Topic selection determines whether your systematic review takes six months or two years, and whether it gets published in a good journal or sits on your hard drive forever.
Check for existing reviews first. Before committing to a topic, search the Cochrane Library, PROSPERO (the international prospective register of systematic reviews), and PubMed for existing systematic reviews on your question. If a high-quality Cochrane review was published in the last two years on your exact question, choose a different topic. If the last review is three or more years old, or if it was poorly conducted, an updated or improved review is justified.
Use the PICO framework to structure your question. Every systematic review starts with a well-defined research question. The PICO framework (Population, Intervention, Comparison, Outcome) provides the structure. A vague question like "does exercise help depression?" is unsearchable. A PICO-structured question like "In adults with major depressive disorder (P), does aerobic exercise (I) compared to pharmacotherapy alone (C) reduce depression severity scores (O)?" is precise enough to build a search strategy around. Use Research Gold's free free PICO question generator to structure your question before meeting with your mentor.
Choose a topic with 15 to 40 eligible studies. This is the sweet spot. Fewer than 10 studies and you may not have enough data for a meaningful synthesis (and definitely not enough for a meta-analysis). More than 50 studies and the screening and extraction workload becomes unmanageable for a student team. Check the approximate number of potentially eligible studies by running a quick PubMed search with your main terms before committing.
Target a gap in the literature. The most publishable systematic reviews address questions where multiple primary studies exist but no one has systematically synthesized them. Look for topics where clinical practice varies because the evidence has not been aggregated, where recent trials have changed the landscape since the last review, or where existing reviews are methodologically weak. Your faculty mentor can help you identify these gaps within their specialty.
Register your protocol on PROSPERO. Once your team agrees on the research question, register your protocol on PROSPERO before starting the search. Registration is free, takes about an hour, and serves two purposes: it prevents duplication (other teams can see your planned review) and it demonstrates methodological rigor to journal editors and peer reviewers. Many journals now require PROSPERO registration for systematic review submissions.
Once you have your team, mentor, and registered protocol, the actual work follows a structured sequence. Here is the workflow, with specific tools and tips for each stage.
Step 1: Develop Your Search Strategy
Your search strategy is the foundation of the entire review. A poorly designed search misses relevant studies or retrieves thousands of irrelevant results. Work with your medical librarian to develop a comprehensive strategy using Boolean operators (AND, OR, NOT), MeSH terms, and free-text keywords.
Build your strategy for PubMed first, then translate it to Embase, CINAHL, and the Cochrane Central Register of Controlled Trials. Most systematic reviews search at least three databases. Use Research Gold's free free search strategy tool to construct and document your Boolean search strings, and refer to our systematic review search guide for detailed instructions on database-specific syntax.
Step 2: Screen Titles and Abstracts
Import all search results into a reference manager (Zotero or Mendeley are free) and deduplicate. Then transfer the deduplicated set to a screening tool. Rayyan is free and designed specifically for systematic review screening. Covidence is the gold standard but requires a paid subscription (some medical schools provide institutional access).
Two reviewers independently screen every title and abstract against your pre-defined inclusion and exclusion criteria. Use Research Gold's free inclusion and exclusion criteria tool to document your criteria clearly before you start. Disagreements are resolved by discussion or by a third reviewer (typically your faculty mentor). Track your inter-rater agreement; a Cohen's kappa above 0.80 indicates excellent agreement.
Step 3: Full-Text Review
Retrieve full texts of all articles that passed title and abstract screening. Both reviewers read each full text independently and apply the inclusion criteria again. Record the reason for excluding each article at this stage, as PRISMA 2020 (Page et al., 2021) requires you to report the number of articles excluded at full-text screening and the reasons for exclusion.
Generate your PRISMA flow diagram using Research Gold's free free PRISMA flow tool. This tool produces a publication-ready diagram that tracks the number of records at each stage of screening.
Create a standardized data extraction form before you begin. The form should capture study characteristics (author, year, country, design), participant demographics, intervention details, comparator details, outcomes, and results. Use Research Gold's free data extraction template builder to generate a customized form for your review, and read our data extraction guide for best practices.
Both reviewers extract data independently from each included study. Compare extractions and resolve discrepancies. This dual extraction process is tedious but essential for accuracy. A single extraction error can invalidate your entire analysis.
Step 5: Assess Risk of Bias
Use a validated risk of bias tool appropriate for your study designs. For randomized controlled trials, use the Cochrane Risk of Bias tool (RoB 2). For observational studies, use the Newcastle-Ottawa Scale or ROBINS-I. Two reviewers independently assess each study, and disagreements are resolved by consensus or a third reviewer.
Step 6: Synthesize Your Findings
If your included studies are sufficiently similar in population, intervention, comparator, and outcome, you can conduct a meta-analysis to pool their results statistically. This is the most technically demanding step and where most medical students need help. A meta-analysis requires selecting the appropriate effect measure (odds ratio, risk ratio, mean difference, standardized mean difference), choosing a statistical model (fixed-effect vs. random-effects), assessing heterogeneity (I-squared, Cochran's Q), and testing for publication bias (funnel plots, Egger's test).
If a meta-analysis is not appropriate (because studies are too heterogeneous), you conduct a narrative synthesis, organizing your findings thematically and describing patterns across studies without pooling results statistically.
For a detailed walkthrough of the entire process, see our comprehensive guide on how to write a systematic review step by step.