Tuberculosis (TB) is the world’s most deadly infectious disease and causes enormous public health problems. The comorbidity with HIV and the rise of multidrug-resistant TB strains impede successful therapy through drug-drug interactions and the lack of efficient second-line treatments. The aim of this thesis was to support the improvement of anti-TB therapy through development of pharmacometric models, specifically focusing on the novel drug bedaquiline, pharmacokinetic interactions and methods for pooled population analyses. A population pharmacokinetic model of bedaquiline and its metabolite M2, linked to semi-mechanistic models of body weight and albumin concentrations, was developed and used for exposure-response analysis. Treatment response was quantified by measurements of mycobacterial load and early bedaquiline exposure was found to significantly impact the half-life of bacterial clearance. The analysis represents the first successful characterization of a concentration-effect relationship for bedaquiline. Single-dose Phase I studies investigating potential interactions between bedaquiline and efavirenz, nevirapine, ritonavir-boosted lopinavir, rifampicin and rifapentine were analyzed with a model-based approach. Substantial effects were detected in several cases and dose-adjustments mitigating the impact were suggested after simulations. The interaction effects of nevirapine and ritonavir-boosted lopinavir were also confirmed in patients with multidrug-resistant TB on long-term treatment combining the antiretrovirals and bedaquiline. Furthermore, the outcomes from model-based analysis were compared to results from conventional non-compartmental analysis in a simulation study. Non-compartmental analysis was found to consistently underpredict the interaction effect when most of the concentration-time profile was not observed, as commonly is the case for compounds with very long terminal half-life such as bedaquiline. To facilitate pooled analyses of individual patient data from multiple sources a structured development procedure was outlined and a fast diagnostic tool for extensions of the stochastic model components was developed. Pooled analyses of nevirapine and rifabutin pharmacokinetics were performed; the latter generating comprehensive dosing recommendations for combined administration of rifabutin and antiretroviral protease inhibitors. The work presented in this thesis demonstrates the usefulness of pharmacometric techniques to improve treatment of TB and especially contributes evidence to inform optimized dosing regimens of new and old anti-TB drugs in various clinical contexts.

Pharmacokinetic and pharmacodynamic models can be applied to clinical study data using various modelling approaches depending on the aim of the analysis. In population pharmacokinetics for instance, simple compartmental models can be employed to describe concentration-time data, identify prognostic factors and interpolate within well-defined experimental conditions. The first objective of this thesis was to illustrate such a ‘semi-mechanistic’ pharmacokinetic modelling approach using mavoglurant as an example of a compound under clinical development. In particular, methods to accurately characterise complex oral pharmacokinetic profiles and evaluate the impact of absorption factors were investigated. When the purpose of the model-based analysis is to further extrapolate beyond the experimental conditions in order to guide the design of subsequent clinical trials, physiologically-based pharmacokinetic (PBPK) models are more valuable as they incorporate information not only on the drug but also on the system, i.e. on mammillary anatomy and physiology. The combination of such mechanistic models with statistical modelling techniques in order to analysis clinical data has been widely applied in toxicokinetics but has only recently received increasing interest in pharmacokinetics. This is probably because, due to the higher complexity of PBPK models compared to conventional pharmacokinetic models, additional efforts are required for adequate population data analysis. Hence, the second objective of this thesis was to explore methods to allow the application of PBPK models to clinical study data, such as the Bayesian approach or model order reduction techniques, and propose a general mechanistic modelling workflow for population data analysis. In pharmacodynamics, mechanistic modelling of clinical data is even less common than in pharmacokinetics. This is probably because our understanding of the interaction between therapeutic drugs and biological processes is limited and also because the types of data to analyse are often more complex than pharmacokinetic data. In oncology for instance, the most widely used clinical endpoint to evaluate the benefit of an experimental treatment is survival of patients. Survival data are typically censored due to logistic constraints associated with patient follow-up. Hence, the analysis of survival data requires specific statistical techniques. Longitudinal tumour size data have been increasingly used to assess treatment response for solid tumours. In particular, the survival prognostic value of measures derived from such data has been recently evaluated for various types of cancer although not for pancreatic cancer. The last objective of this thesis was therefore to investigate different modelling approaches to analyse survival data of pancreatic cancer patients treated with gemcitabine, and compare tumour burden measures with other patient clinical characteristics and established risk factors, in terms of predictive value for survival.

Comment s’adaptent des organisations lorsqu’elles font face à des changements qui les dépassent? De cette question a émergé une recherche voulant comprendre comment et pourquoi des organisations de santé décident d’adapter (ou non) leurs services aux besoins et aux caractéristiques des populations migrantes accueillies sur leur territoire. Pour y répondre, cette thèse s’est intéressée à la gouvernance multiniveaux appliquée aux organisations de santé fournissant des services à des populations migrantes. Dans un contexte de régionalisation de l’immigration, la dynamique du processus migratoire est de mieux en mieux documentée, mais les capacités organisationnelles d’adaptation le sont beaucoup moins. Nous avons réalisé une étude de cas multiples à l’aide d’entrevues semi-directives auprès d’acteurs provenant de deux CSSS montéregiens (région au sud de Montréal, Québec) et des paliers locaux, régionaux et nationaux. Les résultats de cette étude ont permis (1) de mettre en évidence les différents acteurs impliqués dans ce processus d’adaptation, dont des acteurs de connectivité; (2) de cerner huit leviers d’action, divisés en trois catégories de leviers : administratif, émergent et d’habilitation. La possible imbrication de ces trois catégories de leviers facilite l’apparition de structures de connectivité, légitimant ainsi l’adaptation de l’organisation; et (3) de montrer l’ambigüité de l’adaptation à travers des facteurs d’influence qui favorisent ou entravent le processus d’adaptation à plusieurs niveaux de la gouvernance. Cette thèse est construite autour de quatre articles. Le premier, de nature conceptuelle, permet de circonscrire les concepts d’adaptation et de gouvernance multiniveaux à travers la lentille des théories de la complexité. Nous campons ainsi notre sujet dans une problématique liée à la vulnérabilité et la migration tout en appréhendant l’adaptation du système et son opérationnalisation au niveau local. Il en ressort un cadre conceptuel avec six propositions de recherches. Le second article permet quant à lui de comprendre les jeux des acteurs au sein d’une organisation de santé et à travers son Environnement. Le rôle spécifique d’acteurs de connectivité y est révélé. C’est dans un troisième article que nous nous intéressons davantage aux différents leviers d’action, analysés selon trois catégories : administrative, émergente et d’habilitation. Les acteurs peuvent les solliciter afin de d’adapter leurs pratiques au contexte particulier de la prise en charge de patients migrants. Un passage des acteurs aux structures de connectivité est alors rendu possible via un espace : la gouvernance multiniveaux. Enfin, le quatrième et dernier article s’articule autour de l’analyse des différents facteurs pouvant influencer l’adaptation d’une organisation de santé, en lien avec son Environnement. Il en ressort principalement que les facteurs identifiés sont pour beaucoup des leviers d’action (cf. article3) qui à travers le temps, et par récursivité, deviennent des facteurs d’influence. De plus, le type d’interdépendance développé par les acteurs a tendance soit à façonner un Environnement « stable », laissant reposer les besoins d’adaptation sur les acteurs opérationnels; soit à façonner un Environnement plus « accidenté », reposant davantage sur des interactions diversifiées entre les acteurs d’une gouvernance multiniveaux. De cette adéquation avec l’Environnement à façonner découle l’ambigüité de s’adapter ou non pour une organisation. / How do organizations adapt when faced with changes that exceed their current capacities? More specifically how and why do healthcare organizations choose to adapt- or not- their services to the needs and characteristics of new and established migrant populations? In this thesis, we attempt to answer these questions using a conceptual model of multilevel governance, applicable to healthcare organizations that provide primary care to migrant population. In a context of regionalized immigration, the dynamics of the migration process are well documented, however organizations abilities to adapt are less so. We conducted a multiple case study, collecting data from semi-structured interviews with providers from two healthcare organizations from Montéregie region (South shore of Montréal, Québec) as well as stakeholders from local, regional and national scale. The data collected has allowed us, (1) to identify the most significant stakeholders in adaptation process and to highlight the connectivity between them; (2) to reveal eight action levers that we divided in three categories: administrative, emerging and enabling. The possible imbrications of these three categories facilitate the creation of connectivity structures that legitimate the organization’s adaptation; and (3) to show the ambiguity of adaptation through the influence of facilitating or hindering factors at several levels of governance. This thesis is structured on four articles. The first article is conceptual: adaptation and multilevel governance are therein defined through the framework of complexity theory. We anchor our subject in the vulnerability and migration schemata, while apprehending the system’s adaptation at the local scale. The result is a conceptual framework with six research propositions. The second article elaborates on the dynamics among stakeholders within healthcare organizations, its networks and the Environment. The specific role of the connectivity between stakeholders is highlighted. In the third article, we focus on the different action levers, which are analysed in three categories: administrative, emerging and enabling. The stakeholders seek to adapt their practices to the particular context of providing timely and appropriate care to migrant patients. A transition from connectivity between actors and structures is then made possible through multilevel governance. Finally, the fourth article is an analysis of the different factors that influence a healthcare organization’s adaptation with regards to the Environment. The results show that many identified factors are firstly action levers (see article 3), and become through time and recursively of influence factors. In addition, the type of interdependence developed between the stakeholders tends to shape either “stable” Environment which transfers the responsibility of the adaptation to the operational stakeholders or “rugged” Environment based on diverse and decentralized interactions between stakeholders through multilevel of governance. From this constant adjustment with the Environment appears the ambiguity of the adaptation for a healthcare organization.

L’analyse de la causalité, qui consiste à déterminer si la prise d’un médicament est la cause de la survenue d’un événement, est la problématique centrale de la pharmacovigilance et de la pharmaco-épidémiologie.La première partie de ce travail aborde l’étude de la causalité au plan individuel, au travers des méthodes d’imputabilité. Nous avons d’abord comparé une méthode d’imputabilité récemment développée, la méthode logistique, et la méthode d’imputabilité officiellement utilisée en France à un jugement consensuel d’experts pris comme référence. Les résultats montrent que la méthode française d’imputabilité tend à sous-coter la responsabilité du médicament (faible sensibilité) alors que la méthode logistique tend à la surestimer (faible spécificité). Par la suite, une nouvelle version de la méthode française d’imputabilité visant à améliorer sa sensibilité et son pouvoir discriminant a été proposée. Le travail de validation portant sur cette méthode réactualisée montre une amélioration de sa sensibilité et des résultats se rapprochant plus du jugement consensuel d’experts. Pour la méthode logistique, les critères d’imputabilité et leurs poids ont été réévalués sur un échantillon plus important d’observations que celui ayant servi à la pondération initiale. La validité de cette nouvelle version et celle de l’un des algorithmes les plus couramment utilisés en pharmacovigilance, la méthode Naranjo, ont été comparées à un jugement consensuel d’experts. Les résultats concernant la validité interne et les qualités prédictives de la méthode Naranjo ne sont pas satisfaisants alors que la méthode logistique présente une spécificité améliorée ainsi qu’une bonne sensibilité et valeurs prédictives. Cette dernière méthode présente donc des caractéristiques qui devraient améliorer l’évaluation de la responsabilité des médicaments dans la survenue des événements indésirables. La mise en place de méthode d’imputabilité spécifique à une classe thérapeutique et/ou à un type d’événement indésirable pourrait aussi améliorer l’évaluation des événements indésirables. Nous proposons une grille d’imputabilité adaptée aux accidents hémorragiques sous antithrombotique. Dans la deuxième partie de cette thèse, l’analyse épidémiologique de la causalité est abordée en proposant deux méthodes : l’analyse populationnelle des cas individuels, en particulier leur délai de survenue après exposition médicamenteuse, et l’approche cas-population. Bien que beaucoup moins robustes que les méthodes classiques, elles sont testées sur des problématiques réelles de pharmacovigilance et les résultats montrent qu’elles peuvent être utiles pour une première exploration d’une association causale potentielle. En conclusion, ce travail méthodologique pourrait aider à mieux évaluer la responsabilité des médicaments dans la survenue d’événements indésirables après leurs autorisations de mise sur le marché. / The analysis of causality, which consists of determining if drug intake is the cause of the event occurrence, is the central issue of pharmacovigilance and pharmacoepidemiology. The first part of this work deals with the study of causality assessment methods at the level of individual cases. We first compared the recently developed logistic causality assessment method and the method officially used in France, to consensusual expert judgement taking as a reference. The results showed that the French causality assessment method tended to underestimate the responsibility of the drug (low sensitivity) whereas the logistic method tended to overestimate it (low specificity). Subsequently a new version of the French causality assessment method aiming to improve its sensitivity and discriminating power was proposed. The validation phase of this updated method showed improved sensitivity and a performance closer to consensual expert judgement. For the logistic method, the criteria of causality assessment and their weights were re-evaluated on a larger sample of drug-event pairs that had been used in the initial weighting. The validity of this method and that of one of the most commonly used algorithms in pharmacovigilance, the Naranjo method, were compared to consensual expert judgement. Results concerning the internal validity and the predictive qualities of the Naranjo method were not satisfactory while the logistic method presented an improved specificity and good sensitivity and predictive values. The logistic method now presents characteristics that should improve the assessment of drug responsibility in the occurrence of adverse events. The implementation of causality assessment method specific to a therapeutic class and / or to a type of adverse event could also improve the assessment of adverse events. We proposed a scale adapted to hemorrhages with antithrombotics and derived from the French causality assessment method. In the second part of this thesis, the epidemiological analysis of causality was tackled by proposing two methods: the populational analysis of individual cases, in particular their time to onset after drug exposure, and the case-population approach. Although less robust than the conventional methods, these were tested on real problems of pharmacovigilance and the results indicate that they may be useful for an initial exploration of a potential causal association. In conclusion, this methodological work could help to better assess drug causality in the occurrence of adverse event in post maketing surveillance.