pec: ipcw – R documentation

Pricing

Become an expert in R — Interactive courses, Cheat Sheets, certificates and more!

Get Started for Free

Documentation

pec

ipcw

Estimation of censoring probabilities

Description

This function is used internally by the function pec to obtain inverse of the probability of censoring weights.

Usage

ipcw(
  formula,
  data,
  method,
  args,
  times,
  subjectTimes,
  subjectTimesLag = 1,
  what
)

Arguments

`formula`	A survival formula like, `Surv(time,status)~1`, where as usual status=0 means censored. The status variable is internally reversed for estimation of censoring rather than survival probabilities. Some of the available models (see argument `model`) will use predictors on the right hand side of the formula.
`data`	The data used for fitting the censoring model
`method`	Censoring model used for estimation of the (conditional) censoring distribution.
`args`	A list of arguments which is passed to method
`times`	For `what="IPCW.times"` a vector of times at which to compute the probabilities of not being censored.
`subjectTimes`	For `what="IPCW.subjectTimes"` a vector of individual times at which the probabilities of not being censored are computed.
`subjectTimesLag`	If equal to `1` then obtain `G(T_i-\|X_i)`, if equal to `0` estimate the conditional censoring distribution at the subjectTimes, i.e. (`G(T_i\|X_i)`).
`what`	Decide about what to do: If equal to `"IPCW.times"` then weights are estimated at given `times`. If equal to `"IPCW.subjectTimes"` then weights are estimated at individual `subjectTimes`. If missing then produce both.

Details

Inverse of the probability of censoring weights (IPCW) usually refer to the probabilities of not being censored at certain time points. These probabilities are also the values of the conditional survival function of the censoring time given covariates. The function ipcw estimates the conditional survival function of the censoring times and derives the weights.

IMPORTANT: the data set should be ordered, order(time,-status) in order to get the values IPCW.subjectTimes in the right order for some choices of method.

Value

`times`	The times at which weights are estimated
`IPCW.times`	Estimated weights at `times`
`IPCW.subjectTimes`	Estimated weights at individual time values `subjectTimes`
`fit`	The fitted censoring model
`method`	The method for modelling the censoring distribution
`call`	The call

Author(s)

Thomas A. Gerds tag@biostat.ku.dk

Examples

library(prodlim)
library(rms)
dat=SimSurv(30)

dat <- dat[order(dat$time),]

# using the marginal Kaplan-Meier for the censoring times

WKM=ipcw(Hist(time,status)~X2,
  data=dat,
  method="marginal",
  times=sort(unique(dat$time)),
  subjectTimes=dat$time)
plot(WKM$fit)
WKM$fit

# using the Cox model for the censoring times given X2
library(survival)
WCox=ipcw(Hist(time=time,event=status)~X2,
  data=dat,
  method="cox",
  times=sort(unique(dat$time)),
  subjectTimes=dat$time)
WCox$fit

plot(WKM$fit)
lines(sort(unique(dat$time)),
      1-WCox$IPCW.times[1,],
      type="l",
      col=2,
      lty=3,
      lwd=3)
lines(sort(unique(dat$time)),
      1-WCox$IPCW.times[5,],
      type="l",
      col=3,
      lty=3,
      lwd=3)

# using the stratified Kaplan-Meier
# for the censoring times given X2

WKM2=ipcw(Hist(time,status)~X2,
  data=dat,
  method="nonpar",
  times=sort(unique(dat$time)),
  subjectTimes=dat$time)
plot(WKM2$fit,add=FALSE)

pec

Prediction Error Curves for Risk Prediction Models in Survival Analysis

v2020.11.17

GPL (>= 2)

Authors

Thomas A. Gerds

Initial release