What Is a Systematic Review? When to Use It and How It Works

A systematic review is research about existing research. The team decides which studies could answer its question, then uses documented methods to find and assess them. Cochrane’s introduction describes the effort to assemble eligible evidence while minimizing bias. No search can guarantee that every relevant study has been found.

Consider two teams reading the same fifty articles. One selects examples that fit its argument. The other applies prespecified eligibility rules, records exclusions, assesses limitations and synthesizes findings using an explicit method. The difference lies in the decisions and how they can be examined, not the length of the bibliography.

That evidence trail should connect each conclusion to the studies behind it. A reader needs to understand why a study was eligible, how several reports of it were linked and where an extracted result came from. Keeping those records makes judgment open to scrutiny; it does not make every decision mechanical or guarantee that another team will agree.

“Literature review” is a broad label. It can mean a discussion in a thesis, a critical account of a field or a review using systematic methods. Read the purpose and methods rather than relying on the label alone. A narrative review can develop a useful argument without claiming to represent every eligible study.

A scoping review maps the extent, characteristics or concepts of an evidence base. It has its own methods and is not a lower-quality version of an effects review; see JBI’s scoping-review guidance. Meta-analysis occupies a different level: it is a statistical technique that may be used within a systematic review. It does not perform the search or appraisal on its own.

Start with the decision the result should inform. In a fictional training example, a curriculum team wants to know whether a specified program, compared with usual training, changes assessment scores among first-year trainees at 12 weeks. It needs an estimate of the difference and its uncertainty, together with an appraisal of the supporting evidence. Population, comparison, outcome, time point and eligible study designs define the question before results influence the choices.

Now change the question to “How have digital training programs been studied across professions?” The desired output is a map of populations, designs and research topics, so a scoping review may fit better. Questions about experience, mechanisms or implementation may call for qualitative or mixed-methods synthesis. Do not force every question into an intervention-effect framework.

Before starting a new review, look for existing reviews and protocols. Would new studies, a changed setting or limitations in earlier methods justify an update? A different title alone is not a reason to repeat the work. Write down who will use the answer and what remains uncertain; that statement helps determine both the scope and the expertise you need.

For the training question, a protocol would specify eligibility, information sources, selection, data collection, appraisal and synthesis. Record later changes with dates and reasons. A study can be eligible for the review without supplying the data for every planned synthesis, a distinction explained in Cochrane’s eligibility guidance.

Search sources appropriate to the question and preserve the strategies, dates and exports. Databases, registers and citation searching may each contribute evidence; a search-engine ranking is not a complete sampling frame. An information specialist can help design the search. Cochrane’s search guidance explains the methodological choices.

The fictional example below then follows four reports: a registry record, main paper and follow-up report from Study A, and one paper from Study B. They describe two studies, not four independent investigations. Link them before collecting results so that a later report can add information without double-counting the participants.

After selection, define the extraction fields and use an appraisal method appropriate to the designs and results. The table shows what each stage leaves behind for this example; it contains no observed training results. These records also support the final PRISMA report, whose job is to communicate the work that actually took place.

A systematic review does not have to calculate a pooled effect. Pooling may be unsuitable when studies answer substantially different questions, measure incompatible outcomes or lack usable estimates and uncertainty. The decision follows the question and evidence. When quantitative combination is plausible, the meta-analysis explainer shows what to consider before choosing a model.

In the training example, a score measured immediately after the course does not automatically answer the 12-week question. Keep immediate and later outcomes distinguishable, state how studies were grouped and explain the method used to compare their findings. A structured presentation should include the actual estimates, uncertainty and limitations where available; this teaching example supplies none of those observed results.

An explicit synthesis is still needed when there is no pooled number. Simply counting how many studies reached statistical significance is unreliable because significance also depends on precision. Cochrane’s guidance on other synthesis methods describes alternatives; the broader evidence-synthesis guide helps connect different synthesis tasks to a workable toolchain.

A reader needs to examine both the included studies and the review process. A trial can have missing outcomes, while a review can introduce a different problem by restricting its search or selecting results after seeing them. Evaluating the studies does not automatically evaluate all of the review’s decisions; Cochrane’s bias chapter explains the distinction.

Risk of bias is also different from certainty across a body of evidence. Certainty considers the support for a conclusion, including inconsistency, imprecision, indirectness and missing evidence where relevant. A large sample is not a universal quality score. Use an established framework when appropriate, such as the approach described in Cochrane’s certainty guidance.

For the curriculum team, even a precise short-term estimate may not answer a question about lasting learning or a different trainee group. Read the limitations and the last search date, not just the publication year. The following four questions keep the appraisal tied to the decision you want to make.

• Can I follow how evidence was found, selected and linked to individual studies?

• Which limitations in the original studies or the review process could affect the result?

• How uncertain is the conclusion, and does it distinguish insufficient evidence from evidence of no important effect?

• Do the population, setting, outcome, follow-up and age of the evidence fit my intended use?

A useful first milestone is an agreed question and a realistic route to answering it. Before drafting a full protocol, write a short feasibility note like the fictional example below. The entries identify work still to do; they do not assert that a search, appraisal or review has already been completed.

Use the note to decide whether to proceed, narrow the question, update an existing review or seek additional expertise. Then develop the full protocol and test ambiguous criteria on a small, varied set of records. If software will assist screening or extraction, the next task is to evaluate systematic-review tools against those requirements.