[R] Cox proportional hazards model

Frank E Harrell Jr f.harrell at vanderbilt.edu
Thu Sep 23 12:01:52 CEST 2004


Min-Han Tan wrote:
> Good afternoon,
> 
> I am currently trying to do some work on survival analysis.
> 
> - I hope to seek your advice re: 2 questions (1 general and 1 specific)
> 
> (1) I'm trying to do a stratified Cox analysis and subsequently
> plot(survfit(object)). It seems to work for some strata, but not for
> others.
> 
> I have tumor grade, which is a range of 1 - 4. 
> 
> When I divide this range of 1:4 into 2 groups, it works fine for
> strata(grade>2) and strata(grade > 3). However, if I do a
> strata(grade>1), there is an error when I do a survfit( <coxph object>
> )
> 
> Call: survfit.coxph(object = s)
> 
> Error in print.survfit(structure(list(n = as.integer(46), time = c(22,  : 
>         length of dimnames [1] not equal to array extent
> 
> (2) As a general question, is it possible to distinguish between
> confounding and interaction in the Cox proportional hazards model?
> 
> Thanks!
> 
> Min-Han

The Design package provides another way to use the survival package.

library(Design)  # also issues library(Hmisc) so need to install both
d <- datadist(mydataframe); options(datadist='d')
f <- cph(Surv(...) ~ ... + strat(z), surv=TRUE)   # note strat instead 
of strata
survplot(f, z=vector of strata values to plot survival curves for)

survplot has many options

-- 
Frank E Harrell Jr   Professor and Chair           School of Medicine
                      Department of Biostatistics   Vanderbilt University




More information about the R-help mailing list