simMSM (version 1.1.42)

simeventhistories: Simulate Event Histories

Description

Simulates n individual event histories.

Usage

simeventhistories(n, mpl, max.time, change.times, X, states.at.origin = NULL, 
Xstruc, partial.markov.x = NULL, partial.markov.eta = NULL)

Arguments

n

number of individuals.

mpl

model parameter list as generated (only a skeleton that has to be suitably completed) by the function mplskeleton (see examples below).

max.time

maximum entry time.

change.times

vector giving the times of change of the time-change covariates.

X

design matrix.

states.at.origin

state-types at origin (default is all possible entry state-types, which is internally calculated).

Xstruc

X structure matrix. See Examples for more information.

partial.markov.x

function defining how the partial Markov covariates are generated (see example below).

partial.markov.eta

list of lists (as generated by the function pmeskeleton in close analogy to mpl) defining how the partial Markov linear predictors are generated (see example below).

Value

Three data frames named msm.bascis, ttsce, tt.indicators are returned organized within one list. The three data frames and their respective variables will be described in the next lines.

msm.bascis contains the following variables variables:

id

id (1, ..., n) of the individual

entry

entry times

exit

exit times

from

values of initial states

to

values of final states

delta

non-censoring indicator function

x1

values of first covariate (additional covariates follow). If partial Markov objects are supplied, the generated covariates are attached as additional variables.

The second data frame ttsce contains a transition-type specific covariate expansion (as well for partial Markov covariates in the case of a partial Markov set-up).

The third data frame tt.indicators contains the values of transition-type indicator functions. For censored observations, all values of one data line are equal to zero (as e.g. needed in a BayesX full likelihood analysis).

Details

The example below provides an intuitive description of how to use the different input arguments. The idea of partial Markov covariates is based on the definition in Commenges (1991). A description of this idea directly in the context of illness-death models is described on pp. 224-225 in Beyersmann et al. (1999).

References

Daniel Commenges (1991) Multi-state Models in Epidemiology. Lifetime Data Analysis, Vol. 5, No. 4.

Jan Beyersmann, Martin Schumacher, Arthur Allignol (2012) Competing Risks and Multistate Models with R. Springer Series 'UseR!'.

See Also

mplskeleton

Examples

# NOT RUN {
## An example for a time-varying setup without partial Markov effects:
tra2 <- matrix(ncol = 2, nrow = 2, data = FALSE)
tra2[1, 2] <- tra2[2, 1] <- TRUE
mpl <- mplskeleton(tmat = tra2)
mpl[[1]]$bhr[[2]] <- mpl[[2]]$bhr[[1]] <- function(t){return(0.5)}
mpl[[1]]$eta[[2]] <- function(x.i, t){ ## time-varying x2 and time-varying f(x2)
  ifelse(t < 5, 
         return(1.0 * x.i[1] + 0.5 * x.i[2]), 
         return(1.0 * x.i[1] + 1.0 * x.i[3]))}
mpl[[2]]$eta[[1]] <- function(x.i, t){ ## time-varying x2 and time-varying f(x1)
  ifelse(t < 5, 
         return(-0.5 * x.i[1] + 0.5 * x.i[2]), 
         return( 1.0 * x.i[1] + 0.5 * x.i[3]))}
set.seed(123)
N <- 2
X <- matrix(nrow = N, ncol = 2, rnorm(2 * N))
X <- cbind(X, X[, 2] + runif(N)/10)
colnames(X) <- c("x1", "x2.t1", "x2.t2")
Xstruc <- matrix(ncol = 2, nrow = 2, data = 0)
rownames(Xstruc) <- c("t1", "t2")
colnames(Xstruc) <- c("x1", "x2")
Xstruc[, 1] <- 1
Xstruc[, 2] <- c(2, 3)
d <- simeventhistories(n = N, mpl = mpl, X = X, max.time = 10, 
                       change.times = c(5), Xstruc = Xstruc)
head(d$msm.basics)
# }
# NOT RUN {
## An Illness-Death model example with time-varying setup and partial Markov 
## effects:
traIDM <- matrix(nrow = 3, ncol = 3, FALSE)
traIDM[1, 2] <- traIDM[1, 3] <- traIDM[2, 1] <- traIDM[2, 3] <- TRUE
mpl <- mplskeleton(tmat = traIDM)
mpl[[1]]$bhr[[2]] <- mpl[[1]]$bhr[[3]] <- mpl[[2]]$bhr[[1]] <- 
                     mpl[[2]]$bhr[[3]] <- function(t){0.25}
mpl[[1]]$eta[[2]] <- mpl[[1]]$eta[[3]] <- mpl[[2]]$eta[[1]] <- 
  mpl[[2]]$eta[[3]] <- function(x.i, t){
    ifelse(t < 5, 
      return(0.5 * x.i[1]), 
      return(0.5 * x.i[2]))}
set.seed(123)
N <- 500
X <- matrix(nrow = N, ncol = 1, rnorm(N))
X <- cbind(X, X[, 1] + rnorm(N)/10)
colnames(X) <- c("x1.t1", "x1.t2")
Xstruc <- matrix(ncol = 1, nrow = 2, data = 0)
rownames(Xstruc) <- c("t1", "t2")
colnames(Xstruc) <- c("x1")
Xstruc[, 1] <- c(1, 2)
Xstruc
## Now set-up the partial Markov influences:
## Function 'partial.markov.x' has to take 5 input arguments representig vectors 
## of past history information. They have to take names 'entry', 'exit', 'from', 
## 'to', and 'delta':
partial.markov.x <- function(entry, exit, from, to, delta){
  count.12 <- sum(as.numeric((from == 1) & (to == 2) & (delta == 1)))
  count.21 <- sum(as.numeric((from == 2) & (to == 1) & (delta == 1)))
  return(c(count.12, count.21))}
## List 'partial.markov.eta' is a list of lists in analogy to 'mpl':
partial.markov.eta <- pmeskeleton(traIDM)
partial.markov.eta[[1]][[2]] <- function(x){return( 0.25 * x[1])}
partial.markov.eta[[1]][[3]] <- function(x){return( 0.50 * x[1])}
partial.markov.eta[[2]][[1]] <- function(x){return(-0.50 * x[1] + 0.25 * x[2])}
partial.markov.eta[[2]][[3]] <- function(x){return(0)}
## Event history simulation:
d <- simeventhistories(n = N, mpl = mpl, X = X, max.time = 10, 
                       change.times = c(5), Xstruc = Xstruc, 
                       partial.markov.x = partial.markov.x, 
                       partial.markov.eta = partial.markov.eta)
# }