13 Research methods, statistics, critical review and evidence-based practice13.1 Introductory concepts
13.1.4 Research cataloguing and efficient retrieval
3 min read2 figures
Written by Resident Doctor Committee · Published
What a record makes searchable
Searchable fields include title, abstract, authors, affiliations, journal, date, language and identifiers such as PMID, DOI and trial registration number.
Publication type describes an article's design and subject headings describe its content. Some records are never indexed, so headings alone miss recent or unindexed papers.
Controlled vocabulary and free text
Controlled vocabularies standardise concepts and link older and newer terms: MeSH in MEDLINE, Emtree in Embase and the psychological thesaurus in PsycINFO.
Free text searches the authors' own words in titles, abstracts and keywords, and misses synonyms and spellings nobody anticipated.
Headings miss unindexed or wrongly indexed records, so searches combine headings with free-text synonyms, spelling variants and acronyms.
Exploding a heading adds the narrower headings beneath it. Focusing on major topics or adding subheadings narrows the search and can miss studies.
Operators
Device | Effect |
|---|---|
OR | Retrieves records in either set |
AND | Keeps only the overlap |
NOT | Removes one set from another |
Truncation (schizophreni*) | Finds words sharing a stem; broad stems such as psych* add noise |
Wildcard (randomi?ed) | Replaces one character |
Proximity (adj2 in Ovid) | Finds nearby terms without needing an exact phrase |
Field tags | Restrict matching, such as .ti,ab. in Ovid or [tiab] in PubMed |
Precedence varies between platforms, so use parentheses: schizophrenia AND (clozapine OR olanzapine).
Concept blocks
Concept blocks join synonyms with OR, then join the blocks with AND, for example schizophrenia AND treatment resistance AND clozapine, each block holding headings and free text.
Each extra AND loses records that omit that concept, so two or three blocks are usual. Outcomes are left to screening because abstracts report them inconsistently.
NOT is risky: NOT child* removes mixed-age studies and papers stating that children were excluded.
Recall and precision
Recall, or search sensitivity, is the proportion of relevant records retrieved. The full relevant set is usually unknown, so recall is judged against a benchmark of known papers.
Precision is the proportion of retrieved records that are relevant; it is not statistical specificity.
Systematic reviews and guidelines favour recall, while a question between clinics favours a small set of useful evidence.
Filters, chaining and deduplication
Validated design filters combine tested headings and text terms; filters on publication type alone miss unindexed records.
A trial filter is never added to CENTRAL.
Route | What it does |
|---|---|
Backward chaining | Reads reference lists of papers in hand, reaching older work |
Forward chaining | Finds later papers citing a known study, through Web of Science or Scopus |
Related articles | Ranks records by shared index terms and title words |
Author searching | Finds a group's other work; ORCID identifiers separate authors sharing a surname |
Identifier searching | A trial registration number links papers to one trial, so they count as one study |
Deduplication in a reference manager matches identifiers, then title, author and year. It is separate from linking several reports of one study.
Recording a search
Record each database and platform, search date, full strategy, limits, filters and number retrieved, following PRISMA 2020 and PRISMA-S guidance.
Reruns must allow for delayed indexing.
Record the route actually used, because a library discovery search can behave differently from the database's own interface.