Boolean Operators for Literature Searches: How to Build a Search String Step by Step (AND, OR, NOT, MeSH, Truncation)

Boolean operators are the three words AND, OR, NOT, which combine search terms so a database returns the records you want [1]. The rule for literature searches is simple: OR within a concept, AND between concepts, and NOT almost never.

When a strategy is peer reviewed, the reviewer checks that operators are used correctly, that brackets nest properly and that NOT causes no unintended exclusions [3].

MeSH (Medical Subject Headings) is the National Library of Medicine’s controlled vocabulary for indexing MEDLINE [2]. Subject headings gather records about a concept however the authors worded it, but new records may not be indexed yet. Free-text synonyms in titles and abstracts catch those records and concepts with no heading, so use both for every concept [4].

A PubMed search applies automatic term mapping to untagged terms, matching them to MeSH, journals and authors [1]. Tag a term with a field, such as [tiab], and you control what it searches.

• Truncation finds word endings: in PubMed, smok* retrieves smoke, smoker and smoking [1].
• Phrase searching keeps words together, as in "type 2 diabetes".
• Proximity operators find words within a set distance. PubMed supports proximity in the title, abstract and affiliation fields, written "term term"[tiab:~N] [1]; Ovid uses adjN [6], Embase NEAR/n, Web of Science NEAR/x and Scopus W/n [7, 8, 9]. PRESS asks reviewers whether proximity or phrase searching would improve precision [3].
• Field tags and wildcards differ between platforms, so never paste syntax from one database into another untested.

Example question: do digital tools improve medication adherence in adults with type 2 diabetes?

1. Split the question into concepts: type 2 diabetes, medication adherence, digital tools.
2. List terms for each concept: a subject heading plus free-text synonyms and spelling variants.
3. Join the terms inside each concept with OR, in brackets.
4. Join the concept blocks with AND.
5. Check the result: run each block alone, look at the counts and test the string against studies you know should appear.

The final Boolean search string is (Population) AND (Outcome) AND (Intervention). Each block stays in its own brackets so the OR terms are combined before the AND.

The logic stays the same, but syntax changes. Subject headings, field tags and proximity operators are written differently on each platform, so rebuild and test the search strategy in each database rather than pasting it across [4]. Table 3 shows one concept in a PubMed search and in Ovid MEDLINE; Table 4 shows how platforms differ in proximity and in the order they evaluate operators [1, 6, 7, 8, 9], which is why unbracketed strings behave differently across databases.

1. Known-item check: collect five to ten studies you know are relevant and confirm the string finds them.
2. Block counts: run each concept alone; a block with very few or very many hits is a warning.
3. Add one concept at a time and watch how the count falls with each AND.
4. Read a sample of results for noise, then adjust the free-text terms.
5. Ask for a peer review: a second searcher checks the strategy with the PRESS checklist before the final run [3].

NOT removes every record containing the excluded term, even one that is also about your topic. Exclude “children” to focus on adults and you also lose studies of both. PRESS lists unintended exclusions from NOT as a specific check [3]; handle such cases at screening instead.

A sensitive string retrieves most relevant records plus many irrelevant ones; a precise string misses some. For a systematic review or other comprehensive literature review, favour sensitivity and accept more screening [4]. Every added AND concept narrows the results, so add one only when the question needs it.

A database string finds only what that database indexes. Supplement it with citation searching, trial registers such as the German Clinical Trials Register (DRKS) and grey literature sources [4]. If you plan to analyse retrieved full texts by machine, check the text and data mining rules first.

Save the exact string for each database, the date it was run and the number of records. PRISMA-S asks authors to report full strategies, dates and record counts, and to describe any search peer review [5]. The search log is also the source of your flow counts.

This guide is the hub of our literature review and gap analysis series. Start at the ten steps of a literature review, then use the guides below.