R help - Mar 2008 - A stats question -- about survival analysis and censoring

Dear UseRs,

Suppose I have data regarding smoking habits of a prospective cohort and wish
to determine the risk ratio of colorectal cancer in the smokers compared to
the non-smokers.  What do I do at the end of the study with people who die
of heart disease? Can I just censor them exactly the same as people who become
uncontactable or who die in a plane crash?  If not, why not?

I'm thinking that heart disease isn't independent of smoking even though
a death from heart disease is probably uninformative about colorectal
cancer risk. Hence
I suspect simply censoring these deaths will introduce a bias, but I don't
know
how to correct for it.

Many thanks,

Geoff Russell -- an interested student

Hi Geoff,

I think the answer to such a problem (overall survival vs. disease free 
survival) depends on the regulatory
environment, for example, in a phase III clinical trial, OS would be 
used, whereas in an equivalence study,
DFS would be used.

Best,

Matthias


Geoff Russell schrieb:
> Dear UseRs,
>
> Suppose I have data regarding smoking habits of a prospective cohort and
wish
> to determine the risk ratio of colorectal cancer in the smokers compared to
> the non-smokers.  What do I do at the end of the study with people who die
> of heart disease? Can I just censor them exactly the same as people who
become
> uncontactable or who die in a plane crash?  If not, why not?
>
> I'm thinking that heart disease isn't independent of smoking even
though
> a death from heart disease is probably uninformative about colorectal
> cancer risk. Hence
> I suspect simply censoring these deaths will introduce a bias, but I
don't know
> how to correct for it.
>
> Many thanks,
>
> Geoff Russell -- an interested student
>
> ______________________________________________
> R-help at 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.
>
>

Geoff,

This is a tricky question. Have you done a literature search?

-Max

Geoff Russell formulated the question :
> Dear UseRs,
>
> Suppose I have data regarding smoking habits of a prospective cohort and
wish
> to determine the risk ratio of colorectal cancer in the smokers compared to
> the non-smokers.  What do I do at the end of the study with people who die
> of heart disease? Can I just censor them exactly the same as people who 
> become uncontactable or who die in a plane crash?  If not, why not?
>
> I'm thinking that heart disease isn't independent of smoking even
though
> a death from heart disease is probably uninformative about colorectal
> cancer risk. Hence
> I suspect simply censoring these deaths will introduce a bias, but I
don't
> know how to correct for it.
>
> Many thanks,
>
> Geoff Russell -- an interested student
>
> ______________________________________________
> R-help at 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.

"Geoff Russell" <geoffrey.russell at gmail.com> wrote in
news:93c3eada0803100546l2a604d53jead5b22bf39bf26c at mail.gmail.com: 
> Dear UseRs,
> 
> Suppose I have data regarding smoking habits of a prospective cohort
> and wish to determine the risk ratio of colorectal cancer in the
> smokers compared to the non-smokers.  What do I do at the end of the
> study with people who die of heart disease? Can I just censor them
> exactly the same as people who become uncontactable or who die in a
> plane crash?  If not, why not? 
> 
> I'm thinking that heart disease isn't independent of smoking even
> though a death from heart disease is probably uninformative about
> colorectal cancer risk. Hence,I suspect simply censoring these deaths 
> will introduce a bias, but I  don't know how to correct for it.
> 
This isn't really an R question, of course. Quick answer; don't censor
if
the subject survives a heart disease event. One is still at risk for 
colorectal cancer.

I am not sure there is a completely correct statitical method for 
correcting this sort of informative censoring induced by competing risks 
if your focus is colonic neoplasms.  There would also be other cancers 
(oropharyngeal, laryngeal, lung, kidney, bladder) that would also pose 
similar potentially informative censoring events, especially if the 
subject dies, eh. I think you can take solace in that fact that any 
positive association you find will be an underestimate regarding 
smoking's induction of colorectal neoplasms.

A full treatment might be a Markov model that would allow transitions to 
"heart disease", "colorectal cancer", "other smoking
related cancers" and
"death". You can still transition to  colorectal cancer after heart 
disease and I doubt that the risk of colorectal cancer is much changed 
after making the transition into "heart disease" except for the fact
that
such subjects will be bombarded by messages to stop smoking. I would 
predict that one's risk of staying in the surviving risk set would be 
very much loweer after transitioning to lung or oropharyngeal cancer, 
however.

-- 
David Winsemius