Why English? | Reporting PRISMA-NMA Network Meta-Analyses Through Network Geometry, Transitivity, Inconsistency and Rankings

PRISMA-NMA network meta-analysis reporting matters when a systematic review compares several treatments at once. People searching for network meta-analysis, PRISMA-NMA checklist, treatment ranking or direct and indirect evidence need to understand how a connected evidence network produces comparisons that no single trial may have tested.

The official PRISMA-NMA page, checked on 5 November 2026, says the extension was published in 2015 for systematic reviews comparing multiple treatments through direct and indirect evidence. The original statement and explanatory paper describes a modified 32-item checklist. The PRISMA extensions page currently lists an update of PRISMA-NMA as under development; this article therefore dates the published framework and avoids presenting an announced update as completed guidance.

English is the map of the network. Nodes, edges, effect modifiers, transitivity, heterogeneity, inconsistency, ranking probabilities and certainty are not interchangeable. Clear reporting tells readers which interventions were compared directly, which estimates rely partly or entirely on indirect routes, whether studies are sufficiently comparable, how disagreement was assessed and what uncertainty surrounds every ranking.

Use this guide with eduKate’s explanations of ROB-MEN missing-evidence bias in networks and PRISMA-S literature-search reporting. A network can be statistically connected and still be weakened by missing studies, incomparable effect modifiers or unclear methods.

---

Define the Network Meta-Analysis Question

Specify population, interventions, comparators, outcomes, time points and eligible designs before drawing the network.

Network reporting task

The official PRISMA-NMA extension was published in 2015 for reviews comparing multiple treatments through direct and indirect evidence. For define the network meta-analysis question, connect every technical term to the decision it helps a reader inspect. Nodes represent intervention options, edges represent direct comparisons and the combined network can add indirect evidence only under defensible assumptions. A diagram is useful, but the report also needs words explaining eligibility, effect measures, model choices and what the geometry leaves uncertain.

Worked network example

Trials compare A with B and B with C, creating an indirect A-versus-C estimate even though no trial directly compared A with C. Write the direct evidence, indirect route and combined estimate separately before discussing a ranking. Check whether effect modifiers are similarly distributed across the comparisons that create the indirect route. If the report says treatments are “best” without showing uncertainty, absolute effects or the outcome and time point, revise the statement until the comparison is decision-ready.

Limit and next check

PRISMA-NMA is reporting guidance, not a method for proving transitivity or inconsistency is absent. In define the network meta-analysis question, reporting does not validate the model by itself. The next step may be a transitivity table, local or global inconsistency check, risk-of-bias display, certainty assessment or sensitivity analysis. Readers should know whether a claim comes from direct evidence, indirect evidence or both, and whether the network contains enough connected information to support it.

---

Use the Published PRISMA-NMA Extension With Current Main Guidance

Date the 2015 extension, pair it transparently with the current main PRISMA statement and watch for the announced update.

Preserve the comparison logic

As of 5 November 2026, the PRISMA site lists an update of PRISMA-NMA as under development, so the published extension remains the official reporting reference. A trustworthy account of use the published prisma-nma extension with current main guidance names the population, interventions, outcome, time point and effect measure before results appear. It then shows how the eligible studies form a connected network. This prevents a common language error: treating every intervention in one review as if every pair had been directly tested against each other.

Inspect assumptions through prose and tables

Older, more severe patients receive A-versus-B trials while younger patients receive B-versus-C trials; severity threatens the transitivity assumption. Ask what must be similar for the indirect comparison to be credible and which observed effect modifiers were checked. Describe inconsistency as disagreement between evidence sources, not simply as statistical “noise”. When a model uses random effects, priors or multi-arm trial adjustments, report them clearly enough that another analyst could understand and reproduce the choice.

Communicate the decision boundary

The 2015 extension predates PRISMA 2020; authors should use the current main statement together with relevant extension guidance and explain their approach. End use the published prisma-nma extension with current main guidance by separating relative effects, absolute effects, ranking probabilities and certainty. A high rank can coexist with wide uncertainty, sparse evidence or a clinically small difference. PRISMA-NMA improves visibility; it does not promise that the highest-ranked option is right for every person, school, service or budget.

---

Register the Protocol and Explain Amendments

Preserve planned nodes, outcomes, effect modifiers and analyses before study results shape decisions.

Follow the evidence around the network

The modified extension contains 32 reporting items and adds network-specific detail to the systematic-review report. Under register the protocol and explain amendments, trace one comparison from eligibility through search, study appraisal, synthesis and presentation. This route reveals whether the same outcome definition and decision time are being compared, whether multi-arm trials are handled correctly and whether disconnected or weakly connected parts of the network are being hidden behind one pooled result.

Make uncertainty readable

A sparse network places one small study on the only edge connecting a treatment, making that treatment’s rank unstable. Report estimates with intervals and explain what the interval means for plausible decisions. If ranking metrics are used, pair them with the underlying effects and certainty rather than presenting a league table as a verdict. Good English names the reference treatment, direction of benefit, scale and time horizon so that readers do not reverse or exaggerate the result.

Responsible reporting action

A connected network is necessary for a joint analysis but does not guarantee credible indirect evidence. For register the protocol and explain amendments, finish with a check that matches the risk: compare direct and indirect estimates, inspect network geometry, test alternative assumptions, or qualify the conclusion. Reporting should make disagreement and missing evidence easier to see. It must not be used to decorate an analysis whose eligibility, assumptions or model remain unexplained.

---

Define Nodes Without Hiding Important Differences

Explain whether doses, delivery modes, combinations or classes are merged and why that decision is defensible.

Network reporting task

Network geometry should reveal intervention nodes, direct-comparison edges and the amount of evidence supporting each part of the network. For define nodes without hiding important differences, connect every technical term to the decision it helps a reader inspect. Nodes represent intervention options, edges represent direct comparisons and the combined network can add indirect evidence only under defensible assumptions. A diagram is useful, but the report also needs words explaining eligibility, effect measures, model choices and what the geometry leaves uncertain.

Worked network example

Direct A-versus-C evidence favours A while the indirect route through B favours C; the report should show and investigate the disagreement. Write the direct evidence, indirect route and combined estimate separately before discussing a ranking. Check whether effect modifiers are similarly distributed across the comparisons that create the indirect route. If the report says treatments are “best” without showing uncertainty, absolute effects or the outcome and time point, revise the statement until the comparison is decision-ready.

Limit and next check

A ranking probability is not the probability that a treatment will work for a particular person. In define nodes without hiding important differences, reporting does not validate the model by itself. The next step may be a transitivity table, local or global inconsistency check, risk-of-bias display, certainty assessment or sensitivity analysis. Readers should know whether a claim comes from direct evidence, indirect evidence or both, and whether the network contains enough connected information to support it.

---

Report Eligibility for Studies and Network Connections

State which comparisons and designs can enter and how disconnected evidence is handled.

Preserve the comparison logic

Transitivity requires that studies contributing to different comparisons are sufficiently comparable on important effect modifiers for indirect inference. A trustworthy account of report eligibility for studies and network connections names the population, interventions, outcome, time point and effect measure before results appear. It then shows how the eligible studies form a connected network. This prevents a common language error: treating every intervention in one review as if every pair had been directly tested against each other.

Inspect assumptions through prose and tables

A treatment ranks first for one outcome but has wide intervals, few events and a small absolute benefit, so rank alone is not decision-ready. Ask what must be similar for the indirect comparison to be credible and which observed effect modifiers were checked. Describe inconsistency as disagreement between evidence sources, not simply as statistical “noise”. When a model uses random effects, priors or multi-arm trial adjustments, report them clearly enough that another analyst could understand and reproduce the choice.

Communicate the decision boundary

Statistical consistency cannot rule out unmeasured differences in effect modifiers. End report eligibility for studies and network connections by separating relative effects, absolute effects, ranking probabilities and certainty. A high rank can coexist with wide uncertainty, sparse evidence or a clinically small difference. PRISMA-NMA improves visibility; it does not promise that the highest-ranked option is right for every person, school, service or budget.

---

Search for Every Eligible Comparison

Design the search around the review question rather than a convenient subset of treatment pairs.

Follow the evidence around the network

Inconsistency concerns disagreement between direct and indirect evidence and can be assessed with local and global approaches. Under search for every eligible comparison, trace one comparison from eligibility through search, study appraisal, synthesis and presentation. This route reveals whether the same outcome definition and decision time are being compared, whether multi-arm trials are handled correctly and whether disconnected or weakly connected parts of the network are being hidden behind one pooled result.

Make uncertainty readable

Three-arm trials share participants across comparisons; analysis must preserve the resulting correlation instead of treating every contrast as independent. Report estimates with intervals and explain what the interval means for plausible decisions. If ranking metrics are used, pair them with the underlying effects and certainty rather than presenting a league table as a verdict. Good English names the reference treatment, direction of benefit, scale and time horizon so that readers do not reverse or exaggerate the result.

Responsible reporting action

The network estimate should be interpreted with risk of bias, missing evidence and certainty for each comparison. For search for every eligible comparison, finish with a check that matches the risk: compare direct and indirect estimates, inspect network geometry, test alternative assumptions, or qualify the conclusion. Reporting should make disagreement and missing evidence easier to see. It must not be used to decorate an analysis whose eligibility, assumptions or model remain unexplained.

---

Show the Study-Selection Flow

Keep records, reports, studies and network-eligible comparisons distinct through screening and exclusion.

Network reporting task

Treatment rankings should be reported with their uncertainty and interpreted beside effect estimates, certainty and clinical relevance. For show the study-selection flow, connect every technical term to the decision it helps a reader inspect. Nodes represent intervention options, edges represent direct comparisons and the combined network can add indirect evidence only under defensible assumptions. A diagram is useful, but the report also needs words explaining eligibility, effect measures, model choices and what the geometry leaves uncertain.

Worked network example

A network diagram shows edge thickness by study count and node size by participant count, making sparse comparisons visible. Write the direct evidence, indirect route and combined estimate separately before discussing a ranking. Check whether effect modifiers are similarly distributed across the comparisons that create the indirect route. If the report says treatments are “best” without showing uncertainty, absolute effects or the outcome and time point, revise the statement until the comparison is decision-ready.

Limit and next check

Results from health interventions do not automatically transfer to educational or career choices without matching the question and evidence design. In show the study-selection flow, reporting does not validate the model by itself. The next step may be a transitivity table, local or global inconsistency check, risk-of-bias display, certainty assessment or sensitivity analysis. Readers should know whether a claim comes from direct evidence, indirect evidence or both, and whether the network contains enough connected information to support it.

---

Extract Effect Modifiers and Network Characteristics

Collect variables that could influence relative effects and threaten indirect comparison.

Preserve the comparison logic

Risk of bias within studies and bias from missing evidence can affect particular comparisons and the network as a whole. A trustworthy account of extract effect modifiers and network characteristics names the population, interventions, outcome, time point and effect measure before results appear. It then shows how the eligible studies form a connected network. This prevents a common language error: treating every intervention in one review as if every pair had been directly tested against each other.

Inspect assumptions through prose and tables

A school project compares study techniques across trials with different baseline achievement, duration and test formats, prompting an effect-modifier table. Ask what must be similar for the indirect comparison to be credible and which observed effect modifiers were checked. Describe inconsistency as disagreement between evidence sources, not simply as statistical “noise”. When a model uses random effects, priors or multi-arm trial adjustments, report them clearly enough that another analyst could understand and reproduce the choice.

Communicate the decision boundary

A report should distinguish evidence description from clinical or policy recommendation. End extract effect modifiers and network characteristics by separating relative effects, absolute effects, ranking probabilities and certainty. A high rank can coexist with wide uncertainty, sparse evidence or a clinically small difference. PRISMA-NMA improves visibility; it does not promise that the highest-ranked option is right for every person, school, service or budget.

---

Display Network Geometry

Report node and edge meaning, study counts, participant counts and any disconnected or weakly connected regions.

Follow the evidence around the network

The official PRISMA-NMA extension was published in 2015 for reviews comparing multiple treatments through direct and indirect evidence. Under display network geometry, trace one comparison from eligibility through search, study appraisal, synthesis and presentation. This route reveals whether the same outcome definition and decision time are being compared, whether multi-arm trials are handled correctly and whether disconnected or weakly connected parts of the network are being hidden behind one pooled result.

Make uncertainty readable

Trials compare A with B and B with C, creating an indirect A-versus-C estimate even though no trial directly compared A with C. Report estimates with intervals and explain what the interval means for plausible decisions. If ranking metrics are used, pair them with the underlying effects and certainty rather than presenting a league table as a verdict. Good English names the reference treatment, direction of benefit, scale and time horizon so that readers do not reverse or exaggerate the result.

Responsible reporting action

PRISMA-NMA is reporting guidance, not a method for proving transitivity or inconsistency is absent. For display network geometry, finish with a check that matches the risk: compare direct and indirect estimates, inspect network geometry, test alternative assumptions, or qualify the conclusion. Reporting should make disagreement and missing evidence easier to see. It must not be used to decorate an analysis whose eligibility, assumptions or model remain unexplained.

---

Describe Direct, Indirect and Network Evidence

Tell readers which source supports each comparison and how evidence routes combine.

Network reporting task

As of 5 November 2026, the PRISMA site lists an update of PRISMA-NMA as under development, so the published extension remains the official reporting reference. For describe direct, indirect and network evidence, connect every technical term to the decision it helps a reader inspect. Nodes represent intervention options, edges represent direct comparisons and the combined network can add indirect evidence only under defensible assumptions. A diagram is useful, but the report also needs words explaining eligibility, effect measures, model choices and what the geometry leaves uncertain.

Worked network example

Older, more severe patients receive A-versus-B trials while younger patients receive B-versus-C trials; severity threatens the transitivity assumption. Write the direct evidence, indirect route and combined estimate separately before discussing a ranking. Check whether effect modifiers are similarly distributed across the comparisons that create the indirect route. If the report says treatments are “best” without showing uncertainty, absolute effects or the outcome and time point, revise the statement until the comparison is decision-ready.

Limit and next check

The 2015 extension predates PRISMA 2020; authors should use the current main statement together with relevant extension guidance and explain their approach. In describe direct, indirect and network evidence, reporting does not validate the model by itself. The next step may be a transitivity table, local or global inconsistency check, risk-of-bias display, certainty assessment or sensitivity analysis. Readers should know whether a claim comes from direct evidence, indirect evidence or both, and whether the network contains enough connected information to support it.

---

Explain the Transitivity Assumption

Compare effect-modifier distributions and justify why indirect comparisons are credible for the target question.

Preserve the comparison logic

The modified extension contains 32 reporting items and adds network-specific detail to the systematic-review report. A trustworthy account of explain the transitivity assumption names the population, interventions, outcome, time point and effect measure before results appear. It then shows how the eligible studies form a connected network. This prevents a common language error: treating every intervention in one review as if every pair had been directly tested against each other.

Inspect assumptions through prose and tables

A sparse network places one small study on the only edge connecting a treatment, making that treatment’s rank unstable. Ask what must be similar for the indirect comparison to be credible and which observed effect modifiers were checked. Describe inconsistency as disagreement between evidence sources, not simply as statistical “noise”. When a model uses random effects, priors or multi-arm trial adjustments, report them clearly enough that another analyst could understand and reproduce the choice.

Communicate the decision boundary

A connected network is necessary for a joint analysis but does not guarantee credible indirect evidence. End explain the transitivity assumption by separating relative effects, absolute effects, ranking probabilities and certainty. A high rank can coexist with wide uncertainty, sparse evidence or a clinically small difference. PRISMA-NMA improves visibility; it does not promise that the highest-ranked option is right for every person, school, service or budget.

---

Assess Risk of Bias Within Studies

Show how primary-study concerns attach to particular outcomes and comparisons in the network.

Follow the evidence around the network

Network geometry should reveal intervention nodes, direct-comparison edges and the amount of evidence supporting each part of the network. Under assess risk of bias within studies, trace one comparison from eligibility through search, study appraisal, synthesis and presentation. This route reveals whether the same outcome definition and decision time are being compared, whether multi-arm trials are handled correctly and whether disconnected or weakly connected parts of the network are being hidden behind one pooled result.

Make uncertainty readable

Direct A-versus-C evidence favours A while the indirect route through B favours C; the report should show and investigate the disagreement. Report estimates with intervals and explain what the interval means for plausible decisions. If ranking metrics are used, pair them with the underlying effects and certainty rather than presenting a league table as a verdict. Good English names the reference treatment, direction of benefit, scale and time horizon so that readers do not reverse or exaggerate the result.

Responsible reporting action

A ranking probability is not the probability that a treatment will work for a particular person. For assess risk of bias within studies, finish with a check that matches the risk: compare direct and indirect estimates, inspect network geometry, test alternative assumptions, or qualify the conclusion. Reporting should make disagreement and missing evidence easier to see. It must not be used to decorate an analysis whose eligibility, assumptions or model remain unexplained.

---

Specify Effect Measures and Model Structure

Name the scale, reference group, fixed or random effects, multi-arm handling and estimation approach.

Network reporting task

Transitivity requires that studies contributing to different comparisons are sufficiently comparable on important effect modifiers for indirect inference. For specify effect measures and model structure, connect every technical term to the decision it helps a reader inspect. Nodes represent intervention options, edges represent direct comparisons and the combined network can add indirect evidence only under defensible assumptions. A diagram is useful, but the report also needs words explaining eligibility, effect measures, model choices and what the geometry leaves uncertain.

Worked network example

A treatment ranks first for one outcome but has wide intervals, few events and a small absolute benefit, so rank alone is not decision-ready. Write the direct evidence, indirect route and combined estimate separately before discussing a ranking. Check whether effect modifiers are similarly distributed across the comparisons that create the indirect route. If the report says treatments are “best” without showing uncertainty, absolute effects or the outcome and time point, revise the statement until the comparison is decision-ready.

Limit and next check

Statistical consistency cannot rule out unmeasured differences in effect modifiers. In specify effect measures and model structure, reporting does not validate the model by itself. The next step may be a transitivity table, local or global inconsistency check, risk-of-bias display, certainty assessment or sensitivity analysis. Readers should know whether a claim comes from direct evidence, indirect evidence or both, and whether the network contains enough connected information to support it.

---

Report Heterogeneity Across the Network

Explain the common or comparison-specific assumptions and quantify uncertainty without relying on one statistic.

Preserve the comparison logic

Inconsistency concerns disagreement between direct and indirect evidence and can be assessed with local and global approaches. A trustworthy account of report heterogeneity across the network names the population, interventions, outcome, time point and effect measure before results appear. It then shows how the eligible studies form a connected network. This prevents a common language error: treating every intervention in one review as if every pair had been directly tested against each other.

Inspect assumptions through prose and tables

Three-arm trials share participants across comparisons; analysis must preserve the resulting correlation instead of treating every contrast as independent. Ask what must be similar for the indirect comparison to be credible and which observed effect modifiers were checked. Describe inconsistency as disagreement between evidence sources, not simply as statistical “noise”. When a model uses random effects, priors or multi-arm trial adjustments, report them clearly enough that another analyst could understand and reproduce the choice.

Communicate the decision boundary

The network estimate should be interpreted with risk of bias, missing evidence and certainty for each comparison. End report heterogeneity across the network by separating relative effects, absolute effects, ranking probabilities and certainty. A high rank can coexist with wide uncertainty, sparse evidence or a clinically small difference. PRISMA-NMA improves visibility; it does not promise that the highest-ranked option is right for every person, school, service or budget.

---

Assess Global Inconsistency

Describe whether the network as a whole shows disagreement between direct and indirect evidence.

Follow the evidence around the network

Treatment rankings should be reported with their uncertainty and interpreted beside effect estimates, certainty and clinical relevance. Under assess global inconsistency, trace one comparison from eligibility through search, study appraisal, synthesis and presentation. This route reveals whether the same outcome definition and decision time are being compared, whether multi-arm trials are handled correctly and whether disconnected or weakly connected parts of the network are being hidden behind one pooled result.

Make uncertainty readable

A network diagram shows edge thickness by study count and node size by participant count, making sparse comparisons visible. Report estimates with intervals and explain what the interval means for plausible decisions. If ranking metrics are used, pair them with the underlying effects and certainty rather than presenting a league table as a verdict. Good English names the reference treatment, direction of benefit, scale and time horizon so that readers do not reverse or exaggerate the result.

Responsible reporting action

Results from health interventions do not automatically transfer to educational or career choices without matching the question and evidence design. For assess global inconsistency, finish with a check that matches the risk: compare direct and indirect estimates, inspect network geometry, test alternative assumptions, or qualify the conclusion. Reporting should make disagreement and missing evidence easier to see. It must not be used to decorate an analysis whose eligibility, assumptions or model remain unexplained.

---

Assess Local Inconsistency

Use suitable methods to examine particular loops or comparisons and explain sparse-data limitations.

Network reporting task

Risk of bias within studies and bias from missing evidence can affect particular comparisons and the network as a whole. For assess local inconsistency, connect every technical term to the decision it helps a reader inspect. Nodes represent intervention options, edges represent direct comparisons and the combined network can add indirect evidence only under defensible assumptions. A diagram is useful, but the report also needs words explaining eligibility, effect measures, model choices and what the geometry leaves uncertain.

Worked network example

A school project compares study techniques across trials with different baseline achievement, duration and test formats, prompting an effect-modifier table. Write the direct evidence, indirect route and combined estimate separately before discussing a ranking. Check whether effect modifiers are similarly distributed across the comparisons that create the indirect route. If the report says treatments are “best” without showing uncertainty, absolute effects or the outcome and time point, revise the statement until the comparison is decision-ready.

Limit and next check

A report should distinguish evidence description from clinical or policy recommendation. In assess local inconsistency, reporting does not validate the model by itself. The next step may be a transitivity table, local or global inconsistency check, risk-of-bias display, certainty assessment or sensitivity analysis. Readers should know whether a claim comes from direct evidence, indirect evidence or both, and whether the network contains enough connected information to support it.

---

Report Pairwise and Network Estimates

Present direct, indirect and combined estimates where informative, always with intervals and direction.

Preserve the comparison logic

The official PRISMA-NMA extension was published in 2015 for reviews comparing multiple treatments through direct and indirect evidence. A trustworthy account of report pairwise and network estimates names the population, interventions, outcome, time point and effect measure before results appear. It then shows how the eligible studies form a connected network. This prevents a common language error: treating every intervention in one review as if every pair had been directly tested against each other.

Inspect assumptions through prose and tables

Trials compare A with B and B with C, creating an indirect A-versus-C estimate even though no trial directly compared A with C. Ask what must be similar for the indirect comparison to be credible and which observed effect modifiers were checked. Describe inconsistency as disagreement between evidence sources, not simply as statistical “noise”. When a model uses random effects, priors or multi-arm trial adjustments, report them clearly enough that another analyst could understand and reproduce the choice.

Communicate the decision boundary

PRISMA-NMA is reporting guidance, not a method for proving transitivity or inconsistency is absent. End report pairwise and network estimates by separating relative effects, absolute effects, ranking probabilities and certainty. A high rank can coexist with wide uncertainty, sparse evidence or a clinically small difference. PRISMA-NMA improves visibility; it does not promise that the highest-ranked option is right for every person, school, service or budget.

---

Present Rankings Without Creating a League Table Verdict

Pair rank probabilities or SUCRA-style summaries with effects, uncertainty, certainty and practical relevance.

Follow the evidence around the network

As of 5 November 2026, the PRISMA site lists an update of PRISMA-NMA as under development, so the published extension remains the official reporting reference. Under present rankings without creating a league table verdict, trace one comparison from eligibility through search, study appraisal, synthesis and presentation. This route reveals whether the same outcome definition and decision time are being compared, whether multi-arm trials are handled correctly and whether disconnected or weakly connected parts of the network are being hidden behind one pooled result.

Make uncertainty readable

Older, more severe patients receive A-versus-B trials while younger patients receive B-versus-C trials; severity threatens the transitivity assumption. Report estimates with intervals and explain what the interval means for plausible decisions. If ranking metrics are used, pair them with the underlying effects and certainty rather than presenting a league table as a verdict. Good English names the reference treatment, direction of benefit, scale and time horizon so that readers do not reverse or exaggerate the result.

Responsible reporting action

The 2015 extension predates PRISMA 2020; authors should use the current main statement together with relevant extension guidance and explain their approach. For present rankings without creating a league table verdict, finish with a check that matches the risk: compare direct and indirect estimates, inspect network geometry, test alternative assumptions, or qualify the conclusion. Reporting should make disagreement and missing evidence easier to see. It must not be used to decorate an analysis whose eligibility, assumptions or model remain unexplained.

---

Assess Missing-Evidence and Small-Study Bias

Explain comparison-level and network-level concerns rather than applying one funnel plot mechanically.

Network reporting task

The modified extension contains 32 reporting items and adds network-specific detail to the systematic-review report. For assess missing-evidence and small-study bias, connect every technical term to the decision it helps a reader inspect. Nodes represent intervention options, edges represent direct comparisons and the combined network can add indirect evidence only under defensible assumptions. A diagram is useful, but the report also needs words explaining eligibility, effect measures, model choices and what the geometry leaves uncertain.

Worked network example

A sparse network places one small study on the only edge connecting a treatment, making that treatment’s rank unstable. Write the direct evidence, indirect route and combined estimate separately before discussing a ranking. Check whether effect modifiers are similarly distributed across the comparisons that create the indirect route. If the report says treatments are “best” without showing uncertainty, absolute effects or the outcome and time point, revise the statement until the comparison is decision-ready.

Limit and next check

A connected network is necessary for a joint analysis but does not guarantee credible indirect evidence. In assess missing-evidence and small-study bias, reporting does not validate the model by itself. The next step may be a transitivity table, local or global inconsistency check, risk-of-bias display, certainty assessment or sensitivity analysis. Readers should know whether a claim comes from direct evidence, indirect evidence or both, and whether the network contains enough connected information to support it.

---

Report Additional and Sensitivity Analyses

Show what happens when node definitions, priors, models, high-risk studies or assumptions change.

Preserve the comparison logic

Network geometry should reveal intervention nodes, direct-comparison edges and the amount of evidence supporting each part of the network. A trustworthy account of report additional and sensitivity analyses names the population, interventions, outcome, time point and effect measure before results appear. It then shows how the eligible studies form a connected network. This prevents a common language error: treating every intervention in one review as if every pair had been directly tested against each other.

Inspect assumptions through prose and tables

Direct A-versus-C evidence favours A while the indirect route through B favours C; the report should show and investigate the disagreement. Ask what must be similar for the indirect comparison to be credible and which observed effect modifiers were checked. Describe inconsistency as disagreement between evidence sources, not simply as statistical “noise”. When a model uses random effects, priors or multi-arm trial adjustments, report them clearly enough that another analyst could understand and reproduce the choice.

Communicate the decision boundary

A ranking probability is not the probability that a treatment will work for a particular person. End report additional and sensitivity analyses by separating relative effects, absolute effects, ranking probabilities and certainty. A high rank can coexist with wide uncertainty, sparse evidence or a clinically small difference. PRISMA-NMA improves visibility; it does not promise that the highest-ranked option is right for every person, school, service or budget.

---

Translate Relative Effects Into Absolute Effects

Use a relevant baseline risk and preserve uncertainty so readers understand possible real-world differences.

Follow the evidence around the network

Transitivity requires that studies contributing to different comparisons are sufficiently comparable on important effect modifiers for indirect inference. Under translate relative effects into absolute effects, trace one comparison from eligibility through search, study appraisal, synthesis and presentation. This route reveals whether the same outcome definition and decision time are being compared, whether multi-arm trials are handled correctly and whether disconnected or weakly connected parts of the network are being hidden behind one pooled result.

Make uncertainty readable

A treatment ranks first for one outcome but has wide intervals, few events and a small absolute benefit, so rank alone is not decision-ready. Report estimates with intervals and explain what the interval means for plausible decisions. If ranking metrics are used, pair them with the underlying effects and certainty rather than presenting a league table as a verdict. Good English names the reference treatment, direction of benefit, scale and time horizon so that readers do not reverse or exaggerate the result.

Responsible reporting action

Statistical consistency cannot rule out unmeasured differences in effect modifiers. For translate relative effects into absolute effects, finish with a check that matches the risk: compare direct and indirect estimates, inspect network geometry, test alternative assumptions, or qualify the conclusion. Reporting should make disagreement and missing evidence easier to see. It must not be used to decorate an analysis whose eligibility, assumptions or model remain unexplained.

---

Connect Comparison-Level Certainty to the Network

Rate confidence for important comparisons instead of assigning one certainty label to the whole diagram.

Network reporting task

Inconsistency concerns disagreement between direct and indirect evidence and can be assessed with local and global approaches. For connect comparison-level certainty to the network, connect every technical term to the decision it helps a reader inspect. Nodes represent intervention options, edges represent direct comparisons and the combined network can add indirect evidence only under defensible assumptions. A diagram is useful, but the report also needs words explaining eligibility, effect measures, model choices and what the geometry leaves uncertain.

Worked network example

Three-arm trials share participants across comparisons; analysis must preserve the resulting correlation instead of treating every contrast as independent. Write the direct evidence, indirect route and combined estimate separately before discussing a ranking. Check whether effect modifiers are similarly distributed across the comparisons that create the indirect route. If the report says treatments are “best” without showing uncertainty, absolute effects or the outcome and time point, revise the statement until the comparison is decision-ready.

Limit and next check

The network estimate should be interpreted with risk of bias, missing evidence and certainty for each comparison. In connect comparison-level certainty to the network, reporting does not validate the model by itself. The next step may be a transitivity table, local or global inconsistency check, risk-of-bias display, certainty assessment or sensitivity analysis. Readers should know whether a claim comes from direct evidence, indirect evidence or both, and whether the network contains enough connected information to support it.

---

Explain Results for Students and Families

Separate an average network estimate from a personal choice, local context and professional advice.

Preserve the comparison logic

Treatment rankings should be reported with their uncertainty and interpreted beside effect estimates, certainty and clinical relevance. A trustworthy account of explain results for students and families names the population, interventions, outcome, time point and effect measure before results appear. It then shows how the eligible studies form a connected network. This prevents a common language error: treating every intervention in one review as if every pair had been directly tested against each other.

Inspect assumptions through prose and tables

A network diagram shows edge thickness by study count and node size by participant count, making sparse comparisons visible. Ask what must be similar for the indirect comparison to be credible and which observed effect modifiers were checked. Describe inconsistency as disagreement between evidence sources, not simply as statistical “noise”. When a model uses random effects, priors or multi-arm trial adjustments, report them clearly enough that another analyst could understand and reproduce the choice.

Communicate the decision boundary

Results from health interventions do not automatically transfer to educational or career choices without matching the question and evidence design. End explain results for students and families by separating relative effects, absolute effects, ranking probabilities and certainty. A high rank can coexist with wide uncertainty, sparse evidence or a clinically small difference. PRISMA-NMA improves visibility; it does not promise that the highest-ranked option is right for every person, school, service or budget.

---

Finish With a PRISMA-NMA Reporting Audit

Verify network geometry, assumptions, model, direct and indirect sources, inconsistency, rankings and limitations.

Follow the evidence around the network

Risk of bias within studies and bias from missing evidence can affect particular comparisons and the network as a whole. Under finish with a prisma-nma reporting audit, trace one comparison from eligibility through search, study appraisal, synthesis and presentation. This route reveals whether the same outcome definition and decision time are being compared, whether multi-arm trials are handled correctly and whether disconnected or weakly connected parts of the network are being hidden behind one pooled result.

Make uncertainty readable

A school project compares study techniques across trials with different baseline achievement, duration and test formats, prompting an effect-modifier table. Report estimates with intervals and explain what the interval means for plausible decisions. If ranking metrics are used, pair them with the underlying effects and certainty rather than presenting a league table as a verdict. Good English names the reference treatment, direction of benefit, scale and time horizon so that readers do not reverse or exaggerate the result.

Responsible reporting action

A report should distinguish evidence description from clinical or policy recommendation. For finish with a prisma-nma reporting audit, finish with a check that matches the risk: compare direct and indirect estimates, inspect network geometry, test alternative assumptions, or qualify the conclusion. Reporting should make disagreement and missing evidence easier to see. It must not be used to decorate an analysis whose eligibility, assumptions or model remain unexplained.

---

Frequently Asked Questions

What is PRISMA-NMA?

It is the PRISMA extension for reporting systematic reviews that incorporate network meta-analysis of multiple interventions.

When was the extension published?

The official PRISMA page states that it was published in 2015. An update is listed as under development as of 5 November 2026.

What is direct evidence?

Evidence from studies that directly compare the two interventions of interest.

What is indirect evidence?

Evidence inferred through a connected comparator, such as A versus C through A versus B and B versus C, under assumptions including transitivity.

What is transitivity?

The assumption that the different comparison sets are sufficiently comparable on important effect modifiers for an indirect comparison to be valid.

What is inconsistency?

Important disagreement between direct and indirect evidence beyond what the model and uncertainty reasonably explain.

Does the top-ranked treatment automatically win?

No. Rankings must be read with effect sizes, intervals, certainty, harms, absolute effects and decision context.

Is PRISMA-NMA a quality score?

No. It is reporting guidance; appraisal of methods, risk of bias, missing evidence and certainty requires complementary work.

---

Useful Next Reading

• Read the official PRISMA-NMA page and checklist.
• Read the original PRISMA-NMA statement record.
• Examine missing evidence in network meta-analysis with ROB-MEN.
• Report literature searches with PRISMA-S.

Closing thought: PRISMA-NMA shows why English is the connective tissue of a network: it names every route, assumption and uncertainty so a diagram becomes evidence rather than decoration.