How to generate KM curve for cause-specific competing risk model

blueblue — Tue, 30 Jan 2024 05:54:49 GMT

Hi, I need to generate KM for Cox cause-specific competing risk. However, it seems that every approach I tried, all come out to be the same, that the KM curve or CIF (both generated from LIFETEST) look very differently from the CIF generated by PHREG (Cox cause-specific). Even for FG competing risk model, the CIF generated from LIFETEST does not share similar pattern/shape to the CIF generated by PHREG. How to explain these? I know KM would not be exactly the same as CIF, yet the shape/pattern should not differ so much between the CIF generated from LIFETEST or PHREG, right?

Here are the approaches I tried for KM curves for Cox cause-specific competing risk:

1.) PROC LIFE TEST with: TIME T*Status(0 2);

2.) PROC LIFE TEST with: TIME T*Status(0) / eventcode(COX)=1; -- ERROR message

3.) PROC LIFE TEST with: TIME T*Status(0) / eventcode=1;

4.) PROC LIFE TEST with: TIME T*Status(0) / failcode=1 2;

In PHREG, I use the following codes for the Cox cause-specific competing risk:

proc phreg data=BMT plots (overlay)=cif;

class disease;

model T*Status(0) / eventcode(COX)=1;

baseline covariates=risk;

run;

And how to generate KM curve for cause-specific competing risk model?

topic How to generate KM curve for cause-specific competing risk model in Statistical Procedures

How to generate KM curve for cause-specific competing risk model