A search strategy for a systematic review is the documented set of terms and logic you run in each database to retrieve every potentially eligible study. It pairs controlled vocabulary such as subject headings with free-text synonyms for each concept of the question, joins them with Boolean operators, and is written out line by line so the search can be peer reviewed, translated across platforms, and rerun exactly.
What separates a strategy from a quick search
A quick search optimises for precision: a short, relevant-looking list. A systematic strategy optimises for sensitivity: capturing eligible studies even at the cost of many irrelevant hits you will discard during screening. That trade is deliberate. Missing one eligible trial can change a review’s conclusion, whereas extra irrelevant records only cost screening time. The strategy is fixed in the review protocol before searching, which is what stops it from being quietly tuned to produce a convenient answer.
The anatomy of a search strategy
Concept blocks
Each concept from the question becomes a block of terms. You derive the concepts from a framework such as PICO, then build one block per concept you intend to search. A typical strategy has two or three blocks, because adding more concepts narrows the search and risks excluding relevant studies. The general method for assembling these blocks sits in our guide to running a systematic literature search.
Two kinds of term in every block
Inside each block you combine subject headings, the indexer-assigned vocabulary covered in MeSH terms and subject headings, with free-text terms that match the title and abstract. Subject headings catch records even when authors used unusual wording; free-text catches recent records not yet indexed and concepts with no good heading. Using both is what gives a strategy its reach.
Boolean combination
Synonyms inside a block are joined with OR; the finished blocks are joined with AND. Truncation and proximity operators stretch single terms to cover plurals, spellings, and adjacent phrasing. The full mechanics, including common pitfalls, are in Boolean operators for literature searching.
Testing and refining the strategy
Before you trust a strategy, validate it against a small set of known relevant studies you expect it to find. If the strategy misses one, work out which term or heading would have caught it and add it. This iterative tuning continues until the strategy retrieves all of your test set without becoming hopelessly broad. Keep a note of each change, because a transparent development history is part of reporting the search well.
From one database to many
The strategy you perfect in your lead database has to be translated for every other source, since syntax and controlled vocabularies differ. Which databases you target is a strategic choice covered in selecting databases for systematic reviews, and database searching is then supplemented with forward and backward citation searching and grey literature. Once finalised, the complete strategy for every source is reported as set out in documenting and reporting your search and counted into the PRISMA flow diagram.
A common structure, illustrated
A simple two-concept strategy reads as a population block OR-ed together, an intervention block OR-ed together, and the two blocks joined with AND. For example, a population block might gather a heading plus free-text synonyms for the group of interest, while the intervention block gathers a heading plus synonyms for the treatment. Only records matching at least one term in each block are retrieved, which keeps the search focused on the question while remaining sensitive within each concept.
Deciding how many concepts to search
The number of concepts you search is the single biggest lever on sensitivity, and the safe default is to search fewer concepts than the question contains. A four-part question framed with a structured framework such as PICO does not become a four-block search. In practice:
- Search the population and the intervention, the two concepts almost always worth a block, because both are usually well described and well indexed.
- Leave out the outcome in most reviews; outcomes are inconsistently indexed and reported, so an outcome block drops eligible studies that simply phrased the result differently.
- Leave out the comparator unless it is a specific named treatment, since a placebo or usual-care comparator is rarely searchable.
Every extra block you join with AND narrows the result and raises the chance of excluding a relevant study, so reviewers treat each added concept as a deliberate cost rather than a free improvement in precision.
Adding methodological filters with care
Some reviews bolt a study-design filter onto the strategy, for instance a validated randomised-trial filter, to limit the records to a design. These published filters are useful, but they trade sensitivity for precision, so a filter is only safe when it has been validated for the platform you are running it on and when the lost precision is worth the risk. Avoid building an informal design filter from a few keywords, because an unvalidated filter quietly discards eligible studies. The same caution applies to language and date limits: each one is a decision that must be justified in the protocol and reported, not applied silently to trim the result count.
Common faults that weaken a strategy
Three recurring errors account for most weak strategies. The first is a thin synonym list, where a concept block relies on one or two obvious terms and misses the variants authors actually use; mining a few known papers for their wording is the cure. The second is over-narrowing with AND, where a reviewer adds the outcome or comparator as a block and the result count collapses, taking eligible studies with it. The third is treating the strategy as finished after one database, when in fact it still needs translating, supplementing with unpublished and grey sources, and submitting for independent search peer review. A strategy that survives all three is the foundation for a clean study-selection flow and a defensible review.