R help - Jan 2012 - Scoring using cox model: probability of survival before time t

Dear Members,
I required to score probability of survival before specified time using
fitted cox model on scoring dataset.
On the training sample data I am able to get the probability of a survival
before time point(t), but on the scoring dataset, which will have only
predictor information I am facing some issues. It would be great help for me
if you tell me where am I going wrong!
Here is the sample script!

#########################################################
library(survival) 

n = 100
beta1 = 3; beta2 = -2; 
lambdaT = .01 
lambdaC = .6  
x1 = rnorm(n,0) 
x2 = rnorm(n,0) 

T = rweibull(n, shape=1, scale=lambdaT*exp(-beta1*x1-beta2*x2)) 
C = rweibull(n, shape=1, scale=lambdaC)   
time = pmin(T,C)  
event = time==T   
train_sample=data.frame(time,event,x1,x2) 

rm(time,event,x1,x2)

fit_coxph <- coxph(Surv(time, event)~ x1 + x2, data= train_sample,
method="breslow") 

#Save model to some directory
save(fit_coxph, file =
file.path("C:/Desktop","fit_coxph.RData"))

#I can get probabilities on train_sample as below:
library(peperr)
pred_train <- predictProb.coxph(fit_coxph, Surv(train_sample$time,
train_sample$event),  train_sample, 0.4)
head(pred_train)

#            [,1]
#[1,] 5.126281e-03
#[2,] 4.324882e-01
#[3,] 4.444506e-61
#[4,] 0.000000e+00
#[5,] 0.000000e+00
#[6,] 3.249947e-01


#In the same line, I need probabilities on scoring_data. Now, close the
earlier session and run the below script in the new #R session, it gives
error.

library(survival) 
library(peperr)


load(file = file.path("C:/Desktop","fit_coxph.RData"))
 		
n = 1000
set.seed(1) 
x1 = rnorm(n,0) 
x2 = rnorm(n,0) 
score_data <- data.frame(x1,x2)
pred_score <- predictProb.coxph(fit_coxph, Surv(time, event),  score_data,
0.04)

#Error in Surv(time, event) : Time variable is not numeric

#After  creating dummy place holder for Surv(time, event), it gives another
error:

time <- rep(2, n)
event <- rep(1, n)
pred_score <- predictProb.coxph(fit_coxph, Surv(time, event),  score_data,
0.04)

#Error in inherits(x, "data.frame") : object 'train_sample'
not found
########################################################

Appreciate your help, is there any other way to get these probabilities on
newdata. 
Thanks in advance!!!!
~Aher



--
View this message in context:
http://r.789695.n4.nabble.com/Scoring-using-cox-model-probability-of-survival-before-time-t-tp4302775p4302775.html
Sent from the R help mailing list archive at Nabble.com.

--begin included message ---

Dear Members,
I required to score probability of survival before specified time using
fitted cox model on scoring dataset.
On the training sample data I am able to get the probability of a
survival
before time point(t), but on the scoring dataset, which will have only
predictor information I am facing some issues. It would be great help
for me
if you tell me where am I going wrong!
Here is the sample script!

------------------

 Your example isn't complete: the error comes from a function
predictProb.coxph() which I have never heard of, and I wrote the
survival library.  I obviously can't comment on why it fails -- you
might want to contact the author of the function.
 
 Using only the survival library (one of the more recent versions), you
need to know the fact that Pr(survival to t) = exp(-expected events by
t), then you can use
	predict(fit_coxph, type="expected", newdata=....)
where newdata has a time variable that contains the desired time point
for prediction.

  Note, by default the coxph result does not store all of the data
needed to make predicted survivals (it needs to keep the entire X
matrix).  You can override this by adding "model=TRUE" to the original
call.  If you do not, then it needs to look up the orginal data set to
do the prediction.  

Terry Therneau