R help - Jun 2010 - coxph and completely missing strata/subsetting

Hi everyone,

I'm doing some coxph() analyses with a large and complex dataset. The data
was collected in different centers, so I am using strata(centers) to stratify
the analysis.

My main issue is, not all centers collected all the variables, so for a model
such as:

coxph(Surv(days, cancer) ~ varA + sex + strata(centers), data)

I might have 1 or more centers that have NA for varA (in practice, all the
individuals monitored at those centers come without varA).

coxph() obviously warns me that a number of individuals have been excluded --
would that be equivalent to doing the analysis on a subset of the data or not?

I ask because I have many centers and many variables, and if the automatic
exclusion of individuals missing the variable in analysis *is not* equivalent to
subsetting I might have some serious work to do.

Best,

Federico

--
Federico C. F. Calboli
Department of Epidemiology and Biostatistics
Imperial College, St. Mary's Campus
Norfolk Place, London W2 1PG

Tel +44 (0)20 75941602   Fax +44 (0)20 75943193

f.calboli [.a.t] imperial.ac.uk
f.calboli [.a.t] gmail.com

> I'm doing some coxph() analyses with a large and complex dataset. The
> data was collected in different centers, so I am using
> strata(centers) to stratify the analysis.
> 
> My main issue is, not all centers collected all the variables, so for
> a model such as:
> 
> coxph(Surv(days, cancer) ~ varA + sex + strata(centers), data)
> 
> I might have 1 or more centers that have NA for varA (in practice,
> all the individuals monitored at those centers come without varA).
> 
> coxph() obviously warns me that a number of individuals have been
> excluded -- would that be equivalent to doing the analysis on a
> subset of the data or not?
It seems so by definition to me.  I am confused as to what else would be 
happening.  You can always create a smaller, easier to understand 
dataset and try things out for yourself.  Perhaps I do not understand 
your question though.