Writing a systematic review involves defining a focused research question, searching for all relevant studies using a reproducible strategy, screening and appraising those studies for quality, synthesizing the findings, and reporting everything transparently. The process typically takes 6 to 18 months depending on the topic’s breadth and your team size. Unlike a traditional literature review, a systematic review follows a strict, predefined methodology designed to minimize bias, and every decision you make along the way needs to be documented so another researcher could replicate your work.
Start With a Focused Research Question
Your research question drives every other decision in the review, from which databases you search to which studies you include. The most common way to structure the question is with the PICO framework: Population (who is affected), Intervention (what treatment or exposure you’re examining), Comparison (what the intervention is being measured against), and Outcome (what result you’re looking for). Not every question fits neatly into PICO, and that’s fine. Qualitative or public health reviews sometimes use variations like PEO (Population, Exposure, Outcome) or SPIDER. The goal is to break your question into clearly defined components so you can build precise search terms later.
Before committing to a question, search existing databases like the Cochrane Library and PROSPERO to check whether a recent systematic review already covers the same ground. If one exists from the last few years and the evidence base hasn’t changed significantly, you’ll need to refine your angle or choose a different question entirely.
Write and Register a Protocol
A protocol is your review’s blueprint. It locks in your methods before you start collecting data, which prevents you from unconsciously shifting your approach based on what the results look like. Your protocol should include the rationale for the review, your research question broken into its structured components, inclusion and exclusion criteria, your planned search strategy for both published and unpublished literature, your approach to data extraction and management, how you’ll assess risk of bias in individual studies, your plan for data synthesis, and how you’ll grade the overall strength of evidence for each question.
Once your protocol is written, register it with a review registry. PROSPERO is the most widely used registry for systematic reviews. Registration is free. A member of your team creates an account, then enters key information about the review design, methods, team members, funder, and timeline through an online form. One important detail: a PROSPERO record is not considered a full protocol on its own, but you can upload a PDF of the complete protocol as part of your registration. If you’re conducting a scoping review rather than a systematic review, PROSPERO won’t accept it. The Open Science Framework (OSF) is an alternative that accepts both systematic and scoping review protocols and also lets you share supporting documents like search strategies and data extraction forms. Protocols on OSF are not reviewed by an editorial board before acceptance. You can also publish your protocol as a standalone peer-reviewed paper through journals like those on BioMed Central.
Design a Comprehensive Search Strategy
The search strategy is what separates a systematic review from a regular literature review. You need to search multiple databases, not just one, and document every search string so it can be reproduced. For health sciences topics, that usually means PubMed/MEDLINE, Embase, and the Cochrane Central Register of Controlled Trials at minimum, but your field may have its own key databases.
Build your search strings from the components of your PICO question. For each concept (population, intervention, outcome), generate a list of synonyms, related terms, and controlled vocabulary terms specific to each database. Connect synonyms within a concept using OR, and connect different concepts using AND. For example, if your population is “adolescents,” you’d also include terms like “teenagers,” “youth,” and “young adults,” joined by OR. Then you’d combine that group with your intervention terms using AND.
Work with a librarian if you can. Most university libraries have research librarians experienced in systematic review searches, and their involvement strengthens the credibility of your strategy. You should also search for unpublished studies and grey literature (conference abstracts, dissertations, clinical trial registries, government reports) to reduce publication bias, which is the tendency for studies with positive results to be published more often than those with null findings. Finally, hand-search the reference lists of included studies and key review articles to catch anything your database searches missed.
Screen Studies in Two Stages
Screening happens in two rounds. First, you review titles and abstracts against your predefined inclusion and exclusion criteria. Studies that clearly don’t meet the criteria get excluded. Anything that might be relevant moves to the second round, where you read the full text and make a final decision.
At least two reviewers should independently screen each study at both stages. When the two reviewers disagree, a third reviewer or a discussion between the pair resolves the conflict. This dual-screening process is one of the key safeguards against bias in a systematic review.
Specialized software makes screening far more manageable. Covidence is one of the most popular tools, offering a streamlined interface for title/abstract screening and full-text review. Rayyan is a free alternative with useful ranking and sorting features, though it has a steeper learning curve. Subscription-based platforms like DistillerSR, Eppi-Reviewer, and PICO Portal offer additional functionality and sometimes provide trial access for a single project. You can technically manage references with citation managers like EndNote or Zotero for collecting and deduplicating records, but they’re significantly more cumbersome for the screening workflow itself.
Extract Data Systematically
Once you’ve identified your final set of included studies, you need to pull out the relevant data from each one using a standardized extraction form. Build the form before you start extracting, and pilot it on a few studies to make sure it captures everything you need without ambiguity. Common fields include study authors, publication year, study design, sample size, participant characteristics, intervention details, comparison details, outcome measures, results, and funding sources.
Two reviewers should independently extract data from each study, then compare their forms and resolve discrepancies. This sounds tedious, and it is, but it catches errors that a single extractor would miss. Record your extraction in a spreadsheet or within your review management software so the data feeds directly into your synthesis.
Assess Risk of Bias
Every included study needs to be evaluated for methodological quality. You’re not just asking “what did this study find?” but “how much should I trust the finding?” The tool you use depends on the study designs in your review.
For randomized controlled trials, the Cochrane Risk of Bias tool (RoB 2) is the standard. It evaluates each study across a fixed set of domains: bias from the randomization process (was allocation truly random and concealed?), bias from deviations from intended interventions (did participants or providers know who got what, and did that knowledge change behavior?), bias from missing outcome data (did enough participants complete the study?), bias in measurement of the outcome, and bias in selection of the reported result. For each domain, you answer a series of signaling questions, and an algorithm proposes a judgment of “low risk of bias,” “some concerns,” or “high risk of bias.” A study is rated low risk overall only if it scores low across every domain. A single high-risk domain, or concerns across multiple domains that collectively undermine confidence, triggers a high-risk rating for the study as a whole.
For observational studies, tools like the Newcastle-Ottawa Scale or ROBINS-I serve a similar function. For diagnostic accuracy studies, there’s QUADAS-2. Choose the tool that matches your study designs and report which one you used.
Synthesize the Evidence
Synthesis is where you bring the findings together. If your included studies are sufficiently similar in design, population, intervention, and outcome measurement, you can perform a meta-analysis, which is a statistical method that pools results across studies to produce a single summary estimate. Meta-analysis also lets you quantify heterogeneity (how much the results vary across studies) and explore potential reasons for that variation through subgroup analyses or meta-regression.
If the studies are too different from one another to pool statistically, you’ll do a narrative synthesis instead. This means organizing the findings thematically, describing patterns and discrepancies across studies, and explaining how study quality and design differences might account for conflicting results. Narrative synthesis still needs structure. Present findings in tables, group studies by relevant characteristics, and clearly describe the direction and consistency of results.
Regardless of approach, assess the overall certainty of the evidence using a grading framework. GRADE (Grading of Recommendations, Assessment, Development, and Evaluations) is the most widely used system. It rates the body of evidence for each outcome as high, moderate, low, or very low based on factors like risk of bias, inconsistency, imprecision, indirectness, and publication bias.
Report Using PRISMA Guidelines
The PRISMA 2020 statement is the accepted standard for reporting systematic reviews. It consists of a 27-item checklist covering everything from the title and abstract through the methods, results, and discussion. The checklist ensures you provide a transparent, complete account of why you did the review, exactly what you did, and what you found.
PRISMA also includes a flow diagram that visually maps how many records you identified, how many duplicates you removed, how many you screened, how many you excluded at each stage (and why), and how many made it into the final synthesis. Journals that publish systematic reviews expect this diagram, and reviewers will check your manuscript against the PRISMA checklist item by item.
An expanded version of the checklist provides detailed reporting recommendations for each item, and there’s a separate abstract checklist to guide how you summarize the review. Using these tools from the start of your writing process, rather than trying to retrofit your manuscript later, saves significant revision time.
Build the Right Team
A systematic review is not a solo project. At minimum, you need two reviewers for screening, extraction, and quality assessment. Most teams also benefit from a subject matter expert who understands the clinical or theoretical landscape, a methodologist who can guide the statistical analysis, and a research librarian who can design and validate the search strategy. For your first review, consider partnering with someone who has published one before. The learning curve is steep, and an experienced collaborator will help you avoid methodological missteps that could surface during peer review.