[R] Running cox models

[Thomas Lumley]

On Sun, 3 Sep 2006, Geoff Russell wrote:

> Hi,
>
> I'm reading van Belle et al "Biostatistics" and trying to run a cox test 
> using a dataset from:
>
> http://faculty.washington.edu/~heagerty/Books/Biostatistics/chapter16.html
>
> (Primary Biliary Cirrhosis data link at top of the page),
>
> I'm using the following code:
>
> --------------- start of code
> library(survival)
> liver <- scan("liver2.txt",list(age=0,albumin=0,alkphos=0,ascites=0,bili=0,
>        cholest=0,edema=0,edmadj=0,hepmeg=0,obstime=0,platelet=0,protime=0,
>        sex=0,sgot=0,spiders=0,stage=0,status=0,treatmnt=0,
>        triglyc=0,urinecu=0))
> fit<-coxph(Surv(obstime,status)~bili+edmadj+albumin+protime+age,data=liver)
> summary(fit)
> ----------------- End of code
>
> but the answer is rather different from that in the book (p.688 - for
> anyone with the book).

A little further up the page (or on a previous page -- I don't have a 
printed copy with me) the example specifies that in the model BILI, 
ALBUMIN and PROTIME are log transformations of the data and AGE is in 
ten-year units.

I must admit that the need to take the 312 records with value for 
treatment is not explicit in the book, though.

If you try
coxph(formula = Surv(obstime, status) ~ log(bili) + edmadj + log(albumin) 
+ log(protime) + I(age/10), data = liver, subset = !is.na(treatmnt))

you will get something much more like the book.


 	-thomas