Most methods advise searching two to four databases for a systematic review, chosen by topic, plus trial registers and gray sources.
You came here for a clear number, not a shrug. There is no single magic count, because coverage varies by field and question. Still, you can pin a usable range. Most teams land on a core of two to four bibliographic databases, then add trial registers and gray sources that match the topic. That mix balances recall, time, and duplicate handling.
What A Thorough Search Means For Database Coverage
Thorough search work is about reducing missed studies, not chasing every index on earth. Core health reviews often pull from MEDLINE and Embase, while education leans on ERIC and social science draws from APA PsycInfo or Sociological Abstracts. STEM topics may point to Web of Science or Scopus. Multidisciplinary projects blend a few of these to widen reach without burning days on marginal gains.
| Database | Best Fit | Add When |
|---|---|---|
| MEDLINE/PubMed | Biomedicine, clinical | Any health review |
| Embase | Drugs, devices, European journals | Interventions, adverse events |
| Cochrane CENTRAL | Randomized trials | Intervention questions |
| Web of Science Core | Multidisciplinary citation reach | Cross-field topics |
| Scopus | Broad science and social science | When Web of Science is missing |
| APA PsycInfo | Behavioral and mental health | Psych outcomes or exposures |
| CINAHL | Nursing and allied health | Practice, workforce, settings |
| ERIC | Education | School, pedagogy, training |
| EconLit | Economics | Costs, incentives |
| AGRIS/AGRICOLA | Food and farming | Nutrition, food systems |
| LILACS | Latin American literature | Regional breadth |
| Global Health | Public health, LMIC scope | Global programs |
How Many Databases To Search For A Systematic Review: Real Ranges
Start with two core databases that index your field well, then judge the gap. Add a third or fourth source if pilot searches show poor recall, if outcomes sit in a niche database, or if stakeholders demand high sensitivity. Past audits found strong recall from sets like MEDLINE plus Embase, bolstered by Web of Science or Scopus. Some projects add CENTRAL for trials or APA PsycInfo for mental health angles.
Transparent reporting beats a fixed count. Methods checklists ask you to name every source, show full strategies, and give dates. That makes the work reproducible and lets readers see trade-offs. If you used a multi-database platform, still list each database it queried.
What The Methods Literature Says About Counts
Major handbooks stress multiple sources and full reporting, without locking you to a number. The Cochrane Handbook lists bibliographic databases, trials registers, and other routes, with advice on sensitivity versus precision in search strings. PRISMA-S lays out how to report information sources and strategies step by step. Method papers that tracked yield show that two to four well chosen databases capture the bulk of records for many topics, with niche files adding new items in some fields.
For wording you can cite, see the Cochrane Handbook search guidance and the PRISMA-S checklist and explanation. Both outline sources to search, strategy design, and full reporting of every database and register.
In short: use more than one, tailor to the question, and document every choice. That mix gives coverage without unmanageable screening loads.
Build Your Set Step By Step
Define The Field And Study Types
List the main concepts, populations, and designs. Trials, cohorts, and qualitative work live in different corners of the literature. If randomized trials drive your question, CENTRAL and trials registers help. If policy or social outcomes lead, add databases that index those streams.
Pilot Two Databases, Then Check Gaps
Draft a broad search in the top two databases. Screen the first few hundred titles and note misses. If a stack of seed papers are not retrieved, find out where they sit and add that database. If yield is thin, expand subject headings or text words and add a third database that covers adjacent fields.
Add Trials Registers And Gray Routes
Search ClinicalTrials.gov or the WHO ICTRP for ongoing and completed trials. Pull regulatory documents when relevant. Scan conference abstracts, theses repositories, and society sites. These routes catch time-lag bias and non-indexed work.
Decide When To Stop
Stop when another database adds little beyond duplicates and noise. Record the cut-off date and save full strategies for every source. If the topic moves fast, set alerts and plan an update window.
Search Quality Moves The Needle More Than Raw Count
Poor strings across six databases waste hours. Sharp strings in three hit more. Use both controlled vocabulary and free text. Map synonyms, acronyms, and spellings. Combine with tested filters for study design only when needed and only after checking loss of recall in pilot sets.
Use Peer Review For Search Strings
Have an information specialist or trained peer check your strings. Peer review catches missed terms and logic errors. Small fixes often add missed studies without adding a new database.
Document Everything Cleanly
Save the exact strategies, with line numbers and counts, for each database and date. Report platform names, years indexed, and any limits. If you used a multi-search tool that queries several databases at once, still list each database by name.
De-Duplication And Screening Volume
Expect high overlap between broad databases. Use a reference manager or screening tool with strong de-duplication. Keep a log of numbers at each step so the PRISMA flow is easy to build. Set daily screening goals to keep progress steady.
Reporting And Transparency That Reviewers Expect
Make reporting as strong as the search. In your methods, list every database by name and platform, the years indexed, the last search date, and any language or document limits. Provide at least one full strategy in an appendix with field tags, operators, and line numbers. If you adapt strings across databases, show the changes so readers can trace logic.
Spell out gray routes as well: registers searched, regulators checked, and sites scanned, with dates. If you reran searches before submission, give the refresh date and counts. These details match what PRISMA-S and major handbooks ask for and they save rounds of peer review back-and-forth.
Gray Literature And Trial Registers Count As Sources
Your database count does not tell the whole story. Trial registers, regulatory files, conference sets, and theses fill in gaps and cut bias from non-publication. Plan time for these routes and keep a log with links and dates.
When Fewer Databases Are Reasonable
Scoping reviews, rapid reviews, and updates may cap the search set to a tight core. Pick the two best databases for the field, add a third if pilot screening shows holes, and offset with stronger gray routes. State the limits up front and explain the trade-off in precision and recall.
When You Need More Than Four
Some topics sprawl across fields. Device safety, rare diseases, or method questions might need five databases plus registers. You will see this in pilot checks: many papers sit outside your first picks. Add the niche files that hold them and plan a careful de-duplication pass.
Time, Tools, And Team
Plan the workflow before you start. Pick a reference manager and a de-duplication method. Decide who builds strings, who runs them, and who screens. Set batch sizes so screening stays steady. Good process trims wasted effort more than any single database choice.
Recommended Core Sets By Topic And Aim
The right mix depends on the question. Use the table below as a starting point, then adjust with pilot data. The core sets here mirror common picks seen in protocols and method texts.
| Review Aim | Minimum Core Set | Add If |
|---|---|---|
| Clinical intervention effects | MEDLINE + Embase + CENTRAL | Trials register, Web of Science or Scopus for cross-checking |
| Public health programs | MEDLINE + Global Health | Regional file (LILACS), trials register |
| Mental health outcomes | MEDLINE + APA PsycInfo | Embase, Web of Science, trials register |
| Nursing and practice | MEDLINE + CINAHL | Scopus or Web of Science |
| Education interventions | ERIC + Scopus | MEDLINE for health-ed cross-over |
| Economics or cost | EconLit + MEDLINE | Web of Science for citations |
| Food and farming | AGRICOLA/AGRIS + MEDLINE | Global Health for programs |
| Cross-disciplinary topics | MEDLINE + Embase + Web of Science | Scopus or a niche file |
Worked Example: From Two Databases To Four
Say your review tests a drug-device combo for outpatient care. You start with MEDLINE and Embase using subject headings and text words. After screening, you notice missing trial reports cited in a major guideline. You add CENTRAL for trial records and run a trials register search. A quick check in Web of Science pulls several conference papers that fill gaps. The final set uses four databases plus two registers, with full strategies in an appendix.
Search Cost, Time, And Yield
Each extra database adds run time, export time, and de-duplication time. The payoff is new records. Track yield per database during pilot runs. If a source adds near-zero new items after de-duplication, drop it and document why. Spend saved time tuning strings and searching gray routes where fresh yield often sits.
Quick Decision Guide
- Two core databases span the field; run pilots and tally misses.
- Add a third or fourth when pilots show gaps or niche outcomes.
- Always search at least one major trials register and list dates.
- Stop adding databases once new records are near-zero and mostly duplicates.
- Log yields per source to guide updates and later planned replications.
Bottom Line For Your Protocol
State the planned sources and why you chose them. Most protocols name two to four databases keyed to the field, plus at least one trials register and a short list of gray routes. Add a note on peer review, dates, and updates. That plan sets up a transparent, reproducible search that meets handbook and journal expectations.