Silvia Maria Lavezzi (1), Jianping Zhang (1), Paolo Magni (2), Giuseppe De Nicolao (2), Laura Iavarone (1)
(1) Quantitative Clinical Development Department, PAREXEL International, (2) Department of Electrical, Computer, and Biomedical Engineering, University of Pavia, Italy
Objectives: Joint pharmacokinetic (PK) modelling of a parent drug and its metabolites is fundamental to evaluate absorption, distribution, metabolization, and elimination of all analytes. This is particularly relevant when metabolites are formed during first pass through the liver and may be pharmacologically active or lead to toxicity episodes. Different joint drug-metabolites PK models have been proposed in the literature: in our study, a priori and a posteriori identifiability analyses were performed on published models including first passage through the liver [1,2].
Methods: Three models were evaluated: Model 1, with one depot per analyte [1]; Model 2, with a single depot and different absorption rates for each analyte [1]; Model 3, with a single depot and absorption rate to a liver compartment [2].
A priori identifiability was explored (similarly to [3]), via DAISY [4], assuming the drug is given intravenously (IV) and orally, and parent responses to both IV and oral administration are available, while two metabolites are measured after oral administration only.
A posteriori identifiability was explored via bootstrap analysis in NONMEM 7.4.3 (FOCE method), fixing a priori non-identifiable parameters. 500 bootstrap datasets were obtained from a re-scaled dataset of a real case study involving two metabolites.
Results: None of these models were a priori identifiable unless prior knowledge on some parameters was assumed. All models were a priori identifiable after fixing metabolites volumes (or clearances), one parameter for parent drug distribution/elimination, and one parameter for drug absorption.
For a posteriori identifiability analysis, metabolites volumes (Vm1, Vm2), bioavailability (F), and intercompartmental drug clearance (Q) were fixed. In particular, Vm1 and Vm2 were set equal either to 1 (scenario i) or to drug central volume, i.e. Vm1=Vm2=V (scenario ii).
– Model 1: During bootstrap, 75% and 83% of runs failed to converge in scenario i and ii, respectively. In both scenarios, fixed effect and inter-individual variability (IIV) for fraction of metabolite 2 absorbed (Fm2) did not change from their initial estimates and a high IIV on fraction of drug metabolized to metabolite 2 (FMm2) was estimated with low precision. In scenario ii, also a high IIV on total fraction metabolized (FM) was estimated with poor precision, as well as both fixed effect and IIV for absorption rate constant.
– Model 2: 63% and 56% of runs failed to converge in scenario i and ii, respectively. In scenario i, metabolite 2 clearance (CLm2), FM, and IIV on FMm2 estimates displayed low precision. In scenario ii, IIV on FM and FMm2 was high on average and poorly estimated.
– Model 3: 39% and 52% of runs failed to converge in scenario i and ii, respectively. In scenario i, all parameters were estimated with reasonable precision, except for FM; IIV on this parameter was fairly high. In scenario ii, this IIV term became higher (on average) and was estimated with low precision.
For all models, estimation issues were rarer in scenario i (Vm1=Vm2=1) compared to scenario ii (Vm1=Vm2=V).
Among common parameters, V and metabolite 1 clearance (CLm1) were on average consistently estimated (i.e. with similar bootstrap medians) across models but not across scenarios. Parent drug clearance (CL) and peripheral volume of distribution (Vp) were consistently estimated across both models and scenarios.
Conclusions: A priori identifiability analysis helps detecting parameters that are non-identifiable in practice: this is a necessary but not sufficient requirement for a posteriori identifiability. A priori identifiability analysis performed on three PK models for a parent compound and two metabolites including first pass effects demonstrated structural over-parametrization. Even after fixing non-identifiable parameters, practical identification issues were still highlighted by the bootstrap analysis (especially for fractions of drug metabolized). This means that additional information would be needed to obtain reliable model estimates. Despite a posteriori identifiability results are dataset-dependent, it is of some interest that Model 3 showed the lowest rates of failed runs, and, when successfully estimated, reasonable precisions for almost all parameters.
References:
[1] Bertrand, Julie, Céline M. Laffont, France Mentré, Marylore Chenel, and Emmanuelle Comets. “Development of a complex parent-metabolite joint population pharmacokinetic model.” The AAPS journal 13, no. 3 (2011): 390-404.
[2] David R. Taft, Ganesh R. Iyer, Leon Behar and Robert V. Digregorio. “Application of a First-Pass Effect Model to Characterize the Pharmacokinetic Disposition of Venlafaxine after Oral Administration to Human Subjects”. Drug Metabolism and Disposition, 25, no. 10 (1997): 1215-1218.
[3] Zhou, X., Zhang, F., & Ma, P. Structural and Practical Considerations in the Development of a Parent-Metabolite Model. PAGE 27 (2018) Abstr 8660 [www.page-meeting.org/?abstract=8660]
[4] Bellu, Giuseppina, Maria Pia Saccomani, Stefania Audoly, and Leontina D’Angiò. “DAISY: A new software tool to test global identifiability of biological and physiological systems.” Computer methods and programs in biomedicine 88, no. 1 (2007): 52-61.
Reference: PAGE 28 (2019) Abstr 8863 [www.page-meeting.org/?abstract=8863]
Poster: Methodology - Model Evaluation