Funnel plots following meta-analysis of proportions

[Bruce Weaver]

Hello folks.  I've been asked to help with a meta-analysis of studies reporting proportions.  When it was suggested that we need a funnel plot, I unthinkingly started making one.  But then this question popped into my head:

Q1. What do we hope to learn from a funnel plot showing prevalence estimates? 

In a meta-analysis of studies with some measure of treatment effect as the effect size estimate, funnel plots can show evidence of publication bias when one of the bottom corners is empty (because small studies not showing an effect are less likely to get published).  But for studies reporting prevalence estimates, there is no treatment effect.  So it's not clear to me which corner of the funnel might be relatively empty when there is publication bias.  And then another question popped into my head:

Q2. How is publication bias even defined for prevalence studies? 

Subsequently, I discovered that some other folks have argued that funnel plots are not useful in meta-analyses of proportions.  The authors of that letter recommend using something called a Doi plot.  They wrote (highlighting added):

"When applied to real-life meta-analyses, the Doi plot and its associated Luis Furuya-Kanamori (LFK) index were superior to the funnel plot and Egger's test for the detection of publication bias in terms of both sensitivity and specificity [3]."

But given that I am not sure how publication is defined for prevalence studies, I'm not at all convinced (yet) that Doi plots will be more helpful than funnel plots! 

Happy to hear the thoughts of others who started thinking about all this before I did and who have figured it all out!  ;-) 

Cheers,

Bruce

PS- My preferred software is Stata.