05-24-2017 09:37 AM
I want to create time-interaction covariate to test whether covariates in model meet proportional hazard assumption.
First few rows of input dataset is as follows:
T1 & T2 correspond to entry and exit time of counting process style of input.
Status is time-dependent variable while COV1-COV3 are time-fixed covariates (continuous or categorical).
Event (0 or 1) and Time indicates censoring status and censored time, respectively.
PHREG model in SAS looks like below:
proc phreg data=have; class status(ref='A') COV3(ref='F'); model (T1,T2)*Event(0)=Status COV1 COV2 COV3 COV1_t; COV1_t = ; run;
At this point, how can I express time-interaction covariate in programming statement of PHREG procedure?
a) COV1_t = (T2-T1)*COV1
b) COV1_t = T2*COV1
c) COV1_t = Time*COV1
I think c) is right way when compared with usual phreg model, but b) seems plausible as well.