Selecting an appropriate staging system is crucial to predict the outcome of patients with hepatocellular carcinoma (HCC). The optimal prognostic model for HCC is under intense debate. This study investigated the prognostic ability of the 5 currently used staging systems, Barcelona Clinic Liver Cancer (BCLC), Cancer of the Liver Italian Program (CLIP), Japan Integrated Scoring (JIS) system, tumor-node-metastasis (TNM), and Tokyo score, for HCC.
Between 2002 and 2008, 1713 prospectively enrolled HCC patients were compared for their long-term survival by using the Akaike information criterion (AIC) according to the staging or scoring methods of these 5 models.
The mean and median follow-up duration was 18 and 14 months, respectively. Among all patients, the CLIP staging system had the lowest AIC value in comparison with other systems in the Cox proportional hazards model, followed by the Tokyo score, JIS score, BCLC staging system, and TNM staging system. Patients undergoing curative treatment had a significantly better survival in comparison with patients undergoing noncurative treatment (P < .001). When the predictive accuracy of the staging systems was analyzed according to treatment strategy, the CLIP staging system had the lowest AIC value and remained the best prognostic model in patients undergoing curative (801 patients) and noncurative (912 patients) treatment.
Hepatocellular carcinoma (HCC) is the fifth most common cancer worldwide with rising annual incidence.1, 2 The major prognostic factors for HCC encompass liver function (serum bilirubin and albumin levels, and prothrombin time), tumor burden (the number and size of nodules, presence of vascular invasion, and evidence of extrahepatic metastasis), and general health status. To date, at least 8 staging systems have been proposed from independent study groups to predict the outcome of patients with HCC.3, 4 Complex interactions among these variables may make the selection of a balanced staging system for HCC quite difficult.5
Among the currently used systems, the Barcelona Clinic Liver Cancer (BCLC) system was shown to have excellent discrimination of survival in patients with HCC.6, 7 Because the BCLC staging system includes performance status and the rationale for treatment allocation, it was recommended as the primary reference system for HCC.8 Conversely, the Cancer of the Liver Italian Program (CLIP) was shown to have a better discriminatory ability and has been considered a more appropriate staging system for HCC.9-11 The Japan Integrated Scoring (JIS) system, however, was considered an equal or superior model in comparison to the CLIP staging system.12, 13 Alternatively, the Tokyo score, which was recently proposed from another independent group in Japan, was hailed as a better model than other staging systems.3, 14 The traditional tumor-node-metastasis (TNM) system has been a prevailing staging model for HCC patients undergoing surgical resection.15, 16 Because patient selection and preferred treatment modality in different published studies highly varied, choosing an optimal staging system for HCC has been an intense debate over the past decade. Moreover, the feasibility of the newly proposed Tokyo scoring system has not yet been validated. In this study, we have prospectively followed up a large cohort of patients from early to advanced cancer stage undergoing different treatment strategies, and investigated the long-term prognostic ability of the currently used 5 staging systems (TNM, CLIP, BCLC, JIS, and Tokyo score) for HCC.
MATERIALS AND METHODS
Our hospital, Taipei Veterans General Hospital, provides primary to tertiary medical care to the residents of northern Taiwan, an area of about 12 million inhabitants. Between 2002 and 2008, 1713 patients with newly diagnosed HCC were prospectively enrolled and retrospectively analyzed to investigate the prognostic ability of the 5 cancer staging systems. The baseline information, including patient demographics, etiology of chronic liver disease, serum biochemistries, severity of cirrhosis, performance status, cancer stage, and treatment, was collected at the time when the diagnosis was established. This study complies with the standards of Declaration of Helsinki and current ethical guidelines.
Diagnosis and Definitions
The diagnosis of HCC was histologically confirmed or based on the findings of typical radiological features in at least 2 imaging examinations including ultrasound, contrast-enhanced dynamic computed tomography (CT), magnetic resonance imaging (MRI) and hepatic arterial angiography, or by a single positive imaging technique associated with serum α-fetoprotein (AFP) level >400 ng/mL.8, 17 The underlying etiology of HCC was attributed to hepatitis B virus (HBV) infection if serological detection of hepatitis B surface antigen (HBsAg) was positive (RIA kits, Abbott Laboratories, North Chicago, Ill). Hepatitis C virus (HCV) infection was diagnosed if patients were seropositive for antibody against HCV (anti-HCV) by a second-generation enzyme immunoassay (Abbott Laboratories). Alcoholic hepatitis was diagnosed in subjects with consumption of at least 40 g of alcohol daily for 5 years or more.18 The diagnosis of cirrhosis was defined as typical ultrasound or computed tomography characteristics of cirrhosis plus at least 1 of the following: hypersplenism, ascites formation, esophageal or gastric varices, or hepatorenal syndrome.19 Ascites was recognized as free peritoneal fluid identified by ultrasound or computed tomography. The diagnosis of hepatic encephalopathy was according to the West Haven criteria after exclusion of intracranial lesions, metabolic, traumatic disorders, or drug intoxication.20 Vascular invasion was defined by the presence of adjacent thrombus to the tumor in portal vein with blurring boundary confirmed by at least 2 imaging modalities or by the direct malignant tissue obtainment from the thrombus.21 Performance status was evaluated using the Eastern Cooperative Oncology Group (ECOG) performance scale: 0 (asymptomatic) to 4 (confined to bed). HCC characteristics including number of nodules, maximal diameter of the largest nodule, extrahepatic spreading, and vascular invasion were detected by dynamic contrast-enhanced CT or MRI. Supplementary studies such as bone scan and brain CT were arranged if clinically indicated.
Clinical data needed to fulfill these 5 staging systems were collected for classification according to the sixth edition American Joint Committee on Cancer (AJCC)/International Union Against Cancer (UICC) TNM classifications,22 the CLIP score,11 the revised BCLC staging system,8 the JIS score,12 and the Tokyo score3 at the time of diagnosis (Table 1). The TNM classification does not contain the factor of severity of cirrhosis. Performance status is exclusively included in the BCLC staging system, whereas serum AFP level is exclusively included in the CLIP staging system.
Table 1. Variables Contained in 5 Staging Systems for Hepatocellular Carcinoma
TNM indicates tumor-node-metastasis; BCLC, Barcelona Clinic Liver Cancer; CTP, Child-Turcotte-Pugh; CLIP, Cancer of the Liver Italian Program; JIS, Japan Integrated Scoring.
No. of nodules, tumor size, portal vein thrombosis, metastasis
Classification I, II, III, IV (6th ed)
Tumor size and no. of nodules, portal vein thrombosis
Stage 0, A, B, C, D
<400 or ≥400 ng/mL
Tumor extent greater or less than 50% area of liver, portal vein thrombosis
Score 0 to 6
TNM defined by Liver Cancer Study Group of Japan
Score 0 to 5
No. of nodules, tumor size <2 or >5 cm and between 2 cm and 5 cm
Score 0 to 8
Criteria of surgical resection for patients with HCC were1 patients with tumor involving no more than 3 Healey's segments,2 Child-Turcotte-Pugh (CTP) class A with less than 25% retention of indocyanine green 15 minutes after injection, and3 no main portal vein trunk involvement or distant metastasis. Liver transplantation was considered in patients fulfilling the Milan criteria with coexisting CTP class B or C cirrhosis.23 For patients who had unresectable lesions, locoregional therapy included transarterial chemoembolization (TACE), percutaneous acetic or ethanol acid injection (PAI, PEI), or radiofrequency ablation (RFA) was performed depending on the size and number of tumor nodules as previously reported.24, 25 Systemic chemotherapy or targeted therapy was given to selected patients with preserved liver functions and distant tumor metastasis. For HCC patients with CTP class C, main portal vein thrombosis or evidence of distant metastasis, or when treatment efficacy was considered limited or associated a substantially high treatment-associated risk, only best supportive care was given. In this study, curative treatment was defined as patients undergoing aggressive treatment, including surgical resection, RFA, PAI, PEI, or liver transplantation. Other treatment modalities were mostly palliative and collectively defined as noncurative treatment.
The Kaplan-Meier method with log-rank test was applied to compare the survival distributions of different staging systems. The discriminatory ability of each staging system was examined by using the Cox proportional hazards model, and the consequences of the Cox model were expressed with the Akaike information criterion (AIC), which reveals how the staging systems affected the dependent variable (patient survival) and represents an overall assessment of a certain staging system.26, 27 The lower the AIC, the more explanatory and informative the model is.28 All statistical analyses were conducted with the SPSS for Windows version 14 release (SPSS, Inc., Chicago, Ill), SAS version 9.1 (SAS, Cary, NJ), and MedCalc for Windows version 4.2 (MedCalc Software, Mariakerke, Belgium). A p value less than 0.05 was considered statistically significant.
The baseline demographics of the 1713 patients were shown in Table 2. The clinical characteristics and short-term outcome in part of these patients had been reported in our previous study.29 The majority of them were men (77%) and the mean age was 64 years. The most common etiology of chronic liver disease was hepatitis B (50.2%), followed by hepatitis C (26.5%) and alcoholism (13.4%). Coinfection of HBV and HCV was found in 81 (4.7%) patients. The mean CTP score was 6.1 and 74.8% of patients were classified as CTP class A. In total, 61% of patients had a single tumor and 58% of patients had a tumor size of 5 cm or larger in the longest axis. Vascular invasion, ascites formation, and hepatic encephalopathy were present in 34.4%, 18.7%, and 2.2% of patients, respectively, at the time of diagnosis. Curative treatments, which included resection, RFA, PAI/PEI, and liver transplantation, were performed in 46.8% of patients; TACE and systemic chemotherapy/targeted therapy were administrated in 30.1% and 8.1% of patients, respectively. The remaining 15.2% of patients received best supportive care.
Table 2. Baseline Demographics of the Study Patients
HBV indicates hepatitis B virus; HCV, hepatitis C virus; INR, International Normalized Ratio; CTP, Child-Turcotte-Pugh; PEI, percutaneous ethanol injection; PAI, percutaneous acetic acid injection; RFA, radiofrequency ablation; TACE, transarterial chemoembolization; TNM, tumor, node, metastasis; CLIP, Cancer of the Liver Italian Program; BCLC, Barcelona Clinic Liver Cancer; JIS, Japan Integrated Scoring.
No. of patients
Age, y, mean±SD
Etiology of cirrhosis
Serum biochemistry, mean±SD
INR of prothrombin time
Hemoglobin, g/dL, mean±SD
Platelet, /mm3, mean±SD
Mean CTP score [range]
No. and size of tumor
≤5 cm/>5 cm
PEI or PAI
Chemotherapy or targeted therapy
BCLC staging system
7 or 8
Comparison of the Prognostic Ability for Long-Term Survival
All patients had been follow-up for at least 3 months. The mean and median follow-up duration was 18 and 14 months, respectively, at the time of analysis. The comparisons of long-term survival distribution according to the staging or scoring methods in the 5 systems were given in Figure 1 (TNM), Figure 2 (CLIP), Figure 3 (BCLC), Figure 4 (JIS), and Figure 5 (Tokyo). Patients with a Tokyo score of 7 or 8 were classified in the same group because of a relatively small case number (n = 45). Significant survival difference was found across all groups of these staging systems, with the exceptions for the comparison between BCLC classifications 0 versus A (P = .06), JIS score 4 versus 5 (P = .067), Tokyo score 0 versus 1 (P = .061), Tokyo score 5 versus 6 (P = .085), and Tokyo score 6 versus 7 or 8 (P = .286).
For all patients, the CLIP score showed the lowest AIC value, followed by the Tokyo score, JIS, BCLC, and, last, TNM (Table 3). Patients undergoing curative treatment had a significantly better survival in comparison with patients undergoing noncurative treatment; the 3-year and 5-year survival rates were 75% and 47% versus 29% and 14%, respectively (P < .001; Fig. 6). When the patients were stratified and compared according to treatment strategy, the CLIP score was identified as the most accurate system, followed by the Tokyo score, JIS, BCLC, and TNM according to the AIC values among 801 patients undergoing curative treatment. Of the 912 patients undergoing noncurative treatment, the CLIP score still showed the best predictive accuracy, followed by BCLC, JIS, Tokyo score, and TNM according to the AIC values (Table 3).
Table 3. Comparison of the Prognostic Ability of the 5 Staging Systems
Akaike Information Criterion
TNM, tumor-node-metastasis; CLIP, Cancer of the Liver Italian Program; BCLC, Barcelona Clinic Liver Cancer; JIS, Japan Integrated Scoring.
Patients with curative treatment, n=801
Patients with noncurative treatment, n=912
In this study, we have comprehensively compared the long-term prognostic ability among the TNM, CLIP, BCLC, JIS, and the newly proposed Tokyo systems. Previous studies reported that the BCLC staging system was the best model for cancer staging.4, 7, 8 In contrast, other studies found that the JIS scoring system was a better model in comparison with the CLIP and BCLC staging systems.12, 13, 30 Several factors may contribute to these discrepancies. The characteristics of tumor-related variables, the relative score weighted for each variable, preferred treatment modality in different centers, the number of patients analyzed, the pattern of patient referral and the etiology of cirrhosis could all or in part influence the final results. Importantly, the current study has enrolled a large cohort of patients from early to advanced cancer stage undergoing different treatment modalities. Our results indicate that the CLIP staging system is the best prognostic model for HCC in terms of long-term outcome prediction. These findings confirm our earlier reports which disclosed that the CLIP was a better prognostic model based on the sequential evaluation over time to investigate the predictive accuracy in different systems.29, 31 In addition, our results are consistent in the subgroup analysis among patients undergoing curative and noncurative treatments, suggesting its predictive accuracy is highly stable and is independent of the treatment strategy.
In this study, all 5 staging systems generally demonstrated good prognostic ability to predict the mortality for HCC patients. However, we found that the BCLC staging system and the JIS and Tokyo scoring systems may not be able to differentiate the outcome of patients at very early or far advanced cancer stages. The significance of the survival difference between patients with BCLC classifications 0 and A, and between patients with Tokyo scores 0 and 1 was considered marginal (P = .06 and 0.061, respectively). This finding is consistent with a recent study from Italy that showed the lack of ability of the BCLC to differentiate the prognosis in patients at very early stage.32 Among patients with advanced and far advanced stages, the significance of survival difference was either marginal or insignificant between patients with JIS scores 4 and 5, and between patients with Tokyo scores 5, 6, and ≥7. In contrast, the CLIP model showed excellent prognostic stratification for patients from very early to far advanced cancer stage (7 categories; Fig. 2). These findings were confirmed by the AIC analysis, which can provide the information of the predictive accuracy of a certain prognostic model. The AIC analysis, which is based on the Cox proportional hazards model, represents an overall assessment of the prognostic system and is the most important reference for the comparison across different staging systems. In this study, we found that the CLIP model had the lowest AIC value in comparison with other systems in the entire cohort, suggesting its prognostic power is homogeneously stable across different cancer stages.
The performance of different staging systems has been further stratified and compared according to the treatment strategy. As expected, we found that patients undergoing curative treatment had a more favorable outcome in comparison with the patients undergoing noncurative treatment. Numerous studies have shown that the treatment modality is directly linked with the survival of patients with HCC.8, 17, 23, 33 Therefore, to reflect the true clinical scenario, the strategic approach to select an appropriate staging system for patients with favorable and unfavorable outcomes might be different. Interestingly, the CLIP staging system was further identified as an ideal model to stage patients undergoing either curative or noncurative treatment. This result has several important implications. Although the CLIP staging system was originally derived from a large unselected patient population and the majority of them had been treated conservatively, in our study, it demonstrated prevailing prognostic ability in a fairly large cohort of patients with early to intermediate stage that underwent potentially curative treatment. This finding is consistent with another independent study group from Taiwan, which showed that the CLIP staging system provided the best prognostic stratification for patients with HCC.34 Alternatively, among 912 patients who were predominantly at intermediate or late stages and underwent noncurative treatment, the CLIP staging system was also identified as the best predictive model. This result was consistently supported by a recent study from Korea, which reported that the CLIP score was the best model among the 7 systems for outcome prediction in HCC patients undergoing palliative TACE.35 Altogether, these lines of evidence indicate that the CLIP staging system is a reliable model that can best differentiate long-term survival and can fit in different patient populations characterized by favorable and unfavorable outcomes.
The Tokyo score was a recently proposed model by the Japanese investigators. Patients enrolled in that study were identified from the screening surveillance and most of them had undergone radical treatment.3 It was shown that the predictive ability of the Tokyo score was equal to the CLIP staging system and better than the BCLC staging system.3 In the current study, we found that the Tokyo score was the second best by comparing the AIC values between different systems in the entire cohort. Among patients undergoing curative treatment, the Tokyo score was still the second best, but its performance in the noncurative group was relatively poor. The feasibility and prognostic accuracy of the Tokyo score for cancer staging require more studies to establish.
In comparison with other systems, the TNM classification had an inferior prognostic ability for long-term survival prediction. This is likely because the factor of severity of cirrhosis is not included in this system. The TNM has been exclusively used to stage patients undergoing surgical resection,16, 36, 37 because these patients usually have a well compensated liver disease and the impact of cirrhosis factor on survival could be minimized. Another shortcoming of the TNM staging system is that it comprises only 4 staging categories (stage I to IV), which may not be enough to categorize and differentiate the outcome of patients with early to advanced stage of cirrhosis and tumor burden. Moreover, only 26% of patients underwent surgical resection in this study and this may also, to some extent, limit the performance of the TNM staging system.
There are a few potential limitations of this study. First, in this single center study, 55% of patients had HBV infection. This feature is distinctly different from the Western countries where HCV infection is the prevailing etiology of chronic liver disease. Second, the majority of our patients belonged to intermediate to advanced cancer stage at the time of diagnosis. Therefore, the generalizability of this study may be limited and our results may not be readily applicable to surgically oriented centers or centers with a high volume of liver transplantation. Third, the initial and subsequent anticancer treatment could be different in the same patient depending on the status of tumor spread and severity of cirrhosis during the follow-up period. This may interfere with the assessment of the impact of any individual anticancer treatment on overall survival.
In conclusion, our results indicate that the CLIP score is the best prognostic model among the 5 currently used staging systems. The performance of the CLIP score is reliable for long-term prognostic prediction and is independent of the treatment strategy. This finding is crucial because selecting an optimal staging system is helpful in improving the design of future clinical trials. The Tokyo score is a potentially promising scoring system, and further studies are needed to confirm the feasibility and prognostic ability of the Tokyo score in HCC patients with different clinical scenarios.
We thank Ms. Ya-Ju Tsai for her help on data collection and manuscript preparation.
CONFLICT OF INTEREST DISCLOSURES
This study was supported by a grant (RD-2010-013) from the National Yang-Ming University Hospital, Yilan, Taiwan, and a grant (V99C1-169) from Taipei Veterans General Hospital, Taipei, Taiwan.