---
name: rwe-protocol-builder
description: Draft or review a research protocol from an approved study brief, choosing among comparative RWE, descriptive RWE, ITC or NMA, economic analysis, and pragmatic randomized trial templates. Use for protocol development and protocol-to-SAP handoff, not dossier marketing copy or autonomous study execution.
---

# RWE protocol builder

Create a reviewable protocol that answers the user's study question and exposes unresolved decisions. Read [the design map](references/design-map.md), then only the relevant template in `assets/`. Read [instructions](instructions.md) for the user-facing workflow and guidance crosswalk.

## Route and draft

Identify the decision, population, intervention or exposure, comparator if relevant, outcomes, time horizon, audience, data availability, and whether treatment is assigned or observed. Use facts already supplied. Ask concise questions only when a missing answer changes the design; draft independent sections while decisions remain open. For mixed goals, explain which linked modules are needed rather than silently forcing one design.

Read actual source documents, data dictionaries, and supplied metadata before operationalizing fields. Verify each new consequential methodological citation against a primary source and record the supported claim and location. State when a source is inaccessible or only its metadata was verified. Guidance is not authorization to access restricted data or publish a study.

Preserve the human owner's methodological decisions. Recommend alternatives with their implications; do not silently choose an estimand, confounder set, outcome algorithm, margin, missing-data assumption, or decision threshold. Mark placeholders as unresolved, with the decision needed and its consequences. Never manufacture dataset fields, sample sizes, validation results, citations, or approvals. Use synthetic examples only when clearly labelled and separate from study facts.

For treatment-effect studies, align eligibility, assignment, and follow-up at time zero. Distinguish a descriptive quantity from a causal effect. For ITC, assess comparability and evidence connectivity before proposing a numeric comparison. For economics, distinguish value from affordability. For randomized studies, preserve allocation, consent, harms, and oversight requirements. Consult the relevant template for the operational details.

Use official protocol guidance for planning and the appropriate reporting checklist for downstream reporting; they are not interchangeable quality certifications. Check the current guidance version and any jurisdiction-specific requirement. Record each applicable item, protocol location, evidence artifact, completeness status, and reviewer.

## Output and review

Return the protocol draft; open decisions with owners; claim/source ledger; guidance crosswalk; and objective-to-analysis-to-output map. Maintain a version/amendment record and disclose what feasibility or outcome results informed changes. Use [the review prompt](prompts/review-protocol.md) for a separate review when requested. A material unresolved design or evidence gap makes the output a draft, not an approved protocol.

Use public, synthetic, or explicitly authorized non-sensitive inputs. Patient-level information belongs in the user's approved analytical environment, not model prompts. Stop before analysis execution, external submission, or publication unless the user separately authorizes that action.