systematic-review git:20260619.8a1ecd4A

Immutable. This exact content never changes and is served at /api/v1/blob/ff46a51e81c97e28.

---
name: Systematic Review
description: Conduct a PRISMA-compliant systematic review from protocol through synthesis.
---

# Systematic Review

A systematic review answers a focused question by finding, appraising, and synthesizing
all relevant studies using a transparent, reproducible method. Follow PRISMA 2020.

## 1. Define the question (PICO)
- **Population**: who/what is studied.
- **Intervention/Exposure**: the factor of interest.
- **Comparison**: the alternative or control.
- **Outcome**: what is measured.
Write the question in one sentence. If it has no clear PICO, refuse and ask for one.

## 2. Register a protocol BEFORE searching
Document eligibility criteria, databases, search terms, screening process, and analysis
plan. State this up front to prevent post-hoc bias. Recommend PROSPERO registration.

## 3. Build the search strategy
- Translate each PICO concept into controlled vocabulary (MeSH) plus free-text synonyms.
- Combine within a concept using OR, across concepts using AND.
- Search at least: PubMed/MEDLINE, Embase, and one field-specific database.
- Record the exact query string and date for every database (reproducibility).
- Supplement with reference-list ("snowball") and citation searches.

## 4. Screen in two stages
1. Title/abstract screening against eligibility criteria.
2. Full-text screening of survivors.
Use two independent screeners; resolve conflicts by discussion or a third reviewer.
Track every count for the PRISMA flow diagram: identified, deduplicated, screened,
excluded (with reasons), included.

## 5. Extract data
Use a piloted extraction form capturing: citation, design, sample, intervention,
comparator, outcomes, effect sizes, funding, and conflicts. Extract in duplicate.

## 6. Appraise risk of bias
Apply a validated tool (RoB 2 for RCTs, ROBINS-I for non-randomized, QUADAS-2 for
diagnostic). Report domain-level judgments, not a single opaque score.

## 7. Synthesize
- If studies are comparable, consider meta-analysis (report heterogeneity via I²).
- If not, conduct a structured narrative synthesis grouped by outcome.
- Assess certainty of evidence with GRADE (high/moderate/low/very low).

## 8. Report
Produce: PRISMA flow diagram, summary-of-findings table, risk-of-bias summary, and a
plain-language conclusion that states what the evidence supports AND its limitations.

## Guardrails
- Never silently drop studies; every exclusion needs a logged reason.
- Distinguish "no evidence of effect" from "evidence of no effect."
- Flag publication bias when few small positive studies dominate.