Meta-Analysis Trim-and-Fill

[Olivia Listrom]

I am trying to run a trim-and-fill for the meta-analysis I'm working on using different random effect models. The data that STATA spits out is the same regardless of if it runs a fixed effect for the iteration process and random(reml) (or other random method) for the pooling or both reml for the iteration and pooling across all different random models. Is this possible or expected? I also want to get a matrix of the estimated imputed effect sizes and standard errors. In the STATA manual it says the meta trimfill stores the results in r() including r(imputed): matrix of effect sizes and their standard errors for imputed studies. How do I access this?

[Leonardo Guizzetti]

The PDF documentation for the trim-and-fill meta-analysis command built-in with Stata 17 discusses the statistical Methods. You can get there from -help meta trimfill- and click the view PDF documentation link at the top of the page that appears. In short, yes, it is possible that there are no studies estimated to be missing, and that's not unusual. In this case, the results -meta trimfill- reports are exactly those from the original meta-analysis.

[Olivia Listrom]

The issue is that there are about 40 studies estimated to be missing.

[Leonardo Guizzetti]

The issue is that there are about 40 studies estimated to be missing.

This was not clear from your first post. And if you have 40 studies missing from this method, that suggests a great number of included studies and also some very concerning publication bias.

You can consider the example from -help meta trimfill- to work with some example data. Immediately after running your -meta trimfill- command, type -return list- to see all of what is returned in r() (and documented). Two matrices are of use here, the table showing the overall pooled estimates with and without these imputed studies, and the imputed matrix with the individual imputed study estimates.

webuse metatrim
meta set stdmdiff se  // deafult is random effects REML estimator
meta trimfill
return list
matlist r(table)
matlist r(imputed)

[Olivia Listrom]

Thank you that worked. Only issue is that it is telling me 40 studies were imputed, but it is only giving me the ES and SE of 33 of those studies. Any advice on this?

[Leonardo Guizzetti]

Sorry, no. Not unless you’re willing to post your data to use directly using dataex.