Strata Academy

Systematic Review Search Strategy: MEDLINE, Embase & CENTRAL

Building sensitive search strings, database selection, MeSH and Emtree, peer review, and documenting searches for PRISMA

• Search MEDLINE, Embase, and CENTRAL for most clinical intervention reviews — not PubMed alone.
• Align search concepts to your registered PICO before building strings.
• Use block structure (population / intervention / design) and document every line.
• Re-run searches before submission if the review ran longer than 12 months.

1. Sensitivity, reproducibility, and why search matters

The search strategy is the foundation of a systematic review. If relevant studies are missed, subsequent screening, risk-of-bias assessment, and meta-analysis cannot recover them — the review inherits selection bias from the outset.

Cochrane methods distinguish sensitivity (finding all relevant records) from precision (minimising irrelevant hits). For systematic reviews, sensitivity wins. A search returning 8,000 references after deduplication is normal; screening applies eligibility criteria later.

Every search must be reproducible: another team should be able to run the same strategy and obtain comparable results. That requires documented databases, full strings, date limits, language filters, and search dates.

Before you open a database, stabilise your review question. Draft your PICO using the free PICO builder at /tools/pico-builder — population, intervention, comparator, and outcomes should map directly to search blocks. A vague question produces a vague string and ambiguous screening.

Note: Searching PubMed alone is not equivalent to searching MEDLINE with a structured strategy. PubMed's automatic term mapping can miss Emtree-only concepts and lacks Embase's conference and pharmacovigilance coverage.

2. Core databases: MEDLINE, Embase, and CENTRAL

MEDLINE (via Ovid, EBSCO, or PubMed with documented syntax) indexes biomedical journal literature with MeSH headings. It remains the primary source for most clinical reviews but does not comprehensively cover European journals, many conference abstracts, or drug trade names.

Embase adds Emtree thesaurus terms, broader European coverage, and conference proceedings. For pharmacological interventions, device trials, or reviews where Embase is accessible through your institution, searching both MEDLINE and Embase is standard.

Cochrane CENTRAL (Cochrane Central Register of Controlled Trials) aggregates trial reports from multiple sources including handsearching. It is essential for intervention reviews and often surfaces trials not indexed elsewhere.

Beyond the core trio, add databases matched to your question: CINAHL for nursing and allied health, PsycINFO for mental health, AMED for complementary medicine, trial registries (ClinicalTrials.gov, ISRCTN, WHO ICTRP), and grey literature sources where pre-specified in the protocol.

3. Building the search string

Organise the strategy into concept blocks aligned with PICO elements. Population and intervention blocks usually carry the most synonyms; comparator is often omitted from the search (screening handles it). Outcome terms are included when they improve sensitivity without exploding the yield unmanageably.

Within each block, combine synonyms with OR. Combine blocks with AND. Use truncation ($ or * depending on platform) and adjacency operators where supported. Document which operator syntax applies — Ovid MEDLINE differs from PubMed.

For MeSH, explode broader headings where appropriate and include specific subheadings (e.g. /drug therapy). Pair each MeSH term with a free-text search in title/abstract to capture records not yet indexed. Repeat the pattern with Emtree in Embase — MeSH and Emtree are not identical.

Add a study design filter only if pre-specified and validated for your topic. Cochrane highly sensitive RCT filters exist but may miss quasi-randomised trials; document any filter and run sensitivity analyses without it if yield is low.

• Block 1 — population (diagnosis, condition, setting)
• Block 2 — intervention or exposure (generic and trade names)
• Block 3 — optional outcome terms if they improve sensitivity
• Block 4 — study design filter (if validated and pre-specified)
• Document line numbers for each database export

Tip: Pilot the string: run it, scan the first 50 hits for obvious misses, then refine before peer review. Do not refine after screening based on which studies you wish to include.

4. Peer review and librarian involvement

The PRESS (Peer Review of Electronic Search Strategies) checklist recommends independent review of search strategies before execution. University librarians trained in systematic review methods can identify missing synonyms, incorrect MeSH explosions, and syntax errors that students miss.

Provide your librarian with the registered protocol, PICO, and any preliminary strings. Request feedback on concept blocks, not just spelling. If your institution offers a systematic review service, use it early — rework after screening is far costlier.

For coursework, a documented email exchange or meeting notes with a librarian satisfies reproducibility expectations better than an undocumented solo search.

5. Supplementary search methods

Database searching alone rarely achieves 100% sensitivity. Cochrane recommends citation searching (forward and backward from included studies), handsearching key journals or conference proceedings, and contacting experts where appropriate.

Reference lists of included studies and relevant systematic reviews should be screened. Forward citation tracking (e.g. via Web of Science or OpenAlex) finds newer papers citing your included set.

Trial registry searches identify ongoing or unpublished studies. Grey literature — theses, regulatory documents, conference abstracts — is covered in our dedicated grey literature guide and should be pre-specified, not added post hoc.

• Backward citation — reference lists of included studies
• Forward citation — who cited your included studies
• Handsearching — target journals missed by indexing
• Expert contact — document who was contacted and responses
• Grey literature — see grey-literature guide for thesis and report searching

6. Documenting and reporting the search (PRISMA)

PRISMA 2020 Item 7 requires reporting of all information sources, search dates, and full search strategies for at least one major database. PRISMA-S extends this with structured tables for each database, platform, and supplementary method.

Export results from each database with counts before and after deduplication. Record the reference manager or review software used (Covidence, Rayyan, EndNote). Note any limits applied — date, language, publication type — and justify them in the protocol.

If the review process exceeds 12 months, re-run searches before final submission and document the update search in an appendix. Report how many new records entered screening and whether any changed conclusions.

Structured search documentation is a core competency tested in the Strata Academy certificate course at /academy/course — Part A covers protocol design and Part B walks through screening and synthesis with supervisor-ready outputs.

7. Common student errors

Copying a search string from a published review without adapting concepts to your PICO. Published strings reflect someone else's question, date limits, and database access.

Over-restricting with outcome terms in the search when outcomes were not consistently indexed in primary studies — this drops relevant trials that report unexpected secondary endpoints.

Failing to deduplicate across databases before screening, inflating PRISMA counts and wasting reviewer time.

Reporting 'PubMed search' without the actual string. Examiners and journal reviewers expect appendices with full reproducible strategies.

Frequently asked questions

Is searching PubMed enough for a systematic review?

No. PubMed is a convenient interface to MEDLINE but does not replace Embase or Cochrane CENTRAL. Most rigorous reviews search at least MEDLINE (with documented syntax), Embase where accessible, and CENTRAL, then deduplicate before screening.

Should I include outcome terms in my search strategy?

Often no — or only selectively. Many trials index poorly by outcome, and restrictive outcome blocks reduce sensitivity. Pre-specify in your protocol whether outcomes are searched; Cochrane reviews frequently omit outcome blocks and screen for eligible outcomes instead.

How do I choose between MeSH and free-text terms?

Use both in parallel within each concept block. MeSH (or Emtree) captures indexed records; free-text title/abstract terms capture records awaiting indexing or using non-standard terminology. Explode MeSH headings where broader terms apply.

When should I update my search?

Re-run all database searches before final submission if the initial search was more than 12 months ago, or if you know major trials published since the original date. Document the update search, new records screened, and any impact on conclusions.

Do I need a librarian for a student dissertation review?

Strongly recommended. Even one consultation to peer review your string against PRESS principles reduces missed studies and demonstrates methodological rigour to examiners. Many UK medical school libraries offer systematic review support at no cost to students.

Interactive walkthroughs and quizzes load when JavaScript is enabled — the checklist and tables above are fully readable without it.