Does the modelling strategy make the difference in pharmacometrics? Some examples in oncology and infectious diseases

The use of mathematical models to describe and predict the pharmacokinetics (PK), i.e., what the body does to the drug, and the phar- macodynamics (PD), i.e., what the drug does to the body, is fundamental across all the phases of the drug development process. Among the other things, these models allow to identify the most promising candidates during the preclinical studies, lead the dose selection for the First-In-Human (FIH) clinical trials, and enable to evaluate the effectiveness of a treatment and to simulate in silico different administration protocols. Nowadays, there are several modeling tools, each of which characterized by different features and specific applicability fields. There are models with a strong mechanistic base, such as the Physiologically Based Pharmacokinetic (PBPK) models, which integrate the information on organism anatomy and physiology with the physicochemical drug properties. There are also models with a poor mechanistic base, such as the standard pharmacokinetics/pharmacodynamics (PK/PD) models. Finally, there are models in which the pharmacokinetics is not explicitely modeled (K/PD). The scientific question of this thesis is whether an “optimal” modeling strategy exists for a given problem. In this perspective, the sentence of George Box: “All models are wrong, some are useful” should be kept in mind. Via some case studies, this thesis aimed to investigate two aspects: i) the suitability of a certain modeling strategy for a given problem in terms of model structure, available/required data, working hypotheses and the robustness of the results with respect to the assumptions made; ii) the dependency of conclusions from the adopted modeling approach. In Chapter 1, to set the scene, a brief introduction on both the drug discovery and development process and the importance of mathematical modeling throughout all the phases of this process was given. The features of the modeling strategies considered in this work to describe the pharmacokinetics and the pharmacodynamics were outlined in details. Subsequently, the scientific question underlying this work of thesis was discussed together with the methodology used to address it. In Chapter 2, the predictive performance of the Whole-Body (WB) PBPK models were investigated. To this aim, six "what-if" scenarios, in which data were added progressively into model development, starting from in vitro and animal experiments, up to human clinical trials, were created. Via these scenarios, the accuracy of the exposure predictions in dependence of the available data was evaluated. Ethambutol (EMB), one of the first-line antibiotics used for the treatment of pul- monary tuberculosis, was used as paradigm drug. When the physiological characterization of the subject with the dis- ease is not sufficient or not available, as in the oncology fields, less mechanistic approaches, i.e., the PK/PD and the K/PD models, were used to draw conclusions on the effectiveness of candidates. In Chapter 3 the most important models currently used for cancer drug discovery were surveyed. In Chapter 4 the dependency on the results from the specific mod- eling strategy was investigated using as a case study the predicted effect of two anticancer drug combination (Sunitinib and Irinotecan) in xenograft mice. In Chapter 5 in the attempt to be more mechanistic, additional details on drug behavior were added by considering drug concentration profiles not only in plasma but also into tumor tissue. In Chapter 6 overall conclusions were reported.