4

I am currently conducting a meta-analysis in which I need to use a mixed treatment comparison method. As I understand it, this method works in the following way:

Say you have a group of studies that make the following set of treatment comparisons:

  • Intervention 1
  • Intervention 2
  • Intervention 3
  • Control

You are interested in all possible comparisons between these treatments. So, not only are you interested in intervention 1 versus control, intervention 2 versus control, and intervention 3 versus control, but also intervention 1 versus intervention 2, intervention 1 versus intervention 3, etc. The problem occurs in that not all of the studies in your meta analysis include each intervention type. So, while study 1 may have tested intervention 1, intervention 2, and a control group, study 2 tested intervention 2 and intervention 3 versus a control group. And so on. Mixed treatment comparisons (Caldwell, Ades, & Higgins, 2005; Lu & Ades, 2004; Mills et al., 2011) arose as a way of using the indirect information from your sample of studies to estimate the magnitude of the missing comparisons.

For my study, I am interested in how several different moderators affect the magnitude of the various treatment comparisons. I stumbled across a paper (Nixon, Bansback, & Brennan, 2007) that combines the mixed treatment comparison method with meta-regression. My problem is finding a good software implementation for this method (preferably an implementation in R, since I'm most familiar with R). As far as I can tell, the metafor package isn't able to handle mixed treatment comparisons. Does anybody know whether there's a package out there that's able to handle both mixed treatment comparisons and meta-regression?

Giuseppe Biondi-Zoccai
  • 2,244
  • 3
  • 19
  • 48
Patrick S. Forscher
  • 3,122
  • 23
  • 43
  • Please let me know, if you're you still interested in this topic - I wanted to share some information, but just paid attention to the dates... – Aleksandr Blekh Jan 10 '15 at 07:56
  • I am! I was editing the text of my old questions, and I discovered that doing these edits places the questions on the active list. However, this is definitely still a topic I'm interested in if you have some extra information. – Patrick S. Forscher Jan 10 '15 at 16:24
  • All right then - will post my answer soon. – Aleksandr Blekh Jan 11 '15 at 00:05

1 Answers1

4

I cannot help you with an R implementation of the Nixon et al. paper. However, I remember a talk given by Ian White and colleagues on "Consistency and Inconsistency in Multiple Treatments Meta-Analysis: Models and Estimation". In this talk they presented a "multivariate random-effects meta-regression" which hopefully can be estimated using the mvmeta package. There also seems a paper available (haven't checked it yet): "Consistency and inconsistency in network meta-analysis: model estimation using multivariate meta-regression".

Bernd Weiss
  • 7,044
  • 28
  • 40
  • 1
    Bernd, thanks for your response. It looks like the mvmeta package is very new (published on CRAN on August 24!). The paper link references a Stata version of mvmeta, which seems to incorporate methods of dealing with mixed treatment comparisons. I'll have to see whether that functionality was also ported over to the R version of the package. – Patrick S. Forscher Aug 28 '12 at 18:12
  • It does indeed look like mvmeta has the appropriate functions for network meta-analysis (i.e., mixed treatment comparisons). Additionally, I have discovered several resources that might be useful for analysts wishing to use this method: the [NICE Decision Support Unit website](http://www.nicedsu.org.uk/Evidence-Synthesis-TSD-series%282391675%29.htm) and the [Multiple Treatments Meta-analysis](http://www.mtm.uoi.gr/#) website. – Patrick S. Forscher Aug 30 '12 at 15:20