The presence of hepatic fibrosis demarcates an important phase in liver disease. The evolution of fibrosis to cirrhosis is associated with changes in morphology, function, hemodynamics, and neoplastic propensity. Fundamentally, clinicians are interested in (1) which patients succumb to the hard clinical endpoints of death, hepatic decompensation, and hepatocellular carcinoma; (2) the length of time this may take; and (3) whether intervention can alter this trajectory. The detection of fibrosis at the index biopsy has been shown to inform prognosis in a number of studies across different etiologies of liver disease. In chronic viral hepatitis C (CHC), Yano et al.1 showed the severity of liver fibrosis at baseline altered the rate of progression to cirrhosis at 10 years. Neal et al.2 found liver fibrosis to be an independent prognostic indicator of clinical outcomes in a large, prospective cohort of patients with CHC. In alcoholic liver disease, significant liver fibrosis and alcoholic hepatitis have emerged as important prognostic determinants.3, 4
Hitherto, the vast majority of noninvasive marker biomarker studies have concentrated on assessing performance of biomarkers against a reference standard of histological fibrosis. The limitations of using a histological reference standard in developing diagnostic biomarkers have been previously well documented. Pathologists will rightly point out that semiquantitative, ordinal, histological scoring systems were never designed to be used for assessing noninvasive biomarkers. Furthermore, the continuing development of biomarkers in this area will be restricted if we solely rely on cross-sectional, histological endpoints. If biomarkers show diagnostic imperfection, the question remains of whether this is due to the inadequacy of the biomarker test, the reference standard, or both. Because histology is a surrogate for future clinical outcomes, then why not use the latter as the “reference standard” against which biomarker performance is measured?
The study by Naveau et al.5 in this issue of HEPATOLOGY heralds the possibility of a changing direction in how we judge the performance of biomarkers in liver disease. The authors retrospectively tested a number of panel biomarker tests on a cohort of 218 patients with alcoholic liver disease. The diagnostic performance of three panel markers—FibroTest, FibrometerA, and Hepascore—produced similar areas under the receiver operating characteristic curve (AUROCs) at baseline for the detection of advanced fibrosis (AUROC of 0.83) and cirrhosis (AUROC of 0.92) compared to histology. This study looked at the ability of panel markers to predict clinical outcomes, as judged by clinical outcomes using “liver-related death” and “survival or non–liver related death”. Three panel biomarker tests produced prognostic performance comparable to histology (AUROC ranging from 0.77 to 0.80) and higher than the Child-Pugh score (AUROC 0.67) for the composite endpoint of “survival or non–liver related death”. In a multivariate model, FibroTest (risk ratio of 23.2) and fibrosis staging at biopsy (risk ratio of 1.5) demonstrated an independent association with liver-related death. Interestingly, alcoholic hepatitis diagnosed by biopsy or suggested by another panel test (Ash Test™) did not appear to inform prognosis in the multivariate model. Larger, prospective studies which stratify alcoholic hepatitis and fibrosis may be required to dissect these closely associated pathological entities.
The generalizability of this study to other etiologies of liver fibrosis is an important question. An observation in the study was the stronger prognostic association of the biomarkers to survival and non–liver related death in comparison to overall survival alone. As the authors allude to in the discussion, there is clearly a relationship between alcohol and “nonhepatic” mortality. The significant number of “nonhepatic” deaths, in a relatively small cohort, necessitated this composite endpoint, and this will not necessarily apply to other chronic liver injuries. Emerging data, however, is encouraging that biomarker prediction of clinical outcomes is not etiology-specific. A study by Ngo et al.6 showed that biomarkers at baseline can predict clinical outcomes in CHC. The prognostic AUC for HCV-related death at 5 years was 0.87 for biopsy compared to 0.96 for FibroTest.
The major issue of using clinical outcome measures as endpoints of biomarker studies is the considerable length of time it requires to generate these endpoints sufficient to provide statistical power. In addition, the drop-out rate from longitudinal follow-up requires large cohorts and raises the issue of selection bias. Although this should not detract from performing such studies, it may limit the number of studies investigators are willing to perform using these robust endpoints. What are the alternatives?
Continuing to improve the quality of cross-sectional diagnostic studies is an important goal. A common theme in the systematic reviews of noninvasive biomarkers is the heterogeneity that exists in the current literature.7, 8 Utilizing criteria such as Standards for Reporting of Diagnostic Accuracy (STARD),9 analogous to the CONSORT guidelines in randomized controlled trials, may be useful in improving the quality and transparency of diagnostic studies. A major source of heterogeneity is the prevalence of the different stages of fibrosis. A test that is derived and validated on a population with severe disease will have a different diagnostic performance in a population with mild disease. The study by Naveau et al.5 addresses the issue of prevalence by applying a statistical correction called difference of mean stages observed between advanced fibrosis and nonadvanced fibrosis (known as DANA analysis); the prevalence of advanced fibrosis in this cohort was 63%. The purpose of trying to account for prevalence is justified, but it may not eliminate spectrum bias. Spectrum bias exists when there are a range of clinical, pathological, and comorbid components that affect diagnosis.10 The prevalence of disease severity may be an indication that spectrum bias exists but is not necessarily causal; importantly, spectrum bias can still exist where prevalence of disease severity is identical. This is because sensitivity and specificity statistics are influenced by factors such as the distribution of disease severity (e.g., the numbers with cirrhosis may be the same but the severity of cirrhosis can be very different) and the range of alternative diagnoses (e.g., recruitment criteria of a study may filter or concentrate conditions that are similar to the target diagnosis). Standardization of reporting diagnostic studies will highlight spectrum bias and allow individual clinicians to decide how published research can be translated to their own practice.
Noninvasive biomarkers are continuous variables, and finding a reference measure that is a robust continuous variable has obvious attractions. There has been emerging interest in using hepatic venous pressure gradient (HVPG) as such an endpoint. HVPG has been shown to predict clinical outcome measures in a number of settings. In the context of liver transplantation and HCV, an HVPG greater than 6 mmHg predicted clinical outcome more effectively than biopsy.11 There have been a number of studies to assess transient elastography using HVPG.12, 13 The use of HVPG to assess biomarkers is very promising, and further studies are awaited. One small note of caution is the relationship of HVPG and liver fibrosis which, although associated, is clearly not mutually exclusive. Dynamic alterations in intrahepatic vasculature, necroinflammation, and steatosis may all contribute to overall HVPG measurement.14, 15
Predicting the performance of noninvasive biomarkers in a community setting is difficult at present. Extrapolating the data from the literature performed in hospital clinics may have limitations, because of the caveats above. Performing primary research with an invasive reference test in a low prevalence of severe fibrosis has ethical considerations that cannot be ignored. Therefore, the extension of biomarker studies into the community necessitates a more effective “gold standard”, and the use of clinical endpoints has particular appeal in this setting.
Fibrosis is a dynamic process; the ability to measure this process in a serial manner at short intervals will alter the way we manage liver disease. The requirement for an alternative reference test, to assess emerging biomarkers, is gathering momentum. Providing robust evidence using hard clinical endpoints, such as liver-related death or hepatic decompensation, offers real hope for these noninvasive biomarkers to realize their potential.