Hi all,

I am carrying out a study examining the effect of a medication on breast cancer death.

I have carried out analyses conceptualizing exposure as binary (use/no use) as well as splitting the medication up into various dose categories (e.g. 0-3 months, 3-6 months....1y-2y, 2y-3y, 3y+).

For dose-response analysis, I created a categorical medication variable, with different categories corresponding to different levels of dosage:

Code:
gen medication=1 if timetofmedication==1
replace medication=2 if timetomedication91dddtime==1
replace medication=3 if timetomedication182dddtime==1 
replace medication=4 if timetomedication273dddtime==1 
replace medication=5 if timetomedication365dddtime==1 
replace medication=6 if timetomedication730dddtime==1 
replace medication=7 if timetomedication1095dddtime==1 
replace medication=0 if medication==.

label define medicationlabel 0"0 DDDs" 1"1-90 DDDs" 2"91-181 DDDs" 3"182-272 DDDs" 4"273-364 DDDs" 5"365-729 DDDs" 6"730-1094 DDDs" 7"1095 or more DDDs", replace
label values medication medicationlabel
I then ran a survival model like so:

Code:
stcox i.medication [other confounders...]
Now, I am wondering what the best method is to test whether or not there is a statistically significant dose-response effect for the medication.

Is the
Code:
contrast
command the best way to go about this? i.e. should the code be
Code:
contrast p.medication
? From here (http://www.baileydebarmore.com/epicode/p-for-trend), it seems like it should be, but I just thought I'd check to see what everyone here thought.

Warm regards, Oliver