Multi-state models are natural generalizations of survival models (two-state models). One reason to construct multi-state models in epidemiology is to provide more comprehensive picture of disease processes than that given by considering the onset of a single terminating
event. Another reason is to more effectively handle incomplete disease histories. Multi-state models can be specified via transition intensities between states or in the framework of multivariate counting processes. Assumptions have to be made about the law of the processes
involved: Markov and semi-Markov assumptions are the most common. The purpose of analysing event histories data using multi-state models is to gain insight into the dynamics of the processes under study by: quantifying transition rates and assessing their dependence on
covariates; computing various types of transition probabilities; making predictions. In this tutorial, we will present generalities on multi-states models and discuss some inference problems that may arise in the presence interval censoring. Examples from longitudinal
studies of diseases will be considered using existing R packages.