topic Re: PH assumption in cox model by including interaction term in Statistical Procedures

PH assumption in cox model by including interaction term

Marjolein — Fri, 04 Nov 2016 10:29:54 GMT

Hi all,

I have a question regarding evaluating the PH assumption in a Cox model. I did this by introducing an interaction term for my exposure variable. My exposure variable is a categorical variable.

proc phreg data=mydata;

model time*censor(0)=exposure exposureT;

exposureT=exposure*(log(time));

RUN;

However, I get output but also an error in the log-window with:

An illegal argument is used in the function call in statement number 1 at line 1331 column 1.

What does this mean and does it influence my output?

Thanks!

Re: PH assumption in cox model by including interaction term

Reeza — Fri, 04 Nov 2016 11:28:19 GMT

Please post your log including code submitted.

Re: PH assumption in cox model by including interaction term

JacobSimonsen — Fri, 04 Nov 2016 12:16:26 GMT

The code is correct. The error message must be because you have time values that are either zero or negative.

Re: PH assumption in cox model by including interaction term

Marjolein — Mon, 07 Nov 2016 08:46:58 GMT

There was one, and I took it out, but the error still persisted!

Is it maybe because it cannot handle missings in the other covariates? Or should that not matter?

Re: PH assumption in cox model by including interaction term

JacobSimonsen — Mon, 07 Nov 2016 09:04:11 GMT

I dont think it has to do with missing in covariates.

Try this variant:

proc phreg data=mydata(where=(time>0));

model time*censor(0)=exposure exposureT;

exposureT=exposure*(log(time));

RUN;

I made a typing mistake and have corrected this: it should be ">", not ">=", in the where statement.

Re: PH assumption in cox model by including interaction term

Marjolein — Mon, 07 Nov 2016 12:58:33 GMT

Yes, now it runs! Perfect! Thanks a lot! 😄