Genetic Determinants of Cirrhosis and Hepatocellular Carcinoma Due to Fatty Liver Disease: What’s the Score?

Fatty liver disease (FLD), either alcoholic or nonalcoholic, is the most common liver disorder in high-income countries. In the United States, alcohol-associated liver disease (ALD) has an estimated prevalence of 2%⁽¹⁾ and is the leading indication for liver transplant listing. Nonalcoholic fatty liver disease (NAFLD), which is associated with obesity and insulin resistance, affects 25%-30% of adults worldwide.⁽²⁾ ALD and NAFLD are leading causes of cirrhosis and hepatocellular carcinoma (HCC), and despite having different etiologies, these diseases share histological features and genetic risk factors, suggesting common underlying biological pathways.

Current guidelines recommend specialist referral only for patients with advanced ALD⁽³⁾ or NAFLD.⁽⁴⁾ Surveillance for HCC is only advocated for patients with FLD who have established cirrhosis.⁽⁴⁾ Importantly, risk stratification currently identifies patients after they have advanced fibrosis. A better strategy might identify patients with FLD at increased risk before developing advanced disease and target them for preventive interventions. The genetic risk score (GRS) proposed by Gellert-Kristensen et al.⁽⁵⁾ could aid that approach.

In this fine study, the authors assessed the association of GRS with risk of cirrhosis and HCC in general population cohorts from Denmark and the United Kingdom with more than 400,000 combined participants. The GRS, ranging from 0 to 6, equaled the number of risk alleles from three variants with known effects on FLD (patatin-like phospholipase domain containing protein 3 I148M; transmembrane 6 superfamily member 2 E167K; hydroxysteroid 17-beta dehydrogenase 13 rs72613567). Comparing participants with GRS of 5 or 6 with those with GRS of 0, the investigators found large odds ratios (ORs) for cirrhosis (OR = 12) and HCC (OR = 29). However, only approximately 0.5% of individuals had a GRS in that range. A GRS of 4, which still conveyed large risks (cirrhosis, OR = 5.2; HCC, OR = 3.3), was found in approximately 5% of this population. Despite some exciting findings, the authors are appropriately cautious regarding widespread application of the GRS.

We wish to add our comments on why this GRS should not be applied to a general population and discuss the potential of the GRS for risk stratification among patients with FLD. Consider a “general population” in which few individuals have GRS greater than or equal to 4 (approximately 6% for a population of European ancestry). A key statistic for assessing the usefulness of surveillance of this “high-risk” group for cirrhosis or HCC is the positive predictive value (PPV), which is the probability that someone with GRS greater than or equal to 4 will develop cirrhosis or HCC. The PPV is a risk probability that captures not only the strengths of association of a risk factor with disease but also the outcome’s prevalence (or incidence). The PPV can be calculated as the sensitivity of the test multiplied by the probability of the outcome divided by the probability that the test is positive in the whole population.⁽⁶⁾ Based on the data from Gellert-Kristensen et al., the PPV of the GRS-based test among the Danish population is 0.008 for cirrhosis and 0.003 for HCC. In other words, among 1,000 persons with GRS greater than or equal to 4, only 8 will develop cirrhosis and 3 will develop HCC. In the UK population, the PPV for cirrhosis was 0.003, whereas the PPV for HCC was only 0.0008. Thus, despite strong associations between higher GRS and cirrhosis or HCC, these PPVs are too low to make the GRS useful in a “general population.” These low PPVs are partly explained by the low sensitivities of our “test,” which ranged from 13.8% to 20.4% for cirrhosis or HCC in these cohorts. Other key factors are imperfect specificity and the low incidence of cirrhosis and HCC in a general population. For both outcomes, specificity is approximately 94%, which may seem good, but even a test with “good” specificity yields many false positives when applied to a rare disease. Thus, GRS should not be used for surveillance in the general population because few individuals with a high score would develop the disease.

PPV calculations highlight that clinical decisions should be based on the absolute risk of an outcome (i.e., probabilities) rather than relative risk measures (e.g., ORs). The clinically most relevant measure of absolute risk is the cumulative incidence—namely, the probability that a person with defined risk factors who is free of disease at the age of risk assessment will develop disease over a given future age interval. Absolute risk models that incorporate the GRS with other risk factors might be developed for target populations of patients with FLD.

One could consider applying the GRS to patients with FLD to help identify those who should be referred for hepatology specialist care. Current approaches advocate a sequential combination of serum tests and imaging, which improves specificity (>90%) for detecting advanced disease but has limited sensitivity (approximately 30% for Fibrosis-4). Suppose the sensitivity and specificity of GRS are the same in patients with FLD as in the general Danish population. Although the lifetime risk of cirrhosis for NAFLD or ALD is unknown, we assume for illustration purposes that it is 20% in a patient with ALD. Then, among those with GRS greater than or equal to 4, the PPV is 46%. Using a lifetime risk of cirrhosis in a patient with NAFLD of 4%, among those with GRS greater than or equal to 4, the PPV is 12%. This suggests that the GRS by itself is insufficient to identify patients with FLD at high risk of progression. Perhaps better stratification among patients with FLD could be achieved by combining GRS with other predictors of progression, such as behavioral, clinical, or laboratory factors.

The GRS could also be relevant for HCC surveillance, which is currently advocated for patients with FLD with established cirrhosis. Although HCC occurs in patients with NAFLD without cirrhosis, its incidence is much too low (0.008% per year)⁽⁷⁾ to justify routine screening, and applying the GRS to that population is unlikely to identify a group of patients among whom surveillance is warranted. However, Gellert-Kristensen et al. present potentially useful information on the relationship between GRS and cumulative incidence of developing HCC or dying following onset of cirrhosis after considering competing causes of mortality (Fig. 4). Individuals with cirrhosis and GRS greater than or equal to 5 had a high risk of HCC and death that could well affect clinical management; however, despite the very large study size, there were few such patients, and that result was not statistically precise. In contrast, Gellert-Kristensen et al. found HCC to be very rare among patients with cirrhosis with GRS less than or equal to 2. If these findings were confirmed, GRS, perhaps in conjunction with other markers of HCC risk, might help determine the appropriate intensity of HCC surveillance for individuals with cirrhosis due to FLD.

Before any model is implemented, its performance must be assessed (chapter 6).⁽⁶⁾ An absolute risk model can be evaluated in terms of its ability to accurately predict the number of events arising in a population (calibration). Good calibration is essential for most risk model applications, including counseling of patients regarding their prognosis, designing preventive intervention trials, estimating absolute risk reduction in the population from preventive interventions, and allocating preventive resources.⁽⁸⁾ Another popular measure of model performance is discriminatory accuracy, which measures how well separated the distribution of risk is in those who develop the outcome from the distribution in those who do not. Discrimination is often measured by the area under the receiver operator characteristic curve. Discriminatory accuracy is important for deciding which patients should be under surveillance for a rare outcome.⁽⁸⁾

Gellert-Kristensen et al. have shown that their GRS strongly predicts cirrhosis and HCC in a general population as well as HCC risk among individuals with cirrhosis due to FLD. The GRS highlights the importance of genetic variation in FLD progression; however, it would have limited utility as a sole predictor of risk. Risk models that combine GRS with other predictors of progression should be evaluated to determine the potential of GRS in clinical management and public health.