Meta-analysis: inverse variance heterogeneity model

[Shafiur Rahman]

Dear all,

I am trying to perform a meta-analysis using the inverse variance heterogeneity model as developed by Doi et al (2015). I usually use metan command to perform fixed-effect and random-effect meta-analysis, but I don't know how to perform meta-analysis using the inverse variance heterogeneity model.

Can anyone tell me which command I should use?
Thank you.

Best,
Shafiur

[David Fisher]

Dear Shafiur,

The inverse-variance heterogeneity model can be performed in Stata using the admetan command (which shares a lot of syntax and options with metan, but is more up-to-date). Type ssc describe admetan, or ssc install admetan, at the Stata command line.

You then specify the IVHet model as follows:

admetan varlist [if] [in], model(ivhet) options

Best wishes,

David.

[Shafiur Rahman]

Dear Dr. David,

Thank you so much for your help.
The admetan command is working and I got my expected results.

Best regards,
Shafiur

[Shafiur Rahman]

Dear Dr. David,

I was performing a meta-analysis of proportion using metaprop_one command. I can use fixed effect or random effect model for meta-analysis of proportions in metaprop_one command, but it doesn't support inverse variance heterogeneity (IVhet) model. Is it possible to perform meta-analysis of proportion using admetan command?

Thank you in advance.

Kind regards,
Shafiur

[David Fisher]

Dear Shafiur,

Not yet, I'm afraid. It's on the list for a future update (as are many things!).

As far as I can tell, at least some of the pooling options in metaprop_one are basically standard inverse-variance meta-analysis methods applied to transformations of the proportion data. Therefore, it should be possible to modify this approach to incorporate IVHet. If you know what the transformations are, you could apply them yourself, then apply admetan with the IVHet option, and then back-transform the result. (It might be wise to first do this without the IVHet option, and check that your results agree with metaprop_one .)

(I note that metaprop_one also has options that fit logistic or log-linear models to the data to obtain the pooled estimate. I'm not sure how IVHet might be incoporated here.)

I hope this helps.

Regards,
David.

[Sathish Thirunavukkarasu]

Dear David,

I would like to plot a forest plot for pooled proportions using metaprop. Instead of 'fraction', I would like to have 'percentage' in the plot. Could you please advise how to do this? Also, it is necessary to have 'weight' in the plot?

metaprop n N, random ftt cimethod(exact) label(namevar=study) xlab(0,0.10,0.20,0.30,0.40,0.50,0.60)xline(0, lcolor(black)) xtitle("Proportion",size(2)) olineopt(lcolor(red)lpattern(shortdash)) plotregion(icolor(ltbluishgray)) diamopt(lcolor(red)) pointopt(msymbol(x)msize(0))boxopt(msymbol(S) mcolor(black)) astext(50) texts(150) nowt

Thanks

Sathish