Longitudinal data analysis, causality and care pathways : application to French nationwide claims database ; Analyse de données longitudinales, causalité et parcours de soin : une application aux bases médico-administratives françaises

For several years, the health field has seen a significant increase in the amount of data available. In France, new platforms have been created to centralize health data, starting with data from claims database. Initially collected for pricing purposes, these data are a great opportunity for real-life studies of healthcare consumption and its evolution over time. Treatment effectiveness is a recurrent issue in health data studies. Our first contribution concerns a model for estimating the average treatment effect in treated patients (ATT) in the presence of time-dependent confounders. We proposed an unbiased estimate of ATT based on a generalization of Gran's model in the presence of several continuous, time-dependent confounders and a repeatable outcome, such as rehospitalizations. A simulation study demonstrated the value of the correction and the model was applied to MIMIC-III data. In a second step, a multivariate extension of the INGARCH model was proposed to take into account discrete confounders in Gran's model. The study of care pathways is another issue often studied on real-life data. We propose a process to analysed hospital care pathways in order to assess their association with mortality. Sequence analysis methods taking into account the order of occurrence of hospitalizations have been combined. This analysis process was applied to the study of the association of cardiovascular hospitalisations with death in patients with heart failure. ; Le domaine de la santé connaît depuis plusieurs années une augmentation importante des données disponibles. En France, de nouvelles plateformes ont vu le jour pour centraliser les données de santé, en commençant par les données des bases médico-administratives. Initialement recueillies pour des besoins tarifaires, ces données sont une grande opportunité pour l'études en vie réelle des consommations de soin et leur évolution dans le temps. L'efficacité des traitements est une problématique récurrente dans les études sur données de santé. Notre première contribution ...