Risk scores for hepatocellular carcinoma in chronic hepatitis B

Abstract

Potential conflict of interest: Nothing to report. See Article on Page 1934 [Correction added on May 8, 2015, after first online publication: reference 8 was added; missing in original publication.] Western hepatocellular carcinoma (HCC) management guidelines suggest that patients at risk for HCC should undergo surveillance. The guidelines provide broad criteria for determining who is at risk for HCC, including patients with cirrhosis, regardless of etiology, and some with hepatitis B above certain age cutoffs. These groups were defined by cost‐effectiveness analyses that determined incidence rates of HCC above which surveillance was deemed cost‐effective. However, common experience shows that only a minority of patients in the risk groups ever develop HCC. There has been considerable research into identifying these at‐risk individuals more precisely so that those not at such a high risk could be excluded from surveillance programs. There are now risk scores that have been developed, some targeted at patients with hepatitis B, others at those with hepatitis C, and yet others included broader populations. Table 1 lists some of the risk scores that have been reported thus far. Table 1 - Risk Scores That Have Been Reported Thus Far Risk Score Variables Population Externally Validated? HBV specific Yang et al. (REACH‐B) Age, ALT, HBeAg status, gender, HBV‐DNA concentration Hepatitis B Yes (only in Asia) Wong et al. (CU‐HCC) Age, albumin, bilirubin, HBV‐DNA concentration, cirrhosis (yes or no) No Yuen et al. (GAG‐HCC) Age, gender, HBV‐DNA concentration cirrhosis, core promoter mutation No Broader patient population El Serag et al. ALT, AFP, age platelets Hepatitis C No Chang et al. Age, gender, platelets, AFP, fibrosis, Diabetes (yes or no) Hepatitis C after SVR No Michikawa et al. Age, gender, alcohol consumption, BMI, diabetes (yes or no), coffee consumption, hepatitis B, hepatitis C General population No Flemming et al. (ADRESS‐HCC) Age, Diabetes, Race, Etiology of liver disease, Sex, Severity (CTP score) Liver transplant waitlist Yes Abbreviations: AFP, alpha‐fetoprotein; BMI, body mass index; CTP, Child‐Turcotte‐Pugh; SVR, sustained virological response. Among the models for hepatitis B virus (HBV) patients, the most robust model to date may be the REACH‐B score.1 This was derived from participants who were not known to have cirrhosis in the REVEAL study.2 The score has been externally validated in other Asian populations.1 However, the score has not been evaluated in other populations where the viral genotype, host genetics, or environmental exposure profile may be dissimilar to the REACH‐B population. The other models for HBV patients take into account cirrhosis and/or degree of hepatic decompensation; however, they have not been externally validated.3 More recently, other scores have been developed that apply to hepatitis C and other populations.5 Some of the variables in some, but not all, models include presence and severity of chronic underlying liver disease (e.g., assessed by Child‐Pugh score), age, gender, race, and diabetes. These considerations raise a question of whether a universal risk score would be practically applied in the general population with vastly heterogeneous levels of HCC risk. In this issue of Hepatology, an article by Hung et al.8 presents another model for assessing HCC risk. They analyzed a large number (n = 12,377) of Taiwanese subjects from three different observational cohorts, heavily enriched with subjects with HBV infection. They proposed four models, aiming to develop a universal tool to stratify a general population for mass HCC screening. The first only included age, sex, and alanine aminotransferase (ALT), whereas the second model added history of chronic liver disease (CLD), family history of HCC, and smoking. The third model incorporated all these variables plus hepatitis B surface antigen (HBsAg) data, and the fourth model included both HBsAg and antibodies to the hepatitis C virus (anti‐HCV) data. Models 3 and 4 performed better than those with less information (models 1 and 2). They also use a decision curve analysis (DCA) to determine the level of risk that is associated with benefit to surveillance. Other risk scores developed to date assess predictive accuracy using odds ratios and receiver operating curves without explicitly considering the level of benefit that is sufficient to make the intervention worthwhile. In the work by Hung et al.,12 an important output is the net benefit, a weighted difference between true‐ and false‐positive classifications. When this was plotted against the threshold probability of 2% for developing HCC over 10 years, their model was superior to the current age‐based criteria for subjects without cirrhosis over an age range of 20‐60 years. This study raises the question of whether it is plausible to apply a scheme to measure HCC risk in the general population. The majority of at‐risk subjects can be characterized by their underlyi