Re: st: Meta-analysis of SMRs

[Roger Harbord <roger.harbord@bristol.ac.uk>]

The usual fudge when meta-analysing 2x2 tables is to add 0.5 to all cells of a 
study when any cell contains zero. So analogously for SMRs you could replace 
observed values of zero by 0.5. This is probably reasonable unless many of 
your studies have no or very few observed events. What to do in that case is 
not immmediately obvious to me - it might be possible to fit a suitable 
random-effects poisson model with an offset using -xtpoisson- or -gllamm- but 
don't ask me for the syntax!

Roger.
----------------------------------------------------
Roger Harbord     mailto:roger.harbord@bristol.ac.uk
Department of Social Medicine, University of Bristol


--On 01 September 2004 12:41 +0100 "Fenty, J." <jf48@leicester.ac.uk> wrote:


> Dear statlist
>
> I am conducting meta-analyses of SMRs using the -meta- command which works
> fine until you have observed values of zero. I am using the formulae
>
> 			log(SMR) = log(observed/expected)
> 			se(log(SMR)) = 1/sqrt(observed)
>
> and both are undefined when the observed value=0. Does anyone have any
> suggestions for dealing with this situation other than omitting studies with
> zeroes or combining them with other studies? Are there other formulae to
> calculate se(log(SMR)) which is not based solely on the observed value?
>
> Many thanks,
>
> Justin Fenty
>
> *
> *   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/