Meta-analysis for complex data with many treatments (and often no controls)?

Question

I'm planning to conduct a meta-analysis of the catches of invertebrates from various types of sampling methods - with the question of: 'which method catches the most inverts?'

In a preliminary search, I've found c.20 relevant studies which use a mix of 12 different treatments/methods (but any single study tends to only include 3-8 methods). Most simply report mean and SEs for each method. Only a handful include a control (number caught by passive sampling) but not sure this should matter as relative comparisons made between treatments are what's important (i.e. is this method x better than that method y). It need to include study as a random effect as the number of invertebrates would vary wildly.

Is it possible to conduct a meta-analysis of these data? I've spent a couple of days reading around the subject but am struggling to relate published methods to these types of data. If it is possible, could someone kindly point me in the direction of appropriate methods or published meta-analyses doing something similar?

Many thanks!

Answer 1

For a start, you ideally need meaningful comparisons of methods within studies. Otherwise, I'm not sure there is much you can do. That's because studies probably differ so much that when any methods comparison is always also a study comparison, you have a very hard time disentangling the two.

Let's assume you have meaningful within study comparisons, but you don't have direct comparisons of all methods, then the topic to look into is network meta-analysis. In particular, in the parameterization, where you take the study - and possibly also the method (there's several publications by Piepho on this such as this one) - as random effects.