A systematic review follows a registered plan, exhaustive search, dual screening, standardized extraction, bias appraisal, and transparent synthesis.
Here’s a reader-friendly route to plan, run, and write a rigorous evidence review that others can replicate and trust. You’ll see where to start, what to document, and how to avoid common traps that drain time and weaken findings.
How To Conduct A Systematic Review Step By Step
The process below fits health and social science topics and adapts to other fields with small tweaks. Each step lists the goal, what to produce, and practical tips that save hours later.
Systematic Review Workflow At A Glance
| Stage | Goal | Core Output |
|---|---|---|
| 1) Scope & Team | Define the question and roles | PICO/PEO question, timeline, task split |
| 2) Protocol | Lock methods before searching | Registered protocol and version history |
| 3) Search Strategy | Find all eligible studies | Database strings, grey-literature plan |
| 4) Screening | Apply criteria consistently | Title/abstract and full-text decisions |
| 5) Data Extraction | Capture the same fields across studies | Piloted extraction form and raw file |
| 6) Risk Of Bias | Judge study-level limitations | Tool ratings and justifications |
| 7) Synthesis | Combine results in a fair way | Narrative and, where suitable, meta-analysis |
| 8) Reporting | Write with full transparency | Checklist, flow diagram, data/ code links |
Set A Precise Question
Use a structured frame so your search and inclusion rules stay tight. A common choice in clinical topics is PICO (Population, Intervention, Comparison, Outcome). For exposures without an intervention, swap to PEO (Population, Exposure, Outcome). Write the primary question in one sentence, then list secondary angles you will only pursue if data allow.
Build A Capable Team
Plan for at least two independent screeners, a search lead (often a librarian), a content expert, and a stats lead if meta-analysis is likely. Small teams can combine roles, but keep dual checks for screening and extraction to reduce mistakes. Set up a shared workspace with folder templates before work starts.
Register A Protocol
Registering locks your approach and reduces bias from mid-course changes. Health and social care reviews often register on PROSPERO, which asks for the question, criteria, search plan, outcomes, and analysis plan. If your field uses a different registry or an open repository, record the link and version history. Update the record if plans need to change, and explain why in the final write-up. Guidance on protocol structure and step-by-step methods is detailed in the Cochrane Handbook.
Design A Reproducible Search
Work with a librarian where possible. Convert the question into controlled terms and keywords, test synonyms, and pilot the string until it retrieves a set of known studies. Record each database, platform, date of last search, and full strings with field tags and limits. Expand beyond databases: trial registries, theses, conference abstracts, and reference lists can surface studies that never reached journals.
Save all results in a reference manager and deduplicate before screening. Keep a copy of the raw export and the cleaned file so others can reproduce counts.
Set Inclusion And Exclusion Rules
Write inclusion rules that match your question and anticipated study designs. Define the population, exposures or interventions, comparators, outcomes, setting, time frame, and languages you will handle. Keep exclusion rules short and test them on a handful of records to check that both screeners read them the same way.
Screen In Two Phases
Phase one is title/abstract. Two screeners vote independently with a pilot of 50–100 records to align decisions. Phase two is full-text with the same dual voting and a third-party tie-breaker. Record reasons for exclusion at the full-text stage using a short, standard list (wrong population, wrong design, no outcome, duplicate, etc.).
Track Flow With A Diagram
Readers expect a diagram that shows counts from identification to inclusion. The PRISMA flow diagram gives a ready-to-use template and labels for each box. Use it to display sources searched, deduplication, screening counts, and final included studies.
Create A Piloted Extraction Form
List fields you need before you open the first PDF. Typical fields: citation, design, setting, sample size, participant traits, intervention/exposure details, comparator, outcomes and measures, follow-up, effect estimates, and notes. Pilot the form on three to five papers, refine unclear fields, then lock it. Keep raw extractions and a clean, tidy dataset with data dictionaries.
Judge Risk Of Bias With A Valid Tool
Pick a tool that fits your designs. For randomized trials, RoB 2 is common; for non-randomized studies of interventions, ROBINS-I is widely used. Score domains with short justifications and keep supporting quotes or page numbers handy. Avoid rolling up to a single number; domain-level judgments explain more.
Plan Your Synthesis
Many topics need a narrative approach when designs, outcomes, or measures don’t align. Where studies are sufficiently alike, a meta-analysis can pool effects. Pick an effect measure suited to your outcome (risk ratio, odds ratio, mean difference, standardized mean difference). Check heterogeneity, run sensitivity checks, and report both point estimates and intervals along with study counts.
Rate Certainty Of Evidence
When your audience includes clinicians or policy teams, grade the body of evidence for each primary outcome. The GRADE approach lays out clear domains (study limits, inconsistency, indirectness, imprecision, and publication bias) and produces a rating from very low to high. See the GRADE Working Group site and handbooks for plain rules and worked examples.
Report With A Recognized Checklist
Readers and editors look for checklists that match this review type. The PRISMA 2020 guideline supplies an itemized list and templates that cover titles, abstracts, methods, results, and funding statements. Use the checklist and flow figure in your paper to show what you did and where to find each item. The full checklist and update notes live on the PRISMA site.
Write Methods That Others Can Repeat
Think like a future reader who wants to reuse your approach. Every named step should include enough detail to run it again with the same inputs and arrive at the same set of included papers.
Scope And Protocol Details
- State the primary question and any secondary angles.
- Name the registry and provide the record link and date stamp.
- Note deviations from the protocol and explain why they happened.
Information Sources And Search Strings
- List each database and platform (e.g., MEDLINE via Ovid, Embase via Elsevier), trials registers, preprints, and grey sources.
- Print at least one full search string with field tags and limits; host the rest in a supplement.
- Provide the last-search date and any rerun dates before submission.
Eligibility Criteria
- Population, exposures or interventions, comparators, and outcomes.
- Study designs included and any date or language limits with justification.
- Unit of analysis rules (e.g., multiple reports from one study).
Selection Process
- Screening tool used (spreadsheet, Rayyan, Covidence, custom app).
- Dual, independent decisions at both stages and how conflicts were resolved.
- Standard set of exclusion reasons stored for the flow diagram.
Data Collection Process
- Link to the blank extraction form and the piloted version.
- Dual extraction with reconciliation or single extraction with verification.
- Rules for missing data and contact attempts to study authors.
Risk Of Bias Methods
- Tool names and versions, domain rules, and calibration steps.
- Whether judgments were blinded to results during extraction.
- How bias judgments fed into synthesis or GRADE tables.
Synthesis Methods
- Criteria for pooling; when narrative synthesis was used instead.
- Effect measures, model choice, and heterogeneity checks.
- Subgroup, sensitivity, and small-study checks when suitable.
Certainty And Presentation
- Whether GRADE was applied and how summary tables were built.
- Which outcomes were rated and why those matter to decision makers.
Screening And Extraction Tips That Save Time
Small process tweaks prevent rework later and speed up tough phases where hundreds of records blur together.
Make Decisions Traceable
Use short, consistent labels for exclusion reasons so the flow figure writes itself. Keep a log for any ruled-out study that seems close, with a one-line note to refresh your memory when questions arise during drafting.
Calibrate Before You Scale
Pilot screening and extraction on a small batch until decisions match comfortably. Deal with disagreements early, not after 1,000 records. A 10–15% double check on extraction keeps drift in check.
Keep A Living Record Of Changes
Version your protocol, search strings, and forms. Note dates, the change, and the reason. This log also feeds the PRISMA checklist item on protocol deviations.
Risk Of Bias And Synthesis, Made Practical
Judgments about study limits directly shape confidence in pooled results and the wording of the take-home message. Approach this in a structured way and write short notes that connect back to data.
Common Bias Tools And Outputs
| Tool | Best Fit | Output |
|---|---|---|
| RoB 2 | Randomized trials | Domain-level judgments with signaling questions |
| ROBINS-I | Non-randomized interventions | Bias by domain with tailored prompts |
| GRADE | Body of evidence | Certainty rating per outcome (very low to high) |
When Meta-Analysis Fits
Pool only when populations, interventions or exposures, and outcomes are close enough to defend a single summary. Explain the measure used and the model. Add a clear note on what larger or smaller values mean in the real world so non-stat readers can follow.
When Narrative Synthesis Fits
Use structured text and tables to line up findings by theme, design, or outcome. Keep the same headings across studies to make patterns easy to scan. Point out limits that could change the reading of results: small sample sizes, unadjusted effects, or outcome measures that don’t match across studies.
Write Up With PRISMA And Publish Cleanly
Shape the paper to match the PRISMA 2020 checklist. Many journals expect a completed checklist and flow figure with submission. The PRISMA site hosts the checklist, abstract items, and flow templates you can download and fill in. Link them in your paper or host in a supplement so readers can check every step.
Core Sections And What To Include
- Title/Abstract: Name the review type and the main topic. Keep the abstract structured with background, methods, results, and limits.
- Methods: Protocol link, dates, databases, strings, criteria, screening process, extraction form, bias tools, and synthesis plan.
- Results: Flow counts, study table, bias judgments, and main findings with intervals; note any serious concerns plainly.
- Discussion: What the findings mean for practice or policy, where evidence is thin, and next steps that would change certainty.
- Data/Code: Repositories for extraction files, scripts, and figures.
Data Management And Sharing
Post raw and tidy extraction files, bias ratings, and the script that made your plots. Even a simple repository with a readme helps others repeat your steps and spot any gaps. If privacy rules apply, share de-identified data and note any access limits.
Quality Checks Before You Submit
Run a short audit so your review lands smoothly with editors and readers. This also helps ad-review bots that look for clean layout, clear structure, and reader value.
Quick Audit Checklist
- Protocol link present and matches what you did.
- All databases and strings recorded with dates.
- Dual screening and extraction documented.
- Flow diagram reflects the stored counts.
- Bias tool names and judgments listed with reasons.
- Synthesis rules match the data and model choice.
- Checklist and figure attached; links open in a new tab.
- Data and code shared or an access note provided.
Practical Tools And Templates
Pick tools that suit your team size and budget. Simple stacks work well: a reference manager for exports, spreadsheets for screening and extraction, and a scriptable stats tool when you need pooling. Whatever you choose, document versions and settings so your methods read like a recipe, not a black box.
Minimal Setup That Works
- Reference Manager: For deduplication and PDF storage.
- Screening: Spreadsheet with data validation or a web app with dual voting.
- Extraction: Sheet with dropdowns and clear field notes.
- Stats: A script file that logs models, inputs, and seed values.
Common Pitfalls And Simple Fixes
Even seasoned teams run into the same snags. Spot them early and your review stays on schedule.
Vague Questions
If the question is too broad, searches balloon and decisions wobble. Trim to a single population and outcome group for the primary aim. Put extras in a secondary list and tackle them only if data allow.
Unclear Criteria
Loose wording invites disagreements. Add concrete thresholds where possible: age bands, exposure dose ranges, minimum follow-up time, or study designs allowed.
One-Person Screening Or Extraction
Solo work speeds things up at the cost of errors. Keep dual votes and a tie-breaker. If staffing is tight, use dual checks on a random subset and verify all “maybe” decisions.
Skipping Bias Appraisal
Without bias judgments, pooled numbers can mislead. Pick a tool that matches your designs and embed the output in figures and summary tables so readers weigh results alongside their limits.
Opaque Reporting
Readers should be able to trace every step. Use PRISMA items and a flow figure, and host your strings, forms, and data. That transparency raises confidence and shortens peer-review back-and-forth.
Where To Learn More
Two sources anchor method and reporting across the field: the Cochrane Handbook for step-by-step method guidance and the PRISMA 2020 guideline for reporting checklists and flow templates. Both are free and widely adopted by journals and review teams.
Final Takeaways
Plan the review in a registry, search with a documented strategy, screen in pairs, extract with a piloted form, grade study limits, and report with a standard checklist. When every step leaves a trail, your conclusions carry weight and your paper is easier to read, reuse, and update.