kmconfband (version 0.1)

confband: Compute an Exact Nonparametric Confidence Band for the Population Survivor Function

Description

This function computes an exact nonparametric confidence band for the population survivor function, based on its one-sample Kaplan-Meier estimate. The theory and methods used in the computations are described in the 2013 article by Matthews. The confidence level required can be specified by the user.

Usage

confband(sobj,conf.level=0.95)

Arguments

sobj

a one-sample Kaplan-Meier estimate, provided in the form of a survfit.object.

conf.level

confidence level, a proportion between 0 and 1; the default value is 0.95.

Value

There are two outputs. The first is a scalar, the required quantile from the sample-specific, exact null distribution of the modified Berk-Jones (B-J) statistic. Inverting this B-J statistic generates a matrix of dimension $(k+1) \times 2$, where $k$ represents the number of changes in the Kaplan-Meier estimate, i.e., the number of distinct, complete observations in the original dataset. This matrix is the primary output of the function. Its first column is the lower estimate; the second column is the corresponding upper estimate. If the smallest value of the Kaplan-Meier estimate is 0, then so is the smallest value of the lower bound; otherwise, it has a positive value. Each row in the matrix represents a pair of lower and upper limits for one of the $k+1$ distinct values of the Kaplan-Meier estimate. The ordering of these rows is the same as the ordering of the original survfit.object.

Details

The exact nonparametric confidence band is calculated as lower and upper estimates of the survivor function derived from the survfit.object.

References

Matthews, D. (2013) ``Exact nonparametric confidence bands for the survivor function.'' Int J Biostat 9(1), doi: 10.1515/ijb-2012-0046

See Also

plotbands

Examples

## Calculate a nonparametric, exact, 95% confidence band for leukemia 
## patient remission experience based on data from 20 patients 
## receiving Treatment B
time<-c(1,1,2,2,3,4,5,8,8,9,11,12,14,16,18,21,27,31,38,44)
status<-c(rep(1,16),0,1,0,1)
fit<-survfit(Surv(time,status)~1)
bands<-confband(fit)

## Separately display the 95% (default) lower and upper confidence 
## band values 
bands[,1]
bands[,2]

## Repeat the same calculations, but for 80% confidence
bands<-confband(fit,0.80)

## Display the lower and upper confidence band values separately
bands[,1]
bands[,2]