Research databases and indexes: choose the source, then document the search

1. Separate a database, an archive and a discovery service

PubMed provides citations and abstracts from several NLM resources. MEDLINE is its largest component and adds controlled MeSH indexing; PMC is a full-text archive. A PubMed record therefore does not, by itself, establish that the whole journal is currently indexed for MEDLINE. Check the record and the journal’s actual status separately.

Scopus is an abstract and citation database covering multiple disciplines and publication types. Web of Science is a platform: record the collection you actually searched, such as Core Collection, rather than saying only “Web of Science.” The comparison below identifies practical roles, not a ranking or a complete inventory of each service.

Reference: NLM: about PubMed and its components · Elsevier: Scopus content and coverage · Clarivate: Web of Science Core Collection search fields

On small screens, scroll the table horizontally, or focus it and use the arrow keys.

2. Build the source plan around your review question

Write down the disciplines, study designs, languages and publication types your question may require. A clinical effectiveness review, a humanities bibliography and a map of engineering conference work need different source plans. Ask a subject librarian to review gaps, including relevant regional repositories or specialist databases. Availability through your institution is a practical constraint to document, not proof that the accessible set is complete.

DOAJ focuses on open access journals. Google Scholar reaches diverse scholarly materials, but its help states that a query displays at most 1,000 results and bulk access is not provided. If you use Scholar, define and report what you actually examined. Do not convert its displayed estimate into a count of records exported or screened.

Reference: DOAJ: purpose of the open access journal index · Google Scholar: search help and coverage limits

3. Translate concepts across interfaces, not just words

Field labels and search rules differ. PubMed offers title/abstract fields and controlled vocabulary; Scopus TITLE-ABS-KEY searches titles, abstracts and keywords; a Core Collection Topic search includes title, abstract, author keywords and Keywords Plus. Identical words in those fields do not describe identical searches.

The following small example concerns telemedicine and diabetes. It only illustrates field syntax. It has not been run here and is not a complete or validated systematic-review strategy. Develop synonyms, relevant indexing terms and any justified limits with the review team, then test the final query in each actual interface. Preserve the database’s interpretation and warnings.

Reference: NLM: PubMed search, history and export help · Elsevier: Scopus advanced search · Clarivate: Web of Science Core Collection search fields

On small screens, scroll the table horizontally, or focus it and use the arrow keys.

4. Preserve a search record at the time of execution

PRISMA-S calls for enough information to understand and reproduce the search, including databases and platforms, strategies, dates, limits and record management. Build that record as you work. Save the exact final query, selected collection, coverage or timespan, filters, execution date and export settings alongside the raw export.

Use separate fields for the displayed result count, downloaded records, imported records and duplicates removed. If an export is interrupted, record it as incomplete and reconcile it before screening. Do not fill an unavailable count with zero. Keep a second, dated record for an update search so new records can be traced to the run that found them.

Reference: PRISMA-S: reporting literature searches

5. Distinguish duplicate records from multiple reports of a study

Two exports can contain the same article, while a conference abstract and later full paper can describe the same underlying study. These are different relationships. Keep one record of an exact duplicate while preserving its source provenance. Link potentially related reports for assessment rather than deleting them because their titles or author lists resemble one another.

In a fictional import, source A contains 120 records and source B contains 90. The team verifies 30 exact duplicate pairs and no other removals, leaving 180 distinct records. That arithmetic does not establish 180 studies or 180 eligible papers. Screening and linking reports to studies still remain. If duplicate status is uncertain, retain the record for review.

6. Check an indexing claim at the correct level

Ask for the exact journal title, ISSN, claimed index and coverage period. Use the provider’s current source record rather than a badge on the publisher’s website. A journal title can change, coverage can be limited to certain years, and an individual article may be discoverable through a route that does not establish journal-wide selection.

NLM explains distinct selection and participation pathways for MEDLINE and PMC. For a publisher statement, keep a dated link and specify the resource and scope actually verified. If the current record cannot be accessed, mark the claim unverified and seek clarification. Do not infer misconduct or scientific invalidity solely from absence in one index.

Reference: NLM: information for publishers

7. Give co-authors a usable search package

Bundle the source plan, executed strategies, raw exports, import reconciliation, duplicate decisions and linked-report notes. Document citation searching separately with the seed papers, direction, service and date. Keep the search package readable without the original browser session or a colleague’s personal account.

For an Arabic methods summary, preserve database names and technical fields exactly while explaining their function naturally. Check that both language versions report the same dates, limits and counts. A librarian can help assess the search design; the author team still needs to judge the studies and explain the limits of the evidence it found.

Questions for a practical final check

• Each source serves a stated part of the question.
• Database, platform and collection are named precisely.
• Full executed queries and raw exports are retained.
• Downloaded and screened counts reconcile without invented zeros.
• Duplicate records and related study reports are handled separately.
• Journal coverage claims link to current, specific evidence.