R help - May 2009 - Function Surv and interpretation

Dear everyone,

My question involves the use of the survival object.

We can have 

Surv(time,time2,event, type=, origin = 0)					(1)

As detailed on p.65 of:

http://cran.r-project.org/web/packages/survival/survival.pdf


My data (used in my study) is 'right censored'  i.e. my variable
corresponding to 'event' indicates whether a person is alive (0) or dead
(1) at date last seen  and my 'time'  indicates time from transplant to
date of last contact (where this is time from transplant to death if person has
died or time from transplant to date last seen if person is still alive).

Now I am using function, rcorr.cens

http://lib.stat.cmu.edu/S/Harrell/help/Hmisc/html/rcorr.cens.html

This function involves use of Surv.

Now here is a section of my syntax:

> time<-data$ovsrecod
> x1<-data$RMY.GROUPS
> death<-data$death
> rcorr.cens(x1,Surv(time,death),outx=FALSE)					(2)
As you can see, I have entered Surv(time,death)...this works (and complies with
the example given in R for rcorr.cens) and all seems to be well...however,
bearing in mind that in (1) we have:

Surv(time,time2,event, type=, origin = 0)

...how does R know that 'death' in *my* syntax (2) is the
'event'...i.e. how does it know that time2 is skipped in my analysis?  I
am a bit perplexed!  The R documentation for Surv says that  Surv(time,event) is
a 'typical usage' as is Surv(time,time2,event, type=, origin = 0)...but
how does it know when we are using the former and not the latter?  I have tried
entering:  rcorr.cens(x1,Surv(time,event=death),outx=FALSE) but it does not like
it saying that "Error in Surv(time, event = death) : argument
"time2" is missing, with no default"....

I hope that this makes sense!

Thank you so much for your advice on this ...it's much appreciated, Kim

Dr Pearce,
 
  1. The heart of your question is why Surv(time, death) works, when the 
standard S action would be to match "death" to the "time2"
argument.
   The answer is that Surv is not conventional -- if only 2 unnamed arguments 
are present, then it assumes that they match to the time and event arguments.  
At least in my work, 99% of the uses of survival are either (time1, time2, 
status) or (time, status); so the function was made convenient for the most 
common case.
   
   2. Q2: why did Surv(time, event=death) give an error message?  This was a 
programming error which has been corrected in the most recent version of the 
code.
   
   3. I'm not familiar with the details of the cuminc function (I assume
from
the cmprsk package).  However, with the newest version of survival you can use 
the alternative:
    fit <- survfit(Surv(ftime, fstatus>0) ~ group, event=fstatus,
data=mydata)

The result is a standard survival curve object, so all of the usual plotting 
methods for survival curves work, e.g.,
    plot(fit[1:3], fun='event', col=c(2,7,5), lty=c(1,2,2))
    
In the survfit formulation 'status' is still 0/1 for
censored/uncensored, and
'event' gives the event type for the deaths.  The event code for the
censors is
ignored and can be any value (but don't use NA - that will cause the
observation
to be dropped).

   	Terry Therneau