Iris K. Minichmayr (1), Elodie Plan (1), Benjamin Weber (2), Sebastian Ueckert (1)
(1) Department of Pharmaceutical Biosciences, Uppsala University, Uppsala, Sweden (2) Translational Medicine and Clinical Pharmacology, Boehringer Ingelheim Pharmaceuticals, Inc., Ridgefield, Connecticut, USA
Objectives: Nonalcoholic fatty liver disease (NAFLD), including its progressive subtype nonalcoholic steatohepatitis (NASH), affects 20-40% of the population in developed countries. Fuelled by the obesity epidemic, NAFLD is alarmingly increasing worldwide and has emerged as a leading indication for liver transplantation. To date, there are no approved drug treatments for NASH and NAFLD. The development of novel therapies has been challenged by the complex, ‘multi-hit’ pathophysiology of NAFLD, inconsistent diagnostic criteria and lack of clarity about appropriate treatment endpoints [1,2]. Currently, NASH diagnosis and FDA-recommended clinical trial endpoints heavily rely on liver biopsies and a histological scoring system for NAFLD comprising 14 features [3,4]. Its most widely and often isolatedly assessed key metrics are (i) the NAFLD activity score (NAS), considering steatosis, lobular inflammation and hepatocellular ballooning, and (ii) the fibrosis stage. The objective of this study was to enhance the understanding of disease processes underlying NAFLD and their relationships with key histological features, and to assess the role of different histological metrics in supporting NASH diagnosis by using item response theory (IRT) modelling.
Methods: Data. The analysis dataset originated from the public National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) NAFLD Adult database [5]. For each patient (nID=914), the first (and often only) available histological assessment was considered. The population spanned the full spectrum of NAFLD (NAS 0-8) and liver fibrosis (0-4) and comprised 52.3% patients with a definite positive and 26.8% with a definite negative diagnosis of NASH.
IRT model. For each item, i.e. histological criterion for liver assessment, a model was developed in accordance with the binary or ordered categorical nature of the data. The models related the probability of the outcome of each item to one or more latent variables, which represent unobserved ’hidden’ disease processes underlying the item responses. Both exploratory models, implying no clear hypothesis about the structure of the item response data, and confirmatory models, assigning the model items to specific latent variables, were developed. The structure of the confirmatory models was guided by cluster dendrograms based on results of the exploratory models. Diverse tools were used for model evaluation, e.g. simulations to assess correlations between items. All modelling activities were performed in R3.6.1 using the R package mirt [6].
Results: Initial model development focused solely on the 3 NAS items steatosis, inflammation and ballooning. The analysis was subsequently extended to 13 histological features suggestive of steatosis, inflammation, hepatocellular injury and other aspects like fibrosis [3]. The 13-item model comprised 5 graded-response and 8 two-parameter logit models. The model allowed for a better distinction between NASH and Non-NASH patients than the sparse 3-item (NAS) model, and a more detailed characterisation of the disease processes underlying NAFLD. Cluster dendrograms and confirmatory models suggested different disease processes (i.e. separate latent variables) for the four cardinal features of NAFLD, i.e. the 3 NAS components—describing reversible features of active injury—and fibrosis, a generally irreversible lesion. Model evaluation revealed adequate performance of the developed multi-dimensional IRT model. Highest correlation (70%) was found between processes reflected by hepatocellular ballooning and fibrosis. The latent variables considering hepatocellular ballooning and fibrosis appeared to best inform NASH diagnosis, followed by the disease processes underlying lobular inflammation, and then steatosis.
Conclusions: An item response theory model based on hepatic histological scoring allowed to jointly characterise disparate disease processes underlying NAFLD, including more rapidly changing and reversible aspects like steatosis, as well as the more slowly progressing and generally irreversible characteristic fibrosis. Our analysis suggests that NASH could be inferred better from a broad panel of histological lesions than from the NAS alone. The model lays the basis for future investigations on the sensitivity of the NAS to changes of different disease processes, e.g. as response to a therapeutic intervention, and on potential non-invasive biomarkers reflecting NAFLD activity.
References:
[1] Wong VW, Chitturi S, Wong GL, Yu J, Chan HL, Farrell GC. Lancet Gastroenterol Hepatol. 2016; 1(1):56–67.
[2] Rinella ME, Tacke F, Sanyal AJ, Anstee QM. J Hepatol. 2019; 71(4):823–833.
[3] Kleiner DE, Brunt EM, Van Natta M, et al. Hepatology. 2005; 41(6):1313–1321.
[4] Food and Drug Administration. Draft Guidance for Industry: Noncirrhotic Nonalcoholic Steatohepatitis With Liver Fibrosis: Developing Drugs for Treatment. June 2019.
[5] Neuschwander-Tetri BA, Clark JM, Bass NM, et al. Hepatology. 2010; 52(3):913–924.
[6] R. Philip Chalmers (2012). mirt: A Multidimensional Item Response Theory Package for the R Environment. Journal of Statistical Software, 48(6), 1-29. doi:10.18637/jss.v048.i06.
Reference: PAGE () Abstr 9406 [www.page-meeting.org/?abstract=9406]
Poster: Drug/Disease Modelling - Other Topics