| name | write-systematic-review |
| description | Use when planning or writing a systematic review of scientific literature — following PRISMA guidelines to search, screen, extract, and synthesize evidence from primary studies with transparent, reproducible methodology. |
| source | Page et al. (2021) "PRISMA 2020 Statement" (BMJ); Cochrane Handbook for Systematic Reviews of Interventions (Higgins et al., 2022); Moher et al. (2009) "Preferred Reporting Items for Systematic Reviews and Meta-Analyses" (PLOS Medicine) |
| tags | ["systematic-review","prisma","evidence-synthesis","literature-review","meta-analysis","research-methodology"] |
Write Systematic Review
Conduct and report a systematic review following PRISMA 2020 standards — with pre-registered protocol, comprehensive search, transparent screening, risk-of-bias assessment, and synthesis of evidence quality.
Why This Is Best Practice
Why best: Systematic reviews with pre-registered protocols and dual independent screening produce reproducible, low-bias evidence syntheses that individual studies and narrative reviews cannot match.
Adopted by: Cochrane Collaboration (the gold standard for systematic reviews), WHO, NIH systematic evidence review programs, NICE (UK clinical guidelines), FDA Evidence-Based Medicine framework, and virtually all clinical guideline bodies require systematic reviews as the primary evidence base for recommendations.
Impact: Ioannidis (2005) "Why Most Published Research Findings Are False" demonstrated that individual studies frequently yield false positives — systematic reviews, by pooling and assessing evidence quality, are the most reliable way to estimate true effect sizes. The PRISMA statement is endorsed by >700 journals (EQUATOR network); systematic reviews using PRISMA have significantly higher citation rates and lower retraction rates than narrative reviews.
Steps
1. Register the protocol before searching
Pre-registration is mandatory for credible systematic reviews:
- Register on PROSPERO (www.crd.york.ac.uk/prospero) — the international register for systematic reviews
- Record: research question, PICO elements, search strategy, inclusion/exclusion criteria, outcomes, planned synthesis method
- Registration prevents selective reporting and outcome-switching — the primary bias in non-registered reviews
2. Formulate the research question using PICO
Structure the question:
- P (Population): who are the participants? What condition/context?
- I (Intervention/Exposure): what is being done or examined?
- C (Comparator): what is it compared to?
- O (Outcome): what outcomes matter? Primary and secondary?
Example: In adults with Type 2 diabetes (P), does GLP-1 agonist therapy (I) compared to SGLT-2 inhibitors (C) reduce cardiovascular mortality (O)?
3. Design the comprehensive search strategy
Search must be comprehensive and reproducible:
- Databases: search at minimum MEDLINE (PubMed), EMBASE, CENTRAL (Cochrane); add discipline-specific databases (PsycINFO, CINAHL, etc.)
- Search terms: combine MeSH terms with free-text synonyms using Boolean operators (AND, OR, NOT); work with a research librarian for complex searches
- Time span: search all years unless specific rationale for cutoff
- Additional sources: grey literature (clinical trial registries, conference abstracts, regulatory submissions); reference lists of included studies; forward citation of key papers
Document the exact search string for each database (required for PRISMA flow diagram).
4. Screen titles/abstracts then full texts
Two-phase screening:
- Title/abstract screening: apply inclusion/exclusion criteria; two independent reviewers screen all records; resolve disagreements by discussion or third reviewer
- Full-text screening: retrieve and assess all articles that passed abstract screening; document reason for exclusion
Use systematic review software (Covidence, Rayyan, DistillerSR) to manage screening and track decisions.
Calculate inter-rater reliability (Cohen's kappa ≥ 0.7 = acceptable agreement).
5. Extract data systematically
Create a data extraction form before extraction begins:
- Study characteristics: authors, year, country, study design, sample size, follow-up duration
- Participant characteristics: demographics, inclusion/exclusion criteria
- Intervention details: dose, duration, comparator
- Outcomes: measures, results, units, time points
- Risk of bias elements (see step 6)
Extract with two independent reviewers for key outcome data; verify against source.
6. Assess risk of bias
Every included study must be assessed for methodological quality:
- RCTs: use Cochrane Risk of Bias 2 (RoB 2) tool: randomization, allocation concealment, blinding, selective outcome reporting
- Observational studies: use ROBINS-I tool
- Diagnostic accuracy: use QUADAS-2
Summarize at the study level and domain level. High-risk-of-bias studies downgrade the certainty of evidence.
7. Synthesize evidence and assess certainty (GRADE)
Narrative synthesis: when pooling is not appropriate (heterogeneous methods/populations); describe patterns, consistencies, and gaps
Meta-analysis: when ≥2 studies report the same outcome with compatible methods; use random-effects model (DerSimonian-Laird) as default for clinical heterogeneity; report I² for heterogeneity assessment
GRADE assessment: rate evidence certainty for each outcome:
- High: consistent, direct, precise evidence from RCTs
- Moderate: one downgrade (risk of bias, inconsistency, imprecision, indirectness, publication bias)
- Low/Very low: multiple downgrades
Common Mistakes
- No pre-registration: Allows post-hoc outcome selection — primary driver of systematic review bias.
- Single-reviewer screening: Increases miss rate for relevant studies; requires dual independent screening.
- Narrative review disguised as systematic: Using the word "systematic" without a registered protocol, comprehensive search, and risk-of-bias assessment is misleading — and journals increasingly desk-reject these.
When NOT to Use
- Research question with <5 existing primary studies: scoping review or narrative review is more appropriate; meta-analysis with too few studies yields unreliable pooled estimates.
