Skip to content

[R-meta] Collate event/proportion data from Pretest-Posttest controlled studies and one-group Pretest-Posttest studies

3 messages · David Pedrosa, Wolfgang Viechtbauer

Original
David Pedrosa
# 
Dear list,

in extension of the help by Wolfgang Viechtbauer from March this year 
(https://stat.ethz.ch/pipermail/r-sig-meta-analysis/2022-March/003918.html), 
we wonder about an additional aspect in our data (so far we have 
restricted our analyses to the effects of different agents on PD 
symptoms). The question we now have is whether we can do a similar 
three-level hierarchical meta-analysis for events/proportions. To 
briefly summarize, we would like to model the occurrence of adverse 
events for all agents and use the study type as a moderator. However, we 
are aware that "RR" or "OR" are poor choices and wanted to ask if 
someone could point us at an alternative or if we should just stick with 
the controlled studies.

Thanks again for the help,

David
3 days later
Wolfgang Viechtbauer
# 
Dear David,

If I understand your query correctly, you are asking if one can combine studies that provide 2x2 table data (for which one can compute measures such as RR and OR) and studies that provide pre-test post-test counts for a single group. If so, for the latter, one can compute measures such as the marginal log risk ratio and the marginal log odds ratio:

https://wviechtb.github.io/metafor/reference/escalc.html#outcome-measures-for-change-or-matched-pairs

Note that you will either need the paired 2?2 tables for such studies or the marginal table but then you have to make a guestimate of the correlation in the paired 2x2 table data.

Best,
Wolfgang
1 day later
David Pedrosa
# 
Thanks Wolfgang.

I had not seen those parts of the documentary yet. This makes things 
much clearer for me.

Best,

David


Am 20.06.2022 um 15:33 schrieb Viechtbauer, Wolfgang (NP):