vignettes/reReg-sims.Rmd
reReg-sims.Rmd
In this vignette, we demonstrate how to use the simSC
function in reReg
package to simulate recurrent event data from a scale-change model. Since the scale-change model includes the Cox-type model and the accelerated mean model as special cases, simSC
can also be used to generate data from these submodels. The simSC
function allows the censoring time to be non-informative (independent given covariate) or informative about the recurrent event process.
Suppose recurrent events can potentially be observed in the time period \([0, \tau]\). For a subject, let \(N_i(t)\) be the number of events in interval \([0, t]\), and \(X_i\) is a \(p\times 1\) covariate vector. Let \(C_i\) be a non-informative censoring time, which is independent of \(N_i(\cdot)\) given \(X_i\). On the contrary, let \(D_i\) be a failure time (informative censoring time), which is associated with \(N_i(\cdot)\) even after conditioning on \(X\). Then the follow-up time is defined as \(Y = \min(C, D, \tau)\). The observed data are independent and identically distributed copies of \(\{N_i(t), Y_i, X_i: t\le Y_i, i = 1, \ldots, n\}\). In the following, we suppress the index for the ease of discussion.
simSC
functionThe function simSC
generates the recurrent times from a recurrent event process, \(N(t)\), with a pre-specified rate function, \(\lambda(t)\), and the failure time, \(D\), from a pre-specified hazard function, \(h(t)\). By specifying the argument type
in the function simSC
, recurrent times and failure time can be generated from the following models:
type = "cox"
for Cox-type models: \[\lambda(t) = Z\lambda_0(t) e^{X^\top\alpha}, h(t) = Zh_0(t) e^{X^\top\beta}.\]
type = "am"
for accelerated mean models: \[\lambda(t) = Z\lambda_0(te^{X^\top\alpha})e^{X^\top\alpha}, h(t) = Zh_0(te^{X^\top\beta})e^{X^\top\beta}.\]
type = "sc"
for scale-change models: \[\lambda(t) = Z\lambda_0(te^{X^\top\alpha})e^{X^\top\beta}, h(t) = Zh_0(te^{X^\top\beta})e^{X^\top\beta}.\]
The \(Z\) is a latent frailty variable. In simSC
currently only allows two covariates, i.e., \(X = (X_{1}, X_{2})^\top\), where \(X_1\) is a Bernoulli random variable with probability 0.5 and \(X_2\) is a standard normal random variable. The non-informative censoring time, \(C\), is generated separately from an exponential distribution with mean 80. The observed follow-up time is then taken to be \(Y = \min(D, C, \tau)\). We further assume the baseline functions \[\lambda_0(t) = \frac{2}{1 + t}, h_0(t) = \frac{1}{8(1 + t)}.\]
The arguments in simSC
are as follow
function (n, a, b, indCen = TRUE, type = c("cox", "am", "sc"),
tau = 60, zVar = 0.25, summary = FALSE)
NULL
The arguments are as follows
n
number of individuala, b
numeric vectors of parameter of length two.indCen
a logical value indicating whether the censoring assumption is imposed. When indCen = TRUE
, we set \(Z = 1\). Otherwise, \(Z\) is generated from a gamma distribution with mean 1 and variance 0.25 (e.g., rgamma(1, 4, 4)
).type
a character string specifying the underlying model.tau
a numeric value specifying the maximum observation time, or \(\tau\) in the above notation.summary
a logical value indicating whether a brief data summary will be printed.In the following examples, we simulate recurrent event using simSC
, with summary = TRUE
.
Cox-type model:
Summary results for number of recurrent event per subject:
Min. 1st Qu. Median Mean 3rd Qu. Max.
0.000 1.000 2.000 3.925 4.250 33.000
Number of failures: 63 (31.5%); Number of censored events: 137 (68.5%)
Number of x1 == 1: 96 (48%); Number of x1 == 0: 104 (52%)
Summary results for x2:
Min. 1st Qu. Median Mean 3rd Qu. Max.
-2.31592 -0.57302 -0.01835 0.06422 0.59445 3.24104
Accelerated mean model:
Summary results for number of recurrent event per subject:
Min. 1st Qu. Median Mean 3rd Qu. Max.
0.00 2.00 4.50 5.02 7.00 15.00
Number of failures: 49 (24.5%); Number of censored events: 151 (75.5%)
Number of x1 == 1: 102 (51%); Number of x1 == 0: 98 (49%)
Summary results for x2:
Min. 1st Qu. Median Mean 3rd Qu. Max.
-2.733513 -0.628024 0.109635 0.008464 0.632278 2.377080
Scale-change model:
Summary results for number of recurrent event per subject:
Min. 1st Qu. Median Mean 3rd Qu. Max.
0.000 2.000 4.000 4.315 6.000 16.000
Number of failures: 58 (29%); Number of censored events: 142 (71%)
Number of x1 == 1: 94 (47%); Number of x1 == 0: 106 (53%)
Summary results for x2:
Min. 1st Qu. Median Mean 3rd Qu. Max.
-2.30480 -0.69522 -0.04912 -0.08015 0.49393 2.37284
The output of simSC
are tibble
objects.
> class(dat.cox)
[1] "data.frame"
> head(dat.cox)
id Time event status x1 x2
1 1 0.07365335 1 0 0 0.8005543
2 1 0.15214310 1 0 0 0.8005543
3 1 0.19034357 1 0 0 0.8005543
4 1 0.22378167 0 1 0 0.8005543
5 2 0.32213835 1 0 0 -0.6868529
6 2 0.92322213 1 0 0 -0.6868529
In this example, subject #1 experienced 3 recurrent events (at time 0.074, 0.152, and 0.190) and died at time 0.224. Similarly, subject #2 experienced 3 recurrent events (at time 1.136, 4.917, and 8.839) and is alive when censored at time 60.