Donnia Robins [1,2], Katharina Krollik [1], Maria Vertzoni [2], Andreas Lehmann [1]
[1] Global CMC Development, Merck KGaA, Frankfurter Straße 250, Darmstadt, Germany; [2] Department of Pharmacy, School of Health Sciences, National and Kapodistrian University of Athens, Zografou, Greece
Objectives: Physiological based pharmacokinetic (PBPK) models are convoluted systems with many non-linear or interacting parameters, which can lead to high variability in the output. Most supplementary tools focus on quantifying the local behavior of parameters with few publications reporting the degree of interactions and often using computationally costly techniques. Such oversight can be disadvantageous when accounting for scenarios or populations with high variability such as the aging population, where small individual physiological changes may result to significant deviations in pharmacokinetics. Furthermore, during drug development, it would be useful to assess interactions between formulation and physiological parameters. The present work evaluates the applicability of an advanced sampling method for a parametric Elementary Effects global sensitivity analysis (GSA) capable of providing estimates of first and second order effects, as well as some insight into the nature of the interaction (eg. collinear parameters, non-linear but non-interacting parameter). Danirixin, a BCS IV zwitterion with age-associated differences in pharmacokinetics, is used as a case study for a cross-platform comparison of parameter sensitivities in young and older people for various formulations and prandial states.
Methods: Whole-body PBPK models for danirixin free base and salt forms were developed in PK-Sim for the young, healthy population and extrapolated to the older and patient population. Intravenous and peroral whole-blood concentration data from clinical studies and published in vitro dissolution studies1,2 were used for model building and verification. To resolve distribution and elimination, generic UGT enzymatic and biliary clearances were added and estimated using the PK-Sim Monte Carlo method. For the GSA a cycle equitable-based (CEQ) sampling design3 was combined with a previous integration of the Morris method with PK-Sim using the OSP suite R package4. Solubility, intestinal permeability, gastric emptying time and gastric pH were analyzed and compared to a PSA from a semi-PBPK3 study in GastroPlus to evaluate the specificity of the new CEQ-Morris method. The GSA was also extended to assess the influence of upper intestinal pH, permeability factors, fractional steady-state fill levels and fraction mucosa, formulation factors and assumptions made during model development. The main effects and cross-derivatives were analyzed using customized graphs generated in R and with Cmax and AUCtEnd as endpoints.
Results: Predicted plasma whole blood profiles were within the standard 2-fold error of observations, and final elimination of the parent drug in bile and urine matched published data. The CEQ-Morris method was successful in identifying second order effects with a relatively low number of simulations. Comparisons of the sensitivities of the parameters between the GastroPlus and PK-Sim models showed similar strong rankings for gastric emptying time and solubility, as well as greater sensitivity in Cmax than in AUCtEnd. However, the CEQ-Morris approach was advantageous in further identifying influential interactions between parameters of different rankings. For instance, intestinal permeability showed a relatively low main effect in the fasted state but participated in some of the strongest interactions. Special scenarios investigated such as food effects also reflected changes in the rankings of parameters and their interactions. Positioning of elementary effects on special plots showed the degree of linearity of the effects, such as in the fed state, the higher order effects observed for gastric emptying time and permeability were mostly due to their interactions with solubility and not with each other.
Conclusions: The CEQ-Morris design successfully estimated parametric first and second order effects and provided deeper insight into the relative importance of each interaction. Results for the danirixin case study are consistent with other sensitivity analysis techniques but provided more in-depth information into the model’s contextual parametric relationships. The technique can be used to evaluate the mechanisms behind changes in model behavior in special scenarios, verify assumptions made during model development and identify key parameters for risk mitigation.
References:
[1] Miller, B.E. et al. (2014) ‘The pharmacokinetics of conventional and bioenhanced tablet formulations of danirixin (GSK1325756) following oral administration in healthy, elderly, human volunteers’, European Journal of Drug Metabolism and Pharmacokinetics, 39(3), pp. 173–181. doi: 10.1007/s13318-014-0179-8
[2] Lloyd, R.S. et al. (2020) ‘Negative Food Effect of Danirixin: Use of PBPK Modelling to Explore the Effect of Formulation and Meal Type on Clinical PK’, Pharmaceutical Research, 37(12), p. 233. doi: 10.1007/s11095-020-02948-z
[3] Fédou, J.-M. and Rendas, M.-J. (2015) ‘Extending Morris method: identification of the interaction graph using cycle-equitable designs’, Journal of Statistical Computation and Simulation, 85(7), pp. 1398–1419. doi: 10.1080/00949655.2014.997235
[4] Robins, D., Lehmann A., Krollik K., Vertzoni M. (2023) Expanding the Application of Global Sensitivity Analysis for Risk Assessment in Special Populations, American Conference on Pharmacometrics [Poster M065], National Harbor, MD, USA
Reference: PAGE 32 (2024) Abstr 10999 [www.page-meeting.org/?abstract=10999]
Poster: Drug/Disease Modelling - Absorption & PBPK