03 · What You Need to Know
Comprehensiveness Comes From Coverage, Not Accumulation
Searching more than one database genuinely matters
The argument against indiscriminately adding databases should not be confused with an argument for searching only one.
Cochrane's current Handbook states that searching two or more bibliographic databases lowers the risk of missing eligible studies. It also explicitly considers MEDLINE alone inadequate for Cochrane intervention reviews because not all known randomized trials are available in MEDLINE and relevant records that are present may still be difficult to retrieve.
For Cochrane intervention reviews, CENTRAL and MEDLINE should be searched, together with Embase when it is available to the review team. Appropriate national, regional, and subject-specific databases are also highly desirable when relevant to the review topic.
The broader lesson is clear: important evidence is distributed across sources. One database rarely represents the entire evidence landscape of a substantial systematic review.
But “more than one” does not imply “as many as possible”
Once several appropriate sources have been selected, the question changes.
Another database is useful when it adds something methodologically valuable: different journals, another discipline, regional literature, additional publication languages, specialist indexing, distinctive document types, or other relevant coverage.
Adding a source merely because it exists is different.
Complementary database
Adds relevant coverage that is absent or inadequately represented in the sources already selected.
Redundant database
Overlaps heavily with existing sources while contributing little additional relevant coverage for the particular question.
A database can also fall somewhere between those categories. Considerable overlap does not mean it contributes nothing, and a single unique eligible study may matter considerably in some reviews.
Database overlap is normal
Major bibliographic databases index many of the same journals and articles. If you search MEDLINE, Embase, Scopus, and Web of Science, substantial duplication should be expected.
This overlap is not evidence that the search was badly designed. It is partly a consequence of selecting reputable sources that cover related scholarly literatures.
At the same time, overlap means that the number of database records retrieved should not be mistaken for the amount of unique evidence identified.
Suppose four searches retrieve:
| Database |
Records retrieved |
Unique contribution after comparison |
| Database A |
2,500 |
Forms much of the initial evidence pool |
| Database B |
2,100 |
Substantial overlap plus useful additional records |
| Database C |
1,800 |
Mostly overlap but several potentially relevant unique records |
| Database D |
900 |
Smaller database but substantial specialist literature absent elsewhere |
The smallest database could be the most important addition if its coverage addresses a gap in the other sources.
Unique relevant records matter more than unique records generally
A database can contribute hundreds of records that appear nowhere else and still add little value if those records are irrelevant to your review.
What matters is its contribution to the evidence you need.
Research by Bramer and colleagues illustrates why this distinction matters. Their prospective study examined database retrieval across 58 published systematic reviews. Among included references identified through database searching, some were retrieved from only one database, demonstrating that unique relevant contributions can occur even when databases overlap substantially. The authors also found that specialist databases could contribute unique included references when their subject coverage matched the review topic.
The practical implication is not that every possible database should be searched. It is that overlap alone cannot tell you whether another source is unnecessary.
A specialist database can add more value than another large general database
Imagine a review concerning psychological interventions for university students.
You have already searched two broad multidisciplinary databases. You now have a choice between adding another broad citation database or APA PsycInfo.
The specialist source may be more valuable because the question directly intersects psychology. Cochrane similarly recommends considering subject-specific databases such as APA PsycInfo for psychological interventions and CINAHL for nursing-related topics. Regional databases should also be considered when relevant.
This is why database selection should begin with the evidence landscape rather than a ranking of database size.
Interdisciplinary questions often benefit from additional databases
Some topics do not belong comfortably to one discipline.
Generative AI in higher education, for example, can appear in education, computer science, psychology, information science, communication, and multidisciplinary journals. Searching only education databases could miss relevant technology research. Searching only computing databases could underrepresent educational and behavioral research.
Cochrane notes that additional databases may be needed for specialized, cross-disciplinary, or emerging topics. The principle applies beyond health research: the more disciplinary boundaries your question crosses, the more carefully your information sources need to represent those different research communities.
This is one reason the decision about how many databases you actually need should be made by mapping relevant disciplines rather than adopting a fixed minimum.
Another database cannot compensate for a poor search strategy
Suppose your search represents a central concept with one unusually narrow phrase. You search five databases and retrieve very little.
Adding a sixth database may not solve the problem.
If the same conceptual weakness is translated into every source, the strategy can repeatedly miss relevant records even when those databases contain them.
Cochrane distinguishes database coverage from search performance for precisely this reason. Relevant records may exist within a database yet remain unretrieved because the search strategy is inadequate. Search strategies should combine appropriate concepts, free-text terms, and controlled vocabulary where available while aiming for high sensitivity.
Watch Out
Do not respond to a weak result set simply by adding databases. First determine whether the problem is source coverage or the search strategy itself. A narrow query can remain narrow in ten databases.
The strategy needs to be adapted to every database
Searching more sources also creates more opportunities for implementation errors.
Databases differ in controlled vocabulary, field codes, phrase syntax, truncation, proximity operators, limits, and interface behavior. A query developed for one platform should not simply be pasted unchanged into another and assumed to behave equivalently.
JBI's systematic-review guidance describes database-specific searching as part of its search process. It expects review protocols to identify the information sources that will be searched and, ideally, to specify the proposed strategy for each bibliographic database.
Adding another database therefore adds another retrieval instrument that needs to be constructed and checked properly.
Every additional database has a workload cost
The cost of an additional source extends beyond the time required to press Search.
| Stage |
Additional work created by another database |
| Search development |
Translate terminology, controlled vocabulary, fields, syntax, and operators |
| Testing |
Check whether the translated strategy retrieves expected records and behaves correctly |
| Documentation |
Preserve the complete strategy, platform, limits, result count, and search date |
| Record management |
Export, merge, normalize, and deduplicate another result set |
| Screening |
Review additional unique records that survive deduplication |
| Updating |
Rerun and document another search when the review is updated |
These costs do not make additional searching undesirable. They make the expected contribution of each source worth considering.
Duplicate records are not the same thing as wasted searching
It can be frustrating to export 3,000 records from another database and discover that most already exist in your reference manager.
But duplication has to be interpreted carefully.
If the database was selected prospectively because it covers a relevant discipline, a high duplicate rate does not prove that searching it was a mistake. You could not necessarily know in advance which eligible records would be unique.
Nor should database selection be decided retrospectively by removing every source that happened not to contribute a unique included study. That approach risks designing future searches around the evidence already known.
Expected coverage is generally a stronger prospective justification than hindsight.
Diminishing returns are real, but they are not a universal stopping rule
As more overlapping databases are searched, the number of new records contributed by each additional source may decline.
This is a form of diminishing returns.
However, there is no universal point at which the return becomes too small. A single unique eligible study might alter an estimate, reveal a neglected population, or change an interpretation. In another review, thousands of additional records from a heavily overlapping source might contribute nothing eligible.
The consequence of missing evidence therefore matters alongside the amount of extra work.
Marginal retrieval
The additional records found by searching another source.
Marginal value
The contribution those additional records make to identifying evidence relevant to the review.
More retrieval is not automatically more value.
Searching more bibliographic databases cannot replace other information sources
This is one of the most important limitations of a database-count mentality.
Suppose you have searched eight bibliographic databases but no trial register. If your review concerns clinical trials, searching a ninth journal database does not necessarily compensate for that omission.
Likewise, adding another citation database does not necessarily replace searching dissertations when unpublished doctoral research is relevant, nor does it replace regulatory sources, organizational reports, reference lists, or other sources required by the question.
Cochrane recommends searching trial registers for intervention reviews and also identifies grey literature, reference-list checking, regulatory sources, and other methods as important parts of comprehensive study identification. JBI similarly frames searching around the widest reasonable collection of appropriate information sources, not databases alone.
After deciding not to restrict the search automatically to peer-reviewed literature, this distinction becomes particularly important. More journal databases do not necessarily provide better access to evidence outside journal publishing.
Database quantity cannot repair a language or regional coverage gap
Imagine searching six major international databases for a question concerning a health intervention widely used in Latin America. If all six sources have weak coverage of regional journals, adding another similar international database may contribute less than adding an appropriate regional source.
The same principle applies to non-English literature.
Cochrane recommends considering national, regional, and subject-specific databases when relevant and explicitly warns that simply removing language restrictions from English-language databases does not substitute for searching relevant non-English-language sources.
This reinforces why language restrictions and multilingual coverage need to be considered as part of source selection rather than merely as filters applied after the databases have been chosen.
Historical coverage creates another dimension of complementarity
Two databases may cover the same discipline but differ substantially in historical depth.
If your review requires older literature, an additional database or historical index may be valuable not because it covers another subject but because it reaches farther back.
Conversely, if the eligible phenomenon began recently, historical depth may contribute little.
The appropriate source set therefore depends partly on how far back the literature needs to be searched.
Search quality should be evaluated at the level of the source portfolio
Instead of asking whether each database is individually comprehensive, consider how the sources work together.
A defensible portfolio might include:
- a major database for the central discipline;
- a complementary multidisciplinary source;
- a specialist database covering an important secondary discipline;
- a regional source where geography matters;
- appropriate registers or grey-literature sources;
- citation and reference-list searching.
The exact configuration varies. The important feature is that each source has a reason for being there.
JBI's guidance captures this logic by emphasizing the widest reasonable collection of information sources considered appropriate to the review. Comprehensiveness is not synonymous with infinity.
Database selection should be planned before you know which source “wins”
It can be tempting after completing a review to examine which databases produced included studies and conclude that the others were unnecessary.
That reasoning is vulnerable to hindsight.
Database selection should normally be justified prospectively from the research question, expected coverage, review methodology, and available evidence about database performance. A source can be reasonable to search even if it ultimately contributes no unique included studies.
The absence of a unique included study is a result of the search, not necessarily evidence that the original source-selection decision was poor.
There is also a point at which resources become part of the methodology
Systematic reviews aim for comprehensive evidence identification, but no search has unlimited resources.
Cochrane explicitly frames thorough searching as occurring within resource limits, while JBI refers to the widest reasonable collection of appropriate information sources.
That does not mean convenience can justify arbitrary omissions. It means source selection involves a defensible balance between expected evidence coverage and practical feasibility.
A useful question is:
If we do not search this source, what relevant evidence are we plausibly accepting a greater risk of missing?
If the answer is clear and consequential, the source deserves serious consideration. If no meaningful coverage gap can be identified, adding it merely to increase the database count is harder to justify.
Information specialists can help distinguish useful expansion from database collecting
Researchers usually understand the substantive scope of their topic. Librarians and information specialists bring knowledge of database coverage, indexing, interfaces, search translation, and overlap.
Cochrane recommends involving an experienced librarian or information specialist from the beginning of the protocol, and JBI similarly recommends expert research-librarian input where possible.
This expertise becomes particularly valuable when deciding whether another database provides genuine complementary coverage or merely another large export file with a promising name.
The final review should report every source actually searched
Regardless of how many sources you use, the process should be transparent.
Cochrane's reporting guidance requires review authors to specify all sources consulted, including databases, registers, websites, and organizations, as well as coverage dates, search dates, limits, and the exact strategies used for each database.
JBI likewise requires detailed search strategies for major databases and identification of the search platforms used where necessary.
This documentation makes the logic of your source portfolio inspectable. A reader can evaluate not simply how many databases were searched but whether the chosen sources and strategies were capable of identifying the intended evidence.
