Re: Survival analysis with clustered observations

Firstly, you can read about state-of-the-art in a publicly available paper:

http://mrvar.fdv.uni-lj.si/pub/mz/mz1.1/xu.pdf
(R. Xu: Proportional Hazards Mixed Models: A Review with Applications to
Twin Models, Advances in Methodology and Statistics, Vol. 1, No. 1, 2004,
205-212)

It's a continuation of Prof. Xu's paper in Statistics in Medicine (Vaida,
F., Xu, R., Proportional hazards model with random effects. Statistics in
Medicine, 2000, 19, 3309-3324.)

BTW -- shameless advertisement -- she'll be among the many distinguished
contributors at our traditional Applied Statistics conference (e.g., this
year A.Agresti and I.Olkin are among the invited speakers) -- see
http://ablejec.nib.si/AS2007/

As you can read in the 1st paper, sticking with the PH model and keeping
things as simple as possible, for a single covariate (which seems to be your
case) it is OK to simply use the twin pair as stratum variable and fit a
stratified Cox model (you can even do it in SPSS via point-and-click). Such
model is, as I understand, equivalent to a simplest frailty model. This and
much more is discussed in O'Quigley, J., Stare, J., Proportional hazards
models with frailties and random effects. Statistics in Medicine, 2002, 21,
3219-3233.

I'll stop here, since Prof. Stare is my boss and I know Prof. O'Quigley and
they're world class authorities on such matters while I'm an amateur.

But again, just as encouragement, usually the simplest approach with a
stratified model is OK. Even if there's more than a single covariate, it's
much better than ignoring the clustering altogether (which can be seen even
in articles published in reputable medical journals). As a rough estimate,
every third of my (way too many) biostatistical consulting tasks involves
survival analysis, and many of those involve Cox regression, and about 1 in
20 of the later involve matched-pairs data, and in all such rare cases
stratification was what I did and it was "good enough for government work"
(i.e., the thesis was approved or the article got published in an SCI
journal).

Regards,

Gaj Vidmar, PhD
Univ. of Ljubljana, Fac. of Medicine, Inst. of Biomedical Informatics


"mark@hsph" <mschult@xxxxxx> wrote in message
news:1182694773.775570.320390@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
Hi All:
I have a dataset with paired observations (twins) with time to
event observations. Much of the data is right censored also. I want to
find out if the event times are a function of an observed covariate. I
know that some theoretical work has been done on this problem, but are
there widely available stat packages available (such as SAS macros)
that will handle it?
Thanks,
Mark



.

Relevant Pages

  • Re: Put your babies on statins
    ... I have never seen used in medicine. ... having a lower rate of infection, ... are a few places where sound statistics is used in medicine, ... I do not claim that these views are those of the Statistics Department or of Purdue University. ...
    (sci.med.cardiology)
  • Re: Put your babies on statins
    ... I have never seen used in medicine. ... having a lower rate of infection, ... are a few places where sound statistics is used in medicine, ... I do not claim that these views are those of the Statistics Department or of Purdue University. ...
    (sci.med.nutrition)
  • Death by Medicine
    ... Death by Medicine By Gary Null, PhD; Carolyn Dean MD, ND; Martin ... Life Extension could cite only isolated statistics to make ... injuries and deaths caused by government-protected medicine. ... It is now evident that the American medical system is the ...
    (alt.politics.bush)
  • Re: Helmets
    ... is (what with his interests in publishing, science, medicine,, ... statistics, law etc.), we'd have been treated to a Last Word ...
    (uk.rec.cycling)