Descriptive Measures for a Network of Studies

Recently the notion of ‘geometry of a network’ has been introduced (Salanti et al., 2008b). Geometry refers to the overall pattern of comparisons between different treatments. Figure 2 presents two trial networks; on the left panel several topical non-steroidal anti-inflammatory for acute pain are competing and in the right panel four smoking cessation therapies.


The size of each node is proportional to the sample size and the weight of each link is proportional to the number of studies addressing each comparison. The features of these two networks are very different and their structure is influenced by several factors including bias and the research agenda in the field. There is no proposed metric to-date for describing, testing and measuring asymmetry in a network of trials, neither to describe the role of each of the nodes and links. Exploring the structure of the network is important, not only for descriptive purposes, as explained below.
Firstly, one can look at measures useful to characterize a network as symmetric or asymmetricand identify ‘missing’ or ‘overrepresented’ treatment comparisons beyond what is expected by chance. On this respect one has to take into account the extent to which different regimens and comparisons are represented in the network, i.e. the number of trials and their sample size per comparison and arm. I have preliminarily explored metrics and tests that have been employed in the ecological literature to characterize ecosystems regarding species co-occurrence and species diversity. These metrics (such as the C-score measuring co-occurrence and PIE index measuring diversity) need further operationalisation to demonstrate how exploration of the network geometry may assist the interpretation of the evidence in an MTM analysis. Under circumstances, significant asymmetry in a network can indicate publication bias.

The structure of the network is also important when evaluating assumptions that underlie MTM. When collating many trials, it is assumed that an intervention in one trial is comparable with the same intervention in another, and therefore they form a common node in the network. One can borrow methodology from social sciences; networks of treatments have similarities with social networks. The methodology examines how a particular intervention is embedded in the trials network by acting as an ‘intermediate’ when offering indirect evidence for other pairs of interventions. In an ‘ego’ analysis of the network there are metrics useful to identify nodes (i.e. interventions) that play a key role in providing indirect evidence for other comparisons.


Salanti, G., Kavvoura, F. K., & Ioannidis, J. P. (2008b). Exploring the geometry of treatment networks. Ann. Intern. Med 148, 544-553

Additional information