Dan Liu (1), Linzhong Li (1), Amin Rostami-Hodjegan (1, 2), Masoud Jamei (1)
(1) Certara UK Limited, Simcyp Division, Level 2-Acero, 1 Concourse Way, Sheffield, S1 2BJ, UK (2) Centre for Applied Pharmacokinetic Research, Division of Pharmacy & Optometry, The University of Manchester, Stopford Building, Oxford Road, Manchester, M13 9PT, UK
Objectives: Sensitivity analysis has been widely used to identify the most influential model parameters affecting pre-specified model outputs. In local sensitivity analysis (LSA) usually parameters are scanned in one or two dimensions while keeping all the rest of model parameters fixed. However, global sensitivity analysis (GSA) allows simultaneously evaluating the relative contributions of each individual parameter to the model output variance by varying all parameters over the entire intended parameter space. Hence the sensitivity or influence of a parameter to model output in GSA is measured without fixing values of all the rest input parameters. This allows ranking the importance of parameters considering their uncertainty and influence on the variation of outputs. GSA is gaining attention in the PBPK modelling and systems biology and pharmacology [1-4]. GAS can provide information about the model structure or driving mechanisms for physiology or biological responses. We present an application of three GSA methods, namely Morris, Sobol, and extended Sobol method to a minimal-PBPK (mPBPK) model of Midazolam. The primary aim is to identify the most influential model parameters affecting the pharmacokinetic (PK) properties of interest. We also investigated the effect of ignoring correlations of model parameters on their rankings. Despite known correlations between biological and drug parameters, they are rarely considered in conduct of GSA when the probability of various parameter values are commonly considered independent of the probability of other parameter values.
Methods: Midazolam, a BCS class II drug that has been widely used in anaesthesia due to its favourable safety profile, and rapid anxiolytic effect or as preanesthetic medication for children [5, 6]. Morris, Sobol, and extended Sobol methods were used to determine the most influential model parameters for the intended PK properties, i.e. Cmax, Tmax, and AUC, of an mPBPK model of Midazolam given orally. Morris and Sobol are GSA methods designed for models where the model parameters are not correlated [7]. Nevertheless, exSobol method is designed to handle a model with correlated model parameters [8]. The influential parameters were determined using these three GSA methods independently. Subsequently, influential parameters picked up by Morris and Sobol methods were compared to those by exSobol to explore how considering correlations could affect appropriately identifying and ranking influential model parameters. System parameters, such as body weight (BW), blood flow rate, tissue volume, tissue to plasma partition coefficient, enzyme abundance, etc., and their correlations were considered in this study.
Results: The exSobol method suggests, Vss (volume of distribution at steady-state), Fg (fraction scape gut wall metabolism), enzyme abundance of CYP3A4 and CYP3A5, and BW, are the most important parameters determining Cmax; ka and Vss are identified as the most significant parameters determining Tmax; enzyme abundance of CYP3A5, CYP3A4, Fg, Vliver (the liver volume), and fa (fraction absorbed into enterocytes), have significant impact on AUC. Compared to exSobol, different sets or ranking of influential parameters were identified by Morris and Sobol due to lack of consideration of parameters correlation, which underestimated the effect of Vliver and the impact of UGT1A4 Abundance on the Midazolam AUC. Further, the qualitative Morris screening was as informative of the quantitative Sobol method, if a lack of correlation of parameters was considered or correlations could be ignored.
Conclusions: Knowing Midazolam’s physicochemical, metabolism and plasma/blood binding properties the determined ranking by exSobol are as expected. Without considering parameters correlation, GSA methods such as Morris and Sobol, can provide biased assessment of their influence on the model outputs of interest. A major weakness of GSA methods assuming independent input parameters is that unrealistic parameter combinations are more likely to be produced due to independent random sampling. Therefore, GSA results should be carefully used when using Morris and Sobol methods or when there are uncertainties around the model parameters correlations. It is important to bear in mind that GSA methods can only provide information about the explored ‘model’ rather than the reality it intends to represent. In other words, if a model mis-specifies the reality or inadequately represent it then the provided GSA outcomes can be either biased or incorrect.
References:
[1] McNally K, Cotton R, Loizou GD (2011) A Workflow for Global Sensitivity Analysis of PBPK Models. Front Pharmacol 2:31.
[2] Scherholz ML, Forder J, Androulakis IP (2018) A framework for 2-stage global sensitivity analysis of GastroPlus compartmental models. J Pharmacokinet Pharmacodyn 45 (2):309-327.
[3] Melillo N, Aarons L, Magni P, Darwich AS (2018) Variance based global sensitivity analysis of physiologically based pharmacokinetic absorption models for BCS I-IV drugs.
[4] Zi Z (2011) Sensitivity analysis approaches applied to systems biology models. IET Syst Biol 5 (6):336-336.
[5] Pang W, Lin RMH, Lin ML, Chen YO, Lin JC, Yang CH, Shih ZM (2018) Midazolam-induced unexpected monoparesis: Not contraindicated for ambulatory general anesthesia, vol 24.
[6] Cote CJ, Cohen IT, Suresh S, Rabb M, Rose JB, Weldon BC, Davis PJ, Bikhazi GB, Karl HW, Hummer KA, Hannallah RS, Khoo KC, Collins P (2002) A comparison of three doses of a commercially prepared oral midazolam syrup in children. Anesth Analg 94 (1):37-43, table of contents
[7] Saltelli A, Ratto M, Andres T, Campolongo F, Cariboni J, Gatelli D, Saisana M, Tarantola S (2008) Global Sensitivity Analysis. The Primer. In. John Wiley & Sons, Ltd.
[8] Kucherenko S, Tarantola S, Annoni P (2012) Estimation of global sensitivity indices for models with dependent parameters. Computer Physics Communications 183 (4):937-946.
Reference: PAGE 28 (2019) Abstr 8875 [www.page-meeting.org/?abstract=8875]
Poster: Methodology - Model Evaluation