odds ratio of proc causalmed

[Well21]

Hi everybody!

Below is the code and result I used to make mediation analysis. I want to know if the OR in proc causalmed has the same meaning as in logistic or cox analysis. That is, if OR <1, the variable plays a protection role on the result.

The problem is that while the HR of interested variable (homa) I put in the proc phreg is >1, the OR of the same variable I've got in proc causalmed is <1. Both of them are significant. And I wonder whether something goes wrong or how can I explain the result?

Thanks!

My code of proc causalmed:

proc causalmed data = homa;
class status(ref = first) drink smoke bmi cvd;
model status = AVGapnea homa;
mediator homa = AVGapnea;
covar drink smoke bmi cvd;
bootstrap CI (PERC) Nboot = 1000;
run;

My code of proc phreg:

proc phreg data = homa;
class drink smoke bmi cvd;
model time*status(0) = AVGapnea drink smoke bmi cvd homa/ risklimits = WALD;
run;

The results:

 

 

[sbxkoenk]

Hello @Well21 ,

I think @SAS_Rob can answer this !

I am not very familiar with ( PROC ) CAUSALMED.

Thanks,

Koen

[Well21]

Thanks for your reply！
Still waiting for the answer...

[sbxkoenk]

---

@Well21 wrote:

Still waiting for the answer...

---

They are just getting out of bed in the US.

Koen

[SAS_Rob · Accepted solution]

If there were no mediator effect and you were running a survival model in CAUSALMED (which can be done in SAS Viya) then I would expect the results to be similar.  But in this case you are trying to compare an odds ratio (that is adjusted for the mediator model) with a hazard ratio from a non-mediation model so no fair comparison could be made.

[Well21]

Thanks a lot!
I will try SAS Viya latter.
So I assume that the odds ratio of homa here can still be interpreted as the protecting factor for the main results? It's just against the clinical reality.
Anyway, thanks for your help!