Multimodality assessment of hepatic fibrosis: ranked paired reading and artificial intelligence identifies fibrosis improvement with aramchol missed by conventional staging

Abstract

Background and aims: Aramchol is a partial inhibitor of hepatic stearoyl-CoA desaturase with direct anti-fibrotic activity in preclinical models and histological improvement in a phase 2b trial. This open-label study explored the speed and extent of fibrosis reduction. We compared different methodologies of fibrosis scoring to optimize the design of a registrational placebo-controlled investigation. Method: 46 Patients (pts) with NASH and fibrosis (28 F3, 11 F2, 7 F1) documented by biopsy were randomized 1:1:1 to receive Aramchol 300 mg BID and underwent a control biopsy at weeks 24, 48 or 72. Biopsies were read by 3 independent pathologists individually, followed by a consensus reading, which determined the final NASH CRN scoring. Three different assessments of the antifibrotic effect were studied on the same slides: 1) a ≥1 stage reduction by NASH CRN; 2) a ranked assessment (improvement/worsening/stable) of paired (pre and post baseline) biopsies, blinded to sequence; 3) an automated and continuous score of Fibrosis Composite Severity (FCS), using FibroNest™, a quantitative digital pathology image analysis and artificial intelligence (AI) Method: a 0.3 reduction in FCS (4 fold higher than the analytical variability) identified any reduction in fibrosis; a 25% relative decline in FCS, a strong reduction in fibrosis. Results: Control biopsies were performed for 26, 15 and 5 pts at 24, 48, and 72 weeks, respectively. Mean (sd) baseline FCS was 5.05 (1.05). Table shows greater fibrosis improvement with longer duration of therapy for both conventional histology and digital pathology readings. Mean FCS reduction was −0.62 (p = 0.017) at Wk24 and −1.74 (p < 0.0001) at Wk≥48. AI evaluation was consistent with paired reading in 21/24 (87.5%) of the pts with fibrosis improvement. When analyzed by AI, 17/23 pts with unchanged NASH CRN stages had any fibrosis response, including 7 with a strong response. Similarly, 13/17 pts with stable ranking had a fibrosis response, including 5 with a strong reduction. No pts with worse CRN stages or worsening ranking had a strong AI fibrosis reduction. Fibrosis reduction Wk24, N = 26% (N) Wk ≥ 48, N = 20% (N) By NASH CRN ≥1 stage 27% (7) 40% (8) By ranked assessment 42% (11) 65% (13) By AI reading, any (delta FCS ≥0.3) 58% (15) 100% (20) By AI reading, strong (-25% FCS) 27% (7) 65% (13) Conclusion: Aramchol resulted in a high proportion of fibrosis improvement using three separate biopsy reading methodologies, with a larger treatment effect with longer duration of therapy. Both ranked assessments and AI evaluations identified more subjects with fibrosis improvement, indicating greater sensitivity to change vs categorical scoring. Digital pathology quantification by AI reveals a high level of fibrosis improvement that would have been missed by conventional histological measurements. AI technologies are promising for the detection of fibrosis changes in future clinical trials.