Janak Wedagedera (1), Hiroshi Momiji (1), Anthonia Afuape (1), Siri Kalyan Chirumamilla (1), Robert Leary (1), Mike Dunlavey (1), Richard Matthews (1), Khaled Abjuljalil (1), Masoud Jamei (1), Frederic Y. Bois (1)
(1) CERTARA UK Limited, Simcyp division, Sheffield, UK
Objectives: PBPK models usually include a large number of physiological and drug specific parameters due to their inherent mechanistic description of the underlying phenomena. The best-practice workflow for PBPK modelling in clinical studies involves in vitro to in vivo extrapolation of parameter values. However, such extrapolations can be uncertain and may benefit from inclusion of clinical observations. Yet, when clinical inter-individual variability is high, or if the data are sparse, it is essential to use a Population PK (PopPK) framework. We compare here the results obtained in such a case using parametric (QRPEM) and non-parametric (NPAG) asymptotic methods implemented in the Simcyp Simulator version 19 to those given by Bayesian numerical estimation.
Methods: The QRPEM and NPAG population PK algorithms were used to analyse a canonical theophylline data set (Trembath 1980; R MEMSS package). A semi-physiological three-compartment model was used to describe the pharmacokinetics of the drug after administration of an immediate-release (IR) or sustained-release (SR) formulation to six healthy subjects. The model includes first-order absorption (rate ka, in 1/h), distribution is characterized by the steady-state volume of distribution (Vss, in L/kg) using a liver to plasma partition coefficient set at 1, saturable metabolism through liver CYP1A2 N3-demethylation (with maximum rate Vmax), and renal clearance fixed at 0.31 L/h. Inter-individual (IIV) and inter-occasion variability (IOV) was estimated for three key parameters (ka, Vss, and Vmax). The same data were analysed in a parametric Bayesian framework using the same structural model using HMCMC (Stan software, Carpenter et al. 2017). We compare parameter estimates and model fits to the data.
Results: For both formulations – IR and SR, Simcyp version19 – QRPEM and NPAG algorithms successfully fitted the PBPK model to observed data to estimate the key parameters. Further, the QRPEM method has estimated IIV and IOV for the parameters. Overall, all three methods showed consistency in the estimates except that the absorption rate constant (ka) estimate was significantly higher with the NPAG method for IR formulation (~4 fold with QRPEM). Run-times for these were within the range of 15 minutes to 1 hr for six subjects. The comparison with HMCMC shows good agreement in the estimates for Vss and Vmax but ka was about two-fold higher compared to the QRPEM estimates for the IR formulation (HMCMC took about 20 minutes for 6 subjects).
Conclusions: The QRPEM and NPAG methods implemented in Simcyp version 19 gave consistent population and individual parameter estimates and were in reasonable agreement with Bayesian HMCMC estimates for this case study. Relatively higher estimate for absorption parameter ka with NPAG and HMCMC could be due sensitivity to prior information on the absorption rate parameter. The run-times of both QRPEM and NPAG estimation methods were reasonable compared to the alternative method HMCMC run with R-STAN package.
References:
[1] Carpenter B et al. 2017. Stan: a probabilistic programming language, Journal of Statistical Software. 76, doi.org/10.18637/jss.v076.i01.
[2] Trembath PW and Boobis SW 1980. Pharmacokinetics of a sustained-release theophylline formulation, British Journal of Clinical Pharmacology 9 (1980) 365-369.
Reference: PAGE () Abstr 9393 [www.page-meeting.org/?abstract=9393]
Poster: Methodology - Estimation Methods