I wanted to conduct the Cox regression through PROC PHREG, however, based on the Kaplan-Meier plot, the treatment seems change over time because they interacted with each other. The treatment is four categories, so I added interaction in the model as follows:
proc phreg data=ps_w_adj;
class treatment(ref='1');
model time*survival(0) = treatment treatment2_ti treatment3_ti treatment4_ti /rl;
treatment2_ti=log(followup_time)*(treatment=2);
treatment3_ti=log(followup_time)*(treatment=3);
treatment4_ti=log(followup_time)*(treatment=4);
weight ps_weight_adj;
run;
In the results, it provided me the HR for main effects and the interaction. But the result is really hard to interpret because in the model, I didn't specify when the slope started to change. So, I checked online, and found someone used R to program similar situation as follows:
Coxph(formula = Surv(survyr, status)~Treatment*tt(Treatment),
data=Dataset, tt=function(x,t, ...) as.numeric(t>7))
The as.numeric(t>7) specified to estimate the slope before 7 years, so if the hazard ratio for treatment is 0.74 and the HR for the interaction is 1.87, they can concluded that before 7 years, the HR for the treatment is 0.74, and after 7 years, the treatment's HR is 0.74*1.87=1.39.
I am wondering that is feasible to achieve the similar result and program in SAS? Thank you so much!!
