I'm curious. I've used the paired-prentice Wilcoxon test for the analysis
of parried survival data. I haven't run into use of the coxph for that
previously, but I have seen it referenced a couple of times in recent web
searches.
I have a data set of subjects like this:
Subject T1 R1 T2 R2
1 32 > 31
2 28 27
3 30 31
...
where Subject is the id of the subject, T1 is the test result from test 1,
R1 is the remark code (> indicates that the test result is the lower end,
right censoring only) and T2 and R2 are the corresponding values for test
2. I would like to know if there is a change from test 1 to test 2.
How can I set up the call to coxph for the paired test?
Thanks much.
Dave
From:
Terry Therneau <thern...@mayo.edu>
To:
"Marcus Michelangeli (Sci)" <marcus.michelang...@monash.edu>
Cc:
r-help@r-project.org
Date:
01/25/2011 09:42 AM
Subject:
Re: [R] Paired data survival analysis
Sent by:
r-help-boun...@r-project.org
--- begin included message ---
Im an honours student at Monash University. I'm trying to analyse some
data for my project, which involved 2 treatments. My subjects were
exposed to both treatments, and i gave them 60 minutes to perform a
certain behaviour. 3 of my subjects performed the behaviour in one
treatment but not the other. Therefore, i need to do a survival
analysis using paired data. Im little confused about how to go about
this in R. Im able to perfrom a normal surival analyses not taking the
paired data into account, but im just wondering if there is some way
to take the pairing into account. I know there are 3 different ways to
deal with grouping in the survival package, strata, cluster and
frailty but i struggle to understand the meaning of these arguments
and therefore do not know which one to use (if any).
--- end inclusion ---
All 3 methods can be defended. Adding cluster(id) to the model is
equivalent to a generalized estimating equations approach (if this were
a glm) or to the variance estimates commonly used in survey sampling (if
this were a linear model). Adding frailty(id) is equivalent to fitting
a linear mixed model. Using strata corresponds to a matched-pair
analysis, and will essentially reduce to a sign test: for each subject
treatment A was better, B was better, or tied. It's overkill in this
case (lower power).
If this were a linear model, you could find strong advocates for
either the GEE and mixed approach being "better". I somewhat prefer the
GEE method myself.
Terry T.
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
[[alternative HTML version deleted]]
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
