How to run instrumental variable analysis in Cox PH model in SAS EAS 9.4

[Lin1110]

Hello everyone, it is good to be here asking some useful help or advice.
I am now trying to run an instrumental variable analysis in a survival analysis-based research.
My topic lies in pharmacoepidemiology field, and the outcome is event of interest, which is not death though.

I have used PROC PHREG twice to mimic 2SLS (two stage least squares, like PROC REG twice in traditional linear regression).
However, it came out extremely weird results (e.g., hazard ratio showing 1x10^8...). It is obviously wrong. 

If there is any possible or practical advice to deal with this situation?

Thanks for your reading and wishing for any suggestions!

[sbxkoenk]

Hello,

( I have moved your topic to the 'Statistical Procedures' board. )

I had never heard before of a 2SLS instrumental variable (IV) approach in a survival context.

As 2SLS IV estimation | regression is quite common in econometrics , I ask @SASCom1  if she has an idea (?).

BR,

Koen

[Lin1110]

Koen, thanks a lot!

I would like to do instrumental variable analysis because I found someone did it in Cox model as well; however, they did it in R package: ivcoxph, which is not an official package as I know. This package came from rdrr.io and it is not allowed for me to use in our analytical department.

Lin

[SASCom1 · Accepted solution]

Hello,

Unfortunately, I am not aware of instrumental variables approach used in the survival analysis context. If you know or other people have discussed how to perform the two stages in the Cox model context, perhaps you can program the two stages yourself following how they performed the two stages. If you are not certain you are following their approach correctly or you are getting strange results using your own data, maybe you can first verify whether you are following the approach correctly by using the same data set that has already been implemented by others in R or another package with reported results.

Sorry I cannot be of more help. Good luck!

[Lin1110]

Thank you for the reply, and I will consider whether to use it in the future.