gms | German Medical Science

Transitivity, also known as similarity, is the cornerstone assumption underlying network meta-analysis [1], [2]. Transitivity states that pre-specified clinical and methodological characteristics of the synthesised trials that act as effect modifiers are similarly distributed across the observed comparisons of the network. The validity of transitivity is necessary to ensure that the results obtained from network meta-analysis are credible [1], [2]. There are currently no universal recommendations for the methods to assess the transitivity assumption regarding the distribution of the effect modifiers.

We propose hierarchical clustering to evaluate the transitivity assumption in a network of interventions. Based on a set of effect modifiers, we will investigate if there are distinct homogeneous clusters of observed comparisons that may signal possible intransitivity in the network. We also propose the framework of pseudo-studies to address comparisons with a single study when performing clustering. Finally, we introduce the network of comparisons as a visualisation tool to aid in detecting hot spots of intransitivity based on dissimilarity measures. We apply our proposed approach to a collection of published systematic reviews with network meta-analysis for the primary outcome.

Our approach offers a simple framework for objectively evaluating the transitivity assumption. The proposed framework considers all effect modifiers simultaneously using well-established statistical methods, offering numerous advantages over multiple statistical testing.

The author declares that she has no competing interests.

The author declares that an ethics committee vote is not required.