Multi-lever meta-analysis of single proportions (prevalence)

[Kristian Berge]

Hello,

I am currently conducting a meta-analysis of the prevalence of having a troponin concentration (a blood marker of cardiac damage) above the upper reference level of normal after participation in sports events. To do this I am using the "meta" package included in stata 18. The variables and commands I use are as follows:

Variables:
studyid // the ID of the study
numberofparticipantsaboveurl // number of participants with measured troponin concentration above upper reference level of normal
numberofparticipants // total number of participants in the study

Stata syntax:
meta esize numberofparticipantsaboveurl numberofparticipants, studylabel(studyid) // Compute effect sizes and declare meta-analysis data
meta summarize, prevalence nostudies // Summarize meta-analysis data, report proportions (for meta-analysis of a single proportion)

So far, so good.

The problem:
However, I discovered that many of the studies use data from the same participants participating in the same sports events. The only difference between some of these studies is that they report results from different assays (laboratory test), such as troponin T and troponin I, and the studies are thus not independent of each other. The study cohort the data is collected from is stored in the variable:

studycohort // the identifyer for the study cohort. this is similar for many of the studies and what I want to specify in the multilevel meta-analysis

In the help file for "meta meregress" (Multilevel mixed-effects meta-regression) there are no examples of multilevel meta-analysis of proportions, but is it somehow possible to use this command (or other) to perform multilevel meta-analysis of proportions (prevalences)? Or are there any other way to achieve what I aim to do?

Any help is highly appreciated!!

Best regards,
Kristian

[Tiago Pereira]

You have try -robumeta- fitting a meta-regression model with an intercept only.

robumeta logit_proportion , study(id) variance(variance_logit_proportion) weighttype(random) rho(.90)