Can Noninvasive Test Dynamics Serve as Surrogate End Points for Histologic Improvement in Metabolic Dysfunction-Associated Steatohepatitis? A Systematic Review and Bayesian Meta-Analysis.

Abstract

Background & aims: Trials on metabolic dysfunction-associated steatohepatitis (MASH) use biopsy for efficacy assessment, despite invasiveness and cost. We aimed to assess the trial-level surrogacy and association of changes in noninvasive tests (NITs), namely liver stiffness measurement (LSM), liver fat content (LFC) by magnetic resonance imaging, Fibrosis-4 score, and Enhanced Liver Fibrosis, with histology in MASH.

Methods: We searched PubMed, Embase, Cochrane Central, and ClinicalTrials.gov up to February 2026 for MASH randomized controlled trials reporting serial NIT data and histologic end points (fibrosis improvement without MASH worsening or MASH resolution without fibrosis worsening). NIT trial-level surrogacy of histologic improvement was quantified through squared Bayesian posterior correlation of treatment effects.

Results: Thirty-five randomized controlled trials and 3 post hoc analyses were included. At the trial level, treatment effects on all NITs showed probable weak correlations with treatment effects on histological improvement, with the exception of LFC, which displayed probable moderate surrogacy for MASH resolution without fibrosis worsening (R² = 62%; 95% Credible Interval [CrI], 10%-95%). In contrast, variations in LSM were near-certainly associated with fibrosis improvement without MASH worsening in F4-excluded studies (odds ratio [OR], 2.67; 95% CrI, 1.33-5.12 per 30% decrease) and near-certainly associated with MASH resolution without fibrosis worsening (OR, 2.46; 95% CrI, 1.16-4.66 per 30% decrease). LFC was near-certainly associated with MASH resolution without fibrosis worsening (OR, 1.92; 95% CrI, 1.14-3.20 per 30% decrease).

Conclusions: LFC and LSM dynamics were associated with histologic response at the outcome level, but surrogacy was limited and imprecise across biomarkers. This supports a lack of consistent trial-level surrogacy, although context-specific utility remains plausible.