COMPARISON OF THE EFFECTS OF SEMAGLUTIDE ON LIVER HISTOLOGY IN PATIENTS WITH NON-ALCOHOLICS STEATOHEPATITIS CIRRHOSIS BETWEEN MACHINE LEARNING MODEL ASSESSMENT AND PATHOLOGIST EVALUATION (VP028)

Background: In trials of non-alcoholic steatohepatitis (NASH), liver biopsies are evaluated by pathologists; however, machine learning (ML) has shown promise in assessing biopsies. PathAI has developed ML models to provide such assessments. This post-hoc analysis included data from a randomized, double-blind, placebo-controlled phase 2 trial (NCT03987451) of patients with compensated NASH cirrhosis to study the effects of semaglutide on individual histological components of NASH assessed by the ML models.

METHODS AND RESULTS: Patients with NASH fibrosis stage 4 were randomized to 48 weeks’ subcutaneous semaglutide 2.4 mg or placebo once weekly. Liver biopsies were obtained at baseline and week 48. After assessment by a single pathologist, biopsies were subsequently digitized for ML evaluation (models deployed December 2021). Evaluations included changes in categorical (pathologist and ML) and continuous (ML only) scores for fibrosis, hepatocyte ballooning, lobular inflammation, and steatosis.
Of 71 patients in the study, 70 had baseline ML results. Percentages of patients with improvement in histology were: fibrosis (semaglutide vs placebo) 13% vs 33% (pathologist) and 15% vs 33% (ML), alignment 73%, Kappa 0.42; ballooning (semaglutide vs placebo) 55% vs 33% (pathologist) and 46% vs 21% (ML), alignment 81%, Kappa 0.64; inflammation (semaglutide vs placebo) 43% vs 38% (pathologist) and 33% vs 13% (ML), alignment 51%, Kappa 0.25; and steatosis (semaglutide vs placebo) 45% vs 33% (pathologist) and 33% vs 4% (ML), alignment 54%, Kappa 0.25. For continuous ML scores, a significant estimated treatment difference (ETD) for semaglutide vs placebo was seen for ballooning (ETD −0.51; 95% confidence interval [CI] −0.90, −0.11; p=0.0120) and steatosis (ETD –0.50; 95% CI –0.84, –0.15; p=0.0047) but not fibrosis (ETD 0; 95% CI –0.24, 0.24; p=0.9884) or inflammation (ETD −0.22; 95% CI −0.48, 0.04; p=0.1030).


Conclusion: In patients with compensated NASH cirrhosis, the effects of semaglutide were consistent across pathologist and ML evaluation, with a lower placebo response measured by ML for inflammation and steatosis and highest concordance between methods for fibrosis and ballooning. Consistent with a phase 2 trial of semaglutide in patients with NASH and F1–F3 fibrosis, semaglutide improved histology assessed by pathologist and ML in the present cirrhotic cohort and ML continuous scoring delivered additional insights to the categorical scoring system.