Type: Package
Title: Multiple-Direction Logrank Test
Version: 0.0.4
Date: 2018-09-28
Author: Marc Ditzhaus and Sarah Friedrich
Maintainer: Sarah Friedrich <sarah.friedrich@alumni.uni-ulm.de>
Depends: R (≥ 3.4.0)
Description: Implemented are the one-sided and two-sided multiple-direction logrank test for two-sample right censored data. In addition to the statistics p-values are calculated: 1. For the one-sided testing problem one p-value based on a wild bootstrap approach is determined. 2. In the two-sided case one p-value based on a chi-squared approximation and a second p-values based on a permutation approach are calculated. Ditzhaus, M. and Friedrich, S. (2018) <doi:10.48550/arXiv.1807.05504>. Ditzhaus, M. and Pauly, M. (2018) <doi:10.48550/arXiv.1808.05627>.
License: GPL-2 | GPL-3
Imports: stats, MASS (≥ 7.3-47)
LazyData: TRUE
BugReports: http://github.com/marcdii/mdir.logrank/issues
Suggests: RGtk2 (≥ 2.20.34), coin
RoxygenNote: 6.1.0
NeedsCompilation: no
Packaged: 2018-09-29 12:57:25 UTC; sarah
Repository: CRAN
Date/Publication: 2018-09-29 15:30:02 UTC

A graphical user interface for the package mdir.logrank

Description

This function provides a graphical user interface for calculating multiple-direction logrank test for the two-sided and the one-sided testing problem.

Usage

calculateGUI()

Two-sample multiple-direction log rank test

Description

The mdir.logrank function calculates the multiple-direction logrank statistic and its corresponding p-values based on a \chi^2-approximation and a permutation approach

Usage

mdir.logrank(data, cross = TRUE, rg = list(c(0, 0)), nperm = 10000,
  dig_p = 3, dig_stat = 3)

Arguments

data

A data.frame, list or environment containing the variables time, event (with values 0 for censored and 1 for uncensored) and group.

cross

logical. Should the weight corresponding to crossing hazards be included? The default is TRUE.

rg

A list (or NULL) containing the exponents c(r, g) of the directions w(x) = x^r (1-x)^g. Both exponents r,g need to be natural numbers including 0. Default is list(c(0, 0)) corresponding to proportional hazards.

nperm

The number of permutations used for calculating the permuted p-value. The default option is 10000.

dig_p

The p-values are rounded to dig_p digits, the default is 3.

dig_stat

The test statistic is rounded to dig_stat digits, the default is 3.

Details

The package provides the multiple-direction logrank statistic for the two sample testing problem within right-censored survival data. Directions of the form w(x) = 1 - 2x (cross = TRUE) and w(x) = x^r * (1-x)^g for natural numbers r,g (including 0) can be specified. The multiple-direction logrank test needs linearly independent directions. A check for this is implemented. If the directions chosen by the user are linearly dependent then a subset consisting of linearly independent directions is selected automatically.

The mdir.logrank function returns the test statistic as well as two corresponding p-values: the first is based on a chi^2 approximation and the second one is based on a permutation procedure.

Value

An mdirLR object containing the following components:

Descriptive

The directions used and whether the directions specified by the user were linearly independent.

p.values

The p-values of the multiple-direction logrank test using the \chi^2-approximation (Approx.) as well as the one using the permutation approach (Perm.).

stat

Value of the multiple-direction logrank statistic.

rg

A list containing the exponents of the direction considered in the statistical analysis.

cross

logical. Was the crossing direction considered in the statistical analysis?

indep

logical. Were the directions specified by the user linearly independent?

nperm

The number of permutations used for calculating the permuted p-value.

References

Ditzhaus, M., Friedrich, S. (2018). More powerful logrank permutation tests for two-sample survival data. arXiv preprint arXiv:1807.05504.

See Also

mdir.onesided (one-sided test)

Examples

library(coin)
data(GTSG)
out <- mdir.logrank(data = GTSG, nperm = 1000)

## Detailed information:
summary(out)

Two-sample multiple-direction log rank test for stochastic ordered alternatives

Description

The mdir.onesided function calculates the multiple-direction logrank statistic for (one-sided) stochastic ordered alternatives and its p-value based on a wild bootstrap approach

Usage

mdir.onesided(data, group1, rg = list(c(0, 0), c(0, 4), c(4, 0)),
  w.user = NA, wild = "rade", iter = 10000, dig_p = 3,
  dig_stat = 3)

Arguments

data

A data.frame, list or environment containing the variables time, event (with values 0 for censored and 1 for uncensored) and group.

group1

The name or the coding for the first group in the data set (neceassary for a one-sided testing problem).

rg

A list containing the exponents c(r, g) of the directions w(x) = x^r (1-x)^g or NA. Both exponents r,g need to be natural numbers including 0. Default is list(c(0, 0), c(0, 4), c(4, 0)) corresponding to the choice of the proportional, early and late direction/weight.

w.user

A list containing the user specified functions or NA (default).

wild

The wild bootstrap approach used for estimating the p-value. The Rademacher (rade, default), the normal distribution (norm) or the centred Poisson distribution (pois) approach can be selected.

iter

The number of iteration used for calculating the wild bootstrap p-value. The default option is 10000.

dig_p

The p-values are rounded to dig_p digits, the default is 3.

dig_stat

The test statistic is rounded to dig_stat digits, the default is 3.

Details

The function provides the multiple-direction logrank statistic for the two sample one-sided testing problem of stochastic ordering within right-censored survival data. The null hypothesis H:F_1=F_2 is tested against the one-sided alternative K:F_1 \ge F_2, F_1 \neq F_2. The first group corresponding to F_1 can be specified by the argument group1. An arbitrary amount of directions/weights of the form w(x) = x^r (1-x)^g for natural numbers r,g (including 0) can be chosen in the list rg. The multiple-direction onesided logrank test needs linearly independent directions. A check for this is implemented. If the directions chosen by the user are linearly dependent then a subset consisting of linearly independent directions is selected automatically. The user can also specify weights of a different shape in the list w.user. But if the user specified own weights in w.user then there is no automatic check for linear independence.

The mdir.onesided function returns the test statistic and the p-value based on a wild bootstrap procedure wild.

Value

An mdirone object containing the following components:

Descriptive

The directions used and whether the directions specified by the user were linearly independent.

p.value

The p-value of the one-sided multiple-direction logrank test using the the using the permutation approach (Perm.).

wild

The wild bootstrap approach which was used

stat

Value of the one-sided multiple-direction logrank statistic.

rg

The argument rg.

w.user

The argument w.user.

group1

The name of the first group.

indep

logical or NA. indep=TRUE/FALSE when the directions specified by rg were linearly independent. indep=NA when rg=NA.

iter

The number of iterations used for calculating the wild bootstrap p-value.

References

Ditzhaus, M., Pauly, M. (2018). Wild bootstrap logrank tests with broader power functions for testing superiority. arXiv preprint arXiv:arXiv:1808.05627.

See Also

mdir.onesided

Examples

library(coin)
data(GTSG)
out <- mdir.onesided(data = GTSG, group1 = "Chemotherapy+Radiation", iter = 1000)

## Detailed information:
summary(out)