A network meta-analysis compares three or more interventions at once by combining direct and indirect evidence, and it reports study selection with a standard PRISMA flow diagram supplemented by the PRISMA-NMA extension. The flow diagram itself is unchanged: records are identified, deduplicated, screened, and assessed exactly as in a pairwise review. What the extension adds is a separate requirement to describe the geometry of the treatment network and to report how many studies and participants inform each comparison. You can build the selection figure in our free PRISMA 2020 flow diagram generator and then add the network-specific reporting alongside it.
The reason a network meta-analysis needs more than the base diagram is that the flow chart alone cannot show how treatments connect. Two reviews can screen identical numbers of records yet produce completely different networks, and the reader needs to see that structure to judge the result.
Why the Base Diagram Is Necessary but Not Sufficient
A pairwise meta-analysis answers one question: does treatment A beat treatment B? A network meta-analysis answers many at once, ranking a whole set of treatments even when some pairs were never compared head to head. The PRISMA flow diagram still documents that the search and screening were comprehensive, which is the foundation of any credible synthesis.
But the diagram counts studies moving through the selection funnel; it says nothing about which treatments those studies compared. A network with twelve included trials might connect six treatments in a well-meshed loop, or it might hang on a single fragile link between two nodes. That distinction decides whether the indirect comparisons are trustworthy, so PRISMA-NMA asks for a network geometry diagram in addition to the flow chart. Our comparison of meta-analysis versus systematic review sets out where pooled quantitative synthesis fits in the broader method.
The two figures answer different questions and both belong in the manuscript. The table makes the division of labour explicit.
| Question | Flow diagram | Network diagram |
|---|---|---|
| What it shows | Records moving through selection | How treatments connect |
| Unit | Records, reports, studies | Treatments (nodes) and comparisons (edges) |
| Phases or elements | Identification to inclusion | Nodes sized by participants, edges by trials |
| Answers | Was the search comprehensive? | Are the indirect links trustworthy? |
| Required by | PRISMA 2020 | PRISMA-NMA extension |
What PRISMA-NMA Adds to the Standard Checklist
PRISMA-NMA is an official extension of the PRISMA statement aimed specifically at reviews incorporating network meta-analysis. It keeps every item of the base checklist and adds items covering:
- A description of the network geometry, usually as a node-and-edge diagram where node size reflects the number of participants and edge thickness reflects the number of trials.
- The assumptions of transitivity, meaning that the studies are similar enough across comparisons for indirect evidence to be valid.
- An assessment of inconsistency, where direct and indirect estimates for the same comparison are checked against each other.
- A ranking of treatments, often with a measure such as the surface under the cumulative ranking curve.
These additions sit beside, not inside, the flow diagram. The flow diagram answers "which studies entered the review"; the extension items answer "how do the treatments relate and can we trust the indirect links". Mapping each checklist item to a section of your manuscript is the same discipline our walkthrough of the PRISMA 2020 checklist items recommends for any review.
Counting Through the Flow Diagram
The selection counts behave exactly as in any review, and they must still reconcile from top to bottom. Records identified minus duplicates equals records screened; reports sought minus those not retrieved equals reports assessed; and full-text exclusions, each with a reason, leave the studies that enter the network. If those arithmetic chains are unfamiliar, our explainer on what goes in each PRISMA box defines records, reports, and studies and works through the subtraction stage by stage.
One nuance matters for a network. The "studies included" box counts distinct studies, but each study contributes to one or more pairwise comparisons within the network. A three-arm trial comparing drug A, drug B, and placebo is a single included study that informs three edges of the network simultaneously. The flow diagram reports the one study; the network diagram reports the three connections.
Drawing and Describing the Network
After the flow diagram establishes the evidence base, the network diagram visualises it. The figure below shows a worked example: five treatments, with most connected through a common comparator and one closed loop providing both direct and indirect evidence.
Good practice is to:
- Draw one node per treatment and one edge per direct comparison that at least one trial provides.
- Scale node and edge size to participant and trial counts so the reader sees where the evidence is thick and where it is thin.
- Flag any treatment connected by only a single small trial, because conclusions about it rest on a weak link.
- State the transitivity assumption explicitly and explain why the populations are similar enough to justify combining direct and indirect evidence.
A sparse, star-shaped network where everything connects only through a common comparator supports far weaker indirect conclusions than a densely connected web with many closed loops. Reporting the geometry honestly is what lets a reader calibrate their trust.
Heterogeneity and Inconsistency
Network meta-analysis inherits the usual concern with heterogeneity between studies and adds a second concern, inconsistency, which is disagreement between direct and indirect estimates of the same comparison. The two are easy to conflate, so the table separates them.
| Heterogeneity | Inconsistency | |
|---|---|---|
| What varies | Effects across studies within one comparison | Direct versus indirect estimates of the same comparison |
| Where it occurs | Any meta-analysis, pairwise or network | Only in networks with closed loops |
| How to detect | I-squared, tau-squared, prediction intervals | Node-splitting, design-by-treatment interaction |
| What it threatens | Precision of a pooled estimate | Validity of combining direct and indirect evidence |
Both must be reported. Our guide to heterogeneity in meta-analysis covers the statistics for the first; for the second, you report node-splitting or a design-by-treatment interaction model that tests whether the direct and indirect routes agree. A network that passes neither check cannot support a credible ranking, regardless of how clean the flow diagram looks.
Frequently Asked Questions
Does a network meta-analysis use a PRISMA flow diagram?
Yes. Study selection is reported with a standard PRISMA 2020 flow diagram, exactly as in a pairwise review. The network meta-analysis adds a separate network geometry diagram and the PRISMA-NMA checklist items, but it does not replace or alter the flow diagram.
What is PRISMA-NMA?
PRISMA-NMA is the official PRISMA extension for systematic reviews that incorporate a network meta-analysis. It retains the full base checklist and adds items on network geometry, the transitivity assumption, inconsistency between direct and indirect evidence, and treatment ranking.
How is a network diagram different from a flow diagram?
A flow diagram tracks records through identification, screening, eligibility, and inclusion. A network diagram shows how the included treatments connect, with nodes for treatments and edges for the direct comparisons trials provide. They answer different questions and both appear in a network meta-analysis.
How do I count a three-arm trial in the flow diagram?
It is one included study in the flow diagram. Within the network it contributes to multiple pairwise comparisons, one for each pair of arms, so the same single study can inform several edges of the network diagram.
What is transitivity in a network meta-analysis?
Transitivity is the assumption that studies across the different comparisons are similar enough in their populations, settings, and methods that indirect evidence is valid. If trials comparing A with placebo differ systematically from trials comparing B with placebo, the indirect A-versus-B estimate may be biased.