In my PH Cox model (using PROC PHREG), one of my continuous covariate (EGFR) violates the PH assumption. Based on the plot of Schoenfeld residual below, I decide to use a heaviside function with the time cut-off at 2 (the 600th rank of time) on this variable.
I believe that, with the heaviside function, I still assume the proportional hazard in each time interval. I tried the ZPH option and ASSESS statement, and I got the note: "Model assessment is not available when there are time-dependent covariates. The ASSESS statement is ignored." I further tried to output Schoenfeld residual (see code below), but I got no observation in the output dataset. Does anyone know how to assess the PH assumption in this model?
PROC PHREG DATA= DF;
CLASS EXPO SEX;
MODEL TIME*SA(0,)= EXPO AGE SEX EGFRT1 EGFRT2 / RL=PL;
EGFRT1=0;
IF TIME<2 THEN EGFRT1=EGFR;
EGFRT2=0;
IF TIME>=2 THEN EGFRT2=EGFR;
OUTPUT OUT=SCHOEN RESSCH=SCHEXPO SCHAGE SCHSEX SCHEGFRT1 SCHEGFRT2;
RUN;
... View more
