How to build a systematic review search strategy
A search is reproducible or it is not a systematic search. Here is how to build one that another researcher could run and get the same results.
A systematic search has one defining property: someone else could run it and obtain the same set of records. Everything else follows from that requirement.
Build it from concepts, not sentences
Break the question into concepts — usually population and intervention, sometimes outcome or design. Each becomes a block of synonyms combined with OR. The blocks are then combined with AND.
Outcomes are often better left out. They are inconsistently reported in titles and abstracts, and including them removes relevant studies that simply did not mention the outcome where the search could see it.
Controlled vocabulary and free text, both
Each block needs database subject headings — MeSH in MEDLINE, Emtree in Embase — and free-text terms searched in title and abstract.
Headings alone miss recent records that have not been indexed yet. Free text alone misses records where the authors used different wording. Neither is sufficient, and the two are combined with OR inside the block.
Translate, do not copy
A strategy written for MEDLINE will not run correctly in Embase, CINAHL or Web of Science. Subject headings differ, field tags differ, truncation and proximity operators differ.
Every database gets its own version, and every version is recorded in full.
Prefer sensitivity to precision
A systematic review search is meant to find everything relevant, accepting that most of what it returns will be screened out. Ten thousand records with high recall is a working search; eight hundred records that missed four eligible studies is a broken one.
If the yield is unmanageable, narrow with justified eligibility limits rather than by quietly dropping synonyms.
Have it peer reviewed
The PRESS checklist exists because search strategies contain errors at a high rate and the errors are invisible in the output — a missing synonym does not announce itself.
A second information specialist or an experienced reviewer checking the translation, the operators and the coverage of each concept is the cheapest quality step in the review.
Record what reporting will require
- The full strategy for every database, verbatim, including limits.
- The date each search was run.
- The interface used, since results differ between them.
- Records retrieved per source, for the PRISMA flow diagram.
- Any supplementary methods: reference-list checking, citation searching, trial registries, contact with authors.
Recording this at the time takes minutes. Reconstructing it six months later is guesswork, and guesswork is what makes a search unreproducible.
Frequently asked questions
How do you write a systematic review search strategy?
Break the question into concept blocks, combine synonyms within each block using OR and the blocks using AND, and search each concept with both subject headings and free text. Translate the strategy for every database rather than copying it, and record each version in full with the date it was run.
Should outcomes be included in the search?
Often not. Outcomes are inconsistently reported in titles and abstracts, so including them removes relevant studies that simply did not mention the outcome where the search could see it.
Why does a search need peer review?
Search strategies contain errors at a high rate and the errors are invisible in the output — a missing synonym does not announce itself. The PRESS checklist exists for this, and a second reviewer is the cheapest quality step in a review.