# sflogistic: 4.7: Two-parameter Spending Function Families In gsDesign: Group Sequential Design

### Description

The functions sfLogistic(), sfNormal(), sfExtremeValue(), sfExtremeValue2(), sfCauchy(), and sfBetaDist() are all 2-parameter spending function families. These provide increased flexibility in some situations where the flexibility of a one-parameter spending function family is not sufficient. These functions all allow fitting of two points on a cumulative spending function curve; in this case, four parameters are specified indicating an x and a y coordinate for each of 2 points. Normally each of these functions will be passed to gsDesign() in the parameter sfu for the upper bound or sfl for the lower bound to specify a spending function family for a design. In this case, the user does not need to know the calling sequence. The calling sequence is useful, however, when the user wishes to plot a spending function as demonstrated in the examples; note, however, that an automatic alpha- and beta-spending function plot is also available.

### Usage

 1 2 3 4 5 6 sfLogistic(alpha, t, param) sfNormal(alpha, t, param) sfExtremeValue(alpha, t, param) sfExtremeValue2(alpha, t, param) sfCauchy(alpha, t, param) sfBetaDist(alpha, t, param) 

### Arguments

 alpha Real value > 0 and no more than 1. Normally, alpha=0.025 for one-sided Type I error specification or alpha=0.1 for Type II error specification. However, this could be set to 1 if for descriptive purposes you wish to see the proportion of spending as a function of the proportion of sample size or information. t A vector of points with increasing values from 0 to 1, inclusive. Values of the proportion of sample size or information for which the spending function will be computed. param In the two-parameter specification, sfBetaDist() requires 2 positive values, while sfLogistic(), sfNormal(), sfExtremeValue(), sfExtremeValue2() and sfCauchy() require the first parameter to be any real value and the second to be a positive value. The four parameter specification is c(t1,t2,u1,u2) where the objective is that sf(t1)=alpha*u1 and sf(t2)=alpha*u2. In this parameterization, all four values must be between 0 and 1 and t1 < t2, u1 < u2.

### Details

sfBetaDist(alpha,t,param) is simply alpha times the incomplete beta cumulative distribution function with parameters a and b passed in param evaluated at values passed in t.

The other spending functions take the form

f(t;alpha,a,b)=alpha F(a+bF^{-1}(t))

where F() is a cumulative distribution function with values > 0 on the real line (logistic for sfLogistic(), normal for sfNormal(), extreme value for sfExtremeValue() and Cauchy for sfCauchy()) and F^{-1}() is its inverse.

For the logistic spending function this simplifies to

f(t;α,a,b)=α (1-(1+e^a(t/(1-t))^b)^{-1}).

For the extreme value distribution with

F(x)=\exp(-\exp(-x))

this simplifies to

f(t;α,a,b)=α \exp(-e^a (-\ln t)^b).

Since the extreme value distribution is not symmetric, there is also a version where the standard distribution is flipped about 0. This is reflected in sfExtremeValue2() where

F(x)=1-\exp(-\exp(x)).

### Value

An object of type spendfn. See Spending function overview for further details.

### Note

The manual is not linked to this help file, but is available in library/gsdesign/doc/gsDesignManual.pdf in the directory where R is installed.

### Author(s)

Keaven Anderson keaven\_anderson@merck.

### References

Jennison C and Turnbull BW (2000), Group Sequential Methods with Applications to Clinical Trials. Boca Raton: Chapman and Hall.

Spending function overview, gsDesign, gsDesign package overview

### Examples

  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 # design a 4-analysis trial using a Kim-DeMets spending function # for both lower and upper bounds x<-gsDesign(k=4, sfu=sfPower, sfupar=3, sfl=sfPower, sflpar=1.5) # print the design x # plot the alpha- and beta-spending functions plot(x, plottype=5) # start by showing how to fit two points with sfLogistic # plot the spending function using many points to obtain a smooth curve # note that curve fits the points x=.1, y=.01 and x=.4, y=.1 # specified in the 3rd parameter of sfLogistic t <- 0:100/100 plot(t, sfLogistic(1, t, c(.1, .4, .01, .1))$spend, xlab="Proportion of final sample size", ylab="Cumulative Type I error spending", main="Logistic Spending Function Examples", type="l", cex.main=.9) lines(t, sfLogistic(1, t, c(.01, .1, .1, .4))$spend, lty=2) # now just give a=0 and b=1 as 3rd parameters for sfLogistic lines(t, sfLogistic(1, t, c(0, 1))$spend, lty=3) # try a couple with unconventional shapes again using # the xy form in the 3rd parameter lines(t, sfLogistic(1, t, c(.4, .6, .1, .7))$spend, lty=4) lines(t, sfLogistic(1, t, c(.1, .7, .4, .6))$spend, lty=5) legend(x=c(.0, .475), y=c(.76, 1.03), lty=1:5, legend=c("Fit (.1, 01) and (.4, .1)", "Fit (.01, .1) and (.1, .4)", "a=0, b=1", "Fit (.4, .1) and (.6, .7)", "Fit (.1, .4) and (.7, .6)")) # set up a function to plot comparsons of all # 2-parameter spending functions plotsf <- function(alpha, t, param) { plot(t, sfCauchy(alpha, t, param)$spend, xlab="Proportion of enrollment", ylab="Cumulative spending", type="l", lty=2) lines(t, sfExtremeValue(alpha, t, param)$spend, lty=5) lines(t, sfLogistic(alpha, t, param)$spend, lty=1) lines(t, sfNormal(alpha, t, param)$spend, lty=3) lines(t, sfExtremeValue2(alpha, t, param)$spend, lty=6, col=2) lines(t, sfBetaDist(alpha, t, param)\$spend, lty=7, col=3) legend(x=c(.05, .475), y=.025*c(.55, .9), lty=c(1, 2, 3, 5, 6, 7), col=c(1, 1, 1, 1, 2, 3), legend=c("Logistic", "Cauchy", "Normal", "Extreme value", "Extreme value 2", "Beta distribution")) } # do comparison for a design with conservative early spending # note that Cauchy spending function is quite different # from the others param <- c(.25, .5, .05, .1) plotsf(.025, t, param) 

gsDesign documentation built on May 21, 2017, 12:38 a.m.
Search within the gsDesign package
Search all R packages, documentation and source code

Questions? Problems? Suggestions? Tweet to @rdrrHQ or email at ian@mutexlabs.com.

Please suggest features or report bugs in the GitHub issue tracker.

All documentation is copyright its authors; we didn't write any of that.