how to derive difference in percentage change from baseline (log-transformed)

[CL52]

Dear all,

I struggled to figure out how to derive SD and confidence interval for difference between groups in percentage change from baseline, analyzed in log-transformation. 

In proc mixed, I have [log(Post) - log(BL)] as response. and treatment as indepent variables, visit, and treatment by visit interaction. So I know, the coefficient beta1 for treatment (and beta0 as intercept), I can transform back to get the percentage change (exp(beta0+beta1) - 1)*100%. 

However what I am interested in is the difference between the group. 

[exp(beta0+beta1)-1] - [exp(beta0)-1] = exp(beta0)*(exp(beta1) - 1)

How can I derive the SD and CI of the above with the LSmeans output? 

Any reference I can go to? 

Many thanks!

Carrie 

[StatDave · Accepted solution]

The LSMEANS statement cannot be used to estimate or test nonlinear combinations of model parameters. For that, you can use the NLEST macro.