Challenges and barriers in hepatocellular carcinoma (HCC) surveillance for patients with non-alcoholic fatty liver disease (NAFLD)

The proportion of hepatocellular carcinoma (HCC) cases due to NAFLD is expected to increase, paralleling the rise in NAFLD due to the obesity epidemic. Early detection is critical, as it potentially enables curative treatment. Current guidelines recommend ultrasound imaging with or without serum AFP measurement in patients with cirrhosis. Unfortunately, several challenges and barriers impede the effective surveillance of HCC in patients with NAFLD. In this review, we focus on four main challenges and barriers: the scale of the NAFLD epidemic, the lack of accurate risk stratification tools, the limitations of available surveillance tools themselves, and the existing disparities in access to care for chronic liver disease. We describe potential solutions, including public health approaches to obesity, improving clinical risk scores using genomic and metabolomic data, improved imaging techniques and blood-based biomarkers, and focusing on underserved groups with liver disease.


INTRODUCTION
Numerous challenges and barriers impede effective surveillance for hepatocellular carcinoma (HCC) in patients with non-alcoholic fatty liver disease (NAFLD). In this review, we focus on four main challenges and barriers: the scale of the NAFLD epidemic, the lack of accurate risk stratification tools, the limitations of available surveillance tools themselves, and the existing disparities in access to care for chronic liver disease. We also provide initial suggestions for potential solutions to address these barriers [ Figure 1].

THE SCALE OF THE PROBLEM
The scale of the NAFLD epidemic makes HCC surveillance difficult and expensive. It is estimated that NAFLD affects one-quarter of the world population and about one-third of the US population [1] . Regretfully, the incidence and prevalence of NAFLD are projected to increase, paralleling increases in diabetes and other features of the metabolic syndrome. One recent study projected a 21% increase in NAFLD prevalence in the United States from 2018 to 2030.This translates into a NAFLD prevalence of 33.5% (or more than 117 million adults) in the US by 2030 [2] . Of equal concern is the increase in pediatric obesity and features of the metabolic syndrome in children, which will increase the frequency and severity of adult NAFLD and resultant cirrhosis in the coming years. Multiple large studies have associated weight gain in childhood and adolescence with severe liver damage and liver-related mortality in adulthood [3,4] .
Although HCC related to viral hepatitis is declining due to improved interventions to prevent and treat hepatitis B and C [5] , the rise in NAFLD will neutralize this benefit by promoting cirrhosis, HCC, and liverrelated mortality. The risk of HCC in NAFLD cirrhosis is about 1.5% per year [6] , with estimates ranging from 0.5%-2.6% [7] . Although NAFLD patients without cirrhosis have a lower risk, their risk remains substantially higher than those with some other etiologies of chronic liver disease.
Given the benefit of HCC surveillance for early liver cancer detection and survival suggested by a recent meta-analysis [9] , the scale of the NAFLD epidemic per se should not prevent its adoption. In fact, HCC surveillance in patients with cirrhosis displays similar cost-effectiveness to colonoscopy [10] , and the latter is recommended for all average-risk adults ages 45 and older. As with improving non-invasive methods for colon cancer screening, such as stool-based tests, it is imperative to find inexpensive and rapid HCC surveillance tools.

LACK OF RISK STRATIFICATION TOOLS
AASLD guidelines recommend twice-yearly HCC surveillance for all patients with NAFLD cirrhosis [11] . A recent meta-analysis estimated the incidence of HCC in non-cirrhotic NAFLD patients at 0.03 per 100 person-years [12] , and an AGA 2020 clinical practice update recommends that patients with advanced fibrosis (F3) on liver biopsy or two distinct non-invasive measures undergo HCC surveillance [13] , but this has not been widely adopted. Two important questions therefore remain unanswered: (1) Do some patients with NAFLD cirrhosis need more intensive HCC surveillance than guidelines recommend? (2) Which patients, if any, with NAFLD and F0-F3 fibrosis warrant HCC surveillance? Below, we discuss clinical, tumor marker, genetic and metabolomic strategies for risk stratification, which could collectively help answer these questions.
Clinical risk factors provide a rudimentary risk stratification system. The strongest predictor of liver-related events including HCC in NAFLD is the advanced fibrosis stage [14] . Although liver biopsy remains the gold standard for assessing fibrosis, non-invasive methods are gaining popularity. Scoring systems such as the Fibrosis-4 (FIB-4) score, AST to platelet ratio index (APRI,), and NAFLD fibrosis score (NFS) are based on readily available laboratory results and clinical data and have excellent performance characteristics. In one European cohort of 1,173 NAFLD patients, NFS had the best performance for predicting the development of HCC, and both NFS and FIB-4 predicted liver-related events [15] . A retrospective study of almost 30,000 patients in Germany found that a FIB-4 score greater than or equal to 1.3 was a strong predictor of HCC over 10 years [16] . More recently, a cohort of 81,108 patients was assessed over a mean-follow up of 34 months, and FIB-4 independently predicted overall mortality and liver-related adverse outcomes [17] . For this reason, recent AASLD practice guidance recommends risk stratification with FIB-4 followed by transient elastography for patients with scores indicative of advanced fibrosis [18] .
Beyond the fibrosis stage, male sex, older age, smoking, diabetes, and obesity are known to increase the risk for HCC [6] . In a large retrospective analysis of European patients (n = 136,703) with NAFLD, diabetes was the strongest predictor of cirrhosis and HCC [19] . Hypertension, hyperlipidemia, and alcohol intake have also been associated with increased HCC risk in NAFLD [20,21] . Although race and ethnicity may affect susceptibility to, and progression of NAFLD, our understanding of this relationship remains underdeveloped. Available data suggest that within commonly described racial and ethnic groups such as "Hispanic", the prevalence of NAFLD may differ significantly. For example, one study found that the prevalence of NAFLD was 33% in patients of Mexican heritage, but only 16% in patients whose heritage was from the Dominican Republic [22] . This may be due to genetic, epigenetic, or environmental factors, or a combination thereof. Considering more specific ancestry in statistical analyses of NAFLD patients may allow for improved phenotyping and risk stratification for HCC.
Non-invasive biomarkers including genetic, proteomic and metabolomic data may aid in risk stratification for HCC surveillance for both patients with and without NAFLD cirrhosis. One example of the use of genetic markers was recently published using individuals from the UK Biobank and Denmark. Authors combined single nucleotide polymorphisms (SNPs) on PNPLA3, TM6SF2, and HSD17B13 into a risk score and found that the number of risk alleles was associated with the risk for HCC, with 6 risk alleles corresponding to an odds ratio of 29 compared to those with 0 risk alleles [23] . A SNP on MBOAT7 has also been associated with HCC risk in non-cirrhotic NAFLD specifically [24] . Transcriptomic data may also aid in risk stratification of patients with NAFLD. A 32-gene expression signature previously developed in patients with cirrhosis due to HCV was associated with an annual incidence of HCC in patients with cirrhosis of any etiology, and the association remained significant in the subgroup with NAFLD [25] . More recently, the same group found that a 133-gene liver tissue-based expression signature predicted the longitudinal development of HCC in a cohort of NAFLD patients. This was translated into a 4-component serum panel, which was associated with HCC development independent of clinical risk factors [26] . Additional gene expression-based risk stratification biomarkers have been developed in patients with other etiologies of cirrhosis (largely viral hepatitis) and should be validated in patients with NAFLD to determine their utility. These include fatty acids [27] , osteopontin [28] , serum glycome (Cirrhosis Risk Score), and IGF1 [29] . Gene expression may fluctuate over time, which is an advantage compared to static data such as SNP genotypes (because changes in gene expression may capture disease progression) but introduces complications for clinical applicability, including how and when to capture this data.
Presumably, the most useful tools for patients and clinicians would be risk stratification scores that incorporate clinical and biomarker data. For example, clinical factors with strong predictive value such as age and diabetes status could be combined with gene expression or SNP data to sort patients into low, intermediate, and high-risk groups. Such classification could aid in prioritizing patients for new and improved surveillance strategies when these are developed.

LIMITATIONS OF AVAILABLE SURVEILLANCE TOOLS AND POTENTIAL SOLUTIONS
A third challenge remains the current suboptimal surveillance tools for identifying HCC at an early stage when curative treatments resulting in improved survival are available. Liver ultrasound is currently the standard of care, with a number of serum-based biomarkers in development to improve sensitivity or even replace imaging. One meta-analysis found that liver ultrasound was only 47% sensitive for the detection of early-stage HCC in patients with cirrhosis of any etiology. Detection sensitivity increased to 63% with the addition of alpha-fetoprotein (AFP) measurement [30] . A more recent Cochrane review confirmed ultrasound sensitivity of 53% for resectable HCC, which increased to 79%-89% with the addition of AFP [31] . Unfortunately, obesity further limits the accuracy of ultrasound as fat attenuates ultrasound beams directly [32] . Additionally, ultrasound is operator-dependent and performance characteristics vary significantly. These challenges point to the importance of studying the sensitivity and specificity of this tool, particularly in obese NAFLD patients. Although contrast-enhanced CT and MRI have excellent sensitivity and specificity for detection of HCC, the time and cost of these technologies has prevented their adoption in place of ultrasound. Abbreviated MRI may provide a better option. Abbreviated MRI techniques utilize a limited number of sequences to detect HCC, and thus decrease time and costs while continuing to provide excellent sensitivity over ultrasound [33] . Abbreviated MRI is currently being studied in the Veterans Affairs Health System (PREMIUM study, CSP #2023) and in a multi-center prospective study [FAST (Focused Abbreviated Screening Technique)-MRI Study, NCT04539717]. Such protocols would give a better definition and ability to apply radiomics compared to ultrasound and would limit time and cost. Important additional considerations for optimal surveillance tools are tumor biology and tumor aggressiveness. While imaging may provide information about tumor size and macrovascular invasion into the portal vein or other surrounding structures, these key measures of cancer risk have not been well captured on imaging traditionally. That may change with the advent of radiomics which uses radiographic features to extrapolate biochemical or molecular characteristics of a lesion. Radiomics could offer several advantages, including the ability to account for heterogeneity within a lesion (which is not accurately captured by a random sample tumor biopsy) [34] . Tumor grade, microvascular invasion, aggressiveness, and gene expression have been predicted with reasonable accuracy using cross-sectional imaging characteristics [35][36][37] . Such technologies may contribute important information to the non-invasive characterization of HCC.
Tumor markers including alpha-fetoprotein have been used in combination with imaging for HCC surveillance. Attempts have been made to refine serum tumor marker-based panels to better risk stratify patients. The most notable example is the GALAD score, which was developed using European and Japanese cohorts and combines AFP-L3, AFP, and des-carboxy-prothrombin in addition to age and sex [38] .
Other non-invasive measures have been studied. Hypermethylation of certain genes can be detected in the blood prior to HCC diagnosis and could be used in combination with other modalities [39] . Several studies have examined blood-based methylation profiles, in combination with gene expression or other data, to detect early-stage HCC. Such "liquid biopsies" may become standard of care in the future, given the cost and inconvenience associated with performing abdominal ultrasound as well as the limitations mentioned above. In one multi-center study, a panel using 3 methylated DNA markers as well as 2 protein markers including AFP was able to discriminate HCC cases from age-matched liver disease controls without HCC, with an area under the curve (AUC) of 0.92 [40] . A simplified blood-based panel from the same group using AFP and multiple DNA methylation markers revealed an AUC of 0.91 for any stage HCC and 0.86 for early-stage HCC. In a separate validation cohort, sensitivity for HCC detection was 88% [41] . The funders have subsequently started a large, prospective clinical trial for this panel (Oncoguard®, Exact Sciences Inc, ALTUS Study, NCT05064553). Methylation profiles need further study and validation prior to use in general clinical practice, but methylation profiling is already in use in non-invasive colon cancer screening (Cologuard®) and holds great potential.
Extracellular vesicles (EVs) represent another type of "liquid biopsy" holding promise in HCC. As EVs contain DNA, RNA, proteins, metabolites and lipids from both normal and tumor cells and are present in circulation, several studies have investigated their role in detecting HCC among at-risk patients [42][43][44][45] . A recently published phase 2 biomarker study revealed a 91% sensitivity and 90% specificity for distinguishing early-stage HCC from cirrhosis, of which NAFLD represented 20% [46] . The HCC EV ECG score was calculated from three HCC EV subpopulations (EpCAM + CD63 + , CD147 + CD63 + , and GPC3 + CD63 + ) in a training cohort (n = 106) and independent validation cohort (n = 72). Further validation in larger multi-site trials is needed, but this promising method could augment current methods to detect HCC at earlier, more treatable stages.

DISPARITIES IN HCC SURVEILLANCE
Disparities in access to HCC surveillance pose another major barrier to successful implementation. Overall surveillance rates for patients with cirrhosis are under 10% (ultrasound every 6-12 months) based on a recent analysis of a large US-based commercial database. During the study period from 2007-2016, 45% of patients had no HCC surveillance whatsoever [47] . NAFLD patients may have even lower rates of surveillance; a multi-center study found that patients with NAFLD were less likely to undergo HCC surveillance compared to patients with chronic hepatitis C infection [48] . Similar findings were discovered in a large cohort from the Department of Veterans Affairs, in which patients with NAFLD were less likely to have received HCC surveillance leading up to a diagnosis of HCC [49] . It is possible that undiagnosed NAFLD contributed to this finding.
Although many patients report difficulty in presenting for HCC surveillance imaging due to time constraints, cost, and transportation [50] , disparities in HCC surveillance by race and socioeconomic status are well established. A single-center study from a large urban safety-net study found that Black patients were less likely to receive consistent HCC surveillance compared to White patients [51] . Larger studies from the SEER-Medicare database and the Veterans Affairs population (specifically in patients with HCV) have also found lower rates of HCC surveillance among Black patients [52,53] . The reasons for this are unclear.
Disparities in access to subspecialty hepatology care (which has been associated with higher HCC surveillance rates), socioeconomic disparities including transportation and financial resources for health care, could have contributed. In the abovementioned study, race was independently associated with lower HCC surveillance rates even after adjustment for insurance status and receipt of primary care services. Therefore, financial factors are unlikely to explain the difference. Provider bias may also contribute; in one study of 467 patients followed at a tertiary care center, more non-White patients reported that their doctors had never talked to them about HCC surveillance [54] .
Although telehealth technology may enable greater access to sub-specialty liver disease care, which is associated with better HCC surveillance rates, geographic areas with less access to specialty care also have lower rates of broadband internet coverage [55] . In addition, it is unlikely that telehealth would fully address disparities in HCC surveillance, since barriers such as transportation and cost to obtain regular imaging and/or blood tests will persist. Engagement with primary care providers and community health centers may expand access to surveillance, and initiatives to educate patients and providers should continue.

CONCLUSIONS
Overall, the barriers to HCC surveillance are significant but surmountable. New technologies and better risk stratification will enable targeted interventions. At the same time, measures to increase surveillance rates in the cirrhosis population and in racial minorities are urgently needed.