Nikolaos Tsamandouras (1), Amin Rostami-Hodjegan (1,2), Aleksandra Galetin (1) and Leon Aarons (1)
(1) Centre for Applied Pharmacokinetic Research, University of Manchester, UK, (2) Simcyp Ltd, Sheffield, UK
Objectives: Bayesian approaches are particularly useful for population data analyses with physiologically-based pharmacokinetic (PBPK) models, as often the fitted data are insufficient to inform the estimation of all model parameters and prior knowledge needs to be utilised. In NONMEM 7, an algorithm for MCMC Bayesian analysis is available and thus may serve as an alternative to WinBUGS in which implementation of complex models has always been a challenging task. The purpose of this work is to evaluate the performance of the Bayesian MCMC method in NONMEM in the specific context of PBPK modelling.
Methods: Diazepam population PK data that have been previously successfully analysed in WinBUGS with a whole-body PBPK model [1] were re-analysed with the Bayesian MCMC method in NONMEM and parameter estimates between the two platforms were compared. Specifically, diazepam plasma concentrations were available for 12 individuals after a 7mg IV infusion. Prior information from pre-clinical species (rat) was utilised to facilitate the estimation of the tissue-to-plasma partition coefficients [1]. Additional practical issues regarding application of the MCMC algorithm in NONMEM for Bayesian PBPK modelling were also explored (e.g. selection of subroutine, assessment of chain convergence). Finally, the NONMEM generated posterior distributions were superimposed with the maximum a posteriori (MAP) point estimates (FOCE-I in conjunction to the prior functionality) [2] to assess the degree of agreement.
Results: Comparable results were obtained across the two platforms. As expected only the model parameters to which the plasma output is sensitive were substantially updated from the priors. It was also illustrated that in complex physiologically-based systems (often stiff), the selection of the differential equation solver (subroutine) is particularly critical for the computation time of the Bayesian analysis (ADVAN13 exhibited the best performance). In addition, it was evident that MCMC convergence in such complex models should be ideally assessed with multiple-chain diagnostics (e.g. Rubin-Gelman diagnostic). Finally, it was demonstrated that the modes of the NONMEM generated posterior densities can be very accurately captured by MAP estimation in the same platform with significant reductions in computation time.
Conclusions: The results of this work are providing further confidence for future use of NONMEM for Bayesian PBPK analyses of population data.
References:
[1] Gueorguieva I, Aarons L, Rowland M. Diazepam pharamacokinetics from preclinical to Phase I using a Bayesian population physiologically based pharmacokinetic model with informative prior distributions in Winbugs. J Pharmacokinet Pharmacodyn (2006) 33(5):571-594.
[2] Langdon G, Gueorguieva I, Aarons L, Karlsson M. Linking preclinical and clinical whole-body physiologically based pharmacokinetic models with prior distributions in NONMEM. Eur J Clin Pharmacol (2007) 63(5):485-498.
Reference: PAGE 24 (2015) Abstr 3397 [www.page-meeting.org/?abstract=3397]
Poster: Drug/Disease modeling - Absorption & PBPK