Published
The PRISMA-NMA checklist is the reporting standard that journals expect when a systematic review includes a network meta-analysis. Hutton and colleagues published it in 2015 as an extension to the original PRISMA statement, and it runs to 32 items [1]. Twenty-seven of them are the familiar PRISMA items with wording adapted to networks of treatments, and five are new items about the geometry of the network and the agreement of direct and indirect evidence. This guide paraphrases all 32 items, explains the five new ones in plain terms, and then applies the whole checklist row by row to a real open-access network meta-analysis so you can see what a complete answer looks like and where even a careful team leaves a gap. The PRISMA-NMA checklist measures completeness of reporting and says nothing about whether the science was sound, so a full score is a floor and not a quality mark.
In Simple Words
A network meta-analysis compares three or more treatments at once, using head-to-head trials and indirect links through shared comparators. The PRISMA-NMA checklist asks you to report everything a normal review reports, plus a map of which treatments were compared, how you checked that direct and indirect evidence agree, and how similar the trials were. If a reader cannot find those three things, they cannot judge whether your ranking of treatments can be trusted.
What the PRISMA-NMA Checklist Is and When to Use It
The extension was written by a group of methodologists who noticed that network meta-analyses were being published with the same reporting as two-arm meta-analyses. A pairwise review needs a flow diagram, a forest plot and a heterogeneity statistic. A network review needs those and more, because the result depends on the structure of the evidence. Treatments that were never tested against each other are compared through a common comparator, and the validity of that step rests on assumptions the reader can only judge if the report shows them [1].
Hutton and colleagues describe the PRISMA-NMA checklist as the 27 PRISMA items with added guidance for networks plus five items marked S1 to S5, which sit in the Methods and Results sections [1][8]. The numbering follows the 2009 PRISMA statement that the extension was built on. The PRISMA 2020 statement has since updated the base guideline, including a revised flow diagram and expanded sub-items under its 27 items [3]. The checklist file hosted on the PRISMA website at the time of writing is still the 2015 version, so a network meta-analysis today usually reports against PRISMA-NMA for the network items and uses the PRISMA 2020 flow diagram for the identification and screening record. Our general guide to the PRISMA 2020 checklist covers the base statement.
Choosing the right guideline takes a minute and saves a revision round. The table maps common review designs to the checklist that fits.
| Your review | Guideline to follow | Note |
|---|---|---|
| Systematic review with a network meta-analysis of treatments | PRISMA-NMA, with the PRISMA 2020 flow diagram | All 32 items apply, including S1 to S5 |
| Systematic review with a pairwise meta-analysis only | PRISMA 2020 | See our PRISMA 2020 guide |
| Review of diagnostic test accuracy | PRISMA-DTA | See our PRISMA-DTA worked example |
| Appraising a published systematic review you plan to cite or update | AMSTAR 2 for quality appraisal | See our AMSTAR 2 guide |
| Methods background for the network analysis itself | Cochrane Handbook chapter on network meta-analysis | See the Cochrane chapter and our network meta-analysis explainer |
The 32 Items of the PRISMA-NMA Checklist
The table below paraphrases each item of the PRISMA-NMA checklist hosted on the PRISMA website [8]. Items S1 to S5 are the network-specific additions and are shaded. Item numbers follow the official layout, so the five S items appear where they fall in the report.
| Item | Section | What the item asks for |
|---|---|---|
| 1 | Title | Identify the report as a systematic review that includes a network meta-analysis, or a related form of synthesis. |
| 2 | Abstract | Give a structured summary with aims, sources, eligibility, appraisal, synthesis method, number of studies and participants, estimates with intervals, limitations, funding and the registration number. |
| 3 | Introduction | Give the rationale in light of existing evidence and say why a network meta-analysis was chosen. |
| 4 | Introduction | State the question with participants, interventions, comparisons, outcomes and study design. |
| 5 | Methods | Say whether a protocol exists, where to read it, and give the registration number. |
| 6 | Methods | Specify eligibility criteria and the treatments in the network, stating whether any were clustered or merged into one node and why. |
| 7 | Methods | List every information source with coverage dates and the date last searched. |
| 8 | Methods | Present the full electronic search for at least one database, with limits, so it can be repeated. |
| 9 | Methods | State how studies were selected at each stage. |
| 10 | Methods | Describe data extraction, including duplicate working and contact with investigators. |
| 11 | Methods | List and define all data items and any assumptions or simplifications. |
| S1 | Methods | Describe how the geometry of the treatment network and related biases were explored, and how the evidence base was summarised graphically. |
| 12 | Methods | Describe risk-of-bias assessment, whether at study or outcome level, and how it informed the synthesis. |
| 13 | Methods | State the summary measures, plus any rankings or SUCRA values. |
| 14 | Methods | Describe the analysis for each network, covering multi-arm trials, variance structure, priors for Bayesian models and model fit. |
| S2 | Methods | Describe the statistical methods used to compare direct and indirect evidence, and what was done if disagreement was found. |
| 15 | Methods | Describe assessment of bias across studies, such as publication bias and selective reporting. |
| 16 | Methods | Describe additional analyses and say which were pre-specified. |
| 17 | Results | Report numbers screened, assessed and included, with reasons for exclusion, ideally in a flow diagram. |
| S3 | Results | Provide a network graph of the included studies. |
| S4 | Results | Summarise the network, including abundance of trials, evidence gaps and biases the structure may reflect. |
| 18 | Results | Present study characteristics with citations. |
| 19 | Results | Present risk of bias for each study, and by outcome where available. |
| 20 | Results | For every outcome, present data for each study, with summary data per arm and effect estimates with intervals. |
| 21 | Results | Present each network meta-analysis with intervals, and rankings if explored. Large networks may focus on comparisons against one comparator. |
| S5 | Results | Report the results of inconsistency investigations, such as test statistics, p values or model fit comparisons. |
| 22 | Results | Present results of the assessment of bias across studies. |
| 23 | Results | Report additional analyses, such as sensitivity, subgroup or meta-regression results. |
| 24 | Discussion | Summarise main findings and the strength of evidence per main outcome, and say who the findings matter to. |
| 25 | Discussion | Discuss limitations at study, outcome and review level, and comment on the validity of transitivity and consistency and on any concern about network geometry. |
| 26 | Discussion | Interpret the results against other evidence and state implications for research. |
| 27 | Funding | Describe funding and the role of funders, including any funding from manufacturers of treatments in the network and conflicts of content-expert authors. |
The Five Network-Specific Items Explained
Most of the PRISMA-NMA checklist asks for what any systematic review reports. The five S items are where network reviews differ, and they are the rows reviewers check first.
Items S1, S3 and S4, the geometry of the network
Geometry means the shape of the evidence. Each treatment is a node, each pair of treatments compared directly in at least one trial is an edge, and the thickness of the edge usually shows how many trials or patients sit behind it. Item S1 asks for the methods used to explore this, S3 asks for the graph, and S4 asks for a short narrative about it [1]. A good network graph answers three questions at a glance. Which treatments have the most evidence, which comparisons have none, and does any single comparator hold the network together.
A quick count helps you write S4. With six treatments there are 15 possible pairs. If the graph shows nine edges, six pairs have no head-to-head trial, and any estimate for them rests wholly on indirect evidence. Saying so in one sentence tells the reader which results to treat with more caution.
Items S2 and S5, direct and indirect evidence must agree
Where a treatment pair has both direct and indirect evidence, the two should give similar answers. Item S2 asks you to name the method used to check this, and S5 asks for the result. Typical choices include a global test such as a design-by-treatment interaction model, and local checks that split a comparison into its direct and indirect parts [1][6]. If the check shows disagreement, the report should say what was done about it, for example investigating the trials behind the conflict or presenting results from a model that does not assume consistency.
Transitivity sits inside items 25 and S4
Transitivity is the assumption that trials comparing different pairs of treatments are similar enough in effect modifiers, such as age or disease severity, for an indirect comparison to make sense. The checklist does not give it a separate item. Instead, the explanatory appendix asks authors to present tabulated study and patient characteristics across comparisons so readers can judge the assumption, and item 25 asks authors to comment on its validity [1]. A transitivity table is therefore the most useful single supplement you can add.
Rankings and the certainty of the evidence
Item 13 allows rankings such as the surface under the cumulative ranking curve, known as SUCRA [5]. A ranking reports where each treatment sits, not how sure the estimate is, so pair it with intervals and a certainty rating. The CINeMA framework gives one structured way to rate confidence in each network estimate across within-study bias, reporting bias, indirectness, imprecision, heterogeneity and incoherence [4].
PRISMA-NMA Checklist Worked Example on an Open-Access Network Meta-Analysis
To show what each item looks like in practice, the table below applies the PRISMA-NMA checklist to a published review. Pornsuriyasak and colleagues compared regular and as-needed inhaled treatments for mild asthma in a systematic review and network meta-analysis of randomised trials, published in BMC Medicine in January 2025 [2]. They searched MEDLINE through PubMed, Scopus and ClinicalTrials.gov up to 31 March 2024, included 42 trials, and analysed children aged 6 to 11 years separately from adolescents and adults. The review was registered in PROSPERO as CRD42022352384.
The article is open access under a CC BY 4.0 licence. We confirmed this from the Crossref licence field for the version of record, the Europe PMC licence field and the licence statement in the article XML. We read the main text, the tables, the figures, Additional file 1 and Additional file 3. This post does not judge the clinical result. It reads the report against the 32 reporting items.
Figure 1 is what items S3 and S4 look like when done well. Each panel is a separate network, one per outcome, because the set of treatments tested differs from outcome to outcome. On our reading of panel A, the six treatments give 15 possible pairs, nine are connected by a direct comparison, and the other six, including tiotropium against as-needed short-acting beta-agonist, can only be estimated indirectly. The article reports tiotropium against that comparator at risk ratio 0.44 (95% CI 0.21 to 0.94) [2], and a reader who sees panel A understands at once that this estimate is carried through regular inhaled corticosteroid.
| Item | Section | Verdict | Where in the article | What we found |
|---|---|---|---|---|
| 1 | Title | Reported | Title | The title ends “a systematic review and network meta-analysis”, so the design is named as the item asks. |
| 2 | Abstract | Partly reported Judgement call | Abstract, all four headings | The abstract gives the registration number, 13 trials in children and 29 in adolescents and adults, effect estimates with intervals, rankings and conclusions. It gives no participant totals and no funding statement, both of which the item lists. Funding appears in the end matter. Judgement call because the journal abstract format is short. |
| 3 | Introduction | Reported | Background, second paragraph | The authors say earlier reviews “are restricted to direct comparisons, failing to definitively rank the efficacy of different treatments”, which is the reason for a network analysis. |
| 4 | Introduction | Reported Judgement call | Abstract Background; Methods, “Interventions of interest” and “Outcomes of interest” | There is no single sentence in PICOS form. The aim, populations, treatments, comparator and outcomes are spread across the abstract and Methods. Judgement call because a reader can assemble PICOS but has to look in three places. |
| 5 | Methods | Reported | Methods, first paragraph; Abstract | Registered in PROSPERO with the number CRD42022352384. |
| 6 | Methods | Reported | Methods, “Search strategy and study selection” and “Interventions of interest” | Eligible designs, ages, exclusions and the no-language-limit rule are listed. Treatment classes are defined, and inhaled corticosteroid with a fast-acting long-acting agent in one inhaler or two is treated as one class. |
| 7 | Methods | Reported | Methods, “Search strategy and study selection” | MEDLINE through PubMed and Scopus from inception to 31 March 2024, plus ClinicalTrials.gov for protocols, ongoing and unpublished trials. |
| 8 | Methods | Reported | Additional file 1, Table S1 | Line-by-line strategies for PubMed and Scopus are supplied in the supplement. |
| 9 | Methods | Reported | Methods, “Search strategy and study selection”; Fig. 1 | Two reviewers screened independently. The article says “Two reviewers independently screened titles and abstracts” and the full-text step follows. |
| 10 | Methods | Reported | Methods, “Data extraction” | The article states “Two reviewers independently extracted the following data”. Trials were excluded if data were insufficient after three contact attempts with the authors. |
| 11 | Methods | Reported | Methods, “Data extraction” and “Outcomes of interest” | Study, participant, intervention and outcome variables are listed, and mild asthma is defined per trial in Additional file 1, Table S2. |
| S1 | Methods | Partly reported Judgement call | Figs. 2 and 3 captions; Methods, “Data analysis and synthesis” | The network graphs are defined in the captions, and transitivity tables are in the supplement. No Methods sentence says how geometry was explored. Judgement call because the content is present but not announced as a method. |
| 12 | Methods | Partly reported Judgement call | Methods, “Risk of bias assessment” | RoB 2 was used by two reviewers working independently, and the article states “If at least one outcome in the study exhibited a high risk of bias, the entire study was considered a ‘high risk of bias.’” Study level is therefore specified. How the judgement fed the synthesis is not described apart from the CINeMA certainty rating. Judgement call because CINeMA includes a within-study bias domain. |
| 13 | Methods | Reported | Methods, “Data analysis and synthesis” | Risk ratios, mean differences and standardised mean differences, with SUCRA for ranking. |
| 14 | Methods | Partly reported | Methods, “Data analysis and synthesis” | A two-stage network meta-analysis with a DerSimonian-Laird random-effects model, multivariate pooling and Stata 18 is described, and the analysis is frequentist, so priors do not apply. The article does not state how multi-arm trials were handled or how model fit was assessed. Panel A of Figure 1 shows about 20 study-level comparisons drawn from nine trials, which suggests multi-arm trials were present. |
| S2 | Methods | Reported | Methods, “Data analysis and synthesis” | The article states “The consistency assumption was assessed using a design-by-treatment interaction model”. |
| 15 | Methods | Reported | Methods, “Data analysis and synthesis” | Comparison-adjusted funnel plots and Egger’s test were planned. |
| 16 | Methods | Partly reported Judgement call | Methods, “Data analysis and synthesis” | Meta-regression, subgroup analysis and leave-one-out sensitivity analysis are described. The effect modifiers for transitivity are called prespecified, but the article does not label the other additional analyses as prespecified or post hoc beyond calling subgroups exploratory. Judgement call. |
| 17 | Results | Reported | Results, first paragraph; Fig. 1 in the article | The flow diagram gives every count with reasons. We checked the arithmetic. PubMed 1,351 plus Scopus 1,884 gives 3,235, minus 663 duplicates gives 2,572 screened, minus 2,452 excluded gives 120 full texts, minus 82 gives 38 trials, and four registry-route reports bring the total to 42. |
| S3 | Results | Reported | Fig. 2 (children) and Fig. 3 (adolescents and adults) | Nine panels in all, one per outcome, with study and participant numbers on each edge. |
| S4 | Results | Reported Judgement call | Results, each outcome subsection; Discussion, limitations paragraph | Each subsection states the number of trials and participants in the network, and the Discussion names gaps, such as few trials in children at asthma step 1. There is no dedicated geometry paragraph. Judgement call. |
| 18 | Results | Reported | Additional file 1, Table S5; Results, first paragraph | Characteristics of all 42 trials are tabulated in the supplement, with ranges for age, FEV1 and sample size in the text and citations by number. |
| 19 | Results | Reported | Results, “Risk of bias”; Additional file 1, Table S3 and Fig. S1 | The article reports 21.4% low risk, 59.5% some concerns and 19% high risk. |
| 20 | Results | Reported | Additional file 2 (extracted data); Additional file 1 forest plots | Per-trial data are supplied as a supplementary file, and the Data availability statement says “Extracted data is available as a supplementary file.” |
| 21 | Results | Reported | Results; Tables 1 and 2; Additional file 1, treatment-effect tables such as Table S6 | Tables 1 and 2 give each treatment against as-needed short-acting beta-agonist with intervals and a certainty rating, and the supplement holds full league tables with SUCRA values, for example in Table S6. |
| S5 | Results | Reported | Results, one sentence per network | Each network reports a test, for example “No evidence of inconsistency existed in the network (chi-square = 0.5, p-value = 0.919)”. The nine p values run from 0.199 to 0.919. |
| 22 | Results | Reported | Results, funnel plot sentences; Additional file 1, Figs. S4, S9, S11, S15, S19 | Asymmetry is reported for specific comparisons, and Egger’s test was significant for quality of life in adults (p = 0.023). |
| 23 | Results | Reported | Results, sensitivity sentences | Dropping one trial changed the rank order for adult symptom scores, with AN-ICS/FABA moving to the top. |
| 24 | Discussion | Reported | Discussion, first paragraph; Tables 1 and 2 | Main findings are summarised by age group, and the certainty ratings sit beside each estimate. |
| 25 | Discussion | Reported | Results, “Transitivity assumption”; Discussion, limitations paragraph | The article states “Therefore, the transitivity assumption may not hold for comparing LTRA vs. SABA and LTRA vs. regular ICS on exacerbation outcome.” That is the kind of explicit comment the item wants. |
| 26 | Discussion | Reported | Discussion; Conclusions | Findings are set against earlier reviews, and the limitations paragraph ends with a call for an update when more data are available. |
| 27 | Funding | Reported | End matter, Funding and Declarations (Competing interests) | Funding is “None” and the article states “The authors declare no competing interests.” This answers the item, including the manufacturer and conflict questions, by implication. |
Tally. Of the 32 items of the PRISMA-NMA checklist, 27 are reported, 5 are partly reported and none could not be located. Six rows carry a Judgement call label, namely items 2, 4, S1, 12, 16 and S4, because reasonable reviewers could score them either way. Two of the six, items 4 and S4, are scored as reported, and the other four are scored as partly reported. Shaded rows are the five network-specific items, and four of the five are fully reported, which is a strong result for a review of this size.
A checklist mismatch worth learning from
The Methods state that the review “followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) extension statement for network meta-analysis (NMA)”. The checklist supplied as Additional file 3, however, is the PRISMA 2020 list of 27 items with sub-items, and its abstract row refers to the separate PRISMA 2020 for Abstracts checklist. It has no rows for network geometry or inconsistency. A reader who relies on that file cannot find items S1 to S5, which is why we located them in the text. The lesson is practical. If you claim the extension, attach the PRISMA-NMA checklist, and fill in page numbers for the S items.
The same article shows three small consistency points that a final read-through would catch. The abstract says the checklist is in Additional file 2, while the supplement list shows Additional file 2 as the extracted data and Additional file 3 as the checklist. An unpublished trial is cited as NCT1316380 in the Results and NCT1316360 in the Discussion, and registry numbers normally carry eight digits after the letters. Table 1 gives 0.39 (0.27, 0.56) for regular ICS/LABA against as-needed short-acting beta-agonist on severe exacerbation without the asterisk used elsewhere for significant results, although the interval excludes 1. None of these changes the findings. They show why identifiers and table markers deserve a last check.
Preparing Your Own PRISMA-NMA Checklist
The example suggests a short routine for any network review. Start with the registration record and decide the node definitions before searching, because merging treatments after seeing the data is a common source of criticism. Then work through the PRISMA-NMA checklist in three passes. A completed PRISMA-NMA checklist with a page number against every item is the file editors expect to see with the submission.
- Pass one, the base items. Fill items 1 to 27 as you would for any review, with the PRISMA 2020 flow diagram for item 17. Our PRISMA flow chart guide explains every box.
- Pass two, the S items. Write one sentence in Methods for S1 and S2, and make sure the network graph, a transitivity table and the inconsistency results each have a named location in the paper or supplement.
- Pass three, the awkward items. Item 14 deserves a sentence on multi-arm trials and model fit, and item 16 deserves a sentence on what was pre-specified. These two were the gaps in the example.
Then match the certainty rating to the claim. If a ranking is the headline, give the confidence rating for it and link the appraisal tool you used. For the trials feeding the network, our guide to RoB 2 explains the five domains, and our guide to quality assessment in network meta-analysis covers the certainty step. When a network is large, move league tables into an appendix and keep the main text focused on comparisons against one reference treatment, as item 21 permits.
Editverse editorial work follows the reporting guidelines journals name, and our standards align with COPE, ICMJE and GPP 2022, with institutional and country-specific ethics followed.
Free Tools for Your Network Meta-Analysis Report
Free Editverse Tools
- GraphPro
Build publication-ready charts, 25 types, with built-in t-test, ANOVA, correlation, and chi-square p-values. Free, no signup. - StatPro
Free statistical calculators and a test-selector tool covering group comparisons, categorical outcomes, time-to-event analysis, and diagnostic accuracy. - Flowchart Pro
Generate PRISMA 2020, PRISMA-ScR, CONSORT, STROBE, STARD, MOOSE, and ARRIVE flow diagrams directly in your browser.
Professional Support for Network Meta-Analysis Reporting
A PRISMA-NMA checklist with every S row filled is easier for an editor to check, and the services below help authors reach that point. If the analysis is the hard part, Editverse biostatisticians can run and write up the network. If the whole review is still ahead of you, the full systematic review service covers protocol to manuscript. Both are listed with the turnaround and price shown on their pages.
Network Meta-Analysis Service
From $2,000 · 14 to 21 days depending on network size
Network geometry and transitivity assessment, a justified choice of frequentist or Bayesian framework, consistency and inconsistency checks, network diagrams, league tables, forest plots and SUCRA or P-scores, with reproducible R code and written methods and results. The service page states that analysis and reporting follow PRISMA-NMA. It works from your extracted data.
Meta-Analysis and Systematic Review Services
$3,500 · 8 to 16 weeks depending on scope
Protocol development and PROSPERO registration, search strategy, screening, data extraction, risk of bias, statistical analysis, GRADE certainty assessment and a PRISMA-compliant manuscript. The page lists network meta-analysis among the supported analysis types.
For a review that is still at the question stage, see our systematic review writing service page, which sets out the evidence synthesis support we offer.
Frequently Asked Questions About the PRISMA-NMA Checklist
How many items does the PRISMA-NMA checklist have?
It has 32 items. Twenty-seven correspond to the PRISMA items with extra guidance for networks, and five, labelled S1 to S5, are new. They cover the geometry methods, the inconsistency methods, the network graph, the network summary and the inconsistency results [1].
Does the PRISMA-NMA checklist replace PRISMA 2020?
No. It is an extension, so it adds to the base guideline rather than replacing it. Authors of network reviews commonly use the extension checklist for the report and the PRISMA 2020 flow diagram for the record of identification and screening [3].
Which journals ask for the PRISMA-NMA checklist?
Instructions to authors differ and change, so read the target journal page before you submit. Many journals that endorse PRISMA ask authors to attach the checklist for the extension that matches the design, and a network meta-analysis falls under the PRISMA-NMA checklist.
Does the PRISMA-NMA checklist change for Bayesian models?
Yes for item 14. A Bayesian analysis must report the prior distributions and how model fit was judged, while a frequentist analysis such as the example above reports the estimation method and software. Both report the same S items on geometry and inconsistency.
Conclusion
The PRISMA-NMA checklist asks for the same transparency as PRISMA, plus a clear picture of the network and evidence that direct and indirect results agree. When we applied the PRISMA-NMA checklist to a recent open-access review, 27 of 32 items were fully reported and five were partly reported, with the gaps sitting in multi-arm handling, model fit, pre-specification and the abstract. The one structural lesson was the mismatch between the extension claimed in the text and the checklist file attached. If you are planning a network review, build the node map first, write the S items as you go, and attach the right checklist. If you want the analysis or the manuscript handled by specialists, the services above are ready to start from your protocol.
References
- Hutton B, Salanti G, Caldwell DM, Chaimani A, Schmid CH, Cameron C, et al. The PRISMA extension statement for reporting of systematic reviews incorporating network meta-analyses of health care interventions: checklist and explanations. Ann Intern Med. 2015;162(11):777-84. doi:10.7326/M14-2385. PMID 26030634. PubMed record
- Pornsuriyasak P, Sa-nguansai S, Thadanipon K, Numthavaj P, McKay GJ, Attia J, et al. Regular versus as-needed treatments for mild asthma in children, adolescents, and adults: a systematic review and network meta-analysis. BMC Med. 2025;23(1):21. doi:10.1186/s12916-025-03847-z. PMID 39838356. PMCID PMC11752773. CC BY 4.0. Full text
- Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021;372:n71. doi:10.1136/bmj.n71.
- Nikolakopoulou A, Higgins JPT, Papakonstantinou T, Chaimani A, Del Giovane C, Egger M, et al. CINeMA: an approach for assessing confidence in the results of a network meta-analysis. PLoS Med. 2020;17(4):e1003082. doi:10.1371/journal.pmed.1003082. Article
- Salanti G, Ades AE, Ioannidis JPA. Graphical methods and numerical summaries for presenting results from multiple-treatment meta-analysis: an overview and tutorial. J Clin Epidemiol. 2011;64(2):163-71. doi:10.1016/j.jclinepi.2010.03.016. Article
- Chaimani A, Caldwell DM, Li T, Higgins JPT, Salanti G. Chapter 11, Undertaking network meta-analyses. In Higgins JPT, Thomas J, Chandler J, et al., editors. Cochrane Handbook for Systematic Reviews of Interventions. Cochrane; current version. Chapter online
- PROSPERO International prospective register of systematic reviews. Record CRD42022352384. Record
- PRISMA NMA checklist of items to include when reporting a systematic review involving a network meta-analysis. PRISMA Statement website. Checklist PDF