Society: AASLD

LIVE STREAM SESSION

Background: Current non-invasive tests (NITs) utilized to screen and predict liver fibrosis were developed in predominantly Caucasian populations, and their performance in non-Caucasians is not well known. The purpose of this study is to evaluate the performance of NITs compared to Transient Elastography (TE) across different ethnic groups from a large diverse national dataset.

Methods: Data were derived from the National Health and Nutrition Examination Surveys (NHANES) 2017-2020, which included a total of 7,767 adults with valid TE measurements. Participants with excessive alcohol consumption, alternative etiologies of liver disease, and/or steatogenic medications were excluded. Patients of multiracial origin or without ethnic information were also excluded. Liver stiffness measurements ≥12 kPa were used to define presence of advanced fibrosis. Performance of the Fibrosis-4 (FIB-4) Index, NAFLD Fibrosis Score (NFS) and the AST to Platelet Ratio Index (APRI) were assessed by calculating the area under the receiver operator characteristic curves (AUROC).

Results: A total of 6,294 adults were included. Participant demographics are shown in Table 1. The prevalence of advanced fibrosis was 3.0% in the overall population based on TE. Prevalence of advanced fibrosis was higher in Caucasian and Mexican-American participants compared to non-Hispanic Black and Asian participants (3.9% vs. 3.5% vs. 2.1% vs. 1.7%, respectively, p=0.003). The FIB-4 and APRI performed significantly worse in non-Hispanic Blacks compared to other ethnicities (Table 2). FIB-4 performed significantly better in Mexican-American and Asian participants and APRI had the highest AUROC among Asian participants. The NFS performed similarly across all ethnicities (Table 2) but tended to overestimate fibrosis in all groups compared to TE. Further analysis revealed that among participants with advanced fibrosis, non-Hispanic Black participants had significantly lower plasma AST and ALT compared to all other ethnic groups. No differences were observed in age, platelets, or albumin in non-Hispanic Black participants with advanced fibrosis compared to other ethnic groups.

Discussion: In a large diverse national dataset, the performance of laboratory based NITs compared to TE showed significant differences across ethnic groups, with significant worse performance in non-Hispanic Blacks. Our findings suggest this may be related to non-Hispanic Blacks with advanced fibrosis having lower AST and ALT levels compared to other ethnic groups. Given that NITs are now widely recommended for use in screening patients for advanced fibrosis, it is imperative that the scores are equitable across ethnic groups. Prospective studies to validate these findings against liver biopsy are warranted.

Background
Recent literature raises concern that Fibroscan may be inaccurate in patients with elevated BMI and NAFLD/NASH. However, FIB-4 appears to perform well in accuracy regardless of BMI stage. Therefore, it is unclear whether existing algorithms of FIB-4 followed by Fibroscan for routine liver fibrosis screening performs well in patients with higher BMIs. We aimed to assess the degree of discordance between FIB-4 and Fibroscan stratified by BMI and how well either test performs relative to biopsy.

Methods
We performed a retrospective review of the National Health and Nutrition Examination Survey (NHANES) 2017 – March 2020 database. We included all patients between the age of 35-65 who had completed Fibroscan and lab testing sufficient to calculate FIB-4 (AST, ALT, and platelets). Patients who were pregnant, consumed excessive alcohol, or had Hepatitis B or C were excluded. The primary outcome was disagreement between FIB-4 and Fibroscan (i.e. if one test reported low risk for advanced fibrosis and the other test reported high risk). We used FIB-4 cutoffs of 1.3 and 2.67 to separate low, intermediate, and high risk for advanced fibrosis, and 9.7 kPa from Fibroscan to separate low and high risk. We then compared our findings to patients locally with biopsy-proven NAFLD/NASH. Statistical analysis was done with SAS 9.4.

Results
A total of 3085 patients from the NHANES database were included. Average age was 50.3 and 54.2% were female. Average BMI was 30.2. Patient demographics are included in Figure 1A and breakdown of % of patients in each FIB-4 and Fibroscan category are listed in 1B. There was overall 4.78% disagreement. Increasing BMI was significant associated with heightened disagreement, with only 0.87% disagreement at BMI<25 going up to 25.1% at BMI>40 [Figure 1C] and odds ratio of 1.15 for disagreement on logistical regression (1D, p<0.0001).

In comparison, at our center between 2018-2022, we included 241 patients. Average BMI was higher at 35.3 with higher FIB-4 and Fibroscan values [Figure 2A]. Disagreement occurred 23.7% of the time and trended similarly when stratified by BMI [Figure 2B]. Out of 57 cases in which disagreement occurred, 51 (89.5%) were due to Fibroscan calling high risk for advanced fibrosis whereas FIB-4 called low risk. In 43/57 (75.4%) cases, FIB-4 was more accurate, and the majority (95.3%) were due to Fibroscan inaccurately labelling advanced fibrosis. For the 82 patients with intermediate risk of advanced fibrosis on FIB-4, Fibroscan was incorrect in 25 (30.5%) cases [see Figure 2C for stratified by BMI].

Conclusions
Disagreement between FIB-4 and Fibroscan increased at higher BMIs, with FIB-4 being more accurate ¾ of the time. Most disagreement is due to Fibroscan overcalling advanced fibrosis. More nuanced algorithms for liver fibrosis screening may be warranted for patients with higher stages of obesity.

Background:
In patients with non-alcoholic fatty liver disease (NAFLD), liver fibrosis stage 2 or higher (F2+) indicates fibrotic non-alcoholic steatohepatitis (NASH) and portends poor long-term prognosis, such as liver-related and overall mortality. Non-invasive detection of F2+ is more difficult compared to advanced fibrosis; whereas existing models such as FIB-4 and NFS (NAFLD fibrosis score) were optimized for F3+. In this work, we applied novel machine learning (ML) methods to distinguish F2+ from earlier stage fibrosis (F0-1) in patients with NAFLD.

Methods:
The Superlearner is an ensemble method that combines many individual ML (‘base’) models into a single ‘super’ model using an optimal weighted combination of the base models. Two non-overlapping data sets consisting of biopsy-proven NAFLD patients from the NASH CRN were analyzed including (1) subjects in the NASH CRN observational study (n=635) and (2) participants in a randomized trial for NASH (‘FLINT’ trial, n=280). The former was used to train, cross-validate and tune the Superlearner and the latter to independently validate the model, in comparison to existing models, such as FIB-4, NFS, Forns, APRI, BARD, and SAFE, using the area under the receiver-operating characteristics curve (AUROC). These analyses were implemented with the R programming language.

Results:
In the training and validation data sets, F2+ was present in 45% and 59%, respectively. The Superlearner employed 90 base models, based on 22 demographic, clinical and laboratory variables, including multivariate adaptive polynomial spline regression (relative model weight: 0.423), neural network (0.135), generalized boosted model (0.134), random forest (0.075), support vector machine (0.057), regularized generalized linear model (0.055), Bayesian generalized linear model (0.033), generalized linear model (0.032), multivariate adaptive regression splines (0.026), bagging trees (0.022), and recursive partitioning tree (0.007). In the Figure, the AUROC for diagnosing F2+ of the Superlearner model is compared against existing models in the validation data set. The Superlearner performed the best, except for the SAFE (steatosis-associated fibrosis estimator), which was recently developed and validated for the diagnosis of F2+ (Hepatology 2022, https://doi.org/10.1002/hep.32545).

Conclusion:
The Superlearner, thought of as the “best-in-class” machine learning prediction, performed better than existing models commonly used in practice, such as FIB-4, in detecting fibrotic NASH. In addition, the SAFE score performed at least as well as the Superlearner, which may be helpful as an initial test in the evaluation of patient with NAFLD.

Introduction: Accurate non-invasive tests (NITs) for identification of cirrhosis and portal hypertension are highly desirable in patients with NAFLD. Since spleen stiffness measurement (SSM) correlates with severity of portal HTN, we assessed the performance of SSM for detection of NAFLD cirrhosis and its portal hypertensive manifestations in comparison to other NITs [(liver stiffness measurement (LSM), aspartate aminotransferase-to-platelet ratio index (APRI), and fibrosis-4 (FIB4)].
Methods: From a prospective database of patients undergoing FibroScan 630 expert, we identified 154 individuals with known NAFLD diagnosed through imaging and/or a liver biopsy who also underwent an upper endoscopy within the year. Of these 154 patients, 125 had a valid SSM while 146 had a valid LSM. Area under the receiver operating characteristic (AUROC) analyses were used to compare the ability of SSM, LSM, APRI, and FIB4 to discriminate participants with cirrhosis, esophageal varices, high-risk esophageal varices, and portal hypertensive gastropathy.
Results: The median age was 59 years, 46% were males, 88% were Caucasian, and the mean BMI was 34.6 kg/m². The prevalence of cirrhosis was 60.4% in this cohort. For discriminating cirrhosis, APRI had the largest AUROC (0.90), FIB4 had the highest sensitivity (93.8%) and NPV (76.2%), while SSM, LSM, and APRI had a similar specificity (87.5%), and SSM had the best PPV (93.5%). For discriminating esophageal varices, FIB4 had the highest AUROC (0.79), sensitivity (96%), and NPV (91.3%) while LSM had the greatest specificity (90.3%) and PPV (76.9%). For high-risk esophageal varices, SSM had the largest AUROC (0.85), sensitivity (83.3%), specificity (76.9%), PPV (73.3%), and NPV (78.6%) (Table 1 and Figure 1).
Conclusion: Compared to other NITs, SSM is most useful for detecting high risk esophageal varices.

Figure 1. Area Under the Curve for Identifying NASH Cirrhosis and Portal Hypertensive Complications

Figure 1. Area Under the Curve for Identifying NASH Cirrhosis and Portal Hypertensive Complications

Table 1. Detection of NASH Cirrhosis and Portal Hypertensive Manifestations

Table 1. Detection of NASH Cirrhosis and Portal Hypertensive Manifestations

Dheeksha Ranginani

Appearances