Meta-analyses and systematic reviews are used in a wide range of application areas in order to summarise the evidence base for questions of scientific interest. In particular, they are extensively used in medicine. One of the most frequently used, although sometimes maligned, methods for performing meta-analyses is the method proposed by DerSimonian and Laird in 1986. For example, Google Scholar currently records over 16,000 citations for this paper.
Network meta-analysis is a much more recent development. In conventional meta-analyses that use the DerSimonian and Laird method, two treatment arms may be compared. However in network meta-analyses, more than two, and sometimes very many treatments are compared simultaneously in the same analysis. This raises further challenges. In particular we must assess if the direct evidence for treatment effects in the network is consistent with indirect evidence. When this is not the case, the network is said to be “inconsistent” or even “incoherent”.
A growing, multi-skilled working group looking into network meta-analysis was formed at the MRC Biostatistics Unit (BSU) approximately six years ago. The most recent publication from the original members of this “supergroup” (Dr Ian White, Dr Dan Jackson, Dr Jessica Barrett and Professor Julian Higgins) is especially exciting as it provides two new ways to extend DerSimonian and Laird’s methodology to the network meta-analysis setting. One of these new methods requires the consistency assumption and the other allows this to be relaxed. The paper (Extending DerSimonian and Laird’s methodology to perform network meta-analyses with random inconsistency effects) is now available in Statistics in Medicine’s early view. In addition to the four original members of the BSU working group, three further authors were involved in this paper. Two of these are also from BSU (Mr Martin Law, who prepared the web supplementary materials, and Dr Rebecca Turner) and the third is Professor Georgia Salanti from Bern. The opportunity to collaborate with Professor Salanti is another fantastic consequence of the formation of the working group at BSU.
This exciting and rapidly developing area of research led by BSU will progress further. In particular, Ian White is continuing to develop his suite of Stata programs for implementing network meta-analyses. Martin Law is working on further estimation methods and Dan Jackson’s focus is now firmly fixed on multivariate network meta-analyses, where multiple treatments and outcomes are included in the same meta-analysis. Finally, Rebecca Turner is exploring methods for incorporating informative priors for heterogeneity variances in network meta-analysis, with others at BSU.
In the future, the team anticipate further methodological development for network meta-analysis developing for many years to come.