Build the search strategy and document it with PRISMA

The results of a bibliometric analysis can only be as good as the records you analyse. Reviewers scrutinise the search more than any other part of a bibliometric paper, because a biased or undocumented search undermines every map that follows. This lesson shows how to build a search that is complete, focused and reproducible.

Choosing the database

• Web of Science Core Collection and Scopus are the standard choices. Both export complete metadata including cited references, which co-citation, coupling, historiograph and RPYS analyses require. Scopus covers more journals; Web of Science has a longer, more selective back catalogue.
• PubMed is the reference database for biomedicine and has rich MeSH indexing, but no citation counts or reference lists.
• Dimensions, Lens.org and OpenAlex offer broad, free coverage, with references available as identifiers.
• Google Scholar has no bulk export and inconsistent metadata; avoid it as a primary source.

Many studies use one database for simplicity and consistent reference formats. Using two (for example Scopus and Web of Science) increases coverage but requires merging and de-duplication (Lesson 3).

Building the search string

1. Identify the core concepts of your topic. For a review of artificial intelligence in digital health: concept A = digital health, concept B = artificial intelligence.
2. List synonyms and variants for each concept, using known key papers, thesaurus terms and keywords of relevant articles: “digital health”, mhealth, “mobile health”, telemedicine, telehealth, eHealth; “artificial intelligence”, “machine learning”, “deep learning”.
3. Combine with Boolean operators: OR within a concept, AND between concepts. Use quotation marks for phrases and truncation (telemedic*) where the database supports it.
4. Restrict the fields to title, abstract and keywords, so that papers that only mention a term in passing are not retrieved.

TITLE-ABS-KEY ( ( "digital health" OR mhealth OR "mobile health" OR telemedic* OR telehealth OR ehealth ) AND ( "artificial intelligence" OR "machine learning" OR "deep learning" ) ) AND PUBYEAR > 2007 AND PUBYEAR < 2026 AND ( LIMIT-TO ( DOCTYPE , "ar" ) OR LIMIT-TO ( DOCTYPE , "re" ) ) AND ( LIMIT-TO ( LANGUAGE , "English" ) )

Testing the search

A good search is sensitive (it finds the relevant papers) and specific (it does not drown them in irrelevant ones). Test it: make a list of 10 to 20 papers you know are relevant and check that the search finds them all. Then read a random sample of 50 results and estimate how many are off-topic. Refine by adding missing synonyms or excluding ambiguous terms, and record each version of the string.

Inclusion and exclusion criteria

Decide in advance and justify: the time span (often from the first relevant publication, or a meaningful event, to the last complete year), document types (articles and reviews are common, because conference papers and editorials have incomplete metadata and different citation patterns), language, and subject areas. Avoid filtering by subject area unless necessary; interdisciplinary fields are often indexed in unexpected categories.

Documenting the search with PRISMA

The PRISMA 2020 statement (Page et al., 2021), originally for systematic reviews, is now expected in most bibliometric papers. Its flow diagram reports the number of records at each step:

1. Identification: records found in each database, and duplicates removed.
2. Screening: records excluded by filters (years, document type, language) and by title and abstract screening, with reasons.
3. Included: the final collection analysed.

Also report the database, the exact search string, the date of the search (databases change daily) and every filter. Save the raw export files; they are your data.

The Studio fills in the PRISMA counts from the import and filter steps automatically: records per file, duplicates removed and records excluded by your filters. You can edit the boxes for steps done outside the Studio, such as manual screening of titles, and download the diagram for your paper.

Manual screening: is it needed?

For a pure bibliometric analysis of a broad field, a well-tested search is often analysed without reading every record. When the search terms are ambiguous (for example, “trust” in finance and in medicine), screen titles and abstracts, ideally with two independent reviewers on a sample, and report the agreement (the Cohen’s kappa tool computes it). Record the excluded records and the reasons in the PRISMA diagram.