Nonalcoholic fatty liver disease (NAFLD) is present in up to 30% of Americans and 25% of Asians and therefore is present in millions of individuals, making it the commonest liver condition in the world.1, 2 From a hepatologist's viewpoint, the task of evaluating and treating this enormous population is frightening! Fortunately, the large majority of individuals with NAFLD will not develop liver-related morbidity or die from their liver disease.3 Thus, the difficulty facing the managing physician is predicting which patients are at greatest risk of developing cirrhosis, thus identifying those who will benefit most from specific treatments, more intensive therapy, and monitoring.
Natural history cohort studies have provided us some information on prognostic factors in patients with NAFLD.3–5 Comorbid diabetes is associated with increased all-cause and liver-related mortality rates.3, 4 Diabetes and obesity have also been associated with higher rates of fibrosis progression.6, 7 Unfortunately, given the high rates of obesity and diabetes among patients with NAFLD and the prevalence of these conditions in the general community, these clinical factors are not sufficiently specific to predict those who will develop cirrhosis or its complications.
A more direct measure of prognosis is liver histology. Several studies have demonstrated that hepatic steatosis without evidence of inflammation or fibrosis is associated with low liver-related death rates of 0%–3% over a one-to two-decade period.4, 8, 9 In contrast, subjects with nonalcoholic steatohepatitis (NASH) are more likely to develop morbidity, with a cohort study of 131 subjects demonstrating a liver-related death rate of 17.5% over nearly two decades of follow-up.4 In this study, the hazard ratio for liver-related mortality associated with NASH was twice that compared to diabetes (13.9 versus 6.7, respectively), reinforcing the prognostic significance of histological assessment.4 However, it is not entirely clear whether the prognostic significance of NASH stems from the presence of steatohepatitis defined by lobular inflammation and ballooning or from the associated fibrosis. For example, Ekstedt et al. found that 18% of patients with NASH and portal fibrosis at baseline developed end-stage liver disease over time, whereas no patient with NASH decompensated in the absence of portal fibrosis.5
Liver biopsy is currently the accepted standard for determining the presence of NASH and fibrosis, but has well-documented problems of sampling and interpretation variability as well as procedural-related complications. These limitations have led to the development of noninvasive methods of histological assessment including clinical, biochemical, and radiological methods.10, 11 Simple clinical indices such as body mass index (BMI) and diabetes have been combined with simple liver function tests12 as well as more direct markers of fibrogenesis (hyaluronic acid, tissue inhibitor of metalloproteinase-1)13 or apoptosis (cytokeratin-18), to predict different degrees of liver injury and fibrosis.14 In contrast, transient elastography (TE) uses sonic detection of liver stiffness to predict hepatic fibrosis. It has been validated in patients with chronic hepatitis C as an accurate predictor of cirrhosis. In this issue of HEPATOLOGY, Wong and colleagues present a much anticipated study examining the accuracy of TE among patients with NAFLD.15
The study population consisted of 246 individuals originating from two centers in France and Hong Kong who underwent liver biopsy and TE. Liver stiffness increased significantly with fibrosis and provided a high level of accuracy for detecting significant fibrosis (defined as at least perisinusoidal and portal/periportal fibrosis), advanced fibrosis (septal or bridging fibrosis) and cirrhosis, with area under the receiver operator characteristic (AUROC) curve values of 0.84, 0.92, and 0.97, respectively. Importantly, in subjects in whom a full set of 10 successful readings could be obtained, the accuracy of TE was not affected by BMI or steatosis grade. Prior reports of falsely high readings due to acute hepatitis16 were not observed, with accuracy not influenced by alanine aminotransferase (ALT) levels or the histological NAFLD activity score, which reflects the relatively indolent inflammatory nature of NASH.
The accuracy of TE was also compared to five clinical and biochemical noninvasive measures; aspartate aminotransferase (AST)/ALT ratio, AST-to-platelet ratio index, FIB-4, NAFLD fibrosis score, and BARD score (derived from three variables: BMI, AST/ALT ratio, diabetes). After excluding subjects with invalid TE measurements, the AUROC values of TE were significantly higher than the clinical/biochemical indices for detecting advanced fibrosis and cirrhosis. However, when the diagnostic characteristics were compared using an “intention to diagnose” approach with the inclusion of subjects who had unsuccessful TE acquisition, the sensitivity and specificity values were not dissimilar from the clinical/biochemical models, although 95% confidence intervals were not provided for statistical comparison. Therefore, when TE measurement acquisition was successful, it was more accurate at predicting advanced fibrosis and cirrhosis than the alternative noninvasive models. Further comparative studies with models that use more direct markers of fibrogenesis such as hyaluronic acid are required before definitive conclusions can be reached regarding the relative accuracy of serum markers and TE in NAFLD.
Based on the performance characteristics of TE, the authors proposed two possible algorithms for determining advanced liver fibrosis. Using a cutoff point of 8.7 kPa, those with a reading below this had a negative predictive value (NPV) of 94.6% and therefore did not require biopsy. The prevalence (or pretest probability) of advanced fibrosis in community practice is likely to be lower than in this study, and thus the NPV is likely to be even better in this setting. The positive predictive value (PPV), however, was not high at 59.5%, and these patients required biopsy for accurate staging. The other proposed algorithm used two cutoffs. When the liver stiffness measurement was <7.9 kPa, subjects had an excellent 96.6% NPV which could be applied to 60% of the population. Those with a reading between 7.9 and 9.6 were deemed indeterminate and required liver biopsy, whereas those with a reading above 9.6 kPa had a 72.4% PPV of having advanced fibrosis. However, the clinical utility of a PPV of only 72% needs to be questioned. The PPV will fall further in settings where the prevalence of advanced fibrosis is less. For example, if the prevalence of advanced fibrosis is 10%, the PPV of TE for predicting advanced fibrosis falls to 50%. Similarly, the strength of TE for assessing cirrhosis in patients with NAFLD was for excluding F4 disease, with very high NPVs between 97%–99% but with modest PPVs between 46%–49%. Although this information is useful to the managing physician and reassuring to the patient, can we reassure patients with a low TE score (and thus low likelihood of cirrhosis or advanced fibrosis) that they are not at risk of developing liver-related morbidity and mortality? As outlined above, natural history studies would suggest that a “lower histological threshold” of NASH or significant (F2+) fibrosis distinguishes those at risk. In the present study, the lowest cutoff point of 5.8 kPA provided the greatest sensitivity (91%) and thus NPV (89%) for F2+ fibrosis; however, it is not clear how many subjects fell below this threshold and thus could be reassured of a relatively benign prognosis. Unfortunately, detection of this “in-between” degree of fibrosis remains the Achilles' heel of both serum-based and TE-based noninvasive algorithms, with the majority of individuals falling within indeterminate zones for the prediction of significant fibrosis.17, 18
Another limitation of TE is the potential for unsuccessful measurements. Just over 10% of subjects did not have valid TE measurements defined as a minimum of 10 successful acquisitions. Valid measurements were less likely to be obtained as BMI increased, with 25.5% of individuals with a BMI ≥30 kg/m2 having unsuccessful measurements compared to 1.6% of individuals with a BMI <25 kg/m2. This likely reflects reduced propagation of the vibration and ultrasound signals due to increased subcutaneous fat levels. It is noteworthy that a relative minority of patients in the study (28.5%) had a BMI ≥30 kg/m2. The overall failure rate would likely have been higher if the prevalence of subjects with a BMI ≥ 30 kg/m2 matched the 67% prevalence found in community-based patients with NAFLD.1 The development of a specific “obese probe” may improve the accuracy of TE in this subgroup of patients with NAFLD, which is particularly important given that obesity is a risk factor for fibrosis.6, 7
In summary, Wong and colleagues have provided valuable data regarding the use of TE in patients with NAFLD. Its strength appears to be for excluding advanced fibrosis and cirrhosis; however, there are a number of issues that need to be clarified before it is routinely used in the clinical setting. Its utility in obese and morbidly obese populations requires further validation, given the potential for unsuccessful acquisition rates and the promise of new “obesity probe” technology. Furthermore, cutoff values are likely to vary between centers due to differing patient populations, different underlying fibrosis prevalence, as well as interobserver variability which increases with obesity and hepatic steatosis.19 For example, Yoneda et al. determined a cutoff of 17.5 kPa to be optimal in predicting cirrhosis in a study of 97 Japanese patients with NAFLD20 compared to a range of 10.3–11.5 kPa in the Wong study. Lastly, if noninvasive markers are going to form part of the routine assessment of the millions of individuals with NAFLD, the expense and availability of each modality may play a decisive role in which noninvasive method is most appropriately taken up by community physicians and specialty hepatologists.