# Indirect comparison and network meta analysis protocol

Use this original working template to plan a systematic evidence review and a feasible indirect comparison. PRISMA-P addresses protocol reporting; ISPOR ITC/NMA guidance addresses conduct and appraisal; PRISMA-NMA addresses the completed network review. A dossier section requesting a comparison is not evidence that the comparison is scientifically defensible.

## Decision question and review scope

Define population, intervention nodes, comparator nodes, outcomes, timing, setting, and eligible study designs. State the target decision population and target relative or absolute effect. Explain whether the intended contrast is conditional or marginal and why it matters for the planned effect scale.

Define treatment nodes before viewing comparative results: doses, combinations, background therapy, line of treatment, treatment duration, and permitted pooling. Set outcome definitions and common assessment windows. Decide which trial phases, designs, follow-up durations, and data sources are admissible.

## Systematic identification and extraction

Search sources, dates, full reproducible strategies, language limits and justification, registry/unpublished evidence sources, duplicate screening, conflict resolution, and update cutoff: [complete]. Identify multiple reports of the same trial and overlapping populations; avoid double counting.

Extract study identifiers, arms and denominators, randomization, eligibility, prior therapy, potential effect modifiers and prognostic factors, outcome definitions, time points, effect directions, uncertainty, and risk-of-bias information. Preserve source page/table and any transformations. Resolve inconsistencies before synthesis; document missing information requests.

## Feasibility and decision to synthesize

Draw the treatment network for each outcome and time horizon. Record which evidence directly connects nodes, whether populations overlap, and which estimates share a comparator or multi-arm trial. Specify minimum evidence/precision needed for the decision; absence of a significance test is not proof of transitivity or consistency.

Compare effect-modifier distributions across comparisons, alongside design, eligibility, setting, follow-up, outcome measurement, and risk of bias. Distinguish measured similarity from an assumption of unmeasured similarity. Prespecify what heterogeneity would make a pooled estimate misleading and when to present a narrative evidence gap instead.

Route the analysis: a valid common-comparator anchored comparison; a connected network model; a population-adjusted comparison with appropriate patient-level data and target-population information; or no defensible quantitative comparison. Unanchored comparisons need substantially stronger control of prognostic factors and effect modifiers. No method creates absent overlap or recovers unavailable confounders.

## Synthesis and diagnostics

Define direction explicitly, for example log ratio for B versus A and C versus A implies B versus C by subtraction. Specify effect measure, likelihood, link, fixed/random effects, heterogeneity model/prior if Bayesian, multi-arm dependence, zero events, and missing uncertainty. Ratios cannot generally be subtracted on their original scale.

Prespecify transitivity assessment, heterogeneity exploration, and inconsistency checks when the network has suitable evidence. In a tree with no independent loops, statistical inconsistency tests cannot establish agreement. Report estimates and uncertainty before rankings; ranks alone do not show clinically meaningful superiority.

For MAIC, specify target population, balancing moments, overlap/weight diagnostics and effective sample size. For STC or model-based standardization, average predictions over a justified target covariate distribution and align the marginal/conditional estimand; predicting at mean covariates is generally insufficient for nonlinear models. Document covariate information unavailable in aggregate reports.

Sensitivity analyses: [node definitions, bias exclusions, effect modifiers, models, time windows, priors, missing data]. Separate protocol amendments from planned analyses. Predefine limits on using observational evidence with randomized evidence and explain any design-specific synthesis.

## Deliverables and guidance crosswalk

Planned outputs: reproducible search, selection flow, excluded-study reasons, extraction ledger, risk-of-bias assessment, feasibility table, network, relative effects and uncertainty, diagnostics, and interpretive limitations. Specify the relevance/credibility appraisal for payer use.

PRISMA-P: [https://doi.org/10.1186/2046-4053-4-1](https://doi.org/10.1186/2046-4053-4-1).
ISPOR conduct: [https://doi.org/10.1016/j.jval.2011.01.011](https://doi.org/10.1016/j.jval.2011.01.011).
ISPOR-AMCP-NPC appraisal: [https://doi.org/10.1016/j.jval.2014.01.004](https://doi.org/10.1016/j.jval.2014.01.004).
PRISMA-NMA reporting: [https://doi.org/10.7326/M14-2385](https://doi.org/10.7326/M14-2385).

Complete the current official item-level crosswalk for the relevant stage. Do not present this editorial planning aid as the official checklist.

## Document control and review

Study title: [complete]. Sponsor and scientific lead: [complete]. Protocol ID and version: [complete]. Evidence cutoff and data cutoff: [complete]. Intended audience and jurisdiction: [complete]. Status: working draft for review.

Record whether investigators have accessed aggregate feasibility results, treatment-specific outcomes, or individual records, with dates. Distinguish feasibility-informed decisions from decisions made before any outcome inspection. Keep the original specification and a dated amendment log.

Decision owner and approval date: [complete]. Unresolved decisions: [list with consequence, owner, and due date]. Ethics, privacy, data-use permissions, registration, funding, conflicts, patient involvement, and dissemination: [complete or justify not applicable]. Do not infer consent, legal permission, or approval from use of this template.

## Evidence and implementation traceability

For every consequential definition, input, or assumption, record: source ID and exact page/table; population and date; the statement it supports; applicability and limitations; implementation artifact; responsible reviewer; verification status. Keep unsupported choices marked unresolved. Store full citations with stable DOI or official URLs.

For each analysis, specify the objective, population, outcome and time horizon, target quantity, estimator, uncertainty, diagnostics, sensitivity analyses, planned table or figure, and responsible programmer. Link the SAP, data dictionary, versioned code lists, data-fitness assessment, phenotype evidence matrix, and deviation log where applicable.

## Review disposition

Scientific design: [reviewer, date, concerns]. Clinical interpretation: [reviewer, date]. Data and programming feasibility: [reviewer, date]. Statistics or economics: [reviewer, date]. Governance: [reviewer, date]. Remaining material limitations: [complete]. A populated template is not approval of the design.