For a continuous variable like Age, you would have to use the ESTIMATE statement. The trick with figuring out the coefficients to use in the ESTIMATE (or CONTRAST) statement is always to write what you want in terms of the model you have specified. Since you are using the AT option, I assume that your model has Age interacting with a categorical variable (FTV). So, assuming that your logistic model in GENMOD is
class ftv;
model y=age ftv age*ftv ... ;
and you want to estimate the change in odds for a 10 unit change in Age at each level of FTV, then let's start with the above model and doing the estimation at the first FTV level. An odds ratio is a difference in log odds and the log odds is what your model directly estimates. So, the desired log odds ratio is (assuming FTV has three levels)
logodds(age=2, ftv=1) - logodds(age=0, ftv=1) = (b0+b1*10+b2+b5*10) - (b0+b1*0+b2+b5*0) = b1*10+b5*10
where the model parameters are b0 (is the intercept), b1 (for age), b2-b4 (for ftv), b5-b7 (for age*ftv). Note that if you are instead using a log-linked binomial model to model the log probability instead of the log odds, then all this works the same but the above estimates the difference in log probabilities which is the log relative risk. Now, translating the above as an ESTIMATE statement and using the EXP option to turn the log odds ratio estimate into an odds ratio (or the relative risk for log binomial model):
estimate "(OR Age at FTV=1)" age 10 age*ftv 10 0 0 / exp;
If your model is different, you need to go through the same exercise to properly determine the coefficients. Do NOT make the common mistake of trying to intuit the coefficients... you will very likely get it wrong.
