Free scoping read. A written assessment of your manuscript in two business days, at no cost.

Start one

Home›Services›Physician writing›Literature Search and Citation Service

Literature Search and Citation Services

Literature Search and Citation Service

Pubrica’s literature search and citation service supports clinicians, researchers, review teams, regulatory and HTA teams and institutions with documented, reproducible searches and verified references. Our information specialists build each strategy in the database’s own controlled vocabulary, such as MeSH in MEDLINE and Emtree in Embase, test it against studies you already know belong in the results, and report every strategy with its date, interface and result count so that it can be re-run. We then select your citations, verify them against source and format them to your journal’s style.

One search at a glance · the same question, two databases Strategies rebuilt for each database

MEDLINE strategy pasted into EmbaseFails silently

Rebuilt in EmtreeRetrieves the correct set

Tested against 4 studies you already have4 of 4

Saved with date, interface and countRe-runnable

Every citation opened and read against its claimRead in full

Question → concepts → per database → test → run → cite

Every search
  • Built per database
  • Tested against known items
  • Saved with its count
  • Limits declared
  • Citations verified to source
From $90A documented search in one database, in 3–4 business days
Since 2009Searches and reference work for clinical and regulatory teams
PRISMA-SEvery strategy reported with its date, interface and result count
ISO 9001:2015Quality management certified · cert. H3CR9TCU20251212INDQM10N

Service Scope

Our Literature Search and Citation Services

Each service is available individually, whether you need a new search, a review of an existing strategy, or verification and clean-up of a reference library that does not match its sources.

Search strategy design The concepts identified from the question, the synonyms found from what other authors in the field actually wrote rather than from a thesaurus alone, the controlled terms selected from each database’s own vocabulary, and the truncation set per term. Built once and then rebuilt for each database rather than translated between them.
Search peer review Review of an existing strategy against PRESS, the published checklist for peer review of electronic search strategies. We identify translation errors between databases, missing synonyms, truncation that broadens a concept and filters that lose records. It is a quick, low-cost check that can significantly improve a review.
Running the search and deduplicating Executed across the databases the question needs, with every strategy saved with its date, interface and result count. Deduplicated with the rate reported, because the deduplication rate tells a reader how much overlap there was — and delivered as a RIS or a reference manager library rather than as a screenshot of a count.
Grey literature and trial registries ClinicalTrials.gov, the WHO ICTRP, conference proceedings, dissertations, government and agency reports, and preprint servers where the field uses them. This is where completed-but-unpublished trials are found, which is where publication bias actually lives — a funnel plot with nine studies detects very little.
Citation chasing, forward and back Backward through the reference lists of every included study and every existing review, forward through Scopus and Web of Science from the key papers. It reliably finds studies the database search missed, and the number it finds is itself a useful check on how good the strategy was.
Reference verification Each cited work opened and read against the statement it is attached to, rather than the citation matched to the sentence. These are different jobs. The second catches formatting errors; the first catches a reference that does not support the claim — which is how errors propagate through a literature for a decade, each author citing the one before.
Citation formatting and library management Formatted to your journal’s style — Vancouver, AMA, APA, Harvard, Chicago — with in-text citations inserted and the bibliography built. Delivered as an EndNote, Zotero or Mendeley library with clean records, or as RIS. Migration between managers where a project has inherited three.
Annotated bibliography and evidence summary Each paper summarised with its design, population, outcome, finding and principal limitation in a table you can actually use, rather than a list of abstracts. Where a paper is weak, the note says how and in which direction that would push its result.

Who we serve

Clients We Support with Literature Searches

We support five groups of clients, each of whom needs a search that another person could run and obtain the same result.

Clinicians and healthcare professionalsA question from practice, a departmental presentation, a background section, or a check on whether a case has been reported before. These searches usually cover one or two databases and have a fixed deadline.
Researchers and postgraduatesThe search behind a review or a thesis, where the strategy itself will be examined and has to be reported in enough detail for somebody to repeat it.
Review teams that have got stuckTeams that have screened thousands of records and suspect the search was flawed, or that have been asked by a reviewer why a relevant paper is missing. Peer review of the strategy against PRESS is the most economical way to identify the problem.
Regulatory and HTA teamsSearches inside a clinical evaluation, a performance evaluation or a payer dossier, where the search is auditable, version-controlled and has to be re-runnable to a stated date because the obligation recurs.
Institutions and departmentsOngoing search support where throughput is the constraint, with a standard approach to databases, citation style and reference manager agreed once for all projects.

Ingredient and Product-Claims Research

Ingredient validation and product-claims searching for nutraceutical, functional food and herbal formulation work, including EFSA and AYUSH compliance, is a separate product regulatory research service with a different evidence base. It is not offered through this page. Please request it by name and we will scope it separately; it is handled by our medical and regulatory writing team.

Methodology

How Our Literature Search Process Works

We test every strategy against papers you already have before running it, because this check prevents far more problems than faster screening later.

Stage one

Concept Analysis of the Research Question

Most questions need two or three concepts, occasionally four. Each concept combined with AND is a filter, and four filters routinely produce twelve records — not because the evidence is thin but because the terms have to co-occur. Where a question needs four concepts, the usual answer is that one of them belongs in the inclusion criteria rather than in the search.

Stage two

Literature-Derived Synonyms and Search Terms

Search terms are taken from the words other authors used. We read the titles and abstracts of the papers you already have and collect their terms, including older terminology, regional variants and the abbreviations authors actually use. This step largely determines the quality of the search and requires specialist judgement.

Stage three

Database-Specific Strategy Design and Known-Item Testing

We use MeSH in MEDLINE, PubMed and CENTRAL, Emtree in Embase, and free text only in Scopus and Web of Science. Each strategy is then tested against three or four papers that definitely belong in the results. If it misses one, the strategy is corrected before it is run, which takes minutes rather than the effort of screening thousands of records.

Stage four

Search Execution, Saving and Deduplication

Every strategy is saved with its date, interface and result count at every line. That is what PRISMA-S asks for and what makes a search re-runnable rather than merely described. Deduplicated with the rate reported, and delivered as a library rather than as a number.

Stage five

Citation Verification Against Source

Each reference is opened and read against the claim it supports. Then formatted to the target style, inserted in text, and delivered as a clean reference manager library. A manager reproduces whatever is in the record it imported, including its errors, so the check is against the source rather than against the record.

After delivery

Search Updates and Re-Runs

Many searches need to be repeated. A clinical evaluation under the MDR is updated rather than finished; a systematic review submitted eighteen months ago will be asked for a search refresh at revision; a guideline is reviewed on a cycle. A strategy saved properly can be re-run in an afternoon. One that was not has to be rebuilt from memory, and the numbers will not match.

Responsible Use of AI

Responsible Use of AI in Reproducible Literature Searching

The value of a professional search is that another person can run it again and obtain the same result. A tool that returns references without a documented strategy produces a reading list rather than a search.

What we use it forPrioritisation and deduplication. Suggesting controlled vocabulary a specialist then verifies against the thesaurus. Checking every citation against the record it claims to cite.
What we do not use it forGenerating references without verification, which is the single commonest failure mode and produces citations that do not exist; deciding the search concepts; and delivering a strategy nobody has peer-reviewed.
What goes in your declarationThe tool, the stage it was used at, and the human oversight applied — because journals, funders and the ICMJE January 2026 update all now expect it, and an undeclared tool found later is treated as a research integrity matter rather than a style preference.

Why we publish no accuracy figure of our own

Accuracy figures for AI tools are usually reported by the companies that sell them.

So Pubrica quotes no self-generated accuracy number, because ours would carry exactly the same conflict. What we will do is tell you which tool was used on your project, at which stage, and show you what a human changed after it.

Request a Free Search Strategy Assessment

Three papers that belong in your results are the most useful material you can send. They show the vocabulary authors in your field use and provide the test the strategy must pass before it is run. Within two business days and at no cost, we provide the concepts, the databases your question needs and a first strategy for you to review.

What you provide
The questionIn a sentence, however rough
Three known papersThat must appear in the results
The purposeReview, paper, submission or decision
Any searchStrategies already run, with their counts
AccessWhich databases your institution has
The styleJournal citation style, and which manager you use

Database access is worth saying at the start. Embase is a paid subscription many institutions do not hold, and a strategy built around it is no use to somebody who cannot run it again. Where you do not have access we will say what the search loses without it — usually European trials and conference abstracts — so that the limitation is a stated decision rather than an invisible gap.

Database-Specific Search Syntax

Search Strategy Syntax Across Six Databases

Enter your concepts once and see the strategy in PubMed, Ovid MEDLINE, Embase, Cochrane CENTRAL, Scopus and Web of Science syntax. The tool shows how the six databases differ in field tags and thesauri, and that two have no controlled vocabulary at all, which is why a strategy cannot simply be pasted from one database into another.

01

Your concepts, written out in six database syntaxes, with the limits you would actually have to declare.

Runs in your browser · nothing sent anywhere
Which database?
PubMed · search builder syntax2 concepts, combined with AND
("heart failure"[tiab] OR "cardiac failure"[tiab] OR "ventricular dysfunction"[tiab] OR "Heart Failure"[mh])
  AND ("telemonitoring"[tiab] OR "remote monitoring"[tiab] OR "telehealth"[tiab] OR "Telemedicine"[mh])

Vocabulary in this database. Controlled vocabulary is MeSH. [mh] searches the heading and explodes it by default; use [mh:noexp] where you do not want the narrower terms. [tiab] searches title and abstract; [tw] is broader and noisier.

Known-item testing

Run it against studies you already knowPick three or four papers that definitely belong in your review and check the search retrieves all of them. A strategy that misses a paper you already have is broken, and this is the only cheap way to find that out. It is also the check that catches a MEDLINE strategy pasted into Embase, which otherwise fails silently — it returns records, the count looks plausible, and it is retrieving the wrong set.

What this is and is not. It writes your concepts out in the syntax of the database you chose, with the field tags and limit syntax that database actually uses. It does not design your search: choosing the concepts, finding the synonyms other authors used, deciding where to truncate and picking the controlled terms from each thesaurus is the skilled part, and it is what an information specialist does. Truncation is deliberately left out here because the right truncation point differs by term and by database and a wrong one silently broadens or narrows the whole strategy. Controlled vocabulary differs between databases and is not translatable: MeSH in MEDLINE, PubMed and CENTRAL, Emtree in Embase, none at all in Scopus or Web of Science. Save every strategy with its date, its interface and its result count — PRISMA-S asks for it, and it is what makes a search re-runnable rather than merely described.

Limits of the Search String Builder

Design it. Choosing the concepts, finding the words other authors actually used rather than the ones you would use, deciding where to truncate and picking the right controlled terms out of a thesaurus with fifty thousand entries is the skilled part, and it is what an information specialist does. The builder above writes your concepts out in the correct syntax, which is the mechanical half. It deliberately does not add truncation, because the right truncation point differs by term and by database and a wrong one silently broadens or narrows the entire strategy — cardi* retrieves cardiology, cardiac, cardiomyopathy and cardinal. The builder also cannot suggest a synonym you have not thought of, which is the error that most often weakens reviews. For this reason the known-item test matters more than the syntax: run your strategy against three or four papers you already know belong in the results, and if it does not retrieve them, the strategy needs correcting.

Our experts

Your Literature Search Team

Your project is handled by a named team that stays with it from scoping to delivery. Your project coordinator introduces each member before work starts, and the handover records which member completed each stage.

Dr. AntonyInformation specialist

Finds what the work is built on. Writes the search line by line for each platform and hands it over dated and reproducible, so a reviewer, an editor or an assessor can run it again.

OwnsSearch and retrieval
Reads againstPRISMA-S, PRESS
Second-checked byA second information specialist
Subject expertA subject matter expert assigned from your topic, not from a rota

Decides what the work can claim. Matched to your topic rather than allocated from a rota, and advises you clearly when the evidence does not support the conclusion you expected.

OwnsQuestion, scope, interpretation
Reads againstICMJE, COPE, PRISMA
FieldMatched to your topic
Dr. DivyaPhD in Life Science · second reviewer

Reads independently of the first pass and reconciles, because most errors in this work are transcription errors rather than reasoning ones. Disagreements are logged and adjudicated, never quietly resolved.

OwnsIndependent second read
Reads againstRoB 2, ROBINS-I, QUADAS-3, JBI
Adjudicated byA third reviewer
Mr. JohnM.Sc. in Statistics · statistician

Owns anything the numbers decide. Says before the analysis exists whether the data can answer the question, and hands over the code so the whole thing can be re-run without us.

OwnsAnalysis and certainty
Reads againstGRADE, SWiM, CONSORT
ToolsR, Stata, SAS, RevMan

Sample work

Sample Work: Search Log and Reference Audit

A search log is the document that shows whether a review can be relied on. This sample is a real log with the topic changed. It includes the strategies as run, the counts at each line, the known-item test, and a reference audit in which every citation was opened and read against the sentence it supports. Use the tabs, then open the full log below.

Search log and reference audit

Six sources · 3,914 records · 62 references audited

The concepts · the strategies · the known-item test · the reference audit

Concept 1 · the population
Eleven free-text synonyms and two thesaurus headings. Four of the eleven came from reading the titles of papers the author already had, not from the thesaurus — including one term that stopped being used in about 2005 and appears in a third of the older literature.
Concept 2 · the intervention
Nine free-text synonyms and one heading. This is the concept where terminology has changed most; a search built only on the current term would have missed the entire first decade.
Concept 3 · considered and dropped
The outcome. Moved to the inclusion criteria instead, because outcomes are inconsistently reported in titles and abstracts and adding it as a third AND lost studies that clearly qualified.
Truncation
Set per term, not globally. Two terms were left untruncated because truncating them pulled in an unrelated field entirely.
Limits applied
None on language, none on date, none on study design. Each of those is a decision that would have had to be declared and defended, and none was justifiable here.
Time spent
About four hours on concepts and synonyms, before a single search was run. This is the part that decides the outcome.

Redacted extract · topic and author details changedSearch log, as delivered

A search log records whether a review can be relied on. Use the tabs to see each part.

The known-item test. Two of four known papers were missed on the first attempt, for two completely different reasons: one because the Emtree heading did not cover terminology that fell out of use twenty years ago, one because the paper was too recent to have been assigned MeSH headings at all. Neither would have been visible from the result count, and neither would have been found by screening. Total cost of finding and fixing both: about forty minutes.

The reference audit. Seven problems were found in sixty-two references, none of them visible from the citation. The most serious was a claim traced back through three papers to a source that says the opposite. This happens when authors cite an earlier paper’s citation rather than the original study.

Synonyms harvested from the literatureIncluding the terminology that stopped being used.
Rebuilt per database, never translatedTwo MeSH terms had no Emtree equivalent at all.
Tested against known items before runningWhich caught two failures nothing else would have.
Every reference opened, not matchedSeven problems, including one retraction.

Ask for the sample work and you get a search log built round a question like yours — same field, same databases, same citation style.

Literature search · sample search log

Search log, known-item validation and reference audit

A real search log with the clinical topic, the terminology and the author details changed, and the record counts adjusted so the project cannot be identified. The structure, the failures caught and the audit findings are as delivered. Ask us and we will send one built around a question like yours.

1From question to concepts

ElementDecisionWhy
Concept 1 — population11 free-text synonyms, 2 thesaurus headingsFour synonyms came from reading titles of papers the author already had, including one term that fell out of use around 2005 and appears across a third of the older literature.
Concept 2 — intervention9 free-text synonyms, 1 headingThe concept where terminology has changed most. A search built on the current term alone would have missed the entire first decade of the field.
Concept 3 — outcomeConsidered, then dropped from the searchMoved into the inclusion criteria. Outcomes are inconsistently reported in titles and abstracts, and adding it as a third AND lost studies that plainly qualified.
TruncationSet per termTwo terms were left untruncated because truncating them pulled in an unrelated discipline. A global truncation rule is how a strategy silently broadens.
Language limitNoneA language limit excludes trials that exist and does so unevenly by region. It was not justifiable here, and where it is, it must be declared in the methods and the limitations.
Date limitNone“Last ten years” is a habit rather than a rationale. There was no event in this field that justified a cut-off.
Design filterNonePublication-type indexing is applied by humans and applied inconsistently, so trial filters reliably miss trials — particularly older ones and those in smaller journals.
Time before the first search ranAbout 4 hoursThe part that decides the outcome, and the part that looks like nothing in the finished document.

2Strategies as run, with counts

SourceInterfaceVocabularyRecords
MEDLINEOvidMeSH, exploded where the tree supports it, plus free text in ti,ab,kw with adjacency operators1,182
EmbaseElsevierEmtree, rebuilt from scratch. Two of the MeSH concepts have no Emtree equivalent and are carried by free text here1,507
Cochrane CENTRALWileyMeSH. No design filter, because CENTRAL is already a trials register604
ScopusElsevierNone. Free text only, so synonym coverage carries the entire strategy391
Web of ScienceClarivateNone. Topic field search, for proceedings indexed here and nowhere else230
ClinicalTrials.gov and WHO ICTRP—Registry fields39
Reference lists and citation chasing—Backward from every included study, forward from the key papers8

Each strategy is reproduced in the appendix line by line with the count at every line, which is what PRISMA-S asks for and what makes it re-runnable. Every one is saved with the date it was executed and the interface it was executed through, because the same strategy through a different interface returns different numbers.

3Known-item validation

Known paperFirst runWhat was wrongFix
1 — recent trial, current terminologyRetrieved in 6 of 6Nothing—
2 — specialty journalRetrieved in 5 of 6Not indexed in ScopusNone needed; recorded as expected coverage
3 — older paper, pre-2005 terminologyMissed in EmbaseThe Emtree heading chosen does not cover the older term, and the free-text set did not include it eitherTwo synonyms added; re-run; retrieved
4 — published six weeks earlierMissed in MEDLINEIndexed but not yet assigned MeSH headings. Controlled vocabulary alone would never have found itFree-text arm of the strategy extended; re-run; retrieved

Two of four missed on the first attempt, for two entirely different reasons, neither visible from the result count. Total cost of finding and fixing both: about forty minutes, before anybody screened a record. This is the cheapest quality control in evidence synthesis and it is routinely skipped.

4Deduplication and yield

StagenNote
Total retrieved3,961Across all seven sources.
Duplicates removed1,03826%. Reported, because the rate tells a reader how much the databases overlapped and therefore how much the extra databases actually added.
Unique records2,923Delivered as a de-duplicated library with full records, not as a count.
Unique to Embase312The number that answers “was Embase worth the subscription” for this question. It usually is; for some questions it is not, and saying so is useful.
Unique to registries and grey literature27Including three completed trials with no publication.
Found only by citation chasing8A low number here is reassurance that the strategy was good. A high one is a warning.

5Reference audit

FindingnDetail
References checked62Each cited work opened and read against the statement it was attached to. Not matched to the sentence — read.
Does not support the claim4In each case the claim had been inherited from another paper’s citation rather than read at first hand. One traced back through three papers to a source that says the opposite.
Wrong paper cited2Reference-manager errors: right author, wrong year, different study. A manager reproduces whatever is in the record it imported.
Retracted1Retracted eleven months before the draft was written. Cited from a PDF downloaded before the retraction, which is how almost every citation of a retracted paper happens.
Formatting errors9Corrected silently. These are the ones everybody expects and the least important.
Verified and delivered55Clean records in the target style, with the RIS file so the library can be reused rather than rebuilt next time.

Seven substantive problems in sixty-two references, and not one of them was visible from the citation. The four inherited claims are the ones worth dwelling on: none of the authors in that chain did anything careless, and the error still propagated for years. Checking a reference against its source rather than against its record is a different and slower job, and it is the only thing that catches this.

Prepared by Pubrica. Pubrica is a trading name of Guires®, registered in England, Wales and India.

Request this sample

Packages

Literature Search Packages

Each package is priced on the number of databases searched, the extent of grey literature covered and the number of references verified. These figures are shown with every price so that the scope of each package is clear.

Starter

One database, documented

Short manuscripts, case reports and background sections where you need a defensible search rather than a systematic one

From $901 database · up to 25 references · 1 revision

Delivery 3–4 business days

  • ✓Strategy built in that database’s own vocabulary and tested against known papers
  • ✓Saved with its date, interface and result count
  • ✓Citations formatted to your journal’s style and inserted in text
  • ✓Delivered as a Word file and a clean reference library

Add later: more databases; an annotated summary.

View a sample Select Starter
Standard

Three databases, with a summary

Journal articles, postgraduate work and grant backgrounds needing a structured, reportable search

From $2603 databases · 15–25 sources summarised · 1 revision

Delivery 5–7 business days

  • ✓Everything in Starter, across three databases rebuilt rather than translated
  • ✓Deduplicated, with the rate reported
  • ✓A summary matrix: design, population, outcome, finding and limitation per paper
  • ✓Citations integrated into your manuscript rather than supplied as a list

Add later: grey literature; registries; citation chasing.

View a sample Select Standard
Advanced

Five databases, reported to PRISMA-S

Review articles, systematic review searches and grant proposals where the search itself will be scrutinised

From $4805 databases · grey literature · registries

Delivery 7–10 business days

  • ✓Everything in Standard, across five databases
  • ✓Grey literature and trial registries, where the question needs them
  • ✓Citation chasing forward and backward from the key papers
  • ✓Every strategy reported line by line to PRISMA-S, with counts
  • ✓The search log and the library handed over, not just the report

Add later: search peer review against PRESS; screening support.

View a sample Select Advanced
Premium

With every reference verified to source

High-impact submissions, regulatory dossiers and anything where a reader will open the references

From $7205+ databases · 30+ references opened and checked

Delivery 10–12 business days

  • ✓Everything in Advanced
  • ✓Every cited work opened and read against the claim it supports
  • ✓Citation audit: style compliance, and cross-verification of every reference
  • ✓Originality screening, with every match classified rather than counted
  • ✓Delivered as Word, RIS and a reference manager library

Search refreshes and re-runs are quoted on the saved strategy.

View a sample Select Premium

Additions are quoted alongside any tier: peer review of a strategy somebody else built, against the PRESS checklist; a scheduled search refresh for a review or a clinical evaluation, which is a recurring obligation rather than a one-off; translation and appraisal of non-English studies, which are excluded far more often for language than for relevance; reference library migration between EndNote, Zotero, Mendeley and RefWorks; deduplication of an inherited library; and screening support where your team has run out of capacity partway through.

Quality Assurance

Our Service Commitments and Limitations

A search is a claim about completeness, so we set out our commitments and limitations here before you engage us.

The Pubrica guarantee

Fully Reproducible Search Strategies

Every strategy is saved with its date, interface and count at every line, and the strategy itself is handed over. If you are asked to re-run the search in eighteen months, you or anyone else can do so.

01

Testing Before the Search Is Run

Each strategy is tested against papers you already know belong in the results. If it misses one, it is corrected before any screening takes place.

02

Delivery of All Working Files

You receive the search log, the de-duplicated library, the RIS file and the reference manager file, not only a report describing them. The search and its workings belong to you.

03

Declaration of Every Search Limit

Language, date and design filters are methodological decisions. Each one is reported in the methods with its rationale, so that reviewers can assess it.

04

Correction of Our Errors at No Cost

If an error originates with us, such as a missed database, a mistranslated strategy or an incorrectly formatted reference, we correct it free of charge and inform you first.

Service limitations. No search can find every relevant study. A well-designed search reports its limits transparently: coverage stated, limits declared, strategies published, and the known-item test showing that it retrieves what it should. We do not translate a strategy between databases and describe it as rebuilt, because MeSH and Emtree are different thesauri and a pasted strategy fails without warning. We do not apply a language, date or design limit for convenience, and where you request one we explain its effect. We do not report a count we did not obtain, and where a database was not searched we say so. We do not describe a reference as verified when it has only been formatted; these are different services, priced differently, and only verification identifies a retracted paper.

Free First Search Strategy

Send your question and three papers you already have, and we will provide the concepts, the databases your question needs and a first strategy for you to review, within two business days, at no cost and with no obligation. If you already have a strategy, we will test it against your known papers and tell you what it misses.

Database Access and Coverage

We report which databases we searched, which we did not, and why. Embase is a paid subscription many institutions do not hold; where it was not searched, the log states this and explains what the search may miss, typically European trials and conference abstracts. A stated coverage gap is a reported limitation rather than an undisclosed weakness in the review.

Retraction Checks

Every reference in an audited set is checked against the retraction record before delivery. Most citations of retracted papers occur because the PDF was downloaded before the retraction. The check takes minutes and prevents an error that cannot be corrected after publication.

Confidentiality

A non-disclosure agreement is signed before any material moves. Unpublished protocols, draft manuscripts and commercial research questions are held under the UK GDPR on access-controlled systems within a quality management system certified to ISO 9001:2015, kept only as long as the project needs them, never used to train anything and never shared outside your project team. A search question is often the most commercially sensitive thing a client sends us, and it is treated that way.

FAQs

Literature Search FAQs

Not sure whether your search is sound, or which databases your question needs? Ask an expert. No quote is required.

Chosen from the question rather than counted. For clinical questions about randomised evidence, MEDLINE, Embase and the Cochrane Library cover most of it, with trial registries for completed-but-unpublished trials. Scopus or Web of Science add coverage outside the biomedical core and let you chase citations forward. CINAHL matters for nursing and allied health, PsycINFO for mental health, and regional databases — LILACS, African Journals Online, CNKI, KoreaMed — matter considerably more than most English-language reviews acknowledge. What does not improve a search is adding a sixth database and running the same badly translated strategy in it. Two well-built searches retrieve more than five pasted ones. A useful number to ask for afterwards is how many records were unique to each database: that is what tells you whether the Embase subscription earned its place for your question, and for some questions it plainly does not.

Because the controlled vocabularies are different thesauri, not dialects of one. MEDLINE, PubMed and Cochrane CENTRAL use MeSH; Embase uses Emtree, which is Elsevier’s own thesaurus with different headings and a different tree structure; Scopus and Web of Science have no controlled vocabulary at all. A MeSH term pasted into Embase will usually map to nothing, and the crucial detail is that it fails silently — the search still runs, still returns records, and the count still looks plausible. It is simply retrieving a different set. Field tags differ too, as does limit syntax and the behaviour of truncation and adjacency operators. The only reliable protection is to rebuild the strategy in each database and then test each version against papers you already know belong in the results. In the sample above, two of four known papers were missed on the first attempt for two different reasons, and neither would have been visible from the result count.

Two checks, one free and one cheap. The free one is the known-item test: pick three or four papers that definitely belong in your results and check that the search retrieves all of them in every database. If it misses one, the strategy is wrong rather than the field being thin, and you have found that out in ten minutes. The cheap one is peer review of the strategy against the PRESS checklist, which is the published instrument for exactly this and which catches the things a known-item test does not — a truncation that has quietly broadened a concept, a Boolean operator in the wrong place, a filter losing records, a line that should be OR and is AND. Both are worth doing before the search is run rather than after four thousand records have been screened. A third signal, after the fact: if citation chasing turns up several studies your database search missed, the strategy had a gap.

Only with a reason, and always with a declaration. Restricting to English excludes trials that exist, and it does so unevenly — by topic, by region and by intervention, so the exclusion is not random noise. It is sometimes defensible on resource grounds, particularly for a rapid review with a stated deadline; it is never defensible silently. If you apply it, say so in the methods, repeat it in the limitations, and say what you think it cost. There is a cheaper middle route that most people do not consider: do not limit the search, screen the non-English titles and abstracts with whatever translation is available, and then decide. You frequently find that the non-English studies are few enough to translate properly, and a review that includes them is materially stronger. Where translation is needed, we can arrange appraisal of non-English studies as an addition.

Formatting checks that the citation is correctly written in the target style. Verification opens the cited work and reads it against the statement it supports. They are different services, priced differently, and only verification identifies substantive errors. In the sample audit above, sixty-two references contained nine formatting errors, which were corrected, and seven substantive problems: four claims the cited paper does not support, two citations of the wrong study, and one retracted paper. In each of the four unsupported claims, the claim had been taken from another paper’s citation rather than from the original source, and one traced back through three papers to a source that says the opposite. Reading the original source is the only reliable way to prevent this.

EndNote, Zotero, Mendeley and RefWorks, and we deliver RIS alongside whichever you use so the library survives you changing your mind. Two practical points matter more than the choice. First, keep one library per project rather than one per author — merging three libraries at submission is where duplicate and wrong-year citations come from, and it always happens under time pressure. Second, a reference manager reproduces whatever is in the record it imported, including the record’s errors: a wrong year in a database record becomes a wrong year in your bibliography, and the software has no way of knowing. That is why verification is against the source rather than against the record. Where a project has inherited three libraries in different formats, migration and deduplication is a service we sell on its own, and it is usually a better first purchase than anything else.

Break the question into two or three concepts — rarely more, because each concept combined with AND is a filter and four filters produce twelve records. For each concept, gather free-text synonyms from the literature rather than from a thesaurus: read the titles and abstracts of papers you already have and harvest the words those authors used, including older terminology and regional variants. Then pick the controlled terms from each database’s own vocabulary. Combine synonyms within a concept using OR, and concepts using AND. Set truncation per term rather than globally. Build the strategy separately in each database, then test each against three or four known papers and fix what it misses. Run it, save every strategy with its date, interface and line-by-line counts, deduplicate and report the rate. Then chase citations forward and backward, and search registries and grey literature if the question needs them. The order matters: everything before “run it” is where the quality is decided.

If it was saved properly, in an afternoon. If it was not, it has to be rebuilt from whatever you have — and the numbers will not match the original, which is exactly the problem you were trying to avoid. This is why a search log carries the date, the interface and the count at every line: a strategy run through PubMed and the same strategy run through Ovid MEDLINE return different numbers, and so does the same strategy through the same interface a year later. Re-running matters more often than people expect. A clinical evaluation under the EU MDR is updated rather than finished, so the search recurs on a cycle. A systematic review submitted eighteen months ago will be asked for a search refresh at revision. A guideline is reviewed on a schedule. Where we built the original, a refresh is quoted on the saved strategy and is a fraction of the first search. Where somebody else built it, the first job is usually working out what they actually ran.

Request a Free Search Strategy Review

The concepts, the databases your question needs, and a first strategy you can look at — in two business days, at no cost and with no obligation. If you already have a strategy, we will test it against your own known papers and tell you what it misses.

Get a free quote

Regulations

Check your project against this rule

  1. 1Your work
  2. 2Your details
  3. 3Send
What do you need?

Material you send is held under signed confidentiality, with access limited to the assigned team and encrypted transfer. It is not used to train anything and is not passed to third-party tools. Ask for the agreement before you send anything you would rather cover first.