Network Meta-Analysis

Example

In an oversimplified example in Figure 1, we have two treatments (Treatment 1 and Treatment 2) and usual care comparison groups. One study directly compared Treatment 1 with usual care in relation to an outcome, which allows us to estimate the difference between them. In this example, the mean difference between the treatment group (M = 30) and the usual care control group (M = 20) was 10 units. Another study directly compared Treatment 2 (M = 26) with usual care (M = 22) in relation to the same outcome and reported a four-unit mean difference. Although no study has directly compared Treatment 1 with Treatment 2, NMA allows us to create an evidence network structure as per Figure 1 to conduct an indirect comparison between Treatments 1 and 2 to estimate the difference between their effects in relation to the outcome dependent variable: Treatment 1 is estimated to have a six-unit mean difference to Treatment 2.

The distinction between direct and indirect effects may not be that meaningful once the network grows in complexity (e.g., comparing over 20 treatments): Evidence that is indirect for one comparison may be direct for another, and evidence on a single comparison may impact estimates of the treatment effects throughout the whole network. The most important issue is to [Page 1043]include all the trials comparing the treatments of interest in the target population. The totality of the evidence in the network determines the estimates of the treatment effects between all treatments.

Figure 1 Example of Direct and Indirect Comparisons Between Treatment 1, Treatment 2, and Usual Care Control Group

The NMA network is usually visually represented by a network plot, which can take on different shapes. Similar to other types of network analysis, there are conventions regarding the visual representation of the evidence network. A network comprises nodes (circles) connected by edges (straight lines). In NMA, each node represents each treatment condition, and the size of each node reflects the overall number of individuals in the included studies who received that specific treatment; the larger the number of individuals, then the larger the node for that treatment in the network. The edges are the lines that connect the nodes; solid lines reflect that direct comparison data are available between the connected node, and the thickness of the line reflects the number of studies that have made the direct comparisons, with thicker lines equating to more studies. Broken or dotted lines indicate indirect comparisons between nodes.

...