sim.fmsm: Simulate paths through a fully parametric semi-Markov multi-state model

Description

Simulate changes of state and transition times from a semi-Markov multi-state model fitted using flexsurvreg.

Usage

sim.fmsm(x, trans, t, newdata = NULL, start = 1, M = 10, tvar = "trans",
  tcovs = NULL, debug = FALSE)

Arguments

A model fitted with flexsurvreg. See msfit.flexsurvreg for the required form of the model and the data.

Alternatively x can be a list of fitted flexsurvreg model objects. The ith element of this list is the model corresponding to the ith transition in trans. This is a more efficient way to fit a multi-state model, but only valid if the parameters are different between different transitions.

trans

Matrix indicating allowed transitions. See msfit.flexsurvreg.

Time, or vector of times for each of the M individuals, to simulate trajectories until.

newdata

A data frame specifying the values of covariates in the fitted model, other than the transition number. See msfit.flexsurvreg.

start

Starting state, or vector of starting states for each of the M individuals.

Number of individual trajectories to simulate.

tvar

Variable in the data representing the transition type. Not required if x is a list of models.

tcovs

Names of "predictable" time-dependent covariates in newdata, i.e. those whose values change at the same rate as time. Age is a typical example. During simulation, their values will be updated after each transition time, by adding the current time to the value supplied in newdata. This assumes the covariate is measured in the same unit as time. tcovs is supplied as a character vector.

debug

Print intermediate outputs: for development use.

Value

A list of two matrices named st and t. The rows of each matrix represent simulated individuals. The columns of t contain the times when the individual changes state, to the corresponding states in st.

The first columns will always contain the starting states and the starting times. The last column of t represents either the time when the individual moves to an absorbing state, or right-censoring in a transient state at the time given in the t argument to sim.fmsm.

Details

sim.fmsm relies on the presence of a function to sample random numbers from the parametric survival distribution used in the fitted model x, for example rweibull for Weibull models. If x was fitted using a custom distribution, called dist say, then there must be a function called (something like) rdist either in the working environment, or supplied through the dfns argument to flexsurvreg. This must be in the same format as standard R functions such as rweibull, with first argument n, and remaining arguments giving the parameters of the distribution. It must be vectorised with respect to the parameter arguments.

This function is only valid for semi-Markov ("clock-reset") models, though no warning or error is currently given if the model is not of this type. An equivalent for time-inhomogeneous Markov ("clock-forward") models has currently not been implemented.

Note the random sampling method for flexsurvspline models is currently very inefficient, so that looping over the M individuals will be very slow.

Examples

Run this code

# NOT RUN {
bexp <- flexsurvreg(Surv(years, status) ~ trans, data=bosms3, dist="exp")
tmat <- rbind(c(NA,1,2),c(NA,NA,3),c(NA,NA,NA))
sim.fmsm(bexp, M=10, t=5, trans=tmat)
# }