The core databases for systematic reviews are the major bibliographic indexes a review searches to find eligible studies, typically MEDLINE, Embase, and the Cochrane Central Register of Controlled Trials (CENTRAL) for health questions, supplemented by subject-specific indexes such as CINAHL for nursing or PsycINFO for psychology. No single database indexes all the literature, so a defensible review searches several complementary sources and reports each one separately.
Why one database is never enough
Each database indexes a different and only partly overlapping slice of the literature, with its own journal coverage, its own controlled vocabulary, and its own date range. A study indexed in Embase may be absent from MEDLINE, and vice versa. Relying on one source therefore builds a hidden gap into the review before screening even begins. Searching multiple databases is the most reliable way to raise the chance that your search strategy captures every eligible study, which is the entire purpose of a systematic search.
The usual core for health reviews
MEDLINE
MEDLINE is the backbone of most biomedical reviews. Its rich MeSH indexing lets you build precise subject-heading blocks, and it is accessible through several interfaces. Because its syntax and headings are well documented, it is often where reviewers develop and perfect a strategy before translating it elsewhere.
Embase
Embase has strong European and pharmacological coverage and indexes many conference abstracts that never appear in MEDLINE. Its Emtree vocabulary differs from MeSH, so the strategy has to be translated, not copied, when you move across, a point covered in running a systematic literature search.
CENTRAL
The Cochrane Central Register of Controlled Trials concentrates on randomised and quasi-randomised trials drawn from multiple sources, which makes it especially valuable for reviews of interventions that will feed into a meta-analysis.
Matching databases to the discipline
Outside core biomedicine, the right mix shifts. Nursing and allied health reviews add CINAHL; psychology reviews add PsycINFO; education, social science, and other fields each have their own indexes. The guiding principle is to choose databases by where the relevant literature actually lives, not by habit. For some questions that means searching a regional or non-English index so the review does not skew toward a single language or region.
What database searching cannot do alone
Even a well-chosen set of databases misses studies that were never indexed, published as reports, or registered but never written up. That gap is closed by deliberately searching for grey literature and by following references and citations through citation chasing and snowballing. Treat database searching as the foundation, not the whole building.
Subject-specific and regional indexes worth knowing
Beyond the biomedical core, several indexes earn their place for particular questions, and recognising them is part of choosing sources by where the literature lives:
- CINAHL for nursing and allied health, with its own subject headings distinct from MeSH.
- PsycINFO for psychology, psychiatry, and behavioural research.
- ERIC for education questions, and social-science indexes for sociology and policy.
- Web of Science and Scopus as broad multidisciplinary indexes, valued as much for their forward citation tracking as for subject coverage.
- Regional indexes such as LILACS for Latin American research, which guard against a search that skews toward English-language, high-income-country studies.
The decision rule is consistent: include a database when the question’s evidence is likely to sit there, and justify each choice in the protocol rather than defaulting to a familiar set.
Why interface and platform matter as much as the database
The same database behaves differently depending on the platform you reach it through. MEDLINE searched through Ovid, PubMed, or another interface uses different syntax, field tags, and truncation symbols, and may apply automatic term mapping that quietly changes what you retrieve. That is why a strategy is never simply copied between platforms but rebuilt to match each one, the translation work covered in adapting Boolean logic across platforms. Recording the exact interface, not just the database name, is therefore part of a reproducible search: “MEDLINE” alone does not tell a later reader which platform’s behaviour produced your record count.
Common database-selection mistakes
The errors here are usually about convenience overriding method. The first is searching a single database, which bakes a coverage gap into the review before screening starts. The second is treating Google Scholar as a primary source; it is useful for citation chasing and grey literature but lacks controlled vocabulary, returns unstable result sets, and exports poorly, so it cannot anchor a reproducible search. The third is adding databases after seeing the results to reach a desired set of studies, which turns the source list into a way to chase a conclusion. Fixing the list in a transparent review protocol before searching removes the room for all three.
Recording your database choices
Every database you search, the interface you used, the date you ran it, and the records it returned all go into the search record, as set out in documenting and reporting your search and counted into the PRISMA flow diagram. Naming your databases and justifying the selection is part of a transparent protocol, so reviewers fix the list before searching rather than adding sources to chase a result.