03-10-2013 01:03 PM
I would like to calculate a sample size (with given power i.e. 80%) for a one sample log rank survival study.
1) Is there a way to do this without using a control arm?
2) I can base the assumptions on a published meta analysis of historical control. Is the only way to do this by using the proc power statement with a twosamplesurvival test option and use the historical control data as one of the groups?
Thanks in advance,
03-12-2013 09:17 AM
You might be able to iterate to a solution by using PROC POWER's TWOSAMPLESURVIVAL option. First, set up the types of curves, accrual times, significance level, censoring patterns, etc. as in the documentation examples. Then, assign its option, POWER, to a missing value. Finally, use its GROUPNS option in the following way: Assign a very large number to the sample size of your second (historical control) sample, but vary the number in your first sample till you get the desired statistical power. A macro that changes this latter number across a specified range of sample sizes should help you iterate to this desired power. You will notice that the estimated statistical power does NOT change very much as you increase the sample size of your second (control) sample.
03-12-2013 10:44 AM
Thank you for your reply and for confirming that it is ok to use this procedure with a historical control as one of the groups.
This is, in fact, what I have done since the historical control data that I have actually has a sample size that is approx 6 times the size of the experimental group.
(I was sort of hoping someone would have a response to (1) above - but I guess not
03-13-2013 08:46 AM
Actually, the second sample can be a hypothetical group with a very large sample size, greatly exceeding that of your historical control group, but with the specifications of this group so that you can compare its specifications with those of your experimental group. This second sample mimics a hypothetical group with an infinite population to obtain results comparable to a one-sample test with SAS's PROC POWER procedure, which lacks such a test. As stated previously, the computed statistical power does not change very much as one increases the sample size of this second, hypothetical group.