Forest plot for multilevel meta-analysis

[Shafiur Rahman]

Hi everyone,

I am trying to produce a forest plot for multilevel meta-analysis using Stata (where a single study can contribute multiple effect sizes). Could anyone kindly tell me if there is any relevant command or code to produce a forest plot in Stata 17 or 18?

Thank you.

[Leonardo Guizzetti]

It sounds like you are fitting a custom model outside of the -meta- suite of commands. In this case, you'll need to construct the forest plot yourself but you may dig into -ipdmetan- which has a -forestplot- command that I think can be used if you feed it the correct data. (I don't recall how to do this off-hand, but it shouldn't be too tricky.)

[Tiago Pereira]

Shafiur, hi.

You can check robumeta and fit an intercept only model. It is also possible to fit other models via mixed-effects models (e.g., melogit). However, these models do not calculate study weights. Hence, you have to plot the estimates using an unweighted forest plot, and replace the default summary estimate with the correct estimate obtained in the multilevel models. I typically accomplish that with meta (check the options forest plot with the customoverall suboptions). They are well described in the manuals.

[David Fisher]

Just to add to what Leonardo and Tiago have said: yes, there are various ways to produce a forest plot separately from running a meta-analysis model (or, indeed any other sort of model).

Leonardo mentioned forestplot, which is my own such command. The latest version may be found within either the ipdmetan or metan packages, both of which I maintain. Currently they are available via SSC, but I am intending to have them accessible via GitHub very soon. The advantage of forestplot is that it constructs the plot entirely from the dataset in memory, which you are free to modify as you wish.

Tiago mentioned that the built-in (Stata 16+) meta suite now enables you to replace the default summary estimate with a user-specified estimate, and thereby create a custom plot.

There are also various commands for plotting effect sizes, which may also be used to generate plots similar to a forest plot, of which the most comprehensive is probably coefplot by Ben Jann.

[Shafiur Rahman]

Thank you for your suggestion, Tiago and David. I will try to follow them. Stata 18 has included commands for multilevel meta-analysis, but did not provide any command for forest plot (for a multilevel meta-analysis).