Multi-state models provide a relevant tool for studying the observations of a continuous-time process at arbitrary times. Markov models are often considered even if semi-Markov are better adapted in various situations. Such models are still not frequently applied mainly due to lack of available software. We have developed the R package SemiMarkov to fit homogeneous semi-Markov models to longitudinal data. The package performs maximum likelihood estimation in a parametric framework where the distributions of the sojourn times can be chosen between exponential, Weibull or exponentiated Weibull one. The package computes and displays the hazard rates of sojourn times and the hazard rates of the semi-Markov process. The effects of covariates can be studied with a Cox proportional hazards model for the sojourn times distributions. The number of covariates and the distribution of sojourn times can be specified for each possible transition providing a great flexibility in a model's definition. This article presents parametric semi-Markov models and gives a detailed description of the package together with an application to asthma control.
