Physiologically-based pharmacokinetic (PBPK) modelling is traditionally employed to predict drug concentration-time profiles in plasma and tissues using information from physiology/biology, in vitro experiments and in silico predictions. Model-based analysis of population pharmacokinetic (PK) data is rarely performed in such a mechanistic framework, as empirical compartmental models are mainly utilised for this purpose. However, the combination of traditional PBPK methodologies with parameter estimation techniques and non-linear mixed effects modelling is an approach with progressively increasing impact due to the significant advantages it offers. Therefore, the general aim of this thesis is to illustrate, explore and thus further facilitate the application of physiologically-based pharmacokinetic models in the context of population data analysis. In order to pursue this aim, this work firstly particularly focuses on the population pharmacokinetics of simvastatin (SV) and its active metabolite, simvastatin acid (SVA). The complex simvastatin pharmacokinetics and their clinical significance, due to the association with simvastatin-induced myopathy, provide an excellent case to illustrate the advantages of a mechanistically sound population model. In the current work, both conventional and physiologically-based population models were developed using clinical PK data for SV and SVA. Specifically, the developed model-based approaches successfully quantified the impact of demographics, genetic polymorphisms and drug-drug interactions (DDIs) on the SV/SVA pharmacokinetics. Therefore, they can be of significant application either in the clinic or during drug development in order to assess myopathy and DDI risk. Secondly, in this work the following advantages offered by integrated population PBPK modelling were clearly illustrated through specific applications: 1) prediction of drug concentrations at the tissue level, 2) ability to extrapolate outside the studied population and/or conditions and 3) ability to guide the design (sample size) of prospective clinical studies. Finally, in the current work, further methodological aspects related to the application of this integrated population PBPK modelling approach were explored. Of specific focus was the parameter estimation process aided by prior distributions and the derivation of the latter from different in vitro/in silico sources. In addition, specific methodology is illustrated in this work that allows the incorporation of stochastic population variability in the structural parameters of such models without neglecting the underlying physiological constraints.

Many methods exist in the literature for the prediction of pharmacokinetic parameters which describe drug distribution in rat and human, such as tissue-to-plasma partition coefficients (Kps) and volume of distribution (Vss). However, none of these methods make use of the in vivo information obtained at the early stages of the drug development process in the form of plasma concentration vs. time profiles. The overall aim of the presented study was to improve upon an existing Kp prediction method by making use of the distribution information contained within this experimental data. Chapter 2 shows that Kp values can be successfully obtained experimentally, but that this process is expensive and time-consuming. Chapter 3 compares six Kp prediction methods taken from the literature for their ability to predict the Kp values of 80 drugs. The Rodgers et al. model was found to be the most accurate, with over 77% of predictions within 3-fold of experimental values. This Chapter also discusses the Vss prediction ability of some of these methods, with the Poulin & Theil and Rodgers et al. models shown to be the most accurate predictors for rat Vss and human Vss respectively. Chapter 4 investigates the relationship between muscle Kp and the Kps of all other tissues, to show that experimental muscle Kp can be used as a surrogate from which all other non-adipose Kp values can be predicted. However, the predictions made using this method were shown to be less accurate than predictions made by the Rodgers et al. model for the same dataset of drugs. A relationship was identified between muscle Kp and tumour Kp in rat, suggesting a potential way to predict tumour Kp in the future. In Chapter 5, a novel method is developed whereby Kp predictions made by the Rodgers et al. model are updated using prior information obtained from the in vivo concentration-time profile. These updated values are then used within a physiologically-based pharmacokinetic (PBPK) model and are shown in Chapter 6 to generate improved predictions for other pharmacokinetic parameters such as Vss and clearance in both rat and human. 100% of human Vss predictions made by the most accurate of the novel methods presented here were within 3-fold of experimental values, compared to 68.8% of predictions made by the Rodgers et al. model. The work presented here has highlighted the need for a more accurate method for the prediction of Kp values, and has addressed this need by generating a model which improves upon the most accurate Kp prediction method currently found in the literature. This will lead to an increase in confidence in the use of predicted pharmacokinetic parameters within PBPK modelling.

There are currently two types of drug formulation in clinical use in the locoregional treatment of intermediate hepatocellular carcinoma (HCC). In the emulsion LIPDOX, the cytostatic agent doxorubicin (DOX) is dissolved in the aqueous phase, which is emulsified with the oily contrast agent Lipiodol® (LIP). In the microparticular system DEBDOX, DOX is loaded into the drug-eluting entity DC Bead™. The overall aim of the thesis was to improve pharmaceutical understanding of the LIPDOX and DEBDOX formulations, in order to facilitate the future development of novel drug delivery systems. In vivo release of DOX from the formulations and the disposition of DOX and its active metabolite doxorubicinol (DOXol) were assessed in an advanced multisampling-site acute healthy pig model and in patients with HCC. The release of DOX and disposition of DOX and DOXol where further analysed using physiologically based pharmacokinetic (PBPK) and biopharmaceutical (PBBP) modelling. The combination of in vivo investigations and in silico modelling could provide unique insight into the mechanisms behind drug release and disposition. The in vivo release of DOX from LIPDOX is not extended and controlled, as it is from DEBDOX. With both formulations, DOX is released as a burst during the early phase of administration. The in vivo release of DOX from LIPDOX was faster than from DEBDOX in both pigs and patients. The release from DEBDOX was slow and possibly incomplete. The in vivo release of DOX from LIPDOX and DEBDOX could be described by using the PBBP model in combination with in vitro release profiles. The disposition of DOX and DOXol was modelled using a semi-PBPK model containing intracellular binding sites. The contrast agent Lipiodol® did not affect the hepatobiliary disposition of DOX in the pig model. The control substance used in this study, cyclosporine A, inhibited the biliary excretion of DOX and DOXol but did not alter metabolism in healthy pigs. The disposition of DOX is similar in healthy pigs and humans, which was shown by the ease of translation of the semi-PBPK pig model to the human PBBP model.