Statistical test for difference between three HRs?

[Oliver Scott]

Hi all,

I am studying the effect of a medication on a cancer outcome. I have stratified the exposure into three distinct/different groups, and have obtained HRs and corresponding 95% CIs for all three groups.

Is there some kind of simple statistical test I can use to test for an overall difference between all the groups? Obviously comparing between 1 and 2, 2 and 3, and 1 and 3 is simple, but I wasn't sure about a test for all three groups at once.

Thanks, Oliver

[Maarten Buis]

Estimate them in one model with interaction terms, and use test (a Wald test) to test if all interaction terms are equal to 0

[Oliver Scott]

Estimate them in one model with interaction terms, and use test (a Wald test) to test if all interaction terms are equal to 0

Thanks for your response Maarten, much appreciated.

I grouped the variable I am interested in stratifying by into 4 categories (0, 1, 2 and 3). I then ran an interaction between my medication variable and the categorical variable like so:

stcox binarymedicationvariable##categoryvariable

I then ran the following test:

testparm binarymedicationvariable##categoryvariable

Is this correct? If I just use test, then I am not performing a global test across all groups, correct?

[Maarten Buis]

You can do a global test with test, testparm is just convenient but otherwise equivalent. Now that is not saying anything bad: if two commands are equivalent for your purpose and one is more convenient than the other, then choosing the convenient one is the obvious choice...

[Oliver Scott]

You can do a global test with test, testparm is just convenient but otherwise equivalent. Now that is not saying anything bad: if two commands are equivalent for your purpose and one is more convenient than the other, then choosing the convenient one is the obvious choice...

Thanks again Maarten.

When I run

testparm binarymedicationvariable#categoryvariable

, I get output as I'd expect.

How would I run the test command to derive similar output? If I run

test binarymedicationvariable#categoryvariable

, STATA tells me "binarymedicationvariable#categoryvariable not found". I can run

test 1.binarymedicationvariable#1.categoryvariable

, but that is obviously not a global test...

Oliver

[Øyvind Snilsberg]

How would I run the test command to derive similar output?

You would write out each interaction,

sysuse auto, clear
recode mpg (0/19=0) (20/29=1) (nonmi=2)
reg price foreign#mpg
testparm foreign#mpg
test 1.foreign#0.mpg 0.foreign#1.mpg 1.foreign#1.mpg 0.foreign#2.mpg 1.foreign#2.mpg