Challenges in Histological Endpoints for MASH Therapies: An Exercise in Statistical Modelling

ABSTRACT Background Regulatory‐accepted efficacy endpoints for nonalcoholic steatohepatitis (NASH; recently updated to metabolic‐dysfunction associated steatohepatitis, MASH) clinical trials include fibrosis improvement with no worsening of NASH or NASH resolution with no worsening of fibrosis determined by liver biopsy using the NASH Clinical Research Network criteria. These endpoints involve the scoring of four liver histology parameters, all of which are associated with significant inter−/intra‐reader variability. Since few trials have shown positive results with these endpoints, we evaluated the effects of imprecision in histologic scoring on trial results from a statistical perspective. Methods Estimating the probability (sensitivity) of accurately scoring histology is based on the relationship between measures of agreement and sensitivity. We simulated kappa values for a range of sensitivities. Then, using published kappa values from NASH trials, we selected corresponding sensitivities for histology parameters. Finally, simulations assuming a range of “overscore” and “underscore” probabilities were conducted to estimate the dilution of the true effect size. Results Simulations for 2‐arm trials with sample sizes of 400 (mix of stage 2/3 fibrosis) subjects showed ~50% dilution of the true effect size for both approvable endpoints due to scoring imprecision. Such dilution remains constant regardless of sample size. Conclusion Imprecise histologic scoring disproportionately impacts the ‘superior’ arm as the error is proportional to the true response rate. This dilution of effect size should be considered when weighing the clinical benefit and the overall risk–benefit profile in the review of NASH studies. This argues for the adoption of non‐invasive biomarkers rather than histologic endpoints.