Julie,
If you assume that the rates all come from similar populations, then you
can simply pool them, that is
(x1 + x2...+x12) / (y1 + y2 ... + y12)
and the confidence interval is the same as for any proportion.
However, if you thought the rates varied between populations (a random
'effect' model), then the method needs some adjustment to give more equal
weight to each of the different populations - but the calculations are then
very difficult,
Of course, if they do seem different then is it something to do with the
study methods (such as case defintions, etc)
I hope that helps,
Paul
At 09:52 AM 9/12/99 , Julie Brown wrote:
>Hi again,
>
>I have another seemingly basic question that has me stumped:
>I am looking at infection rates in febrile infants. I want to combine 12
>studies each with x/y infected infants. Is the overall rate of infection
>x1 + x2 + x3 . . . . x12/ y1 + y2 + y3 + . . . y12 ? Or can I account for
>variable sizes of these studies as I would in a meta-analysis of odds
>ratios? Also, can I determine a confidence interval for my summary rate?
>
>Thanks,
>
>Julie
>
>Julie Brown, MD
>Pediatric Emergency Medicine Fellow, box CH-04
>Children's Hospital and Regional Medical Center
>Seattle, WA 98105
>e-mail: [log in to unmask]
>fax: 206 527 3892
Paul Glasziou
Harvard Centre for Risk Analysis
718 Huntington Avenue, E-221
Boston, MA 02115
Fax: 1-617-432-0190 ph: 1-617-4320095
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|