A systematic review is not just a long literature review with more references. It is a structured, reproducible, and pre-registered investigation of every paper that addresses a specific question. The reproducibility is what gives it weight on a CV and in the evidence hierarchy. It is also what makes the process slow and exacting in ways that surprise medical students who agreed to "help out with a review" expecting two weekends of work.
This guide walks through the full pipeline as honestly as possible — what each step is, what it actually costs in time, and where most first reviews quietly go wrong.
Step 1: Frame the question in PICO (1 to 2 weeks)
Before anything else, your question has to be answerable. Vague questions like "Is exercise good for depression?" cannot be systematically reviewed because you cannot define what to include. Use the PICO framework — Population, Intervention, Comparator, Outcome — to force specificity. "In adults with major depressive disorder, does 12 weeks of structured aerobic exercise compared with usual care reduce Hamilton Depression scores at 12 weeks?" is something you can systematically review. The first one is not.
This step takes longer than people expect. You will revise the question two or three times once you start scoping the literature and realize your original framing was either too broad (50,000 hits) or too narrow (eight papers).
Step 2: Check that the review has not already been done (1 to 2 days)
This is the step that most students skip and most reviewers regret skipping. Before committing months of work, search PROSPERO (the international prospective register of systematic reviews) and the Cochrane Library to see whether your exact question is already registered, in progress, or recently published. If a Cochrane review on your topic was published in the last two years, your review is dead on arrival unless you can articulate a meaningfully different angle.
See our walkthrough on checking whether your idea has been done and why PROSPERO matters.
Step 3: Write and register a protocol (1 to 2 weeks)
Once your question survives the duplication check, write a protocol following the PRISMA-P 2015 checklist. The protocol locks in your inclusion criteria, search strategy, risk-of-bias tool, and analysis plan before you start screening. Locking this in advance is what prevents post-hoc rationalization — deciding to exclude papers because they did not fit the conclusion you wanted.
Register the protocol on PROSPERO. Registration is free, takes 24 to 72 hours to be reviewed, and gives you a CRD number that journals will ask for at submission.
Step 4: Build and run the search (1 to 2 weeks)
A real systematic review searches at least three databases. PubMed/MEDLINE, Embase (via Ovid), and Cochrane CENTRAL are the minimum. Add Web of Science, Scopus, and CINAHL if your topic spans biomedicine and another field. For each database, you write a search string from scratch — the syntax differs. See our piece on PubMed search strings for the foundation, and our database comparison for what each one covers.
Export results in a citation-manager-friendly format (RIS or NBIB) and dedupe in EndNote, Zotero, or directly in Rayyan or Covidence. A typical search yields between 800 and 8,000 records before deduplication.
Step 5: Title and abstract screening (3 to 6 weeks)
Two reviewers independently screen every title and abstract against the inclusion criteria. Disagreements are resolved by discussion or a third reviewer. Tools like Rayyan and Covidence automate the blinding and conflict-detection so two reviewers can work in parallel without seeing each other's calls.
Most students vastly underestimate this step. Screening 4,000 records at 30 seconds each is 33 hours per reviewer. With breaks, fatigue, and re-screening when criteria evolve, that is six weeks of evenings if you are a med student doing this around classes.
Step 6: Full-text review (2 to 4 weeks)
The 100 to 300 records that survive title/abstract screening get their full texts pulled and re-screened. Each exclusion needs a reason recorded (PRISMA flow diagrams require it). This is the step where you discover that the full text you needed is behind a paywall your institution does not cover, requiring an inter-library loan that takes three weeks.
Step 7: Data extraction (2 to 4 weeks)
For each included study, two reviewers independently extract pre-specified data fields into a standardized form (typically a spreadsheet or a Covidence template). Fields include population characteristics, intervention details, outcomes, effect sizes, and risk-of-bias judgments. Disagreements are resolved by consensus. Plan for 30 to 90 minutes per included study.
Step 8: Risk-of-bias assessment (1 to 2 weeks)
Use a tool matched to the study design. RoB 2 for randomized trials, ROBINS-I for non-randomized intervention studies, QUADAS-2 for diagnostic accuracy, AMSTAR-2 for overviews of reviews. Two reviewers, independent judgments, consensus on disagreements. Document everything.
Step 9: Synthesis (2 to 4 weeks)
If the included studies are clinically and methodologically similar enough, run a meta-analysis (typically random-effects, using R's metafor or RevMan). If they are too heterogeneous, do a narrative synthesis using SWiM (Synthesis Without Meta-analysis) guidance. Assess certainty of evidence with GRADE. Build a forest plot and a summary of findings table.
Step 10: Write up using PRISMA 2020 (3 to 6 weeks)
Use the PRISMA 2020 checklist to structure the manuscript. Include the full search strategy for every database in the supplementary material. Include the PRISMA flow diagram. Pre-specify any deviations from your protocol and explain them.
Realistic timeline
For a first-time systematic review with a small team (you plus one collaborator plus a supervisor), the honest end-to-end timeline is 6 to 12 months. Cochrane reviews routinely take 18 to 24 months. Reviews that publish in 3 months either had a very narrow topic, a very experienced team, or quietly cut corners. If you are budgeting time as a medical student, plan for at least 8 months and use any slack to write the manuscript carefully.
Where most first reviews go wrong
- Question keeps changing. Lock the protocol early. Every post-hoc change is a credibility hit.
- Single reviewer. Doing screening or extraction alone is technically not a systematic review — it is a one-person scoping exercise. Journals will catch this.
- No PROSPERO registration. Most journals now require it, and reviewers ask for the CRD number.
- Skipping risk-of-bias. A summary table is the bare minimum.
- Cherry-picking the discussion. If three trials say one thing and one says another, your synthesis has to reflect that mix, not the trial that confirms your prior.
A well-conducted systematic review is among the most valuable things a medical student can put on a CV — first authorship on a published SR is a serious credential. But it earns that weight because it is hard, slow, and structurally honest. Treat it accordingly.