Hello @SteveDenham, Thank you for your quick reply. A patient can be followed-up over time between different hospitals with the same unique ID. Regarding this statement: " Next, what about interactions between your independent variables? Are you willing to assume that males on tttA have the same effect on interval as females on tttB? If you can make that assumption, then the model statement looks OK. And last, how many unique patients are involved. To have pat as a continuous effect, you will almost certainly need to sort your data by hospital and patient." I cannot assume that males on tttA have the same effect on interval as females on tttB. I will test sex*treat interaction There are 455 distinct patients (350 on tttA and 105 on tttB) Yes my data are sorted by hospital and pat See below results with and without sex*treat interaction How to interpret the intercept ? Regarding this statement: "Next, I suspect that interval is not a variable with Gaussian error. Intervals like this give rise to Poisson distributed counts, and are generally gamma distributed. If that is the case, you will likely need to shift to PROC GLIMMIX. " Should I run the model using PROC GLMMIX with a gamma link and select the model with lower AIC? Regards
... View more