Trial Designs Based On Fisher's Exact Test

Calculates sample size, effect size and power based on Fisher's exact test

Usage

fe.ssize(p1, p2, alpha=0.05, power=0.8, r=1, npm=5, mmax=1000)
CPS.ssize(p1, p2, alpha=0.05, power=0.8, r=1)
fe.mdor(ncase, ncontrol, pcontrol, alpha=0.05, power=0.8)
mdrr(n, cprob, presp, alpha=0.05, power=0.8, niter=15)
fe.power(d, n1, n2, p1, alpha = 0.05)
or2pcase(pcontrol, OR)

Arguments

p1

response rate of standard treatment

p2

response rate of experimental treatment

d

difference = p2-p1

pcontrol

control group probability

n1

sample size for the standard treatment group

n2

sample size for the standard treatment group

ncontrol

control group sample size

ncase

case group sample size

alpha

size of the test (default 5%)

power

power of the test (default 80%)

r

treatments are randomized in 1:r ratio (default r=1)

npm

the sample size program searches for sample sizes in a range (+/- npm) to get the exact power

mmax

the maximum group size for which exact power is calculated

n

total number of subjects

cprob

proportion of patients who are marger positive

presp

probability of response in all subjects

niter

number of iterations in binary search

OR

odds-ratio

Details

CPS.ssize returns Casagrande, Pike, Smith sample size which is a very close to the exact. Use this for small differences p2-p1 (hence large sample sizes) to get the result instantaneously.

Since Fisher's exact test orders the tables by their probability the test is naturally two-sided.

fe.ssize return a 2x3 matrix with CPS and Fisher's exact sample sizes with power.

fe.mdor return a 3x2 matrix with Schlesselman, CPS and Fisher's exact minimum detectable odds ratios and the corresponding power.

fe.power returns a Kx2 matrix with probabilities (p2) and exact power.

mdrr computes the minimum detectable P(resp|marker+) and P(resp|marker-) configurations when total sample size (n), P(response) (presp) and proportion of subjects who are marker positive (cprob) are specified.

or2pcase give the probability of disease among the cases for a given probability of disease in controls (pcontrol) and odds-ratio (OR).

References

Casagrande, JT., Pike, MC. and Smith PG. (1978). An improved approximate formula for calculating sample sizes for comparing two binomial distributions. Biometrics 34, 483-486.

Fleiss, J. (1981) Statistical Methods for Rates and Proportions.

Schlesselman, J. (1987) Re: Smallest Detectable Relative Risk With Multiple Controls Per Case. Am. J. Epi.