‘Best practice’ for conventional indirect comparisons/network meta-analysis using aggregate RCT data

Network meta-analysis (NMA)

Information on best practice for conventional indirect comparisons and network meta-analysis (NMA) is summarised on this page, with links to useful resources.

For more information describing NMA see here. The GetReal review on NMA methods can be found here and the articles identified in this review can be found here.

Assessing the assumptions of NMA

NMA adopts the same set of assumptions as a usual (pairwise) meta-analysis, but also uses an additional assumption that may be hard to assess, called transitivity (also called similarity or exchangeability) (Ades 2011, Salanti 2012, Efthimiou et al 2016).

  • Transitivity assumes that information for the comparison between treatments B and C can be obtained via another treatment, A, using the comparisons A vs. B and A vs. C.
  • Researchers can assess this assumption by checking the distribution of effect modifiers across comparisons (Jansen et al 2011).
  • They can also use conceptual considerations, for example, checking whether the missing treatments in each trial are ‘missing at random’ or whether the choice of treatment comparisons in the trials is not associated either directly or indirectly with the relative effectiveness of the interventions and that the treatments in the network are ‘jointly randomisable’ (Salanti 2012, Lu et al 2006, Cipriani et al 2013).
  • When the transitivity assumption does not hold, it may lead to inconsistencies in the data. Inconsistency refers to the statistical disagreement between the observed direct and (possibly many) indirect sources of evidence. Checking the network for inconsistencies offers a method for assessing the validity of the transitivity assumption.

Implementing an NMA

There are a range of approaches that can be used to conduct an NMA, including both frequentist and Bayesian methods. More details, including information on models and software codes, can be found in:

Different approaches for assessing the consistency of a network include (but are not limited to):

  • the loop-specific approach
  • the node-splitting approach
  • the Lu & Ades inconsistency model
  • the design-by-treatment inconsistency model
  • the Q-statistic for inconsistency in NMA.

The GetReal review on NMA methodology gives a general review of methods for implementing an NMA model and checking for inconsistencies in the network. It also covers more advanced issues in NMA, such as:

  • the use of different effect measures
  • extensions of NMA to account for effect modifiers
  • investigating potential bias in NMA
  • modelling multiple outcomes
  • definition of nodes in a NMA
  • planning future studies.

Other resources for conducting and appraising an NMA

Key contributor

Orestis Efthimiou, University of Bern and University of Ioannina School of Medicine