Modeling Reductions in Liver Fat: Comparing Noninvasive Tests to Magnetic Resonance Imaging–Proton Density Fat Fraction

Abstract

Background and Aims

Magnetic resonance imaging–proton density fat fraction (MRI-PDFF) is an accurate, noninvasive tool for diagnosing metabolic dysfunction–associated steatotic liver disease, but its feasibility is limited in routine clinical practice. We aimed to assess the clinical utility of alternative, cost-efficient approaches for assessing liver fat changes and their relationship with MRI-PDFF changes.

Methods

This is a secondary analysis of a phase 2a study that included adults with metabolic dysfunction–associated steatotic liver disease who received clesacostat, a selective, reversible inhibitor of acetyl-CoA carboxylase. In this secondary analysis, responders were defined as those in whom a ≥30% decrease in liver fat by MRI-PDFF was observed with clesacostat or placebo. Other endpoints were evaluated for their ability to predict MRI-PDFF responder status, including controlled attenuation parameter (CAP), liver enzymes (alanine aminotransferase, aspartate aminotransferase, and gamma-glutamyl transferase), metabolic dysfunction–associated steatohepatitis–related biomarkers (liver stiffness measurement by vibration-controlled transient elastography, cytokeratin 18-M30, and cytokeratin 18-M65), and markers of hepatic steatosis (hepatic steatosis index and fatty liver index). These relationships were investigated through correlation, univariate, and multivariable regression analyses.

Results

Of 260 participants with a baseline and on-treatment measure at week 12 or week 16, 143 were responders. Based on correlation analyses, a significant but weak positive correlation between MRI-PDFF and CAP measurements of relative percentage change from baseline in liver fat was observed. By combining the selected 6 parameters (CAP, hepatic steatosis index, fatty liver index, alanine aminotransferase, gamma-glutamyl transferase, and cytokeratin 18-M65) through multivariable regression modeling, responders can be predicted with a high level of sensitivity and specificity (mean area under the receiver operating characteristic curve = 0.831 from 10-fold cross-validation).

Conclusion

Modeling multiple noninvasive assessments of liver fat closely aligned with MRI-PDFF measurements. These data support further assessment of its suitability in real-world clinical practice.