Forestplot from metaprop to meta forest

[Gianfranco Di Gennaro]

Dear STATA users,
I conducted a meta analysis of proportions using metaprop.

I used a random effect model and computed CIs by Wilson method.
I subsequently obtained my forest plot.
Since I have to graphically "homogenize" this forest plot with forestplots with plots obtained from meta set and metaforest

I wonder:
is it possible to use _ES _seES _LCI _UCI obtain from metaprop to build a "handmade" forest plot via meta set, or meta forest?

Thank you VERY much in advance.
Gianfranco

PS: my script for my metaprop

metaprop numerator denominator, by (Device) random cimethod(wilson) ftt label(namevar=Author, yearvar=Year) xlab(.25,0.5,.75,1) subti("Outcome" , size(2)) xtitle("Proportion",size(2)) nowt olineopt(lcolor(red)lpattern(shortdash)) plotregion(icolor(ltbluishgray)) diamopt(lcolor(red)) pointopt(msymbol(x)msize(0))boxopt(msymbol(S) mcolor(black)) astext(70) texts(100) lcols(Author Year Country)

[Tiago Pereira]

Giafranco, hi.

This is my experience with that problem. Not sure if others have a better solution.

The answer is "no" if you want something easy. The answer is "yes" if you are willing to work on the codes - and can use the logit transformation.

In any case, -meta- will not plot CIs based on Wilson's method because they are typically asymmetrical. Using a higher tolerance, you can force it, but -meta- will assume that the 95% CIs are symmetrical, and the final results will be a disaster.

So, a possible pseudo-solution is to use the logit transformation. However, -meta- does not have an inverse logit transformation, and you will need to write a bit of additional code via gr_edit to change all logit values to proportions and plan the x-axis strategically.

The most straightforward approach is to plot the estimates directly as a proportion (lower and upper bounds) via -metan-, and use the second() option, which will contain the correct result from -metaprop-. The first summary estimate should be manually removed from the graph.

I hope this helps!

Tiago

[David Fisher]

I basically agree with Tiago. The problem is that meta forest is a command designed to work alongside meta set / meta esize; that is, to work only with analyses built up in the ways described in the help for meta. Since this does not currently include support for single-arm proportions, your ability to work with meta forest is severely limited and will inevitably involve "fudges" such as Tiago describes.

It may be of help to you, however, to point out that:
-- the user-contributed command metan now natively handles single-arm proportions; there is no need to use metaprop;
-- furthermore, the metan package now includes the command forestplot, which is a fully-flexible command for producing forest-plot (and "forest-plot-like") graphs from any data whatsoever, with no dependence upon any particular analysis command. I cannot claim that they are as pretty to look at as Stata's own meta forestplot graphs. However, with a bit of work you can achieve a good approximation.

Best wishes,

David.


(Disclaimer: I am author of forestplot and co-author of metan, and current maintainer of both)