Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: METAGRAPH after META command - is it possible to truncate (manually define ticks)


From   "Tom Trikalinos" <ttrikalin@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: METAGRAPH after META command - is it possible to truncate (manually define ticks)
Date   Fri, 28 Nov 2008 12:30:45 -0500

[sacrificed line]

I presume you do univariate analyses of sensitivity and specificity, I
assume that what you do makes good sense in your case and that you
prefer this way over bivariate meta-analysis.

It is not a very good idea to do the synthesis in the (untransformed)
proportion scale, especially if your proportions are all over the
place:

1. transform the proportions
2. do the synthesis in the transformed proportions
3. back transform the CI's of the studies and the summaries in the
proportion scale and use these values for your forest plot.

transformations commonly used:

A.
logit(sens) with variance =(1/TP + 1/FN)
logit(spec) with variance =(1/TN + 1/FP)

B.
the angular transformation is another option that stabilizes the
variances nicely. But no need to go there.

HTH

t








Thomas A. Trikalinos, MD

Co-Director, Tufts Medical Center Evidence-based Practice Center

Associate Director, Center for Clinical Evidence Synthesis


Mailing address:
Institute for Clinical Research and Health Policy Studies
Tufts Medical Center
800 Washington St
#63
Boston, MA 02111

Physical address:
Office 710
35 Kneeland St
Boston
Phone: (+1) 617-636-0734
Fax: (+1) 617-636-8628
TTrikalinos@tuftsmedicalcenter.org




On Thu, Nov 27, 2008 at 4:04 PM, Wendy Wilkins <wlm054@mail.usask.ca> wrote:
> I have been using STATA for a meta-analysis of diagnostic test data
> (test sensitivity and specificity). I have used metagraph to generate
> forest plots after the meta command. The problem I am having is one of
> "esthetics" - I would like to include the forest plots in my
> manuscript, to be submitted for publication. However, given the wide
> standard errors for some observations, the resulting confidence
> intervals frequently extend well
> beyond the sensible limits of 0 to 100% defining sensitivity and
> specificity e.g. the point estimate for these values for any given
> study will lie between 0 to 1, but on the forest plot the CI's may
> extend from -10 to +10.
>
> My question is this - is there any way to truncate the forest plot to
> limit the graph to display 0 to 1 only?
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index