Network-based Meta-analysis — Comparing Multiple Treatments Simultaneously
Network-based Meta-analysis (Network Meta-Analysis) · Also known as: NMA, network meta-analysis, mixed-treatment comparison, multiple-treatments meta-analysis
Network-based Meta-analysis (NMA) extends conventional pairwise meta-analysis by simultaneously synthesizing evidence across a network of two or more competing treatments, including pairs that have never been compared head-to-head in a single trial. By combining direct and indirect evidence within a coherent statistical model, NMA produces relative effect estimates for all treatment pairs and generates a probabilistic ranking of which treatment performs best on the outcome of interest.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use network-based meta-analysis when the clinical or policy question involves three or more competing treatments and a sufficient body of randomised trial evidence exists, including indirect comparisons. It is the method of choice when no single trial has compared all relevant options head-to-head and when decision-makers need a coherent ranking to guide treatment guidelines or reimbursement decisions. Do not use NMA when the evidence network is very sparse (many disconnected nodes), when the transitivity assumption is implausible (e.g., treatments were studied in populations so different that indirect comparisons are misleading), or when primary studies are predominantly observational — pairwise meta-analysis or qualitative synthesis is then more appropriate.
Strengths & limitations
- Enables simultaneous comparison of all treatments in a therapeutic area from a single coherent analysis.
- Incorporates both direct and indirect evidence, increasing statistical precision over pairwise meta-analysis alone.
- Produces a probabilistic treatment ranking (SUCRA / P-score) that is directly useful for clinical guidelines and health technology assessment.
- Identifies gaps in the evidence network and highlights which head-to-head trials are most needed.
- Bayesian implementation allows prior information about heterogeneity to be formally incorporated.
- Requires the transitivity assumption: patient populations, intervention definitions, and outcome measurements must be sufficiently similar across the network for indirect comparisons to be valid.
- Sparse networks (few trials per comparison) yield wide credible intervals that may not support meaningful ranking.
- Multi-arm trials require special handling to avoid unit-of-analysis errors.
- Publication bias in the underlying trials propagates into the network estimates and is harder to detect than in pairwise meta-analysis.
Frequently asked
What is the difference between network meta-analysis and pairwise meta-analysis?
Pairwise meta-analysis pools results only from trials that directly compared two specific treatments. Network meta-analysis extends this by simultaneously analysing all treatments in a therapeutic area — including pairs that have only been compared indirectly through a common comparator — producing a single coherent set of relative effect estimates and a treatment ranking.
What is the transitivity assumption and how do I check it?
Transitivity (also called the similarity assumption) requires that trials across the network are similar enough in their patient characteristics, intervention definitions, and outcome measurements that indirect comparisons are valid. You check it by tabulating key effect modifiers across trials for each comparison and assessing whether there are systematic differences. Statistical consistency tests (e.g., the design-by-treatment interaction model) detect statistical inconsistency but do not replace the clinical plausibility assessment.
Should I use a Bayesian or frequentist approach?
Both are valid. Bayesian NMA (implemented in R2jags, gemtc, or JAGS) allows formal incorporation of prior distributions for heterogeneity and produces full posterior distributions for treatment effects and rankings. Frequentist NMA (e.g., the netmeta R package) is computationally simpler and easier to reproduce. The choice should be driven by the complexity of the network, the need to incorporate prior information, and the conventions of the target journal or HTA body.
How do I handle multi-arm trials?
Multi-arm trials (those with three or more treatment arms) must be included as a whole unit, with the shared control arm modelled jointly across all contrasts from that trial. Splitting a multi-arm trial into independent pairwise comparisons inflates the effective sample size and artificially increases precision. The netmeta and gemtc packages handle this correctly by default.
What software is available for network meta-analysis?
The most widely used tools are the netmeta R package (frequentist, by Rücker and Schwarzer), gemtc R package (Bayesian, wrapping JAGS), and the mvmeta Stata command. For GRADE-based certainty assessment, the CINeMA web application is the standard tool. Cochrane RevMan also supports NMA through the MetaInsight plugin.
Sources
- Lumley, T. (2002). Network meta-analysis for indirect treatment comparisons. Statistics in Medicine, 21(16), 2313–2324. DOI: 10.1002/sim.1201 ↗
- Salanti, G. (2012). Indirect and mixed-treatment comparison, network, or multiple-treatments meta-analysis: many names, many benefits, many concerns for the next generation evidence synthesis tool. Research Synthesis Methods, 3(2), 80–97. DOI: 10.1002/jrsm.1037 ↗
How to cite this page
ScholarGate. (2026, June 3). Network-based Meta-analysis (Network Meta-Analysis). ScholarGate. https://scholargate.app/en/scientometrics/network-based-meta-analysis
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Bibliometric AnalysisScientometrics↔ compare
- Network Meta-AnalysisEvidence Synthesis↔ compare
- Scoping ReviewScientometrics↔ compare
- Systematic Literature ReviewScientometrics↔ compare
- Umbrella ReviewEvidence Synthesis↔ compare