Power calculation for rate ratio test

#1
Hello all! I'm stumped on a question and hoping someone can help.

I have calculated a rate ratio for the rate of genetic mutations in a patient group vs. control group. The rate for each group is calculated as (mutation rate)/(number of samples * number of callable base pairs) and the significance is calculated for ratio>1. I now am interested in calculating the power I have to detect a certain difference in rate (i.e. how much greater than 1 the rate ratio is) based on the number of samples and number of callable bases in each group.

Does anyone know of a straightforward way to do this? I haven't found an answer from my searching the internet, and the R package I used to do the rate ratio test doesn't give power. Thanks!!
 

hlsmith

Not a robit
#2
What will the results be used for? Is this for class or for a real-life context? There was a recent epidemiology paper on power for the precision on the rates instead of difference that may be of interest. If you are stuck on the ratio part, looking at power calculations for Poisson regression model may be of interest or you could just do a power calculation for Fisher's exact test. Also, this would likely easily be done with a data simulation.
 
#3
What will the results be used for? Is this for class or for a real-life context? There was a recent epidemiology paper on power for the precision on the rates instead of difference that may be of interest. If you are stuck on the ratio part, looking at power calculations for Poisson regression model may be of interest or you could just do a power calculation for Fisher's exact test. Also, this would likely easily be done with a data simulation.
This is for real life (my research). If using a power calculation for Fisher's exact test is valid for this purpose that would be much easier! I was trying to do the calculations here http://www.pmean.com/10/PoissonPower.html but having trouble following. Thank you!
 

hlsmith

Not a robit
#4
Wel! You same patients in your description, is this a randomized study or more observational? If latter, do you need to control for anything to balance sample difference?

it might be easiest for you to plot what you expect the numbers to be and we can play around with simulations. In medicine people are historically inclined to look at RR, but risk differences are more informative. I said fishers test cause i had little to work with on assumed rate s andwas appraid of a little rate situation were something like chisq may not be a good approximate.
 
#5
Wel! You same patients in your description, is this a randomized study or more observational? If latter, do you need to control for anything to balance sample difference?

it might be easiest for you to plot what you expect the numbers to be and we can play around with simulations. In medicine people are historically inclined to look at RR, but risk differences are more informative. I said fishers test cause i had little to work with on assumed rate s andwas appraid of a little rate situation were something like chisq may not be a good approximate.
It is observational, and we are not doing anything to balance for control. The sample sizes are 183 and 777. Unfortunately I don’t have expected rate numbers, but thank you anyways for your help!!
 

hlsmith

Not a robit
#6
Well you can do power calculations with out proposed rate values. Use best guesses based on your knowledge and published/similar studies.