[R-meta] Zero seTE in network MA

Carla Gomez Creutzberg cgomezcre at gmail.com
Wed Sep 20 05:07:57 CEST 2017


Dear members of the netmeta community,

I am conducting a network meta-analysis and have been using a log-response
ratio as the effect size estimator. For each pairwise treatment comparison
in each study,we calculate the ratio as:
[image: Inline image 2]

where Ybar1 and Ybar2 are the means for each of the treatments reported in
the study. We estimate the ratio variance as:
[image: Inline image 1]

where s2 one and s2 two and n1 and n2  are the variances and sample sizes
for each of the treatment means (Ybar1 and Ybar2). For one study in my
analysis the above calculation gives a variance of zero since for a few
pairs of treatments where the same value is reported across the different
replicates and, consequently, the variance ends up being 0.

I have been looking a bit into the netmeta documentation and have found
that for the  0.9-6 version things are set up such that a error is
generated when a seTE of zero is fed into netmeta.
I understand that for many cases it is useful to have this error reported
however, for the study in my analysis I have checked and the authors
legitimately report the same values for the different replicates of some of
their treatments.

I was wondering what would be the best way to address this issue? Would it
be possible (and appropriate) to "switch off" this error and run netmeta
using the 0 variances? Or would this result in any calculation (or
conceptual) errors? And if so, would this then constitute a case where I
need to exclude the study for the analysis or perhaps even consider adding
an infinitesimally small values to the variances?

Thanks for your attention and any advice you can provide.

With best wishes,

Carla Gómez
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://stat.ethz.ch/pipermail/r-sig-meta-analysis/attachments/20170920/27a90d78/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: var.jpg
Type: image/jpeg
Size: 3981 bytes
Desc: not available
URL: <https://stat.ethz.ch/pipermail/r-sig-meta-analysis/attachments/20170920/27a90d78/attachment.jpg>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: LogResponse.jpg
Type: image/jpeg
Size: 4951 bytes
Desc: not available
URL: <https://stat.ethz.ch/pipermail/r-sig-meta-analysis/attachments/20170920/27a90d78/attachment-0001.jpg>


More information about the R-sig-meta-analysis mailing list