topic Re: Survival Analysis with 3 levels of censor in Statistical Procedures

Survival Analysis with 3 levels of censor

Tawfiq — Mon, 27 Mar 2017 01:31:24 GMT

Hi All,

I'm doing survival analysis, and I want to censor data >5 and state=2 or state=3, and the SAS code I'm writing is

proc phreg data=bc.def;

class size (ref ='1') / param=ref;

model survive(<=5)*state(2|3)=size/ties=exact rl;

run;

While

size which has 4 categories (1, 2, 3, 4)

state which has 3 categories (1, 2, 3), and 1=event

survive should be less than or equal 5

I wonder whether it's possible to modify the above code, and get what I intend to do. I appreciate your help!

Re: Survival Analysis with 3 levels of censor

Reeza — Mon, 27 Mar 2017 04:37:43 GMT

Is that working code?

I'm a bit uncertain what your question is here.

Re: Survival Analysis with 3 levels of censor

Tawfiq — Mon, 27 Mar 2017 09:12:14 GMT

Hi Reeza,

The code is not working. What I want is to censor survive >5 for state 2 and state 3 and get survive <=5 for state=1. If I use the model as follows

proc phreg data=bc.def;

class size (ref ='1') / param=ref;

model survive*state(2)=size/ties=exact rl;

run;

the model works; but it doesn't provide survive <=5 and it adds state=3 with state=1 (or event=1) which shouldn't be the case. My purpose is to to use the model with the 2 conditions as ... model survive (<=5)*state(2 &3)=size/ties=exact rl;

Re: Survival Analysis with 3 levels of censor

Reeza — Mon, 27 Mar 2017 11:41:48 GMT

I think you need to precalculate the censor and survival time variables. Use a data step to create new variables that have the values you need.

Re: Survival Analysis with 3 levels of censor

Tawfiq — Mon, 27 Mar 2017 20:04:58 GMT

Hi Reeza,

I created two variables (one for censoring and another for survival_5y) and used the model and it worked!

Thanks for the help!

Re: Survival Analysis with 3 levels of censor

JacobSimonsen — Tue, 28 Mar 2017 11:43:41 GMT

You first need to create a new censoring variable. And you can run your Cox-regression:

data mydata;
   set bc.def;
   if (survive>5)+(state in (2,3)) then censur=1;
  else censur=0;
run;

proc phreg data=mydata;
      class size (ref ='1') / param=ref;
      model survive*censur(1)=size/ties=exact rl;
run;

Good luck:-)

Re: Survival Analysis with 3 levels of censor

Tawfiq — Tue, 28 Mar 2017 20:39:43 GMT

Thanks for your help! It also works!