Message-ID: <Pine.LNX.4.64.0612211405410.17355@tajo.ucsd.edu>
Date: 2006-12-21T22:24:03Z
From: Charles C. Berry
Subject: : newbie estimating survival curve w/ survfit for coxph
In-Reply-To: <1c6126db0612211201i7fdd16e2r709fec1596de11ea@mail.gmail.com>
Spencer,
It always helps to look at the documentation for the objects you use.
?survfit.object
tells you:
-----------
COMPONENTS
strata
if there are multiple curves, this component gives the number
of elements of the time etc. vectors corresponding to the first
curve, the second curve, and so on. The names of the elements
are labels for the curves.
-----------
Note:
> z$strata
rx=1 rx=2
7 3
>
If it is still not obvious what gives, look at
plot(z)
then at z$time, z$strata, and z$surv.
On Thu, 21 Dec 2006, sj wrote:
> I am wondering how to estimate the survival curve for a particular case(s)
> given a coxph model
>
> using this example code:
>
> #fit a cox proportional hazards model and plot the
> #predicted survival curve
> fit <- coxph(
> Surv(futime,fustat)~resid.ds+strata(rx)+ecog.ps+age,data=ovarian[1:23,])
> z <- survfit(fit,newdata=ovarian[24:26,],individual=F)
> zs <- z$surv
> zt <- z$time
>
> I get the following output:
>
> 24 25 26
> [1,] 0.9740399 0.91737529 0.9873785
> [2,] 0.9431988 0.82552974 0.9721557
> [3,] 0.9023088 0.71387936 0.9515702
> [4,] 0.8518865 0.59121499 0.9255099
> [5,] 0.7269607 0.35151309 0.8572808
> [6,] 0.6130966 0.20108501 0.7895814
> [7,] 0.2807720 0.01553603 0.5415119
> [8,] 0.9159643 0.74991615 0.9584975
> [9,] 0.8225510 0.52704511 0.9099799
> [10,] 0.7058027 0.31906859 0.8451399
>
> [1] 59 115 156 268 329 431 638 464 475 563
>
> I am not sure what I am getting here, I was expecting to get three survival
> curves for each scenario in the "newdata" however the survival curves don't
> seem to be monotonically decreasing with time. I get the feeling that this
> has something to do with the use of strata(rx) in the model, but I am no
> sure how to sort it out.
>
> any help is appreciated,
>
> best regards,
>
> Spencer
>
[ irrelevant prior msg deleted ]
Charles C. Berry (858) 534-2098
Dept of Family/Preventive Medicine
E mailto:cberry at tajo.ucsd.edu UC San Diego
http://biostat.ucsd.edu/~cberry/ La Jolla, San Diego 92093-0717