Inverse Transform Sampling To Generate Time-to-event Data From Parametric Models

Random survival times generation for the weibull or log-logistic distributions with parameters `scale` and `shape`.

itsamp(
  n,
  beta = c(2, -1),
  event_scale = 10,
  censor_scale = 4,
  features = data.frame(x1 = rnorm(n, 0), x2 = rnorm(n, 0)),
  shape = 2,
  model = c("ph", "po", "aft"),
  dist = c("weibull", "llogis"),
  censor = TRUE
)

Arguments

n	integer; sample size
beta	vector of regression coefficients
event_scale, censor_scale	event and censoring scale parameters
features	matrix of features (columns)
shape	event and censoring distribution shape
model	either "ph" (default) or "aft" for weibull and "po" or "aft" for log-logistic distribution
dist	"weibull" or "llogis"
censor	logical; if `TRUE`, censoring is required, that is mean(status) > 0

Value

data.frame of `ncol(x) +2` columns in which the survival times are the response variable denoted by `y`, `status` indicates failure (0 = failure) and the features are appended to the next columns.

Details

sim_surv returns weibull (log-logistic) randomly generated survival times. According to Collett (2003), the accelerated failure time model encompasses a wide variety of parametric models, including weibull and log-logistic models.

See also

spbp

Examples

rows <- 200

categorical <- rbinom(rows, size = 3, prob = .5)
x <- data.frame(numerical = rnorm(rows),
           cat0 = as.numeric(categorical == 0),
           cat1 = as.numeric(categorical == 1),
           cat2 = as.numeric(categorical == 2),
           cat3 = as.numeric(categorical == 3))

newdata <- itsamp(n = rows, beta = c(1, -2, .5, .1, 1),
  features = x, model = 'ph', dist = 'weibull')
#> The censoring percentage is 25%, generated from ph weibull