msmbayes

msmbayes is an R package for Bayesian multi-state modelling of intermittently-observed data.

It is similar to the msm package. It supports the following models:

• Markov models for intermittently-observed state data

• Hidden Markov models for intermittently-observed, misclassified (discrete) state data

• Phase-type semi-Markov models for intermittently-observed state data

Models are fitted with Bayesian estimation, via any of the algorithms available in Stan, whereas msm uses only maximum likelihood.

Advantages of msmbayes compared to msm

• Informative priors can represent background information

• Prior information can also help to stabilise model fitting

• Automatic, efficient uncertainty quantification for any model output

• Phase-type models with any number of phases are supported, though these have not been investigated much

Limitations of msmbayes compared to msm

• "Exact death time" observation schemes are not supported (but models can still have absorbing states, or any state structure).

• Continuously-observed processes (exacttimes in msm()) are not supported.

• "Censored states" are not supported.

• Equality constraints and fixed parameters are not supported. However, parameters can be constrained through their prior distributions.

• Time-inhomogeneous models specified through pci in msm() are not supported. However, models with time-varying intensities can still be specified through a time-dependent covariate (e.g. time itself), which assumes that intensities are constant between successive observations of the state.

• Hidden Markov models with general outcome distributions are not supported. The only HMMs supported are those where the observed state space is the same as (or a subset of) the true state space. This includes misclassification and phase-type models.

• Multivariate hidden Markov models are not supported.

• Fewer output functions.

• More limited documentation and worked examples.

Getting started

Examples of using msmbayes are given in: vignette("examples").

Installation

Warning: this package is experimental. Some knowledge of Bayesian analysis is needed to develop and interpret models with it!

(a) Install cmdstan and cmdstanr by following the instructions linked here

(b) Install msmbayes by doing:

## install,packages("remotes") # if need be
remotes::install_github("chjackson/msmbayes")

If you use it, please give feedback on github issues, or by email.