[R] obtainl survival curves for single strata

Terry Therneau therneau at mayo.edu
Fri Feb 1 16:11:02 CET 2013

    I can almost but not quite get my arms around what you are asking.  A bit more detail 
would help.  Like
      cph.approve = what kind of object, generated by function __

But, let me make a guess:
    cph is the result of coxph, and the model has both covariates and a strata
    You want predictioned survival curves, more than 1, of the type "covariates = a, b,c, 
strata=1"  "covariates = d,e, f, strata=2", ... for arbitrary covariates and strata.

Now, the predicted survival curves for different strata are different lengths.
The survfit.coxph routine gets around this by being verbose: it expects you to give it 
covariate sets, and returns
all of the strata for each covariate. This allows it to give a compact result.   You can 
always do:
    newpred <- survfit(cox-model-fit, newdata=something)
    newpred[5,17] #  survival curve for the 5th strata, covariates from the 17th row of 

But, you want to put the results into a matrix, for some pre-specifed set of time points.  
Take it one step further.
     mytimepoints <- seq(0, 5*365, by=30)  # every 30 days
     z <- summary(newpred[5,17], time=mytimepoints, extend=TRUE)$surv

The summary.survfit function's "time" argument was originally written for people who only 
wanted to print certain time points, but it works as well for those who only want to 
extract certain ones.  It correctly handles the fact that the curve is a step function.

Terry Therneau

On 02/01/2013 05:00 AM, r-help-request at r-project.org wrote:
> What is the syntax to obtain survival curves for single strata on many subjects?
> I have a model based on Surv(time,response) object, so there is a single row per subject and no start,stop and no switching of strata.
> The newdata has many subjects and each subject has a strata and the survival based on the subject risk and the subject strata is needed.
> If I do
> newpred<- survfit(cph.approve,new=newapp,se=F)
> I get all strata for every subject.
> Attempting
>> >  newpred<- survfit(cph.approve,new=newapp,id=CertId,se=F)
> Error in survfit.coxph(cph.approve, new = newapp, id = CertId, se = F) :
>    The individual option is  only valid for start-stop data
>> >  newpred<- survfit(cph.approve,new=newapp,indi=T,se=F)
> Error in survfit.coxph(cph.approve, new = newapp, indi = T, se = F) :
>    The individual option is  only valid for start-stop data
> Please, advise if obtaining a single strata for a basic (time,response) model is possible? Due to differing lengths of the surv for different strata this will not go in a "wide" data.frame without padding.
> Thanks everybody and have a great day.
> Stephen B

More information about the R-help mailing list