[R] Sample size calculation proportions with EpiR: Discrepancy to other calculators

[Thomas Lumley]

On Tue, 26 May 2009, Chuck Cleland wrote:

> On 5/26/2009 2:53 AM, Karl Knoblick wrote:
>> Hallo!
>>
>> I have done a sample size calculation for proportions with EpiR. The input is:
>> treatment group rate p=0.65
>> control group rate p=0.50
>> significance level 0.95
>> power 0.80
>> two-sided
>> ration group 1 and 2: 1.0
>>
>> I have done this in the following way:
>> library(epiR)
>> epi.studysize(treat = 0.65, control = 0.5, n = NA, sigma = NA, power = 0.80,
>>    r = 1, conf.level = 0.95, sided.test = 2, method = "proportions")
>>
>> Result:
>> $n
>> [1] 82
>>
>> PASS 2002 and NQuery give both 170 subjects per group without continuity correction. With continuity correction 183 per group.
>>
>> Looking at http://statpages.org/proppowr.html I get 182 subjects per group (with continuity correction, I admit).
>>
>> What am I doing wrong? Can anybody explain this?
>
> epi.studysize(treat = .65, control = .50, n = NA, sigma = NA,
> power = 0.80, r = 1, conf.level = 0.95, sided.test = 2, method = "cohort")
>
>  gives the same sample size as PASS 2002 and NQuery (170 per group).
>

And simulation confirms that the larger numbers are correct. I don't know 
what is happening with epi.studysize(,method="proportion").

epi.studysize(,method="cohort") doesn't seem exactly appropriate, since 
judging from the example on the help page the inputs are supposed to be 
cumulative incidence rather than probabilities.

 	-thomas

Thomas Lumley			Assoc. Professor, Biostatistics
tlumley at u.washington.edu	University of Washington, Seattle