Multistate survival analysis with R Seminar

In multistate models for survival analysis, the life history is represented by a sequence of states and transitions between states. The history may be described in discrete or continuous time. In this presentation I consider histories in continuous time. The movements between states are governed by origin-destination specific transition intensities that vary with age and depend on current personal attributes, past experiences and contextual factors. Multistate models are used for explanation and prediction.

In the seminar I present the R package Biograph . The package was designed to help explore life history data (using descriptive statistics and graphics) and to facilitate the estimation of multistate models. It uses the graphical capabilities of R to visualize life histories of individuals and cohorts.

Biograph facilitates teaching multistate survival analysis. For instance, it has a utility to retrieve for subjects in the sample the observed life histories and the different types of episodes (open, closed; episodes ending in the event of interest, censored episodes) that are considered in the likelihood function. The display of life histories in a Lexis diagram is a powerful method for explaining multistate survival analysis from demographic and process points of view.

Biograph makes the estimation of multistate models more accessible. It prepares input data for packages for survival analysis ( survival , eha ) and specialized packages for multistate modeling ( mvna , mstate and msm ). Several packages for multistate modeling were recently presented in a special issue of Journal of Statistical Software (Vol, 38, Issue 1, January 2011) ( www.jstatsoft.org ). Biograph predicts life-course trajectories using the concepts of synthetic cohort and synthetic biography , which are also central in the life table.

The package is fully described in “ Biograph . Multistate analysis of life histories with R” (forthcoming in the Use R! Series , Springer).