Differences in CIF curves in PHREG vs LIFETEST

[jpsmith]

Hello,

I am analyzing time-to-event data from a dataset with competing risks and would like to present the CIF curves. Could someone please explain to me the difference between producing these curves in LIFETEST vs PHREG? Is it as simple as the PHREG procedure allows for the use of covariates?

Lastly, why do the CIF step curves produced from PHREG move in 'lock-step' with each other when an event is in either exposure category, and those produced by LIFETEST only affect the curve for the exposure in where the event happened. Is this because of the proportional hazards assumption in PHREG? I am attaching a PDF file with two graphs - both the same bivariate analysis of exposure and disease outcome. One is ran with PHREG (lighter CI bands) and the other with LIFETEST (darker bands). Which is more appropriate to present?

I am running SAS 9.4 using the EVENTCODE option for both procedures, with CIF plots requested. Example code is below. Thank you for your time and effort in clearing up the differences between these two procedures!

Jonathan

proc lifetest data=example plots=(CIF(test)) conftype=loglog notable ;
	time time*disease(0)/eventcode=1;
	strata exposure;
run;

proc phreg data=example covs(aggregate) plots(overlay=stratum)=cif;
	model time*disease(0)=exposure/eventcode=1 ties=efron rl;
	baseline covariates=exposure; 
run;

[Reeza · Accepted solution]

Lifetest uses non parametric approach and doesn't include covariates. 

PHREG is a semi-parametric approach and can account for covariates. As part of being semi-parametric approach, it looks like you're using the Fine-Gray model competing risks type approach, there are base assumptions that should be verified. 

You can confirm some of these differences by reading the Overview section of each PROC documentation as well as the details sections if you want to get into the technical differences in how things are calculated. 

http://documentation.sas.com/?docsetId=statug&docsetTarget=statug_introsurv_sect002.htm&docsetVersio...

[jpsmith]

Thank you!

[JacobSimonsen]

Yes, reeza is right. You should be able to get the same from the two procedures if you instead have "exposure" in the strata statement in proc phreg, and delete "exposure" from the model line.