Network Meta-Analysis Explained

A standard (pairwise) meta-analysis compares two treatments at a time. Network meta-analysis (NMA) compares three or more treatments simultaneously, even when some pairs have never been directly compared in a trial — which is exactly why it's increasingly requested by journals and guideline committees making treatment-choice recommendations.

Direct and indirect evidence

If Trial A compares Drug 1 vs Drug 2, and Trial B compares Drug 2 vs Drug 3, NMA can estimate an indirect comparison of Drug 1 vs Drug 3 through their shared comparator (Drug 2) — even though no trial directly compared them. Where direct evidence also exists, NMA combines it with indirect evidence for a more precise pooled estimate.

The transitivity assumption

This is the assumption that makes indirect comparisons valid: trials in the network must be similar enough in their patient populations, settings, and other effect-modifying factors that comparing across them is fair. If Drug 1's trials enrolled much sicker patients than Drug 3's trials, the indirect comparison can be misleading — this is the single most important thing to assess and report before running an NMA, not an afterthought.

Reading a network diagram

NMA results are typically presented alongside a network diagram — nodes represent treatments, and lines connect treatments that have been directly compared, with line thickness often representing the number of trials. A poorly connected network (few direct comparisons, long indirect chains) generally produces less trustworthy estimates than a well-connected one.

Ranking treatments

NMA often produces treatment rankings (e.g., SUCRA scores or rankograms) showing the probability each treatment is best, second-best, and so on. These rankings are frequently over-interpreted — a treatment can have the highest probability of being "best" while still having wide, overlapping confidence intervals with several competitors. Report rankings alongside the actual effect estimates and their uncertainty, not instead of them.

Reporting standards

The PRISMA-NMA extension covers the additional items an NMA needs beyond standard PRISMA 2020 reporting, including the network diagram, transitivity assessment, and (where estimated) inconsistency between direct and indirect evidence.