Key requirements for the translation of computational models are: i) reproducibility of results; ii) reusability and extensibility of models; iii) availability of data; and iv) strategies for stratification and individualization of models. Here we will present a modeling workflow focused on these key aspects applied to liver function tests.
Assessment of liver function is a key task in hepatology but accurate quantification of hepatic function has remained a clinical challenge. Dynamic liver function tests are a promising tool for the non-invasive evaluation of liver function in vivo. These clinical tests evaluate the function of the liver via the clearance of a given test substance, thereby providing information on the metabolic capacity of the liver. We modeled these tests via whole-body physiological models of absorption, distribution, metabolization and elimination using multi-scale SBML models (core and comp package).
One class of such tests are breath tests based on the conversion of 13C-labeled substrates by the liver to 13CO2 subsequently measured in the breath. A commonly applied substrate is 13C-methacetin, converted to paracetamol and 13CO2 via cytochrome P450 1A2 (CYP1A2), used orally in the methacetin breath test (MBT) and intravenously in the LiMAx test. An important clinical question is which factors can affect MBT and LiMAx results. The aim of our study was to answer this question using computational modeling to derive basic information for a better understanding of the methacetin breath test and factors influencing its results.
A second example, the liver function test based on caffeine is known for long, but its clinical usability is hampered by large interindividual variability and dose-dependency. By applying a physiological based pharmacokinetics model (PBPK) for the evaluation of the caffeine clearance test, we were able to assess in silico the hepatic conversion of caffeine to paraxanthine via cytochrome P450 CYP1A2. The model is able to reproduce results from a wide range of reported studies under varying caffeine doses and application routes and accounts for interindividual differences based on distributions of CYP1A2 and modification of CYP1A2 activity via lifestyle factors, e.g. smoking, and pharmacological interactions, e.g. oral contraceptives. Validation was performed with an independent clinical trial (EudraCT 2011-002291-16, ClinicalTrials.gov NCT01788254) demonstrating an improved prediction using individualized models accounting for smoking status and contraceptive use. Hereby, we could reduce the large variability in the test results providing the basis for better sensitivity and specificity in diagnosing subjects with liver problems.