Re: st: Comparing overlapping groups

[Fred Wolfe <fwolfe@arthritis-research.org>]

David,

Thanks again. I avoided more detail in my first email to save
bandwidth. But here are some additional details so it might be easier
to understand what our goals were. A colleague in Germany has access
to a high quality random sample from the German general population. He
(and I) are interested in examining 4 definitions of "widespread pain
(WSP)." Originally used as a criterion for a medical illness called
fibromyalgia (FM), WSP has become of interest to investigators for
other reasons. However, there is no Gold standard for WSP, and it has
not been studied in high quality population studies. The central
questions my colleague and I were asking was, "Is there a clinical
difference associated with different definitions, what is the cause of
the difference, and does it associate with clinical variables like
general health (GH) and psychological status. In our email I called
the definitions FM1-FM4 for ease of understanding, though perhaps I
should have used the term WSP.

At the simplest level we could measure GH (and other "outcomes") with
a series or 4 if clause analyses or in 4 different regression analyses
(with or without covariates). The issue we were interested in the
question I posed was, How can we describe the difference between the
different model results in terms of difference and confidence interval
for the various models, and perhaps with a p-value (that some reviewer
is sure to ask for)?

I hope this makes it clearer. Thanks again.

Fred



On Wed, Oct 3, 2012 at 5:54 AM, David Hoaglin <dchoaglin@gmail.com> wrote:
>
> Fred,
>
> You're welcome.
>
> If the people who have FM according to all four of the definitions
> (FM1, FM2, FM3, and FM4) are the largest subset, you may want to make
> that the "reference category" and use the coefficients for the
> indicators of the other subsets to measure the effect of departures
> from that main category.
>
> One might hope that the four definitions agree for most people.
> Otherwise, diagnosis of FM would seem chaotic.  The relative
> frequencies of the 15 subsets may be of interest in themselves.
>
> If the definitions are based on lists of symptoms or results of tests,
> it may be instructive to use the individual symptoms or tests in the
> analysis, instead of bundling them together as definitions.  This
> approach may require some care if, for example, two definitions use
> the same test but with different thresholds.
>
> I haven't seen a question like yours in the literature, but that's not
> saying much.  Perhaps someone else who reads this list can contribute
> a reference or a link.
>
> David Hoaglin
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/


Fred Wolfe
National Data Bank for Rheumatic Diseases
Wichita, Kansas
NDB Office +1 316 263 2125 Ext 0
Research Office +1 316 686 9195
fwolfe@arthritis-research.org
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/