Risk Scores for Hepatocellular Carcinoma in Patients With Chronic Hepatitis B

Among patients with chronic hepatitis B (CHB), the risk of developing hepatocellular carcinoma (HCC) is highly variable, but cirrhosis is the strongest risk factor. The incidence of HCC is 3% to 4% per year in Asia and 2% to 3% per year in Europe, in those with compensated cirrhosis, whereas HCC rates are 0.6%, 0.5%, and 0.3% per year in East Asia, West Africa, and Europe respectively, among patients with CHB without cirrhosis. Each of the largest liver societies (American Association for the Study of Liver Disease, European Association Study of the Liver, and Asia-Pacific Association for Study of the Liver) recommend HCC surveillance in patients with CHB with cirrhosis. The American Association for the Study of Liver Disease guidance statements on HCC screening in hepatitis B surface antigen (HBsAg)–positive persons are the most expansive and recommend surveillance based on age, sex and race (for Asian and African born), family history of HCC, as well as hepatitis D virus coinfection.

The goal of HCC surveillance is to detect HCC at an early stage, such that more curative treatment options are available and survival odds are improved. Optimal outcomes with surveillance require:

• A sensitive and specific diagnostic tool
• Application of the screening tool to the population at risk
• Streamlined algorithms for management of positive tests
• Therapies that provide opportunity for cure

Increasing uptake of surveillance requires attention to the role of physicians, patients and healthcare systems in conducting surveillance. A more tailored approach to screening is desirable to maximize patient benefits while being mindful of healthcare resources and costs.

Why Consider Risk Stratification for HCC Surveillance?
Risk stratification strives for a more personalized approach to surveillance and may offer 2 specific benefits. First, low-risk or very low–risk individuals may be able to avoid surveillance or require a less frequent or less burdensome form of surveillance. Second, those at high risk may benefit from an intensification of surveillance with more frequent monitoring or the use of more sensitive (and likely more expensive) surveillance tests.

Most HCC risk scoring systems are based on the combination of routine clinical and laboratory parameters that stratify patients with CHB into low-risk, intermediate-risk, and high-risk groups. Recently, liver stiffness measures have been added to risk scores and have shown to enhance prediction, which is not surprising since the use of elastography likely improves the identification of cirrhosis. Finally, validation of risk scores for patients receiving antivirals is important because long-term antiviral therapy is recommended for all patients with CHB and cirrhosis. Several risk scores (eg, PAGE-B and CAMD) have been validated in patients receiving antiviral therapy.

How Are Risk Scores Defined?
It is worth briefly outlining how risk scores are developed. First, a defined cohort of patients with CHB is used to identify independent risk factors associated with HCC using regression methods, that is, either the Cox proportional hazard model if survival analysis is performed or a binary logistic regression model if HCC is considered as a binary variable at a defined time point. Based on the multivariable analysis from this training or derivation cohort, the independent risk factors and their corresponding weights are used to construct a risk score. This derived risk score is then tested in an independent cohort of patients (the validation cohort) to confirm prediction accuracy. If no independent cohort is available, internal validation using bootstrapping or other methods is performed. The performance of the score in measured as discrimination and calibration.

Discrimination can be assessed using an area under the receiver operating characteristic curve, as well as sensitivity, specificity, and negative and positive predictive values. Calibration is evaluated by estimating the observed HCC risk using the Kaplan-Meier method with the same cumulative risk scores.

How Well Do Risk Scores Work?
It is worth noting that all the HCC risk prediction scores perform very well in excluding those at risk for HCC, that is, they have high specificity and negative predictive value, but a low positive predictive value. Thus, the primary clinical utility of these scores is in identifying those at low risk of HCC, with >95% negative predictive values across all the predictor scores. Patients in low-risk groups can have surveillance studies suspended for a 3-year to 10-year period (as defined by the score) or other less intensive monitoring may be considered. By contrast, positive predictive values are low (<20%), indicating a low ability to predict who will develop HCC.

Of importance, the time horizon for the risk prediction score needs to be considered. For example,  the REACH-B score uses a 4-year time horizon, so a low-risk patient for HCC would need to be reassessed in 4 years to ascertain if the risk group has changed.

Certain common variables are dominant in risk prediction scores, as highlighted in Table 1. Age, sex, and the presence of cirrhosis are almost uniformly included (with older age, male sex, and cirrhosis heavily weighted). Markers of more advanced cirrhosis, such as low platelet count, low albumin, elevated bilirubin, and greater liver stiffness contribute additional points to the risk scores. Of the viral factors, hepatitis B e antigen–positive status and high HBV DNA levels are weighted more heavily, with at least 1 study demonstrating that quantitative HBsAg can be substituted for HBV DNA.

Table 1. Variables, Populations, and Predictive Value of HCC Risk Scores

A review of the key patient characteristics in the development of risk scores reveals that the majority were developed and validated in East Asian populations; none was developed in US patients and very few have been validated in a US population. This remains an important area for research: to develop and/or validate existing prediction tools in the untreated and treated populations with CHB within the United States. Other considerations include the patient setting—whether outpatient primary care (such as the general population with CHB) or hospital-based (hepatologist providers)—because prevalence of HCC, and hence the accuracy of the predictive tool, likely varies between these 2 settings. Given these challenges in interpreting HCC risk scores in a US population, there is not currently wide application of these scores in the United States. However, this remains a vitally important area of research as we increasingly focus on personalized medicine. In addition, the intersecting epidemics of viral hepatitis, nonalcoholic fatty liver, and alcohol-associated liver disease make inclusion of patients with dual or triple diagnoses important in future risk models. The emergence of diabetes as part of the CAMD score reflects this evolution.

In summary, risk prediction scores for HCC are most accurate in identifying low-risk patients who are not at risk for HCC. Although their positive predictive value is low, when applied to the moderate-risk to higher-risk population, they may serve to enhance adherence to HCC surveillance by clarifying to healthcare providers and patients the magnitude of the risk. New biomarkers and genetic assays may enhance risk score prediction in the future.

Your Thoughts?
Do you use risk scores to assess HCC risk in your patients with CHB? If so, how do you use risk scores to inform personalized HCC surveillance in low-risk, intermediate-risk, and high-risk patients? Please share your thoughts and elaborate in the comments section.