PRISMA-NMA (PRISMA Extension for Network Meta-Analysis)

What it is

The PRISMA Extension for Network Meta-Analysis (PRISMA-NMA) is a reporting guideline for systematic reviews that synthesize a connected network of interventions using direct and indirect comparisons. Published by Hutton, Salanti, and colleagues in Annals of Internal Medicine (2015), it adds network-specific items on top of the base PRISMA framework: a description of network geometry (the network plot, nodes, edges, and weighting), an explicit account of the transitivity / similarity assumption, the methods used to assess consistency (agreement of direct and indirect evidence), and the presentation of ranking statistics (SUCRA, P-scores, mean ranks). It is maintained as a PRISMA extension under the EQUATOR Network and is intended to sit alongside, not replace, the parent PRISMA 2020 statement. Like all PRISMA products it is a transparency instrument — a list of what a competent report must disclose — not a methods manual, a risk-of-bias tool, or a quality score.

When to use

Use PRISMA-NMA to report any systematic review whose synthesis includes at least one indirect comparison — i.e., any network meta-analysis or mixed-treatment comparison, whether the included evidence is from randomized trials or, increasingly, from observational comparative-effectiveness studies. It applies to journal manuscripts, HTA/payer dossiers that submit an indirect treatment comparison to value the relative effect of a new therapy against multiple competitors, and regulatory or scientific-advice packages that lean on network estimates. Decision rule for choosing the right PRISMA product: if the review pools only head-to-head (pairwise) evidence, the base PRISMA 2020 statement suffices; if it forms a connected network with indirect comparisons, use PRISMA-NMA; if you are registering or publishing the protocol, use PRISMA-P (and pre-register in PROSPERO); a scoping review uses PRISMA-ScR and a diagnostic-test-accuracy review uses PRISMA-DTA. PRISMA-NMA governs the reporting of the synthesis; the methodological appraisal of confidence in the network estimates is a separate, companion step (e.g., CINeMA or GRADE for NMA).

What it requires

Beyond the standard PRISMA items (eligibility, search, selection, data extraction, risk-of-bias assessment of the included studies, summary of findings), PRISMA-NMA enforces a set of network-specific disclosures: - Network geometry — a network plot with nodes and edges, node/edge weighting, the number of studies and participants per comparison, and any disconnected subgraphs. A network reported as connected when it is not is a reporting failure. - Node definition (lumping vs splitting) — how interventions were grouped into nodes (dose, regimen, formulation, drug class) and the sensitivity of conclusions to that grouping. - Transitivity / similarity — an explicit statement of the assumption that potential effect modifiers are distributed comparably across the comparisons being indirectly linked, with the clinical and methodological reasoning (and data) used to defend it. - Consistency / incoherence — the statistical approach used to check agreement of direct and indirect estimates (node-splitting / back-calculation, loop-specific tests, or the design-by-treatment interaction model) and what it showed. - Effect estimates and uncertainty for every contrast in the network, including comparisons informed only indirectly, with between-study heterogeneity (τ²) and, for Bayesian fits, priors and convergence. - Ranking statistics, reported with restraint — SUCRA, P-scores, or mean ranks presented with their uncertainty and explicitly framed as relative summaries, not as effect sizes or definitive orderings. For an observational ("real-world") network the same RWE discipline that governs the individual studies carries up to the synthesis: design transparency and data-fitness-for-use of each contributing database, phenotype/algorithm validation, time-zero alignment, and the estimand (and intercurrent-event handling) of each node must be comparable before the nodes are networked.

When NOT to use — limitations and common misapplications

- It is not a methods or quality instrument. Completing the PRISMA-NMA checklist documents that you reported the network; it does not certify that transitivity holds, that the model is correct, or that the evidence is trustworthy. Confidence in the estimates is assessed by CINeMA / GRADE for NMA, and the included studies' internal validity by RoB 2 / ROBINS-I — PRISMA-NMA does none of these and must not be scored as if it did. - Wrong PRISMA product. Reporting a network meta-analysis against the base PRISMA 2020 checklist silently omits the network-specific items (geometry, transitivity, consistency, ranking) that are the entire reason an indirect comparison needs special reporting. Conversely, applying PRISMA-NMA to a purely pairwise review is over-scoping. - Over-interpreting rankings is the single most common misuse: declaring the "best" treatment from a SUCRA value when credible intervals overlap heavily, when the network is sparse, or when transitivity is doubtful. PRISMA-NMA exists in part to discipline this by forcing rankings to be shown with their uncertainty. - Ignoring transitivity — pooling across populations whose effect modifiers (severity, line of therapy, era, care setting) differ, then reporting a tidy network as if the indirect link were valid. - Lumping clinically distinct interventions into one node merely to force the network to connect, or reporting a disconnected network as connected. - Checklist-as-theater — a complete checklist appended to a report that does not actually contain a network plot, a transitivity argument, or consistency diagnostics.

How it maps to this catalog

PRISMA-NMA is the reporting layer over the synthesis concept network-meta-analysis (and meta-analysis-obs when the evidence is observational); the upstream comparative-effectiveness logic lives in comparative-effectiveness-research-cer-methods and ispor-indirect (the ISPOR good-practices for indirect/mixed-treatment comparisons). The question is scoped with picots-framework-rwe. The transitivity assumption is, at root, a generalizability-transportability-external-validity-rwe problem applied across studies: the indirect link is only valid if effect-modifier distributions are exchangeable across the linked comparisons. For a real-world-data network, node comparability is best secured by building each node from a target-trial-emulation (or at minimum an active-comparator-new-user cohort) with aligned eligibility and time-zero, and by pre-specifying each node's estimands-ate-att-intercurrent-events-rwe so the contrasts being networked are the same estimand. Database-level fitness and contrast construction for claims-derived nodes draw on claims-analysis. Confidence in the resulting network estimates is graded with grade (extended for NMA). PRISMA-NMA tells the reader which of these were done and how; the concepts above tell them how to do each one.

Applied note (claims / EHR / registry RWE)

A payer-facing indirect comparison that networks drug-class contrasts assembled from several claims and EHR databases is the hardest transitivity case: channeling, calendar time, formulary and care-setting differences vary by data source and act as effect modifiers across nodes. PRISMA-NMA's transitivity and consistency items become the place where the analyst must show — not assert — that the observational nodes are comparable: harmonized eligibility and time-zero (target-trial framing), validated outcome phenotypes with reported PPV/sensitivity, the same estimand per node, and a documented sensitivity analysis on node definition. Reporting the network plot with per-edge study/participant counts and the τ² for each contrast lets a reviewer judge whether the indirect evidence is thin or robust before any ranking is read.