Generating a graph from meta-regression output

[jemima spencer]

Hi,

Could someone kindly point me in the right direction re how to generate a graphical display of my meta-regression results?

In my systematic review, I want to see whether any of the heterogeneity in the (log)odds of coronary angiography could be due to the date of the data (avdate) within each study.

Here is my data:

input float(id data_avdate angio_logOR angio_loglci angio_loguci)
4 2007 -.3011051 -.3856625 -.21072103
8 2012 -1.3470737 -1.3862944 -1.3093333
9 2005 -.9162907 -.967584 -.8439701
12 2007 -.12783338 -.12783338 -.11653383
13 2013 -1.0498221 -1.0788096 -1.0216511
16 2010 -.8915981 -1.0216511 -.7550226
18 2007 -.4307829 -.8915981 .029558774
19 2011 -2.207275 -3.218876 -1.2039728
22 2017 -.7765288 -.8439701 -.6931472
25 2012 -.5276328 -.8209805 -.2484614
26 2009 -1.3470737 -1.3862944 -1.2729657
23 2008 -1.5141277 -1.5606477 -1.469676
23 2015 -1.4271164 -1.469676 -1.3862944


I have used the following code to do the meta-regression:

meta set angio_logOR angio_loglci angio_loguci, random(reml) civartolerance(0.1) studylabel (id)

meta regress data_avdate

I can't find anything relevant to generating a graph from this on the STATA "meta regression help" files. I did find something relevant when using the "metareg" command which I have just installed. Perhaps I should be using this instead?

Any/all suggestions would be lovely.

Thanks so much

Jemima

[Bruce Weaver]

Hello jemima spencer. I think you want -estat bubbleplot- after your -meta regress- command. You can find an example on this page:

• https://www.stata.com/features/overview/meta-analysis/

Cheers,
Bruce

PS- In the data you provided, the row with id=12 is causing problems because the point estimate is not greater than the lower limit of the CI. I had to delete that row to get things to run. you also had the same id code (23) on the last two rows. Here is my cleaned up code:

clear *
input float(id data_avdate angio_logOR angio_loglci angio_loguci)
4 2007 -.3011051 -.3856625 -.21072103
8 2012 -1.3470737 -1.3862944 -1.3093333
9 2005 -.9162907 -.967584 -.8439701
12 2007 -.12783338 -.12783338 -.11653383
13 2013 -1.0498221 -1.0788096 -1.0216511
16 2010 -.8915981 -1.0216511 -.7550226
18 2007 -.4307829 -.8915981 .029558774
19 2011 -2.207275 -3.218876 -1.2039728
22 2017 -.7765288 -.8439701 -.6931472
25 2012 -.5276328 -.8209805 -.2484614
26 2009 -1.3470737 -1.3862944 -1.2729657
27 2008 -1.5141277 -1.5606477 -1.469676
28 2015 -1.4271164 -1.469676 -1.3862944
end
* NOTE that you had id=23 on the last two rows.
* I changed them to 27 and 28.

twoway rcap angio_loguci angio_loglci id || ///
scatter angio_logOR id, name(fig1, replace)

* The record with id=12 is causing problems
* because the point estimate is not greater
* than the lower limit of the CI.
* I'll just get rid of that record for now.
* You may want to find a better solution.
drop if id==12

meta set angio_logOR angio_loglci angio_loguci, ///
random(reml) civartolerance(0.1) studylabel (id)

meta regress data_avdate
* Now show a bubbleplot
estat bubbleplot, name(fig2, replace)

[jemima spencer]

Hi Bruce,

Thanks so much, that'd really kind - as yes, does work

I've actually altered the ID12 LCI in my dataset due to this problem. Really tight CI in a huge study and I assume looks asymmetrical due to rounding of decimals.

I've more recently read about metareg as a command which does give a graph, but having problems getting this code to work (metareg angio_logOR data_avdate wwse(angio_logSE), eform graph randomsize noconstant reml) so I think I'll stick with your suggestion.

Thanks again

Jemima