Is there an ideal prognostic model for hepatocellular carcinoma?

We read with interest the paper by Grieco et al (Gut 2005;54:411–8). It is an elegant study that retrospectively compared the prognostic power among the Okuda, Cancer of the Liver Italian Program (CLIP), and Barcelona Clinic Liver Cancer (BCLC) staging systems for patients with hepatocellular carcinoma (HCC). The authors concluded that BCLC and CLIP were good models for non-surgical HCC, and BCLC had better predictive value compared with the others for patients with early stage HCC. As the CLIP system has been prospectively validated and proposed as the primary staging system for HCC,¹ it would be interesting to examine how these commonly used HCC staging systems were derived and explore the potential limitations of the authors’ conclusions.

The main reason why the authors have reached this conclusion is probably related to the distinct characteristics of the study population, as the majority (249/268; 93%) had undergone active treatment (percutaneous ablation or arterial chemoembolisation), suggesting most had early or intermediate stage disease. These characteristics made the BCLC system, which contains treatment derived parameters,² a prevailing model for prognostic prediction. A recent study comparing the various staging systems consistently showed that BCLC was best compared with CLIP, Okuda, and other systems in a surgically oriented referral centre.³ It should be noted that the CLIP and Okuda systems were originally derived from a large unselected patient population and the majority had been treated conservatively.^4,5 Therefore, although the prognostic predictors selected for the currently used staging systems are not mutually exclusive, the derived predictive models from these predictors may have an otherwise variable differentiation power. Certain important risk factors, such as tumour size <3 or 5 cm, used in BCLC, can only be significant in the patient population that predominantly undergo active locoregional therapies.^6,7 In these instances, the predictive power of a given staging model, constructed from selected risk factors, could be drastically impaired if the majority of patients do not have early stage HCC. Such an effect may explain why the BCLC system is better than the CLIP and Okuda systems in the current study because clinical outcome was intimately associated with patient demographics and subsequent treatment strategy. Consistent with this notion is that a Canadian study group demonstrated that CLIP was a good predictive model for their HCC patients in whom more than half (52%) had only been treated conservatively due to a relatively advanced tumour or cirrhotic stage.⁸ Therefore, it is not surprising that BCLC is better that its competitors in an appropriate study environment.

In summary, the BCLC system contains treatment derived parameters and may work well in areas where HCC is diagnosed at a relatively early stage, whereas the CLIP or Okuda system would only prevail in patients with intermediate or late stage disease, under which conditions any aggressive forms of therapy are less likely to succeed. As the clinical presentation of HCC is tremendously heterogeneous, it is necessary to consider all known predictive factors, from early to advanced stages, in building an ideal staging system to fit all patient populations.