Re: st: meta-analysis and ROC

[Tom Trikalinos <ttrikalin@mac.com>]

Hi Andrei,

Summary ROC (SROC) is not exactly a meta-analysis of "ROC" curves. It's 
somehow different.
SROC provides a summary measure of diagnostic performance when you only 
have sensitivity & specificity values per study. Each study is 
represented by a pair of sensitivity-specificity values rather than the 
ROC curve. SROC was devised because often the cutoff points of the 
diagnostic test variables differ across studies, and in fact SROC is 
informative only when the cutoff points differ.

If you really want to meta-analyze ROC curves, you can in principle 
meta-analyze AUC values with an inverse-variance model (you need the 
AUC variance for this).

Otherwise you might be better off with a meta-analysis of sensitivity 
and specificity across studies, after selecting the same cutoffs for 
the patients of each center.

Having all values for all patients, I might opt for different analyses 
rather than a SROC curve.

For the record there's no module in Stata that does SROC analyses, to 
my knowledge at least.  You could find the SROC manually in Stata, but 
it seems you don't want this.

tom




On Dec 12, 2005, at 10:07 PM, Andrei Malinovschi wrote:


> Dear Statalisters,
>
> I am having a database where we are looking at
> different symptoms (categorical variables) and
> different diagnostic tests (continuous variables). The
> data comes from several centers and we are trying a
> meta-analysis approach because we can't pool the data
> (due to differences between methods used in different
> centers).
>
> I would like to perform a meta-analysis on the ROC
> curves and get the AUC for a specific test for all the
> centers for one symptom (and then compare the AUCs for
> different tests for that symptom). It appears that I
> would need to do a SROC, is it right?. I couldn't find
> a program to perform SROC in STATA and I am wondering
> if there is an easy way to do that (the hard way would
> be to use -somersd- package, but it looks complicated
> to me and I was lost somewhere among vectors and
> diagonal matrices)?
>
> I welcome any comments regarding other approaches
> (better ones or easier to perform) to answer my
> question!
>
> Thank you in advance!
>
> With best regards,
> Andrei
>
> __________________________________________________
> Do You Yahoo!?
> Tired of spam?  Yahoo! Mail has the best spam protection around
> http://mail.yahoo.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/
Tom Trikalinos, MD
ttrikalin@mac.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/