R help - Jun 2005 - predict.coxph fitted values for failure times

I would like to extract predicted failure times from a
coxph model in library(survival). However, none of the
prediction options ("lp", "risk", "expected",
"terms")
seem to bear any relationship to failure time.

Perhaps I am asking the wrong question, but can coxph
provide predicted failure times?

Thanks,
Dan Bebber

Department of Plant Sciences
University of Oxford




		
___________________________________________________________ 
How much free photo storage do you get? Store your holiday

On Wed, 22 Jun 2005, Dan Bebber wrote:
> I would like to extract predicted failure times from a
> coxph model in library(survival). However, none of the
> prediction options ("lp", "risk", "expected",
"terms")
> seem to bear any relationship to failure time.
>
> Perhaps I am asking the wrong question, but can coxph
> provide predicted failure times?
Strictly no, as a Cox proportional hazards model does not model the 
baseline hazard.  However, there is a coxph method for survfit() which
allows you to predict the survival distribution for the fitted or new 
data, presumably using Breslow's estimator of the baseline hazard.
There's an example on the help page and more in MASS (the book and its 
scripts).

Unfortunately the help page is misleading, documenting only the
"km" method as if it were the generic (and using \synopsis to avoid
this being flagged).

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

On Wed, 22 Jun 2005, Dan Bebber wrote:
> I would like to extract predicted failure times from a
> coxph model in library(survival). However, none of the
> prediction options ("lp", "risk", "expected",
"terms")
> seem to bear any relationship to failure time.
>
> Perhaps I am asking the wrong question, but can coxph
> provide predicted failure times?
>
It's tricky, because it depends what you mean by "predicted".
It's
typically impossible to estimate the mean survival time for given 
covariates when there is censoring.  You can use survfit() on your Cox 
model to get predicted survival curves.

 	-thomas