PK extrapolation from animal to man: the good, the bad and the ugly. An overview of the performance of different methods applied across the project portfolio at Roche
H.M. Jones(1), K. Jorga(2), T. Lavé(1)
Preclinical (1) and Clinical (2) Modeling & Simulation, F. Hoffmann-La Roche AG, Basel, Switzerland
Background: In order to reduce failures related to pharmacokinetic (PK) issues and to determine the suitability of compounds for an intended dosing regimen, it is important to predict human PK as early as possible. A variety of different approaches are available for this purpose. Allometric scaling has been traditionally used, however recently mechanistic physiologically based PK (PBPK) models have been developed and made more useful and attractive to the pharmaceutical industry.
Objectives: The aim of this work was to compare empirical and PBPK approaches for the prediction of human PK for 19 Roche compounds that reached clinical development in the last 5 years and to identify the main limitations of the current models. In addition the PK modeling approach was compared to the mg/kg or body surface area (BSA) approach for starting dose determination.
Results: PBPK approaches gave more realistic predictions than the classical empirical methods for all 19 compounds; a greater proportion of the predicted parameters (e.g. Cmax, AUC, t1/2) and plasma concentrations were within 2-fold error of the observed values. For example, AUC was relatively well predicted with 76% of compounds within 2-fold error of the observed value using PBPK compared to only 42% using the Dedrick approach. For Cmax, there was a systematic tendency to underestimate the observed value, and the overall success rate was only 47% using PBPK approaches.
Any poor predictions were generally as a result of biliary elimination and/or enterohepatic recirculation processes that were not incorporated into the model.
Conclusion: In addition to improved prediction accuracy, PBPK approaches offer more potential in the early stages of the drug development process. However the prediction of some parameters such as Cmax remains challenging, for a number of reasons. The lack of prediction models for processes such as biliary elimination and active transport processes still limits the accuracy of the predictions and increases the level of uncertainty.