Trim and fill options

[Tom Yates]

I have completed a meta-analysis where the outcome is a proportion. I elected to use a double arcsine transformation to stabilise variance and Doi's IVhet model to account for variability arrising from between study heterogeneity.

To explore the potential for publication bias, we currently plot log odds of the proportion against sample size, as suggested by Hunter (https://pubmed.ncbi.nlm.nih.gov/24794697/). A reviewer has asked for Trim and Fill, as a sensitivity analysis. Is it possible to undertake such an analysis in STATA, whilst also applying the double arcsine transformation and using an IVhet model?

Our main analysis uses metan, e.g.

metan discordant_pairs total_household_pairs, pr model(ivhet) transform(ftukey, iv) study(study) by(tbincidence_category) sortby(tbincidence_id) lcols(proportion) forestplot(astext(40) textsize(100) boxscale(50) spacing(1.2) leftjustify range(0 1) dp(2)) extraline(yes) hetinfo(isq h)

Data and code here - https://github.com/tayates/strain_discordance.

Thanks,
Tom

[Nick Cox]

Not an answer but on Googling "double arc sine" (arc sine rang bells, but not "double arc sine") this came up

Röver C, Friede T. Double arcsine transform not appropriate for meta-analysis. Res Synth Methods. 2022 Sep;13(5):645-648. doi: 10.1002/jrsm.1591. Epub 2022 Jul 22. PMID: 35837800.

https://pubmed.ncbi.nlm.nih.gov/35837800/

[Tom Yates]

Thanks - I was aware of the importance of using an appropriate back transformation (see https://pubmed.ncbi.nlm.nih.gov/34632718/), which is properly implemented in metan, but hadn't seen this criticism. This work is near finalised, but I wonder whether I should run a GLMM analysis, as a sensitivity analysis, to address this concern about potential instability where there is marked variation in sample sizes (see https://pmc.ncbi.nlm.nih.gov/articles/PMC11632795/). I don't think think GLMM can be implemented using an IVhet model (1 vs 2 stage models), but that's okay. Also, may require me to lean on my collaborator, who is running a newer version of STATA? I am still using STATA 14.2.

With best wishes,
Tom