Abstract
The study of the evolution of a patient constitutes an important field in clinical research. Recently, the development of the multi-state models allows to study this dynamics by taking into account several health states. In this manuscript, we use the semi-markovian models. This type of process distinguishes the durations in the states and the trajectories of the transitions, contrary to the traditional markovian approach. We proposed several adaptations to apply this type of model: the interval-censoring, the choice of the distributions of the durations and the introduction of the covariates. A goodness-of-fit statistic is also proposed to check the stationnarity assumption. Lastly, an original method, including the theory of the ROC curves, is presented to define relevant health states. These developments are mainly applied to kidney transplant recipient follow-up (DIVAT database).