A Bayesian approach for the integration of preclinical information into a PBPK model for predicting human pharmacokinetics
Bizzotto R, Nucci G, Poggesi I, Gomeni R
Clinical Pharmacokinetics/Modeling&Simulation, GlaxoSmithKline, Verona, Italy
Objectives: To evaluate the performance of a Bayesian approach for integrating preclinical in vitro and in vivo information into a PBPK model to predict human pharmacokinetics.
Methods: A basic, generic whole body physiologically based pharmacokinetic model was implemented in rats, dogs and humans [1]. The predictive performance of the basic model in these species was initially evaluated using a dataset of 23, 21 and six compounds for rats, dogs and humans, respectively. Since in vivo pharmacokinetic data in animals are always generated before the first study in humans, our aim was to take advantage of this information to improve the predictive performance of the basic PBPK. Our proposal is to use a Bayesian approach to estimate few critical parameters of the PBPK model, to provide a better adherence to the observed pharmacokinetics in rats and dogs, and to eventually use these parameters for predicting the PK in humans. For the application of this proposal the Bayesian parameters identification as implemented in SAAM II was used.
Results: The basic model, applied to the preclinical species showed reliability similar or better than that reported in the literature. Average fold-errors for the main pharmacokinetic parameters were lower than 2, with 91% of the compound parameters predicted within a 3-fold error. The proposed approach increased the reliability of the prediction of the pharmacokinetic data in humans. For the six compounds, for which human oral pharmacokinetic data were available, the average fold-errors for the systemic exposure markedly decreased from 3.6 (basic) to 1.9 (new approach).
Conclusions: During the drug development process, incremental in silico, in vitro and in vivo information is gained before a candidate drug is effectively given to humans. The basic PBPK model proposed by Poulin [1] may be based on too simple assumptions and limited input data for providing predictions in human with the desired level of reliability. In this work the PBPK model prediction were refined applying a Bayesian approach to fine tune few critical PBPK model parameters based on in vivo animal data. If these promising results will be confirmed on a more extensive dataset of compounds, this approach will strongly improve the design and safety of first time in human studies.
References:
[1] Poulin P, Theil FP. J Pharm Sci 2002, 91:1358-1370.
