What is the difference between pairwise and network meta-analysis?

Pairwise meta-analysis pools studies comparing two interventions; network meta-analysis combines multiple comparisons across a connected set of trials, including indirect ones, to compare and rank three or more interventions at once.

What is transitivity?

Transitivity is the assumption that the trials connected through a common comparator are similar enough in effect-modifying characteristics that comparing treatments indirectly is fair; it is the key condition for valid indirect comparison.

Network Meta-Analysis

Network meta-analysis extends pairwise meta-analysis to compare three or more interventions simultaneously by combining direct evidence (from head-to-head trials) with indirect evidence (inferred through common comparators). It allows treatments that have never been tested against each other to be compared and ranked within a single coherent analysis, and it is widely used in health technology assessment to inform choices among competing options.

Definition

Network meta-analysis is a statistical method that synthesises evidence from a connected network of trials comparing multiple interventions, combining direct and indirect comparisons to estimate relative effects between every pair of treatments and to rank them.

Scope

This topic covers the logic and key assumptions of network meta-analysis: how direct and indirect evidence are combined, the central assumptions of transitivity and consistency, treatment ranking, and the cautions that accompany its use. It is a methodological reference, not clinical or prescriptive guidance.

Core questions

How can interventions never directly compared be compared through common comparators?
Is the assumption of transitivity across the trial network plausible?
Do direct and indirect estimates agree (consistency)?
How should treatment rankings be interpreted and their uncertainty conveyed?

Key concepts

Direct versus indirect evidence
Common comparator
Transitivity assumption
Consistency (agreement of direct and indirect estimates)
Mixed treatment comparison
Treatment ranking (e.g. SUCRA)
Network geometry

Mechanisms

When trials form a connected network through shared comparators, the relative effect of two treatments not directly compared can be inferred indirectly: if A versus B and B versus C are each estimated, A versus C follows. Network meta-analysis combines such indirect estimates with any direct evidence into a coherent model that yields all pairwise relative effects and can rank treatments. Its validity rests on transitivity — that trials are similar enough in effect-modifying characteristics for indirect comparison to be fair — and on consistency, the agreement between direct and indirect estimates where both exist. The methodology was formalised in the mixed-treatment-comparison framework of Lu and Ades and elaborated in subsequent methods literature (lu-ades-2004; caldwell-2005; salanti-2012).

Clinical relevance

Network meta-analysis underpins many health technology assessments and guideline comparisons where several treatment options compete and head-to-head trials are incomplete. Its rankings describe relative evidence across options under stated assumptions; they summarise comparative evidence and are not a directive to use any particular treatment for an individual.

Evidence & guidelines

The method's statistical foundations were set out by Lu and Ades and introduced to clinicians by Caldwell and colleagues; Salanti's overview synthesised its benefits and concerns, and the Cochrane Handbook describes its conduct and reporting within systematic reviews (lu-ades-2004; caldwell-2005; salanti-2012; higgins-handbook-2019).

History

Indirect comparison methods were developed in the 1990s and unified into the mixed-treatment-comparison framework by Lu and Ades in 2004. Caldwell and colleagues introduced the approach to a clinical audience in 2005, and as the method spread under several names, Salanti's 2012 overview consolidated terminology, assumptions, and concerns, establishing network meta-analysis as a mainstream evidence-synthesis tool (lu-ades-2004; caldwell-2005; salanti-2012).

Debates

How safe is the transitivity assumption?: Indirect comparison is only valid if trials are similar in effect-modifying characteristics; when transitivity is doubtful or direct and indirect estimates are inconsistent, network results can mislead, so assessing these assumptions is essential.

Key figures

Guobing Lu
Tony Ades
Deborah Caldwell
Julian Higgins
Georgia Salanti

Seminal works

lu-ades-2004
caldwell-2005
salanti-2012

Frequently asked questions

What is the difference between pairwise and network meta-analysis?: Pairwise meta-analysis pools studies comparing two interventions; network meta-analysis combines multiple comparisons across a connected set of trials, including indirect ones, to compare and rank three or more interventions at once.
What is transitivity?: Transitivity is the assumption that the trials connected through a common comparator are similar enough in effect-modifying characteristics that comparing treatments indirectly is fair; it is the key condition for valid indirect comparison.