How to Conduct a Systematic Review

The defining feature of a systematic review is that someone else could repeat it and get the same answer.

A systematic review differs from a narrative literature review in one crucial respect: the method is specified in advance and reported in enough detail to be reproduced. That reproducibility is the whole point, and it is why systematic reviews are treated as evidence in their own right rather than as commentary on evidence.

Write a protocol before you search

The protocol states the question, the inclusion and exclusion criteria, the databases, the search terms, the screening process and the planned synthesis — all before you look at any results. Writing it afterwards, or adjusting criteria once you have seen what the search returns, reintroduces exactly the selection bias the method exists to prevent.

Where the review is for publication, protocols are often registered publicly. Even for coursework, writing one first improves the review substantially and is usually assessed.

Frame the question in a structured form

Most disciplines use a framework to force precision. PICO — population, intervention, comparison, outcome — is standard in health. SPIDER suits qualitative work; SPICE suits service evaluation. The framework matters less than the discipline it imposes.

Each element of the question becomes part of your search strategy and your screening criteria, so vagueness at this stage propagates through the whole review.

Build a search strategy you can report

Search at least two or three databases appropriate to your field, and record the exact strings used, the date searched, and the number of results from each. A search that cannot be reported cannot be reproduced.

• Combine synonyms with OR, concepts with AND.
• Use controlled vocabulary such as MeSH where available, alongside free text.
• Record filters applied — language, date range, study type — and justify each.
• Supplement with reference-list checking and citation searching.

Screen in two stages, against stated criteria

Screen titles and abstracts first, then full texts. Apply the criteria you wrote in the protocol, and record the reason each excluded study was excluded at the full-text stage — that record is required for your flow diagram.

Where possible, have a second person screen a sample independently and compare. Disagreement rates are informative, and reporting how disagreements were resolved strengthens the review.

Report the flow of studies

A PRISMA-style flow diagram shows how many records were identified, how many remained after duplicates were removed, how many were screened, how many full texts were assessed, and how many were included. Readers use it to judge whether your search was adequate and your screening consistent.

Appraise quality systematically

Use an established appraisal tool suited to the study designs you have included — CASP checklists, the Cochrane risk-of-bias tool, JBI instruments. Applying a recognised tool consistently is more defensible than forming your own impression of each paper.

Quality appraisal is not a filter to discard inconvenient studies. Its purpose is to weight findings and to explain disagreement between them.

Extract data into a structured table

Build an extraction form before you start: study design, setting, sample, intervention, comparator, outcomes, results, limitations. Consistent extraction makes synthesis possible and prevents the drift that occurs when you record whatever seems interesting about each paper.

Synthesise rather than list

The weakest systematic reviews describe each included study in turn. Synthesis groups findings by outcome or theme, states where studies agree, and — most importantly — explains disagreement in terms of population, method, setting or measure.

Where studies are similar enough, meta-analysis may be possible. Where they are not, narrative synthesis is entirely legitimate, provided the grouping logic is explicit.

Report limitations of the review itself

Distinguish limitations of the included studies from limitations of your review. Restricting to English-language publications, searching two databases rather than five, or screening single-handed are limitations of your method, and stating them is expected rather than damaging.

Finish by saying what the evidence collectively supports, how confidently, and what study would most usefully be done next. That last judgement is often the review’s real contribution.

Decide the scope you can actually complete

Full systematic reviews are substantial undertakings, often taking a team many months. For a dissertation, a narrower question, fewer databases, or a defined date range makes the work tractable — provided you state the restriction and treat it as a limitation rather than concealing it.

A well-executed narrow review is worth far more than an over-ambitious one abandoned at the screening stage.

Manage the mechanics before the volume arrives

Searches routinely return hundreds or thousands of records, and the practical machinery matters. Export results into reference software, deduplicate systematically, and use a screening tool or a structured spreadsheet with one row per record and a column for the exclusion reason.

Set this up before searching. Retrofitting a system onto two thousand records already scattered across downloads is where reviews stall.

Distinguish the review types

A scoping review maps what exists and what has been studied. A rapid review compresses the method under time pressure, with the compromises stated. A meta-analysis pools quantitative results statistically. Calling your work a systematic review when it is a structured narrative review is a claim examiners check, so use the term that fits what you actually did.