ENG I DEU
Drugs or medical interventions are usually compared head-to-head in randomized controlled trials to identify the better option for treatment.
Nowadays, multiple treatment options are available in many indications. To figure out which intervention is the best, we would need to compare each intervention with all others available in head-to-head studies. Unfortunately, all possible comparisons between all available interventions are usually unavailable. Thus, we cannot directly answer the question which treatment is the best. Nevertheless, if two interventions have been tested against a third one, they could be compared indirectly (Fig. 1).
Such indirect comparisons could result in a highly complex network between all possible interventions. The links of this network are formed by direct comparisons (i.e. studies that compared at least two interventions head-to-head), while some interventions remain unlinked (i.e. they have never been tested against each other).
Network meta-analyses are used to analyse such networks and (most importantly) estimate the differences between interventions that have been never compared. The methods for network meta-analyses have been developed in the late nineties (e.g. Bucher 1997). Their basic principle is simple: if we know that A is better than B and B is better than C, A must also be better than C. In addition, network meta-analyses also take into account variability in the results of individual studies and thus, summarize all available direct and indirect evidence.
Similar to standard meta-analyses, however, network meta-analyses have some limitations and challenges. Most importantly, the analyses are relying on assumptions that need to be matched to produce reliable results: first of all, all treatment options for an indication must be linked to the network, i.e., every intervention must be compared to at least one intervention within the network. This is usually true, because most medical interventions have been compared against placebo or a common standard treatment in their initial studies. Furthermore, all studies have to be as homogeneous as possible, i.e., they should have been conducted with similar patient groups under comparable conditions. This is especially important for network meta-analyses, because comparing apples and oranges usually does not make sense, particularly if strawberries, cherries and raspberries come into play (to take the metaphor further).
Finally, network meta-analyses require consistent direct and indirect evidence. That means that all direct and indirect evidence has to be based on the very same ("true") effect. Yet this assumption might not be met if the studies included were actually not comparable. This might be relevant, e.g., if intervention A showed a substantially stronger (or weaker) effect than B in a specific patient population because its effectiveness had been influenced by an additional factor in this particular study. In this case, the difference between A and B in this particular study is caused by an effect that is not observed in other studies. Therefore, the comparison of only homogenous studies is crucial for network meta-analyses.
Network meta-analyses allow indirect comparison of interventions, even if they have never been compared in a single study. However, the included studies need to be comparable to each other as in conventional pairwise meta-analyses.
Bucher HC, Guyatt GH, Griffith LE, Walter SD. The results of direct and indirect treatment comparisons in meta-analysis of randomized controlled trials. J Clin Epidemiol. 1997; 50(6): 683-691