How to link biomarkers using physiologically based pharmacokinetic models.

Summary and comments ("Research Highlights" in News and Views of the journal "Biomarkers in Medicine") on : Combining the “bottom-up” and “top-down” approaches in pharmacokinetic modelling: Fitting PBPK models to observed clinical data. Nikolaos Tsamandouras, Amin Rostami-Hodjegan, Leon Aarons BJCP, Sep 2013, on line doi: 10.1111/bcp.12234 “How to link biomarkers: by using PBPK models” Translational Medicine approaches impose to use many biomarkers from in vitro ones to clinical endpoints. The integration and articulation of those biomarkers are typically organized by “modelling and simulation” techniques. In this article, the authors delineate the process by which they can combine different approaches of modelling, using the specific example of pharmacokinetics. The article summarizes well the 3 levels of complexity of modelling techniques. The first level is the empirical approach i.e. data are described by a mathematical model chosen on the “best fit”. The second level is best described by being between level 1 and level 3 i.e. it is a so-called “semi-mechanistic” model using in some parts an empirical approach and in other parts a mechanistic approach. The level 3 is the mechanistic model and in pharmacokinetics, it is the complex physiologically based pharmacokinetic (PBPK) model. In this case, all key pathways through which a test drug is expected to go in a living organism, is computed. Recent tendency in drug development is to use more the level 3. This was made possible by improvements in computers and softwares, by increased use of in silico methods and by development of in vitro in vivo extrapolation methods. Numerical sensitivity of a model is often assessed by sensitivity analysis. This type of analysis tries to evaluate the origin of the variability of the model output by assessing different sources of variation included in the model. Ultimately, there is the good advice that all PBPK models should be assessed on the plausibility of the physiological parameters and their assigned range of values. Correlation between parameters must be carefully examined. If there is a high correlation, one of the parameters should be quantified, using a physiologically plausible value. For parameter estimation, 2 aspects are underlined. The authors are suggesting that most of the parameters should be fixed from physiological values and only a few should be estimated by model optimization techniques. Additionally, the reliability of the estimate should be described and this is often a challenging task for non physiologically parameterized values. A very interesting clarification is made between the 2 terms, uncertainty and variability. “Variability refers to differences attributable to environmental or genetic factors” when “Uncertainty is variation that derives from errors in the experimental procedure, measurement, modelling and assumptions of the studied system”. A possibility to incorporate Bayesian statistics approaches into PBPK is discussed. One key advantage is to bring a distribution of probability perspective i.e. sort of variability description. However, a disadvantage is the need for high power, high speed computers and highly (or at least specifically) skilled scientists. Ultimately, good PBPK models are sometimes considered overparameterized because they include many different parameters. The whole body system, sometimes described, encompasses too many dimensions in comparison to the available data. Then, it is reasonable to use techniques such as lumping procedures and semi-mechanistic models instead. The conclusion is driven by common sense and proposes “to develop mechanistically sound models with clinical relevance”. In drug development, mechanistic approaches are often key to resolve difficult hurdles because they permit to identify the key parameter(s) driving the failure of a particular molecule