01-03-2017 05:48 PM
I need to calculate the sample size in person-years for a single group prospective cohort study (surveillence study) aiming to demonstrate non-inferiority. I have some information based on population A from a previous study. I want to obtain the sample size needed to show that the events of interest in population B is non-inferior to population A by some margin or error. In SAS, there is a one-sample one-sided method for analyzing proportions. However, my event is extremely rare. Through quite a bit of searching it seems that the normal approximation is being used to calculate sample sizes such as in my case. At this point, I have decided to use PROC POWER’s non-inferiority test for a proportion. Does anyone know if it is acceptable by the FDA (or the standard) to use the normal approximation assumption to calculate sample size for non-inferiority trials even when the event is rare? If it is not acceptable, does someone have advice on which method/procedure you might use or have used? Or to the software that may help?
08-22-2017 03:19 PM - edited 08-22-2017 03:28 PM
1. When you mention, that you seek to demonstrate non-inferiority, I automatically think about non-inferiority clinical trials, that show that the new treatment is not inferior to current standard of care. If this is what you seek to do, than this guidance may be of some help. https://www.fda.gov/downloads/Drugs/Guidances/UCM202140.pdf
2. On the other hand as I understand you want to compare 2 proportions, this could be done manually (something like this http://udel.edu/~mcdonald/statfishers.html) without having raw data, and after that you can estimate power (https://support.sas.com/documentation/cdl/en/statug/63033/HTML/default/viewer.htm#statug_power_sect0...). In SAS University edition: tasks and utilities>Power and sample size>test of proportions.
3. From regulatory perspective it is wise to seek scientific advice, because if you want to establish non-inferiority, you need to very clearly define the non-inferiority margin and robust rules in doing that are lacking.
4. Trying to establish that population B is non-inferior to population A could result in methodological issues.
If you would clarify the problem, I could perhaps be of more help.