W hen we perform meta-analyses of clinical trials or other types of intervention studies, we usually estimate the true effect size of one specific treatment. We include studies in which the same type of intervention was compared to similar control groups, for example a placebo. All else being equal, this allows to assess if a specific type of treatment is effective.
Yet, in many research areas, there is not only one “definitive” type of treatment-there are several ones. Migraine, for example, can be treated with various kinds of medications, and non-pharmaceutical therapy options also exist. Especially in “matured” research fields, it is often less relevant to show that some kind of treatment is beneficial. Instead, we want to find out which treatment is the most effective for some specific indication.
This leads to new problems. To assess the comparative effectiveness of several treatments in a conventional meta-analysis, sufficient head-to-head comparisons between two treatments need to be available. Alas, this is often not the case. In many research fields, it is common to find that only few-if any-trials have compared the effects of two treatments directly, in lieu of “weaker” control groups. This often means that traditional meta-analyses can not be used to establish solid evidence on the relative effectiveness of several treatments.
However, while direct comparisons between two or more treatments may not exist, indirect evidence is typically available. Different treatments may have been evaluated in separate trials, but all of these trials may have used the same control group. For example, it is possible that two medications were never compared directly, but that the effect of both medications compared to a pill placebo has been studied extensively.
Network meta-analysis can be used to incorporate such indirect comparisons, and thus allows us to compare the effects of several interventions simultaneously (Dias et al. 2013). Network meta-analysis is also known as mixed-treatment comparison meta-analysis (Valkenhoef et al. 2012). This is because it integrates multiple direct and indirect treatment comparisons into one model, which can be formalized as a “network” of comparisons.
Network meta-analysis is a “hot” research topic. In the last decade, it has been increasingly picked up by applied researchers in the bio-medical field, and other disciplines. However, this method also comes with additional challenges and pitfalls, particularly with respect to heterogeneity and so-called network inconsistency (Salanti et al. 2014).
Therefore, it is important to first discuss the core components and assumptions of network meta-analysis models. The underpinnings of network meta-analysis can be a little abstract at times. We will therefore go through the essential details in small steps, in order to get a better understanding of this method.
