Bayesian POP-PK analysis of exposure data from a Phase IIb clinical trial
Henry Pertinez, Leon Aarons
University of Manchester
Objectives: As part of an ongoing PhD project on modelling the pharmacokinetics of “Drug X”, a plasma exposure dataset derived from a phase IIb clinical trial investigating the safety and efficacy of “Drug X” is being analysed. Simulation of the phase IIb study plasma profiles using parameters derived from a 3-compartment empirical model fitted to earlier phase I clinical trial data, revealed that while the steady state exposure during the dosing period was adequately described, an extended terminal phase only visible in the longer phase IIb timecourse was not. This phase is likely to reflect long term, “deep” distribution of “Drug X”, which is of particular interest to describe accurately as it has the potential to influence “Drug X”s pharmacological effects. Furthermore, an accurate description of this terminal phase will be necessary if long term predictions of exposure are required, and also for development of a forcing function to allow open loop PBPK modelling of plasma and tissue data from the phase IIb clinical trial.
Methods: A Bayesian analysis using the WinBUGS software package was carried out to allow the information derived from the phase I studies to be carried forward to allow modelling of the sparsely sampled and noisy phase IIb data and so allow for a more appropriate description of the terminal phase. The richly sampled phase I data consisted of single dose IV infusion data (28 day profile, n=15), single dose PO tablet (28 day profile, n=24 ) and BID dosing PO tablet data (n=10 at 4 dose levels, BID dosing for 25 days with overall profile of 53 days). The pop-PK of these datasets was analysed in a single combined run in WinBUGS using a 3-compartment disposition model with saturable absorption to account for the nonlinear bioavailability seen in the 4 dose levels of the BID PO data. The parameters derived from this analysis were then used as informative priors for the mixed effects modelling of the phase IIb study dataset (n=80 at 3 dose levels, PO tablet BID dosing for 2 years with overall sampling profile of 3.5 years). A 4-compartment disposition model with saturable absorption was used in this analysis to allow the long-term timepoints of the phase IIb data timecourse to be described by an additional exponential phase, with informative priors used on the nested parameters of the model common to the prior 3-compartment analysis. After preliminary investigation, due to the lack of information in the phase IIb profiles on the earlier phases of the PK profile of “Drug X”, the WinBUGS “CUT” function was used so that the phase IIb data would only be used to allow estimation of the parameters related to the 4th exponential phase of the model.
Results: The initial results of the Bayesian analysis are satisfactory (as assessed by visual predictive check), providing a description of the phase IIb dataset that captures the long terminal phase seen on this timescale, while remaining consistent with the modelling of the shorter timescale studies. This improved description of the data, will be more appropriate as a forcing function for future PBPK modelling efforts. Issues remain however regarding poor convergence of the WinBUGS MCMC sampling chains in the analysis and work is ongoing to investigate various options to improve this convergence.
References:
[1] Fryback, D. G., Stout, N. K. & Rosenberg, M. A. An elementary introduction to Bayesian computing using WinBUGS. Int J Technol Assess Health Care 17, 98-113 (2001).
[2] Lunn, D., Best, N., Spiegelhalter, D., Graham, G. & Neuenschwander, B. Combining MCMC with 'sequential' PKPD modelling. J Pharmacokinet Pharmacodyn 36, 19-38 (2009).
