RE: [R] coxph question

[Man, Chris T.]

Thanks Mayeul,

I actually would like to test each variable individually and use those have low 
p-value to build a classifier (not in cox model). Therefore, I need to write a 
function to subset those low p-value variables, instead of putting them as covariates. 
Any ideas?

Chris

-Original Message-
From: Mayeul KAUFFMANN [mailto:[EMAIL PROTECTED]
Sent: Thursday, August 12, 2004 8:17 PM
To: [EMAIL PROTECTED]
Subject: [R] coxph question


 I have many variables to test using cox model (coxph), and I am only
interested in those variables with p value less than 0.01. Is there a
quick way to do this automatically instead of looking at the output of
each variable?
 Chris

I guess you need covariate selection.
for a lengthy discussion of another method (AIC/BIC), look at last month
archive:
https://www.stat.math.ethz.ch/pipermail/r-help/2004-July/subject.html#53519

and try using
 library(MASS)
 stepAIC (...)

Most of the time, it starts removing the covariate with the lower p-value.


Mayeul KAUFFMANN
Univ. Pierre Mendes France
Grenoble - France

__
[EMAIL PROTECTED] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html

__
[EMAIL PROTECTED] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html

Re: [R] coxph question

[Mayeul KAUFFMANN]

Chris,
I understood the following: you want to try every single covariate in a
cox model with only one covariate, then take the best ones according to
p-value.

Assume your columns  look like:
stop status event x1 x2 x3 etc
You want to add column 3 (x1), then 4, etc.

I suggest a for() loop:

z-NULL;for(i in 3:ncol(data))
{coxtmp - coxph(Surv(stop,status)~ data[,i]) #you can modify the formula
for
   #adding covariates
in any case, for instance
beta-coxtmp$coefficients[1]
se-sqrt(diag(coxtmp$var))[1]
z- rbind(z,c(i,beta,se,pvalue=signif(1 - pchisq((beta/ se)^2, 1), 4)))
print (i)}

#then select the covariates according to the p values:
z[z$pvalue.01,i]

Hope it helps.

Mayeul KAUFFMANN
Univ. Pierre Mendes France
Grenoble - France


- Original Message - 
Thanks Mayeul,

I actually would like to test each variable individually and use those
have low p-value to build a classifier (not in cox model). Therefore, I
need to write a function to subset those low p-value variables, instead of
putting them as covariates. Any ideas?

Chris

-Original Message-

 I have many variables to test using cox model (coxph), and I am only
interested in those variables with p value less than 0.01. Is there a
quick way to do this automatically instead of looking at the output of
each variable?
 Chris

__
[EMAIL PROTECTED] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html