How to Structure a Systematic Literature Review

By Kevin Fernando · Sep 17, 2026 · 14 min read

A systematic literature review is structured around a specific research question and documented eligibility criteria. You search multiple databases using a recorded protocol, screen results through transparent stages, extract data consistently, assess study quality, and synthesize findings while reporting your methods completely so others can audit or replicate your work.

A systematic literature review is not a narrative survey of what you already know about a topic. It is a documented workflow designed to find, evaluate, and synthesize all relevant studies on a specific question in a way that another researcher could audit or replicate. That reproducibility is what makes it systematic.

The structure matters because shortcuts that feel efficient, searching one database instead of three, screening studies alone instead of with a partner, skipping the risk of bias step, undermine the core purpose. A systematic review answers the question: what does the totality of evidence show, and how certain should we be? This article walks you through the stages and the checkpoints that distinguish a systematic review from a more casual survey.

What you'll be able to do, and what you need first

After completing a systematic review, you will have a documented, reproducible account of what published evidence shows on a specific question. You will know which studies exist, how their quality varies, where they agree and where they conflict, and how confident the evidence base is overall.

To start, you need access to academic databases (your institution's library typically provides these), time to work through multiple stages, and ideally a partner or team. Many systematic reviews take months to complete. You also need to decide whether a systematic review is the right approach for your question. If you are exploring whether a topic has been studied at all, or mapping the landscape of research without synthesizing findings, a scoping review may be more appropriate. If your studies are qualitative rather than quantitative, the principles of systematic review still apply, though your synthesis method will differ.

Define your research question and eligibility criteria

Before you search, you must define exactly what you are looking for. This boundary work prevents two failures: a search so broad it returns thousands of irrelevant results, and a search so vague that you unconsciously shape it to find studies you expect. Your question and eligibility criteria are your defense against that drift.

Write a specific, answerable question

Start with a broad topic, then narrow it to a question you can actually search. A broad topic is "Does exercise help depression?" A searchable question is "In adults diagnosed with major depressive disorder, does aerobic exercise compared to usual care reduce depressive symptoms as measured by the PHQ-9 scale?"

The difference is specificity. The narrow question tells you what population to include (adults with major depressive disorder, not adolescents or people with subsyndromal symptoms), what intervention to search for (aerobic exercise, not all physical activity), what comparison to use (usual care, not other exercises), and what outcome matters (PHQ-9 scores, not self-reported mood). Each of these choices will become an inclusion criterion.

Write your question down. It should be a single sentence. If you cannot fit it in one sentence, it is still too broad or you are trying to answer multiple questions at once.

Set inclusion and exclusion criteria

For each element of your question, write down which studies you will include and which you will exclude. Make these rules explicit and specific enough that another person applying them would reach the same decision most of the time.

For population: Will you include only adults, or also adolescents? Will you include people with depression diagnosed by a clinician, or also people who score above a threshold on a screening tool? Will you exclude people with bipolar disorder, or include them?

For intervention: Will you include only aerobic exercise, or also resistance training? Will you exclude studies where exercise was one component of a larger program? Will you set a minimum duration or frequency?

For comparison: Will you include studies comparing exercise to no treatment, to usual care, to other treatments, or all of these?

For outcomes: Will you require that studies measure depression using a validated scale, or will you accept self-reported improvement? Will you include studies that measured depression as a secondary outcome, or only primary outcomes?

For study type: Will you include only randomized controlled trials, or also observational studies? Will you exclude studies published only as conference abstracts, or include them?

For language and date: Will you limit your search to English-language publications, or include translations? Will you search back to when research on this topic began, or start from a specific year?

Document all of these decisions before you begin searching. They are your inclusion and exclusion criteria. When you encounter a borderline study during screening, these criteria tell you whether it belongs in your review.

Develop and document your search strategy

Your search must be comprehensive enough to find the relevant studies that exist, transparent enough that someone else could repeat it, and documented enough that you can remember what you did six months later. A search that lives only in your head is not reproducible.

Choose your databases and search terms

Select at least three databases. For medical topics, a combination of PubMed, Embase, Web of Science, and Google Scholar has been shown to provide adequate coverage. For other fields, choose a mix: one large multidisciplinary database (such as Google Scholar or a general academic database your library provides) and two subject-focused ones (PsycINFO for psychology, ERIC for education, EconLit for economics, and so on).

Searching only MEDLINE, even though it indexes abstracts from more than 4,500 journals, is not considered adequate. Depending on your field and question, MEDLINE alone captures only 30% to 80% of known published randomized trials. The gap varies by topic, so you cannot know in advance whether your question is in that 30% or that 80%. Search multiple databases to close the gap.

For each database, construct a search string using the key concepts from your research question. Use the database's controlled vocabulary (MeSH terms in PubMed, for example) where available, and combine keywords with Boolean operators (AND, OR, NOT). For the depression and exercise example, you might search: (depression OR depressive disorder) AND (exercise OR aerobic activity OR physical activity) AND (randomized controlled trial OR RCT). Test your search string in each database and adjust it to match that database's syntax and vocabulary. A search that works perfectly in PubMed may need modification for Web of Science.

Cast the net wide enough to capture relevant studies, but not so wide that you are buried in noise. If your search returns 50,000 results, it is too broad. If it returns fewer than 20, you may have missed relevant studies. Iterate until you find a balance.

Record your search protocol

Write down the exact search string you used in each database, the date you searched, and the number of results returned. Include the database platform and version if relevant (PubMed via NCBI, for example). Save a copy of your search history from each database if the platform allows it.

Also record any limits you applied: date range, language, study type, publication status (peer-reviewed only, or including preprints). If you hand-searched journals or contacted authors for unpublished data, document that too.

This documentation serves two purposes. First, it lets you verify later that you searched consistently across all your databases. Second, it allows someone reading your review to understand exactly what you looked for and to replicate your search if needed. When you write up your methods, this protocol becomes your search strategy section.

Screen studies and extract data

Your search will return hundreds or thousands of records. You now filter those down to the studies that actually belong in your review, pull the relevant information from each one, and organize it for synthesis.

Screen titles and abstracts

Import all your search results into a screening tool (Covidence, DistillerSR, Rayyan, or a spreadsheet if your review is small). Remove obvious duplicates. You now have a list to screen.

Read the title and abstract of each record and decide: does this study match my inclusion criteria? You are not making a final decision yet. You are filtering out studies that clearly do not belong. If the abstract does not mention your population, intervention, and outcome, exclude it. If the study is a review article rather than a primary study, exclude it. If it is in a language you cannot read, exclude it.

When in doubt, include it. A study that looks borderline at the abstract stage should move forward to full-text review, where you can read the methods and make a final decision.

Have at least two reviewers screen titles and abstracts independently. Each reviewer makes their own decision for each record without seeing the other's choice. Then compare. If both reviewers excluded a study, it is out. If both included it, it moves forward. If they disagreed, discuss the study and apply your inclusion criteria to reach a final decision. Document how many studies you included, excluded, and disagreed on at this stage. That number tells you whether your inclusion criteria are clear (high agreement) or ambiguous (frequent disagreement).

Review full texts and resolve disagreements

For each study that passed the title and abstract screen, obtain the full text. Read the methods section carefully. Does the study actually meet your inclusion criteria, or did the abstract mislead you?

Common reasons to exclude at this stage: the study used a different population than you specified, the intervention was not what you were looking for, the study did not measure your outcome, or the study design does not match your criteria (for example, you wanted randomized trials and this is an observational study).

Again, have at least two reviewers review full texts independently and compare their decisions. When reviewers disagree, discuss the study together and refer back to your inclusion criteria. If the criteria do not resolve the disagreement, the criteria were not specific enough. Revise them, document the revision, and apply the new rule consistently to all remaining studies.

At the end of full-text review, you have a final list of included studies. Document how many studies you reviewed at full text and how many you excluded at this stage, along with the reasons for exclusion (wrong population, wrong intervention, wrong outcome, wrong study design, and so on). This accounting is part of your methods and shows readers that you applied consistent rules.

Extract data from included studies

For each included study, pull out the information you need to answer your research question and to assess study quality. Data extraction may include title, author, year, journal, research question, study methods or study type, participant characteristics, intervention details, outcomes measured, and results. You may also extract identification features such as funding sources and publication type, along with study quality factors.

Create a data extraction form (a template with fields for each piece of information you need). Pilot the form on a few studies to make sure it captures what you need and that the fields are clear. Revise it if necessary, then use the final version consistently for all studies.

Have at least two reviewers extract data from each study independently. Compare their extractions. If they agree, move forward. If they disagree, discuss the study, check the original text, and reach agreement. Document any disagreements and how you resolved them. This process reduces errors and catches misreadings.

Organize your extracted data in a table or database so you can sort and compare across studies. You will use this data in your synthesis stage.

Assess the quality and risk of bias in included studies

Not all included studies are equally reliable. A study with poor randomization, high dropout rates, or selective outcome reporting is more likely to give you a biased answer than a well-designed study. Assessing risk of bias lets you weight the evidence appropriately and communicate to readers which findings rest on solid ground and which rest on shakier evidence.

Choose a risk of bias tool that matches your study types. For randomized controlled trials, use the Cochrane Risk of Bias tool (RoB 2), which assesses domains such as selection, performance, detection, attrition, and reporting biases. For non-randomized studies of interventions, use ROBINS-I. For diagnostic accuracy studies, use QUADAS-2. For randomized trials, the Jadad scale focuses on randomization, allocation concealment, and attrition.

Read each study's methods section and rate the risk of bias in each domain. Most tools ask you to rate each domain as low risk, high risk, or unclear risk. Document your reasoning for each rating. Have at least two reviewers assess risk of bias independently and compare their ratings. Disagreements often signal that the study report is ambiguous; discuss and reach agreement.

At the end, you have a risk of bias profile for each study. This profile informs your synthesis: studies with low risk of bias across most domains carry more weight in your conclusions than studies with high risk of bias.

Synthesize findings and present results

You now have a set of included studies, each with extracted data and a risk of bias assessment. Your task is to combine what they show into a coherent answer to your research question.

Decide whether to pool data statistically

You have two broad options: meta-analysis (statistical pooling) or narrative synthesis (summary and discussion).

Meta-analysis combines numerical results from multiple studies into a single pooled estimate. It works when the studies are similar enough that pooling makes sense. For a successful meta-analysis, the characteristics being studied, such as population, intervention, and comparison, must be sufficiently similar, and the same outcome or measure must be calculated the same way at the same time intervals. If the studies differ widely in who they enrolled, what intervention they tested, or how they measured the outcome, pooling their numbers produces a meaningless average.

Meta-analysis is a statistical method for combining the findings from quantitative studies. It evaluates, synthesizes, and summarizes results. But it requires that the underlying studies be comparable.

If your studies are too diverse to pool, or if some studies are qualitative, use narrative synthesis. Organize your findings by theme or outcome, describe what each study found, discuss areas of agreement and disagreement, and explain what the pattern of evidence suggests. Narrative synthesis is not a weaker alternative; it is the appropriate choice when pooling would obscure more than it would clarify.

Present findings with certainty of evidence

Do not simply report what the studies found. Report what the studies found and how confident you should be in that finding. Use the GRADE approach to assign a certainty rating to each key finding.

GRADE uses a structured approach to assign a rating of high, moderate, low, or very low certainty to the evidence. Randomized controlled trials initially start at a high level of certainty while non-randomized studies traditionally start at low level of certainty. Then you adjust based on factors affecting certainty: study design, risk of bias, inconsistency, indirectness, imprecision, and publication bias.

For example, suppose five randomized trials tested your intervention and four showed benefit while one showed no effect. The study design is strong (randomized trials), but there is inconsistency across studies. You might rate the certainty as moderate rather than high. If the trials enrolled only men, but your question asks about all adults, you might downgrade further for indirectness. If the trials were small and the confidence intervals were wide, you might downgrade for imprecision.

Conversely, if observational studies consistently showed a large effect, you might upgrade the certainty from low to moderate.

The GRADE approach provides a framework for explicit consideration of the limitations of evidence included in a systematic review and for incorporating this assessment into the conclusions. When you report your findings, state not just what the evidence shows, but also the certainty rating. "Exercise reduced depressive symptoms (moderate certainty evidence)" tells readers something very different from "Exercise reduced depressive symptoms (very low certainty evidence)." The first suggests you can act on the finding; the second suggests you need more research.

Document your methods and report transparently

Write up your methods in detail: your research question, your inclusion and exclusion criteria, your search strategy (databases, search terms, dates), how you screened studies, how you extracted data, how you assessed risk of bias, and how you synthesized findings. Include the numbers at each stage (how many records you retrieved, how many you screened, how many you included, how many you excluded and why).

Use PRISMA 2020 as your reporting guideline. PRISMA is designed to improve the reporting of systematic reviews. The main PRISMA reporting guideline primarily provides guidance for the reporting of systematic reviews evaluating the effects of interventions. PRISMA 2020 is complemented by various PRISMA extensions, which provide guidance for the reporting of different types or aspects of systematic reviews and other types of evidence synthesis.

Transparency serves two purposes. First, it allows readers to understand your methods and judge the quality of your review. Second, it allows future researchers to replicate your search, update your review, or apply your methods to a different question. A review that hides its decisions is a review that cannot be audited or improved.

Recognize when a review has gone wrong

Several common failures undermine the systematic nature of a review. Watch for these.

Searching only one database or using a search string you did not record. You cannot verify that you were systematic, and no one else can repeat your work.

Screening studies alone, without a second reviewer. You cannot catch your own biases or errors. Disagreement between reviewers is not a problem to avoid; it is a feature that catches problems.

Changing your inclusion criteria mid-review based on which studies you are finding. This is the most insidious failure because it feels like refinement. It is actually bias. If you need to revise your criteria, do so explicitly, document the change, and apply the new rule to all studies, not just the ones you have not screened yet.

Including studies without assessing risk of bias, or assessing bias but then treating all studies as equally reliable in your synthesis. A study with high risk of bias can still be included; you simply weight it appropriately in your conclusions.

Pooling studies that are too different to pool. If your studies differ in population, intervention, outcome measurement, or follow-up time, averaging their results produces a number that answers no one's question.

Reporting findings without certainty ratings. "The evidence shows X" is incomplete. "The evidence shows X with moderate certainty" tells readers whether to act on it.

Not documenting your methods clearly enough that someone else could audit or replicate your work. If your review lives only in your head, it is not systematic, and it is not reproducible.

Do this automatically

A researcher manually executing this workflow, defining eligibility criteria, searching multiple databases with documented terms, screening thousands of abstracts, extracting data from included studies, and assessing risk of bias across each one, faces months of labor that scales with the breadth of the literature. Software can automate the repetitive screening, searching, and documentation steps that consume the bulk of this time while maintaining the transparency and reproducibility that systematic reviews demand. Academia research surveys the literature, extracts the findings and methods described above, and attaches a citation to every claim it reports, across as many subjects as you need, with Mimir attaching a citation to every figure. It will not write your paper for you. It gathers and cites the evidence; the argument stays yours, and deliberately so. Every plan includes it, including the free one.

Start your first report and run it on something you already know well.

This article is general information and research commentary. It is not professional advice, and it does not account for your particular circumstances.