Re: st: clustering in proportional hazards models with stata/mp 10

["E. Paul Wileyto" <epw@mail.med.upenn.edu>]

Cox regression with shared frailty is really, really, really slow.  Like 
get your results tomorrow slow.  Works much better with parametric models.

P

Jeph Herrin wrote:

> I am surprised that Bill Gould did not mention:
>
>   stcox x1 x2, shared(center)
>
> Is there a reason to rule it out of contention?
>
>
>
>
> Roberto G. Gutierrez, StataCorp LP wrote:
> > In response to Bill Gould's <wgould@stata.com> post, Daniel O. Koralek
> > <dkoralek@unc.edu> follows up with:
> >
> > > Thank you for your quite thorough response to my query.  One quick 
> > > question.
> > > Stata will allow me to combine your solutions 1 and 3.  And it does (at
> > > least things move in the directions I would expect) what I would expect
> > > (narrower confidence intervals for the estimates derived using 
> > > solution 3.
> > > But something about it doesn't seem valid.  Any thoughts?
> > Solution 1 in Bill Gould's post refers to fitting a Cox model where you
> > cluster on groups (centers in this example).  Solution 3 refers to 
> > instead
> > stratifying on these centers.  As such, combining 1 and 3 would both 
> > cluster
> > and stratify on center, i.e.
> >
> >    . stcox x1 x2, strata(center) vce(cluster center)
> >
> > Daniel expresses concern that the above may not be valid.  It is 
> > perfectly
> > valid.  As Bill explained, stratifying on center allows for different 
> > baseline
> > hazard functions for each center.  However, the effects of the 
> > covariates x1
> > and x2 are assumed to be proportional on the hazard and, more 
> > importantly,
> > shared across centers.  In estimating the standard errors of these 
> > (overall) covariate effects, one could also choose to cluster on center.
> >
> > You stratify on center because you want the flexibility afforded by 
> > having
> > uniquely shaped hazards for each group.  However, once you stratify, 
> > you are
> > still faced with deciding whether the subjects within each center are
> > to be treated as independent given the shared baseline hazard.  If 
> > you do not
> > wish to make this additional assumption, then you should cluster on 
> > center as
> > well.
> >
> > --Bobby
> > rgutierrez@stata.com
> > *
> > *   For searches and help try:
> > *   http://www.stata.com/support/faqs/res/findit.html
> > *   http://www.stata.com/support/statalist/faq
> > *   http://www.ats.ucla.edu/stat/stata/
> *
> *   For searches and help try:
> *   http://www.stata.com/support/faqs/res/findit.html
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
--
E. Paul Wileyto, Ph.D.
Assistant Professor of Biostatistics
Tobacco Use Research Center
School of Medicine, U. of Pennsylvania
3535 Market Street, Suite 4100
Philadelphia, PA  19104-3309

215-746-7147
Fax: 215-746-7140
epw@mail.med.upenn.edu
http://mail.med.upenn.edu/~epw/ 
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/