Proc power for a case control study with a continuous exposure

[kpberger]

Hi all. Can proc power do calculations for case-control studies with a continuous exposure? I haven't found any examples online. There doesn't seem to be a place where you tell the program what kind of study you have. Does anyone know? Thanks.

[ballardw]

The syntax for proc power except for the Plot statement, basically does use the 'study you have'. Look at the names of the statements:

Coxreg

logistic

multreg

onecorr

onesamplefreq

Cox regression, logistic regression , multiple regression, simple and partial Pearson correlation, single proportion and so forth.

What exact test will you be performing on the data?

You may have to provide more details on how "continuous exposure" is involved.

[kpberger]

Thanks for your reply. I don't see how the types of analysis correspond to study designs. For example, you can use a logistic regression in a case-control or cohort study. I would like to find the min and max ORs my case-control study has the power to detect with a logistic regression. I can't see how to do that though, so I wrote code to get the N needed:

proc power; logistic alpha=0.05 power=0.80 ntotal=. vardist("exposure")=normal(2,13) testpredictor="exposure" testregcoeff=-0.0896 intercept=-0.0725;
run;

It says I need 36 people so something isn't right... The regression coefficient and intercept are from a logistic regression in the study - this is a post hoc calculation requested by a reviewer. Nowhere in this code though have I put in the number of controls per case (1:1 here) which I would think it would need.