A systematic review is a rigorous, transparent, and reproducible method for identifying, evaluating, and synthesizing all available evidence on a specific research question. Unlike narrative literature reviews, systematic reviews follow a predefined protocol, use comprehensive search strategies across multiple databases, and employ explicit criteria for study selection and quality assessment. This makes systematic reviews the gold standard of evidence synthesis in healthcare, social sciences, education, and many other disciplines.
This step-by-step guide covers the complete systematic review process, from formulating your research question to publishing your PRISMA 2020 compliant review.
Step 1: Define Your Research Question
Every systematic review begins with a clearly defined, answerable research question. The PICO framework is the most widely used structure:
- Population: Who are you studying?
- Intervention: What treatment, exposure, or condition are you examining?
- Comparison: What is the alternative (placebo, standard care, no intervention)?
- Outcome: What results are you measuring?
A well-formed PICO question might be: "In adults with type 2 diabetes (P), does aerobic exercise (I) compared to standard care (C) reduce HbA1c levels (O)?"
Step 2: Develop and Register Your Protocol
Before starting the review, write a systematic review protocol that pre-specifies your methods. This prevents post-hoc changes that could introduce bias. Your protocol should include eligibility criteria, search strategy, screening process, data extraction plan, and analysis methods.
Register your protocol on PROSPERO (for health-related reviews) or other appropriate registries. Protocol registration increases transparency and reduces duplication of effort. For detailed guidance, read our guide on writing a systematic review protocol using PRISMA-P.
Step 3: Develop Your Search Strategy
A comprehensive search strategy is the foundation of a systematic review. You should:
- Search multiple databases relevant to your topic (PubMed/MEDLINE, Embase, Cochrane CENTRAL, PsycINFO, CINAHL, Scopus, Web of Science, etc.)
- Use a combination of MeSH/controlled vocabulary terms and free-text keywords
- Apply Boolean operators (AND, OR, NOT) to combine search concepts
- Avoid overly restrictive filters that might miss relevant studies
- Document your full search strategy for reproducibility
Consider consulting a research librarian or information specialist to optimize your search. The search strategy is arguably the most critical methodological component because a flawed search will undermine everything that follows.
Step 4: Search and Collect Records
Execute your search strategy across all planned databases and download the results into a reference management tool (EndNote, Zotero, Mendeley) or systematic review software (Rayyan, Covidence, DistillerSR).
Also search for grey literature if your protocol includes other sources: clinical trial registries, conference proceedings, organizational reports, thesis databases, and reference lists of included studies. Citation searching (forward and backward) can identify additional relevant records.
Step 5: Remove Duplicates
Import all records and remove duplicates using your reference manager or systematic review software. Record the number of duplicates removed, as this number appears in your PRISMA flow diagram.
Step 6: Screen Titles and Abstracts
Screen all unique records at the title and abstract level against your eligibility criteria. Best practice requires at least two independent reviewers, with disagreements resolved through discussion or a third reviewer.
Pilot your screening criteria on a random sample of 50-100 records before full screening to ensure consistency between reviewers. For detailed guidance on this critical phase, read our article on systematic review screening best practices for title and abstract review.
Step 7: Retrieve Full-Text Reports
Obtain the full text of all records that passed title/abstract screening. Document any reports you could not retrieve and the reasons why.
Step 8: Assess Full-Text Eligibility
Two independent reviewers should assess each full-text report against the eligibility criteria. Record the specific reason for excluding each report, since these reasons appear in your PRISMA flow diagram.
Common exclusion reasons include wrong population, wrong intervention, wrong outcome, wrong study design, duplicate data, insufficient data, and conference abstract only (if full publications are required).
Step 9: Extract Data
Develop a standardized data extraction form and pilot it on 2-3 studies. Extract:
- Study characteristics (author, year, country, design, setting)
- Participant characteristics (sample size, demographics, condition severity)
- Intervention details (type, duration, frequency, dosage)
- Outcome data (measures used, timepoints, results)
- Risk of bias/quality assessment data
Two reviewers should independently extract data, with discrepancies resolved by consensus.
Step 10: Assess Risk of Bias
Evaluate the methodological quality and risk of bias of each included study using a validated tool. Common tools include:
- Cochrane RoB 2 for randomized controlled trials
- ROBINS-I for non-randomized studies of interventions
- Newcastle-Ottawa Scale for observational studies
- QUADAS-2 for diagnostic accuracy studies
For a comprehensive comparison of assessment tools, see our guide on quality assessment in systematic reviews.
Step 11: Synthesize the Evidence
Depending on the heterogeneity and nature of included studies, choose an appropriate synthesis method:
- Meta-analysis: Quantitative pooling of effect estimates when studies are sufficiently similar
- Narrative synthesis: Structured textual summary when meta-analysis is not appropriate
- Vote counting based on direction of effect: When summary data are insufficient for meta-analysis
If conducting a meta-analysis, assess statistical heterogeneity using I² and Q statistics, consider subgroup analyses and sensitivity analyses, and evaluate publication bias using funnel plots and statistical tests.
Step 12: Assess Certainty of Evidence
Use the GRADE (Grading of Recommendations, Assessment, Development and Evaluations) framework or a similar approach to assess the overall certainty of evidence for each key outcome. GRADE considers risk of bias, inconsistency, indirectness, imprecision, and publication bias.
Step 13: Report Using PRISMA 2020
Report your systematic review following the PRISMA 2020 reporting items. Create your PRISMA flow diagram using our free PRISMA 2020 compliant flow diagram tool, which supports all four official templates and produces publication-ready PNG exports.
Common Pitfalls to Avoid
- Searching too few databases or using overly narrow search terms
- Failing to register a protocol before starting the review
- Using a single reviewer for screening or data extraction
- Not recording specific exclusion reasons at the full-text stage
- Using the outdated PRISMA 2009 flow diagram format
- Applying vote counting based on statistical significance rather than direction of effect
Frequently Asked Questions
How long does a systematic review take?
A systematic review typically takes 6-18 months, depending on the scope of the research question, number of databases searched, volume of records to screen, and team size. Simple reviews with few included studies may take less time, while complex reviews with thousands of records can take longer.
How many databases should I search?
There is no fixed minimum, but most systematic reviews search at least 2-3 major databases relevant to their field. Health-related reviews commonly search PubMed, Embase, and Cochrane CENTRAL as a minimum. Broader searching reduces the risk of missing relevant studies.
What is the difference between a systematic review and a literature review?
A systematic review follows a pre-specified protocol, uses comprehensive database searching, applies explicit eligibility criteria, and reports transparently using PRISMA guidelines. A narrative literature review is less structured, may not be comprehensive, and does not require reproducible methods.
Can one person conduct a systematic review?
While possible, it is not recommended. Best practice requires at least two independent reviewers for screening, data extraction, and quality assessment to reduce bias and errors. A single-author review should acknowledge this as a limitation.
Do I need to do a meta-analysis?
Not necessarily. A meta-analysis is a statistical synthesis that may be conducted within a systematic review when included studies are sufficiently similar in design, population, intervention, and outcomes. If studies are too heterogeneous, a narrative synthesis is more appropriate. Learn more about the differences between meta-analysis and systematic review.