Bosch J, Chung C, Carrasco-Zevallos OM, et al. - In nonalcoholic steatohepatitis (NASH) patients with cirrhosis, a machine learning (ML) model based on trichrome-stained liver biopsy slides can anticipate clinically significant portal hypertension (CSPH).

• A phase 2b trial included NASH patients with compensated cirrhosis.

• By morphometry, the ML hepatic venous pressure gradient (HVPG) score was more significantly linked to HVPG than hepatic collagen.

• The ML HVPG score distinguished patients with normal (0-5 mmHg) from those with increased HVPG (5.5-9.5 mmHg) with CSPH (median: 1.51 vs 1.93 vs 2.60).

• In the training and test sets, the AUROCs of the ML HVPG score for CSPH were 0.85 and 0.76, respectively.

• The addition of an ML parameter for nodularity, ELF, platelets, AST, and bilirubin enhanced the discrimination of the ML HVPG score for CSPH.

• While the baseline ML HVPG score was not predictive of clinical outcomes, alterations were linked with improved hemodynamic response and fibrosis.