Thèse Modélisation de la Dynamique de Transmission des Bactéries Résistantes aux Antibiotiques en France et de leurs Associations avec l'Utilisation des Antibiotiques à l'Échelle Populationnelle H/F

Doctorat.Gouv.Fr

Le troisième axe utilise le modèle de transmission calibré pour évaluer l'impact sanitaire d'interventions de bon usage des antibiotiques ciblées. S'appuyant sur la stratégie nationale de lutte contre l'antibiorésistance et les plans d'action européens, un ensemble de scénarios prédéfinis sera modélisé, incluant des réductions de la consommation totale d'antibiotiques, des restrictions par classe ciblant les antibiotiques les plus fortement associés à la sélection des BLSE, et des interventions ciblant les groupes d'âge aux profils de prescription les plus influents.

Les trois axes constituent ensemble un cadre analytique cohérent : caractériser ce qui est observé, comprendre pourquoi, et évaluer ce qui doit être fait pour améliorer la santé populationnelle. Les résultats visent à éclairer une politique de bon usage des antibiotiques fondée sur les preuves en France et à contribuer des avancées méthodologiques généralisables au domaine plus large de l'épidémiologie de l'antibiorésistance et de la modélisation épidémique. High levels of global antibiotic use contribute to a growing clinical burden of antibiotic-resistant bacteria, which in 2019 were estimated to cause 1.27 million deaths worldwide [1]. In a world where resources to invest in interventions to mitigate antimicrobial resistance (AMR) are limited, there is a critical need to better understand how and why AMR spreads through human populations, and to estimate the population health impacts of policy interventions targeting reduced antibiotic use [2].

For most clinically relevant bacteria, including Escherichia coli, Klebsiella pneumoniae and others, efforts to uncover the primary drivers of transmission have been limited, especially in the community setting. Ecological studies have helped to quantify associations between antibiotic use and AMR at the population level in some settings [3], but such associations are complex, time-varying and context-specific. Despite investments in AMR surveillance infrastructure in recent years, few studies in France have attempted to disentangle the historical drivers of AMR across population groups and settings [4-6].

Statistical associations are useful to improve understanding of AMR epidemiology, but are insufficient to predict future dynamics and intervention impacts [7]. Predicting AMR dynamics requires carefully designed transmission models that account for the mechanistic dependency of pathogen transmission on carriage prevalence, human behaviour, antibiotic use and other ecological forces [8]. Models have been used previously to predict the transmission dynamics of bacterial pathogens such as Staphylococcus aureus and Streptococcus pneumoniae, but, to date, few models exist for the antibiotic-resistant bacteria of greatest concern today, including ESBL-producing Enterobacterales [9].

These research gaps are a key limitation to predicting the efficacy and efficiency of the interventions proposed to reduce antibiotic use and, in turn, mitigate AMR. Even in healthcare settings, where data are better available and transmission dynamics are better understood, evaluations of AMR interventions tend to omit mechanistic links between antibiotic use and resistance [10-12]. From a policy perspective, this lack of evidence makes it challenging for decision-makers to prioritize investment in AMR interventions relative to other population health priorities.

Motivated by these evidence gaps, this thesis project aims to use statistical and mathematical modelling approaches to better understand: (1) associations between population-level antibiotic use and AMR in France; (2) the transmission dynamics of ESBL-producing Enterobacterales in French community and healthcare settings; and (3) the potential impacts of proposed antibiotic use interventions on AMR burden. This PhD thesis project aims to develop and apply novel statistical and mathematical models to exploit national AMR datasets and predict population-level associations between antibiotic use and resistance in France. This work will explicitly account for trends in different regions, settings (community, hospital) and age groups, with a focus on ESBL-producing Enterobacterales. The expected results are:

(1) Ecological modelling exploring spatiotemporal associations between antibiotic use and resistance across population groups.

(2) Transmission dynamic modelling characterizing the spread of ESBL-producing Enterobacterales in community and hospital settings.

(3) Model-based evaluations of the expected impacts of antibiotic use policy targets on the dynamics and burden of AMR in France. This PhD project is articulated through the following three bodies of work:

(1) Ecological analysis of spatiotemporal associations between antibiotic use and antibiotic resistance in France

National-level datasets describing antibiotic use (SNDS) and antibiotic resistance in clinical isolates (PRIMO, SPARES) in France will be assembled, harmonized and exploited to estimate spatiotemporal associations between human antibiotic exposure and the prevalence of multidrug-resistance in outpatient and inpatient settings. A literature review of recent association studies will be used to form hypotheses regarding antibiotic selection [6,13,14], and will inform the most appropriate statistical modelling framework, including generalized estimating equations and generalized additive models that account for spatial and temporal autocorrelation, exposure-response time lags and region-level fixed effects. This modelling will account for antibiotic co-selection where relevant [15], will consider multivariate outcomes where possible, will account for testing biases in the underlying data, and will include publicly available sociodemographic variables to adjust for residual confounding. These results will provide a national overview of the evolution of antibiotic resistance in France, will quantify associations between antibiotic use and resistance, and will identify regions and population groups bearing the greatest burden. These results will further generate hypotheses for consideration in mechanistic transmission modelling.

(2) Transmission dynamic modelling of ESBL-E. coli to estimate transmission rates and antibiotic selection pressure in community settings

A dynamic transmission model will be developed using ordinary differential equations to describe the spread of antibiotic-resistant bacteria in the community in France across population groups. This model will focus specifically on ESBL-producing E. coli, which is a leading cause of AMR-related morbidity and mortality in France and globally, and which transmits relatively rarely in healthcare settings [1,16]. The model will include latent classes describing unobserved bacterial colonization, and infection classes in both outpatient and inpatient settings, considering stratification by age, region and healthcare exposure. Critically, selection pressure from antibiotic exposure will be incorporated mechanistically into the model, allowing for estimation of the degree to which antibiotic use drives community-level transmission. The model will be fit to the databases exploited in the previous body of work using Bayesian methods (e.g. Hamiltonian MCMC). To account for structural uncertainty, several candidate models will be derived and compared, with rigorous model selection used to select the most parsimonious model that reliably reproduces observed trends.

(3) Policy evaluation of the public health impacts of antibiotic use reduction targets

In global AMR policy, there is a major push towards setting national-level targets for sustainable levels of antibiotic use, high enough to address local infection burden but low enough to limit excess antibiotic exposure and AMR selection [17,18]. Using the data analysed in the first part of the PhD and the transmission model developed in the second, a simulation study will be realized to evaluate impacts of changing antibiotic use on trajectories of antibiotic resistance. A range of scenarios will be explored considering different levels of change of levels of use of key antibiotic classes in outpatient and inpatient settings, including reductions and switches from Watch to Access antibiotics. By simulating transmission trajectories, the impacts of these changes in antibiotic use on the spread of antibiotic-resistant bacteria will be projected forward in time. Clinical parameters (infection incidence, case fatality rates for antibiotic-sensitive and -resistant infections) will be integrated to quantify the potential burden of disease averted by achieving national-level antibiotic use reduction targets.