Date:         Wed, 24 Jun 2009 11:21:30 -0500
Reply-To:     Herbert Morley A <Morley.Herbert@HCAHEALTHCARE.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Herbert Morley A <Morley.Herbert@HCAHEALTHCARE.COM>
Subject:      Sample Size by Simulation
Content-Type: text/plain; charset="us-ascii"

A number of times I have come across people suggesting that the use of a simulation to determine an appropriate sample size. Generally I would use the SAS procedure Proc Power for the appropriate data structure, or look it up in tables.

The current problem. The docs want to do a trial where they are comparing two types of treatment for a piece of vein. They are looking for a non-inferiority outcome. Historically treatment A has a failure rate of 20%. Because treatment B has benefits for the patient, they will accept a failure of 24% (20% worse) as an acceptable outcome. Worse than that, the tradeoff would make it a no-go. These are yes/no outcomes.

How would I do a simulation to get the sample size for power = 0.8 (say). I am looking for the steps, rather than the actual code, which I probably could handle.

I tried a uniform random number generation with a divide at 0.2 to create a dataset with the desired proportion of yes/no's. I then ran a bootstrap to get samples and then calcluated the mean and CI for the distribution.  Is this the right path? If so, now what. If not, what should I be doing?

Thanks in advance for any hints.
Morley Herbert

Biomedical and Surgical Research
Medical City Dallas Hospital
7777 Forest Lane, Ste C-740
Dallas, TX 75230
972-566-6716
Morley.Herbert@HCAhealthcare.com