Systematic Reviews: A Step-by-Step Guide

Systematic reviews sit at the top of the evidence hierarchy, and for good reason. They synthesize all available research on a defined question using rigorous, reproducible methods. Whether you are a graduate student conducting your first review or a senior researcher expanding into evidence synthesis, a clear process makes the difference between a publishable contribution and months of wasted effort.

This guide breaks the process into concrete phases you can follow from start to finish.

Phase 1: Define Your Question

Every strong systematic review begins with a tightly scoped research question. The most widely used framework is PICO: Population, Intervention, Comparison, and Outcome. PICO forces specificity. Instead of asking whether exercise helps mental health, you ask whether aerobic exercise (Intervention) reduces depressive symptoms (Outcome) in adults with clinical depression (Population) compared to standard pharmacotherapy (Comparison).

Spending extra time refining PICO elements at the outset prevents scope creep later. If your question is too broad, your search will return thousands of irrelevant records. Too narrow, and you may not find enough studies to synthesize.

Phase 2: Register Your Protocol

Protocol registration is what separates a systematic review from a literature review. Registering your protocol on a platform such as PROSPERO documents your methods before you begin, reducing the risk of post-hoc bias and increasing transparency for readers and peer reviewers.

Your protocol should specify your eligibility criteria, search strategy, databases to be searched, screening process, data extraction plan, risk-of-bias assessment tool, and synthesis method. Think of it as a contract with the research community: here is what we will do, and here is how we will do it.

Phase 3: Build Your Search Strategy

The search strategy is the backbone of any systematic review. It must be comprehensive enough to capture all relevant studies while remaining manageable.

1. Identify key concepts from your PICO question and generate synonyms, related terms, and controlled vocabulary (MeSH terms for MEDLINE, Emtree for Embase).
2. Combine terms with Boolean operators. Use OR within concept blocks and AND between them.
3. Search multiple databases. At minimum, most reviews require MEDLINE, Embase, and the Cochrane Library. Discipline-specific databases (PsycINFO, CINAHL, Web of Science) may also be necessary.
4. Supplement with grey literature. Conference proceedings, dissertations, trial registries, and preprint servers reduce publication bias.
5. Document everything. Record exact search strings, database interfaces, date ranges, and result counts. PRISMA 2020 requires this level of detail.

Consulting a research librarian at this stage is strongly recommended. Librarians specialize in optimizing search sensitivity and precision, and many journals now require librarian involvement as a quality indicator.

Phase 4: Screen and Select Studies

Import all records into a reference manager or dedicated screening tool and remove duplicates. Then screen in two stages:

• Title and abstract screening: Apply your eligibility criteria broadly. When in doubt, include the record for full-text review.
• Full-text screening: Retrieve the full texts of remaining records and apply your criteria strictly. Record reasons for exclusion at this stage.

Both stages should involve at least two independent reviewers. Disagreements are resolved through discussion or a third reviewer. Inter-rater reliability metrics such as Cohen's kappa add credibility to your process.

Phase 5: Extract Data and Assess Risk of Bias

Design a standardized extraction form covering study characteristics, participant demographics, intervention details, outcomes, and results. Pilot the form on three to five studies and revise as needed before extracting data from the full set.

Simultaneously, assess each included study for risk of bias using a validated tool. The Cochrane Risk of Bias tool (RoB 2) is standard for randomized trials. For observational studies, the Newcastle-Ottawa Scale or ROBINS-I are common choices. Risk-of-bias assessments should also be performed independently by two reviewers.

Phase 6: Synthesize and Report

If studies are sufficiently homogeneous in design, population, and outcomes, a meta-analysis quantifies the pooled effect using statistical models. When clinical or methodological heterogeneity is too high, a narrative synthesis organized by outcome or subgroup is appropriate. Either way, present your findings alongside the certainty of the evidence using a framework such as GRADE.

Report your review following PRISMA 2020 guidelines. The PRISMA checklist contains 27 items covering everything from the title and abstract to funding disclosures. The PRISMA flow diagram visually tracks the number of records identified, screened, excluded, and included. Journals increasingly require both as a condition of submission.

A Checklist You Can Use Today

Before you begin, confirm that you have completed each of these preparatory steps:

1. Formulated a focused PICO question
2. Searched PROSPERO to confirm no duplicate review is registered
3. Registered your own protocol
4. Built and peer-reviewed your search strategy with a librarian
5. Selected screening software and assembled your review team
6. Chosen your risk-of-bias tool and synthesis method
7. Downloaded the PRISMA 2020 checklist and flow diagram template

Systematic reviews are demanding, but they produce some of the most cited and influential work in any field. A disciplined, transparent approach from day one is the foundation of a review that stands up to scrutiny and contributes lasting value to the evidence base.