Résumé
In many fields, describe the evolution of the phenomena in time is of a capital interest, in particular to approach the problems of prediction and research for causal factors. In epidemiology, one has cohort data which inform about a group of patients followed in time. The multi-states Markov type models propose an interesting tool which makes it possible to study the evolution of a patient through the various stages of a disease. First of all, we recall the methodology relating to the homogeneous Markov model. This model is the least complex; it supposes that the intensities of transition between the states are constant in time. In the second time, we study a homogeneous semi-Markov model which supposes that the transition intensities depend on the time spend in the health state. The theory of counting processes is then presented in order to introduce non-parametric estimation methods within the framework of a non-homogeneous Markov model. In this model, the transition intensities depend on time since inclusion in the study. Most of the estimation methods suppose that the mechanism of censoring does not bring any information on the evolution of the disease. This assumption seldom being checked in practice, we propose an estimation method which allows to deal with an informative censoring. We also present a programming guideline aiming to facilitate the implementation of the estimators based on counting processes. These methods are applied in order to study a data base of asthmatic patients. The objective is to help the clinicians with a better understanding of the disease evolution. The results make it possible to highlight the negative impact of overweight on the asthma evolution.