Trim and fill using metatrim

[bibha dhungel]

I looked through other forum posts (post1, post2, post3, post4) and this question has been asked many times in the forum, but it remained unanswered.

I am using the command 'metatrim' in stata.
I am doing a meta-analysis of effective reproduction number. The LFK index shows a major asymmetry and publication bias. So, I try to trim and fill in the effect size.
But the result states: "No trimming was performed: data unchanged". I am not sure why it is not working. Can anyone please help?

metatrim rt lci uci, reffect funnel

where, rt is the effect size and lci and uci are the lower and upper confidence intervals respectively.

I expect to get some trimming and filling. However, in the result, state states:

Note: no trimming performed; data unchanged

For some weird reason, I can reproduce my result using some fake data.

Here I am just manipulating the automiss data, to have an effect size- mpg with its lower and upper 95% confidence lintervals- lci and uci

webuse automiss.dta, clear
keep make mpg rep78
replace rep78 = 2 if rep78 == .
generate lci = mpg - rep78
generate uci = mpg + rep78
drop rep78
drop if mpg == .

Installing metafunnel and meta bias package
package st0061 from http://www.stata-journal.com/software/sj4-2

net sj 4-2 st0061

Installing metabias command
package sbe19_6 from http://www.stata-journal.com/software/sj9-2

net sj 9-2 sbe19_6

I am creating a funnel plot and publication bias test. In my real data, the funnel plot looks assymetrical but the metabias command shows no bias

metafunnel mpg lci uci, ci name(funnel_ci, replace) // funnel plot
metabias mpg se, egger //

I thus will use LFK index to test for bias.

generate se = (uci-lci)/ 3.92
generate logmpg = log(mpg)
generate loglci = log(lci)
generate loguci = log(uci)
generate logse = (loguci-loglci)/ 3.92

admetan logmpg loglci loguci, model(reml) effect("mpg") eform

Now for LFK index, install the file. 'LFK': module to compute LFK index and Doi plot for detection of publication bias in meta-analysis

net describe lfk, from(http://fmwww.bc.edu/RePEc/bocode/l)

Running the code

lfk _ES _seES

Installing metatrim command:

sbe39_2 from http://www.stata.com/stb/stb61

net describe sbe39_2, from(http://www.stata.com/stb/stb61)

metatrim mpg lci uci, ci funnel reffect

Putting in all the same codes as above together for easier copy-pasting in stata:

webuse automiss.dta, clear
keep make mpg rep78
replace rep78 = 2 if rep78 == .
generate lci = mpg - rep78
generate uci = mpg + rep78
drop rep78
drop if mpg == .

generate se = (uci-lci)/ 3.92
generate logmpg = log(mpg)
generate loglci = log(lci)
generate loguci = log(uci)
generate logse = (loguci-loglci)/ 3.92

metafunnel mpg lci uci, ci name(funnel_ci, replace) // funnel plot
metabias mpg se, egger

admetan logmpg loglci loguci, model(reml) effect("mpg") eform

lfk _ES _seES

metatrim mpg lci uci, ci funnel reffect

[Tiago Pereira]

What is your effect size? We cannot help you unless there are clearer examples. Funnel plot asymmetry does not necessarily mean publication bias or small-study bias. Besides, I would not recommend the trim-and-fill approach to deal with funnel plot asymmetry. What about contour-enhanced plots and Egger's test?

[bibha dhungel]

Thank you Tiago Pereira for your response. All my analysis was conducted in log space and then exponentiated using eform option.
Here is my effect size
1.260(1.007–1.577)

Funnel plot:

The egger's test

LFK test to test publication bias (https://pubmed.ncbi.nlm.nih.gov/29621038/)

[Tiago Pereira]

Small-study bias assessment will not be useful in that case.

1. There is little variability in the within-study variances.
2. There is extreme between-study variability.

Check extracted data and make sure all standard errors are correct.