find.n: finding satisfactory sample size for a hierarchical Bayesian...

In hbdct: Hierarchical Bayesian Design for Clinical Trials

Description

generates artificial data from binomial model, samples interested parameters from posterior distributions, simulates the operating characteristics, and chooses a design (sample size) that delivers pre-specified desired operating characteristics.

Usage

find.n(del.cut, k, T, p.c, effect.size, alpha.target = 0.1, 
       power.target = 0.9, tol.b1 = 0.02, tol.b2 = 0.02, tol.a = 0.05, 
	   n.min = 5, n.max = 70, nstep=10, rep = 500, pathcode = path.package("hbdct"), 
	   codefile = "BUGS_Bin_2arm.txt", pathout = getwd(), method = "linear")

Arguments

del.cut	$delta_cut$, efficacy boundary
k	number of tumor types
T	$T_p$, threshold probability
p.c	$pi_C$, a vector of true response rates for k tumor type in control arm
effect.size	$delta_i$, effect size to dectect in treatment group
alpha.target	desired family-wise Type-I error rate, default is 0.1.
power.target	desired level of individual power for each type of tumor, default is 0.9.
tol.b1	tolerance of lower bound of target individual power rate, default is 0.02.
tol.b2	tolerance of upper bound of target individual power rate, default is 0.02.
tol.a	tolerance of upper bound of target family-wise Type\_I error, default is 0.05.
n.min	the minimum sample size needed, default is 5.
n.max	the maximum sample size afforded, default is 70.
nstep	the search precision of "linear" method. Linear method starts searching from nmin, and linearly increases n by nstep each time to find the optimal sample size.
rep	number of repetition of generating posteriors $Pr(delta_i > delta_cut | data) > T_p$, default is 500.
pathcode	path to the BRugs model file. If missing, current working directory will be used to search for BRugs model.
codefile	name of the file containing the BRugs model. Defualt is "BUGS\_Bin\_2arm.txt".
pathout	path for the output files. If missing, current working directory will be used to search for BRugs model.
method	available methods "linear" or "bisect". "linear" method linearly increases the sample size from n.min by nstep each time until it hits the optimal sample size or until it reaches n.max. "bisect" method use the Bisection method to find the optimal sample size. Default is "linear" method.

Value

For each sample size searched ($n_searched$), find.n generates text file "alt5\_0.025\_<$n_searched$>.txt" and "null\_0.025\_<$n_searched$>.txt" to store posteriors ${ Pr(delta_i > 0.025 | data) > 0.85; i=1,...,k }$ at each one of $N_rep$ repetitions under alternatives and null hypothesis. After $N_rep$ repetitions, the Operating Characteristics corresponding to $n_searched$ is output to text file "history\_linear.txt". Once the searching-for-sample-size process is over, find.n outputs a message whether the satisfactory sample size is found or not. Meanwhile, the Operating Characteristics related to the final sample size is print out.

del.cut	efficacy boundary $delta_cut$
T	threshold probability $T_p$
n	final sample size per arm per group
alpha.fw	final family-wise Type-I error rate
power.ind1,...,power.indk	final individual power rates for each type of tumor

Note

find.n calls function bisect, power.fun, and alpha.fun.

Examples

## IMPORTANT: Please install package BRugs
## use Linear method
#find.n(del.cut=0.025, k=3, T=0.85, p.c=c(0.5, 0.4, 0.65), effect.size=0.2,
#        alpha.target=0.1, power.target=0.9, 
#        tol.b1=0.02, tol.b2=0.02, tol.a=0.05,
#        rep=500, pathcode=path.package("hbdct"), codefile="BUGS_Bin_2arm.txt", pathout=getwd())

## use Bisect method
#find.n(del.cut=0.025, k=3, T=0.85, p.c=c(0.5, 0.4, 0.65), effect.size=0.2,
#        alpha.target=0.1, power.target=0.9, 
#        tol.b1=0.02, tol.b2=0.02, tol.a=0.05,
#        rep=500, pathcode=path.package("hbdct"), codefile="BUGS_Bin_2arm.txt", pathout=getwd(),
#        method="bisect")

hbdct documentation built on July 4, 2017, 9:09 a.m.