Network Meta-Analysis (NMA)
Compares MULTIPLE treatments at once — combining direct head-to-head trials with indirect evidence through a common comparator — to estimate and rank options, even pairs never trialled against each other.
An extension of meta-analysis that synthesises a whole network of trials. It blends direct evidence (A vs B head-to-head) with indirect evidence (A vs B inferred because both were trialled against a common comparator C) into one coherent set of estimates — and can rank every option, including pairs never compared directly.
Treatments never trialled head-to-head (A vs C) are bridged indirectly through their shared comparator.
Each node is a treatment; each line a body of evidence. Solid lines are direct head-to-head trials (thicker = more trials); the dashed line is a comparison with no head-to-head data, estimated indirectly because A and C were each trialled against the same comparator. The NMA stitches the whole network together and can output a ranking of all options.
First check the network is connected and ask whether the trials are similar enough to bridge (transitivity). Then check consistency — do the direct and indirect estimates for the same comparison agree? Only then trust the pooled effects. Treat any rank ordering as a fragile summary, not a verdict.
NMA answers the clinician’s real question — “which of these works best?” — when no single trial compared them all. But its credibility rests on assumptions a pairwise meta-analysis never has to make, so a tidy ranking can give false confidence.
Direct + indirectevidence combined across a network- Validity needs
transitivity&consistency - Rank probabilities /
SUCRA— unstable; read with caution
Antiepileptics for established status epilepticus — before the ESETT trial put levetiracetam, fosphenytoin and valproate head-to-head, network meta-analyses pooled the scattered direct comparisons with indirect evidence to estimate which second-line agent stopped seizures best and to rank them. The networks suggested broad equivalence — later borne out when ESETT found no clear winner among the three. The ranking is only as trustworthy as transitivity and consistency allow: trials differed in dosing, timing and seizure definitions, so a confident “best agent” from rank probabilities alone would have over-stated a near-tie.
- Intransitivity — bridging trials/populations too dissimilar to share a common comparator.
- Inconsistency — direct and indirect estimates for the same comparison disagree.
- Over-reading the rankings — treating an unstable SUCRA order as a definitive “best treatment”.
Quick check
What key assumption lets a network meta-analysis make indirect comparisons?
Answer: Transitivity — the trials are similar enough (in populations, co-interventions and design) that the common comparator can validly bridge treatments never compared head-to-head.
Ready to build your plan? EMF Premium gives you all 40,000+ questions, 20 mocks and 1,215 OSCE stations from £29/month — or a one-off 3- or 6-month pass.