Combining odds ratio from different categories of variables in meta-analysis

[Gebretsadik Shibre]

Dear experts,

I kindly ask you to help me on how I should go about a problem in a meta-analysis. I have got for example 10 published articles for my meta-analysis. In two of the papers, parity was classified as 0 (reference), 1-2, 3-4, and 5+. In another couple of articles, it was coded as 1(reference), 2-3, and 4+. Yet, in the other articles, parity was classified as 1(reference), 2-4, 5-9, 10+. For others, it is coded as <5 (ref) and 5+. Finally, as 0 (ref), 1, 2-3, 4 and 5+.
My aim is to get the combined odds ratio of parity using meta-analysis. How do I first harmonize the variables so that they have the same classification of parity with the same reference? Is there any package that does this before creating combined/average odds ratio in meta-analysis in R or Stata?
Thank you for your help.

[David Fisher]

Just to clarify: is "parity" taking on the role of an intervention here? Or are you looking to estimate some intervention effect within parity subgroups?

If you have the actual raw count data for each of your studies, then possibly you could construct some clever model on the raw data based on a GLM with a log(istic)-link.

Otherwise, your best bet is probably to estimate the effect of "low" vs "high" parity by combining categories together. From the information you have provided, "<5 vs 5+" and "<4 vs 4+" would seem to cover all the possibilities. However, either way you will find that a couple of studies are left out. You could specify one analysis as primary and the other as sensitivity and see if you obtain materially different results.

As far as I'm aware there is no package that will do this work for you. Having the raw count data makes things easier, because you can then directly construct the (log) odds ratios you require. If you have odds ratios relative to a reference, you would need to make use of consistency assumptions, and/or the sorts of considerations discussed here: https://doi.org/10.1002/jrsm.1259 (Rücker G, Cates CJ, Schwarzer G. Methods for including information from multi-arm trials in pairwise meta-analysis. Res Synth Methods 2017; 8(4): 389-568)

I hope that helps, if only a little!

[Gebretsadik Shibre]

Thank you sir. Parity in my study is just like any other exposure variables that would be related to my outcome variable. I just wanted to get the pooled odds ratio of parity (for example 5+) compared to a parity that served as reference.
To do this, all the studies in the meta-analysis need have the same classification of parity, with all studies using same reference category. I could not get the raw data since most of the studies are very old that some of the authors might have been even died.