This is an archived version of the Handbook. For the current version, please go to training.cochrane.org/handbook/current or search for this chapter here.

Methods are available for analysing, simultaneously, three or more different interventions in one meta-analysis. These are usually referred to as ‘multiple-treatments meta-analysis’ (‘MTM’), ‘network meta-analysis’, or ‘mixed treatment comparisons’ (‘MTC’) meta-analysis. Multiple-treatments meta-analyses can be used to analyse studies with multiple intervention groups, and to synthesize studies making different comparisons of interventions. Caldwell et al. provide a readable introduction (Caldwell 2005); a more comprehensive discussion is provided by Salanti et al. (Salanti 2008). Note that multiple-treatments meta-analyses retain the identity of each intervention, allowing multiple intervention comparisons to be made. This is in contrast to the methods for dealing with a single study with multiple intervention groups that are described in Section 16.5, which focus on reducing the multiple groups to a single pair-wise comparison.

The simplest example of a multiple-treatments meta-analysis is the indirect comparison described in Section 16.6.2. With three interventions (e.g. advice from dietician, advice from doctor, advice from nurse), any two can be compared indirectly through comparisons with the third. For example, doctors and nurses can be compared indirectly by contrasting trials of ‘dietician versus doctor’ with trials of ‘dietician versus nurse’. This analysis may be extended in various ways. For example, if there are also trials of the direct comparison ‘doctor versus nurse’, then these might be combined with the results of the indirect comparison. If there are more than three interventions, then there will be several direct and indirect comparisons, and it will be more convenient to analyse them simultaneously.

If each study compares exactly two interventions, then multiple-treatments meta-analysis can be performed using subgroup analyses, and the test for subgroup differences used as described in Chapter 9 (Section 9.6.3.1). However, it is preferable to use a random-effects model to allow for heterogeneity within each subgroup, and this can be achieved by using meta-regression instead (see Chapter 9, Section 9.6.4). When some studies include more than two intervention groups, the synthesis requires multivariate meta-analysis methods. Standard subgroup analysis and meta-regression methods can no longer be used, although the analysis can be performed in a Bayesian framework using WinBUGS: see Section 16.8.1. A particular advantage of using a Bayesian framework is that all interventions in the analysis can be ranked, using probabilistic, rather than crude, methods.

Multiple treatment meta-analyses are particularly suited to problems addressed by Overviews of reviews (Chapter 22). However, they rely on a strong assumption that studies of different comparisons are similar in all ways other than the interventions being compared. The indirect comparisons involved are not randomized comparisons, and may suffer the biases of observational studies, for example due to confounding (see Chapter 9, Section 9.6.6). In situations when both direct and indirect comparisons are available in a review, any use of multiple-treatments meta-analyses should be to supplement, rather than to replace, the direct comparisons. Expert statistical support, as well as subject expertise, is required for a multiple-treatments meta-analysis.