[Elen Xr]

Hi David,

So, I have 6 studies with two treatments (DCB and BA) and I want to check them in a year, yr=year and n=sum, and i want to find the OR to make a comparison. This the first command for metan and then the second for metaan (I cant find how to creat the grey box).

Code:
metan yrDCB nDCB yrBA nBA, or label(namevar=STUDY) random

---

metaan yrDCB nDCB yrBA nBA , bdl exp mhor label(STUDY) forest

The first, command gives me OR=0.99 and I-squared= 19.3% (as the admetan) and the second gives me OR=0.91 and I-squared = 72.3%, p=0.003...And, so I dont understand why there is so much dieffernt.





Thank you!

[David Fisher]

Hi Elen,

Thankyou for your code and output. I can see two possible reasons for the differences in your results: one more obvious and one less so.

Firstly: you have specified the Bootstrapped DerSimonian-Laird (bdl) option to metaan, but the standard DerSimonian-Laird model to metan. The Bootstrapped DerSimonian-Laird method, as suggested by Kontopantelis et al (PLoS ONE 2013; 8: e69930), is a simple nonparametric bootstrap of the heterogeneity variance, designed to increase the chance of estimating a nonzero heterogeneity when its true value is close to zero. I will not comment here as to whether it should be recommended; but certainly the two models are different, and hence may give different results.

However, I don't think that accounts for such a large difference. Upon reading the metaan help file, I noticed that the four-variable syntax requests the number of events in group 1 followed by the population size of group 1 (followed by the same for group 2). For metan, however, the four-variable syntax requests the number of events in group 1 followed by the number of non-events in group 1 (followed by the same for group 2). Hence, your code as shown above actually results in different data being analysed by the two commands.

If I have understood the metaan help file correctly, these commands should give identical results (for Standard DerSimonian-Laird, since the bootstrap is not available with metan):

. metan  yrDCB nonyrDCB yrBA nonyrBA, or label(namevar=STUDY) random
. metaan yrDCB nDCB yrBA nBA , dl exp mhor label(STUDY) forest

* where
. nonyrDCB = nDCB - yrDCB
. nonyrBA = nBA - yrBA

I'm assuming here that the calculations at the end make sense given your data.

Does that help at all?

David.

[Elen Xr]

Hi David,

I just now try it and finally it works! Now, I have to find what to do with the high heterogeneity. So, thank so much for your help and your time, it was really needed!


Thank you

Eleni

[David Fisher]

This thread is rather old, but I can't find a more recent one where I've announced any updates for ipdmetan specifically (i.e. not bundled together with metan), so here goes:


With thanks as ever to Kit Baum, an updated version of the ipdmetan package is now available for download from SSC (ssc install ipdmetan, or adoupdate ipdmetan if already installed). The package description from SSC is given below.

This version (4.03) includes various minor bug fixes, and changes so as to work more effectively with the latest version of metan.

The biggest additionis the ability to use prefix commands, and in particular mi estimate, post. I myself am not an expert in the field of MI, so would welcome any comments or feedback. To clarify: the imputation procedure is applied independently to each set of trial data, and the relevant estimate of effect (+ SE) is retrieved. The meta-analysis pooling and presentation then proceeds as for any other two-stage analysis.

Note that you must use the post option to mi estimate, or ipdmetan will be unable to retrieve the relevant effect estimate.

Many thanks,

David.

[Mario Lumbreras]

Hi David David Fisher, I hope you are doing well!
I'm conducting a subgroup analysis for a clinical trial, and I was wondering if it would be possible to add the p-value for the interaction term to the forest plot; below, you will find the code I'm using:

Code:

ipdover, over(gest_eq_less_eleven) or forestplot(favours("Odds of outcome decrease" # "Odds of outcome increase")) : logistic saetot trt

Thanks in advance for your help and guidance!

All the best,

Mario