, the USA and Hong Kong, analysed a very large cohort of patients from the Liver Cancer Study Group of Japan (LCSGJ) (M Makuuchi, personal communication). This analysis has not been published but was presented at the conference. The group seems to have taken the AJCC analysis and applied it to their cohort, using only multiple Kaplan-Meier survival curves to compare outcomes in different groups. No univariate or mulivariate analysis of predictive factors was done. The results were not identical to the AJCC analysis, in that a different tumor cut-off size was determined (<2 cm vs <5cm). They also found that tumor size was a predictive factor in patients without vascular invasion, whereas in patients with vascular invasion tumor size was only predictive in tumors >5 cm in diameter (Table 7).

Table 7. UICC staging system for HCC.

T classification:
Morphology-3 variables:
i) Single tumor
ii) Size <2 cm
iii) No vascular invasion of portal or hepatic vein
Staging:
Stage I	Tl	N0	M0
Stage II	T2	N0	M0
Stage III	T3	N0	M0
Stage IV A	T4	N0	M0 or T any N1 M0
Stage IV B	T/N	any Ml

UICC, United International Consensus Committee.

The UICC system has as its major strength the fact that it was developed using a very large number of patients. In addition, it also conforms to the TNM system. Unfortunately, this system has not yet been published, nor has it been validated, either internally or externally. It does not include measures of liver function, but these may be irrelevant since the system is only applicable to patients who have survived resection, and who therefore probably have well-preserved liver function. There is some doubt as to whether the variables used in this system are truly independent. Tumor size, number of tumors, and vascular invasion seem intuitively to be linked to disease progression. As the tumor grows, vascular invasion occurs and satellites develop.

Recommendations

The Consensus Panel makes the following recommendations based on the above currently available evidence.

1. The primary staging should be clinical staging, which can be applied to all patients. The CLIP system should be the clinical staging system of choice, because it is generally applicable to most patients, it includes easily collected variables. Most importantly, it has been externally and prospectively validated. As a caveat, the CLIP system may not be applicable to patients with chronic hepatitis B.

2. A secondary staging system for patients undergoing resection or liver transplantation is needed. The AJCC version of the modified TNM system should be used because it has been internally validated, although external and prospective validation is still lacking. Furthermore, it conforms to the TNM standard.

3. Since neither of these systems is free of limitations, other factors which might be included in accessing prognosis include treatment-directed variables (according to BCLC), the etiology of the underlying liver disease, and newly discovered factors affecting tumor biology.

4. All studies on HCC where it is appropriate to use staging should use one or both of these staging systems (CLIP or AJCC) to define the patient population. Medical journals considering such manuscripts for publication should insist on cohorts being classified according to these staging systems (excluding, of course, those studies looking at improving staging systems).

5. Further studies on the validation of staging systems and harmonization of the different systems are urgently required.

Discussion

Primary staging – clinical

The clinical staging systems for HCC are clearly in evolution at present and the ‘ideal’ system does not exist. In an effort to bring some order to this field, the panel felt it is important that centers publishing data on HCC should adhere to a published and validated clinical staging system when presenting their data.

The CLIP system has the advantage of using readily available clinical, radiologic, and laboratory data to stage patients. Child's classification, tumor size, major vascular thrombosis, and AFP can be readily obtained. This system has also been prospectively validated. However, it does have shortcomings, such as its apparent inapplicability to some populations of hepatitis B patients.

The BCLC system introduces new and innovative concepts, introducing therapy options, and overall performance status, and the presence/absence of portal hypertension into staging. While the panel agrees that these concepts are important, they increase the complexity of immediate clinical staging. This system has not been prospectively validated as yet.

The other clinical staging systems (Okuda and CUPI) include important components, but were not considered the currently best available systems by the panel.

Secondary staging – pathologic

Secondary staging requires a surgical resection specimen, either partial or complete hepatectomy. The two staging systems available are the 6th edition of the AJCC Cancer Staging Manual¹, and the proposed United International Consensus Committee System, proposed by the IHPBA working group. Both systems give prognosis in resected patients and are useful for both registry purposes and for the study of post-resection and post-transplant adjuvant therapy.

A recommendation to use the AJCC sT system is based on its current acceptance, required use by many pathologists, and internal validation. This system has the advantage that it has been published ¹⁴ and thus has been peer reviewed and thought to be appropriate. Univariate and multivariate analyses were performed with univariate prognostic factors entered into a multivariate model to identify independent predictors of long-term survival. Major vascular invasion had the greatest impact on survival followed by microvascular invasion and severe fibrosis/cirrhosis of the adjacent liver. Multiple tumors and tumors measuring >5 cm were also associated with worse survival.

The UICC staging system uses three morphologic variables: numbers of lesions, size <2 cm, and no vascular invasion to portal vein, hepatic vein, and bile duct. This system has been studied in a large number of patients, and a univariate analysis has been carried out on these parameters, but there is no evidence of a multivariate analysis being performed. This system has not been validated, peer reviewed, or published as yet.