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

Start one
Contents

Academy · Research Data Analytics & AI · review data sources

Sources of Data for a Systematic Review: Databases, Registries, Regulatory Records and Investigators

A systematic review draws on three kinds of source rather than one: bibliographic databases, non-bibliographic records such as trial registries and regulatory documents, and people who hold data that were never published. Most reviews search the first thoroughly, mention the second and omit the third, yet the third is where evidence capable of changing the conclusion tends to be found.

Pubrica AcademyWritten by the medical writing and biostatistics teams Reviewed byPubrica biostatistics team Published9 October 2026 Last checked9 October 2026 Reading time12 minutes Sources30 peer-reviewed

The search determines what a systematic review can conclude, yet it is the component most often described in a single sentence. A statement that “a comprehensive search of PubMed and Google Scholar was performed” gives a reader nothing to verify and, more importantly, conveys nothing about what was structurally invisible to that search. This article provides an inventory of the sources of review data, the proportion of the literature each source holds, and what each systematically misses.

Learning objectives

  • The three tiers of source that a review must consider, and the purpose each serves
  • The contents of MEDLINE, Embase and CENTRAL, and why searching one is not equivalent to searching the others
  • Whether a single-database search is defensible for a given question
  • Identification of outcome data that exist but were never published: registries, regulatory packages, theses and conference abstracts
  • When contacting investigators justifies the effort, given the typical yield
  • Recording sources so that the search is reproducible rather than merely described

Common misconceptions

  • “PubMed is comprehensive enough.” Coverage studies cited by the Cochrane Handbook found that the majority of included studies could be identified via PubMed, with a range of 75% to 92% in one analysis and 68% to 73% across topics in another. A PubMed search is therefore useful but incomplete, and the missing fraction is not random.
  • “Embase is MEDLINE with a different interface.” Embase holds more than 10 million records from approximately 3,000 journals that are not indexed in MEDLINE at all. The overlap is large; the difference is not trivial.
  • “Google Scholar can substitute for a database search.” It cannot be searched with a documented, reproducible strategy, results are not stable between runs, and there is no controlled vocabulary. It is a supplementary tool for citation chasing, not a database of record.
  • “Unpublished data is not available, so it cannot be sought.” Trial registries carry results, regulatory agencies publish review packages (FDA drug approval packages are online for drugs approved since 1997), and investigators sometimes respond to requests. The yield is low, which is a reason to allocate resources for it rather than to omit it.
  • “Grey literature means low quality.” Grey literature means not commercially published. A completed trial reported only in a thesis, a registry entry or a regulatory document is not a lower-quality trial; it is a trial the journal literature happens not to contain.

Three tiers of data source and their functions

“The search” is better understood as several activities than as one. A review draws on three structurally different kinds of source, each of which requires a different method, a different record and a different form of transparency in the report.

Bibliographic databases are indexed collections of journal literature, searchable with controlled vocabulary and documented strategies. They are the backbone of the search and the only tier that is genuinely reproducible: another team running the same strategy on the same platform on the same date should obtain the same set.

Non-bibliographic sources hold records of studies rather than records of articles: trial registries, regulatory submission and review documents, clinical study reports, data repositories and health technology assessment reports. These sources identify studies that were conducted but never published, and provide the protocol and outcome information required to detect selective reporting in the studies that were 12,15.

People and institutions are the third tier: study investigators, manufacturers, specialist librarians, subject experts and the authors of existing reviews. This tier is not reproducible in the same sense, since whether an investigator replies is not a property of the method, but it is the only route to some data, and its omission should be a stated decision rather than an oversight 3.

A review that searches tier one thoroughly and declares that it did only that is a defensible review with a stated limitation. A review that searches tier one and implies it searched everything is not. Our page on reporting guidelines for literature reviews covers what has to be stated either way.

The three tiers of source, and what each tier is and is not good for

TierExamplesWhat it gives youWhat it missesReproducible?
Bibliographic databasesMEDLINE, Embase, CENTRAL, regional and subject databases, citation indexesThe published journal literature, searchable with a documented strategyStudies never submitted or never accepted; results that differ between the paper and the study reportYes — strategy plus platform plus date reproduces the set
Non-bibliographic recordsTrial registries, regulatory approval packages, clinical study reports, HTA reports, data repositoriesStudies that exist without a paper, and the protocol record needed to detect selective outcome reportingStudies that were never registered or never submitted to a regulatorMostly — record the registry, the browse path and the date
People and institutionsInvestigators, manufacturers, trial groups, abstract authors, specialist librariansUnpublished outcome data, clarification of ambiguous reporting, missing subgroup resultsWhatever is not answered — most requests, on the available evidenceNo — but the attempt and its outcome must still be recorded

Structure compiled from the Cochrane Handbook’s treatment of searching for and selecting studies, and of obtaining unpublished data.

Size and coverage of the three core databases

For reviews of health interventions, three databases carry most of the load, and their differences are worth knowing in numbers rather than impressions 1.

MEDLINE contains approximately 30 million references to journal articles in biomedicine and health, with coverage from 1946 onwards, and more than 5,000 journals in about 40 languages are indexed for it 1. Its controlled vocabulary is the most mature in the field, which is what makes a precise, documentable strategy possible.

Embase contains more than 40 million records from 1947 onwards, covering more than 8,000 currently published journals from approximately 100 countries. Crucially for review purposes, it includes more than 10 million records from approximately 3,000 journals that are not indexed in MEDLINE. Its conference abstract coverage and its European and pharmacological strengths are the usual reasons it is searched alongside MEDLINE rather than instead of it.

CENTRAL, the Cochrane Central Register of Controlled Trials, is different in kind: it is a curated register of reports of controlled trials rather than a general bibliographic database, assembled from database searches, hand-searching and other sources. As of April 2024 it contained more than 2,000,000 records 1. For a review of randomised evidence it is the highest-yield single source relative to its size, because the filtering for trial reports has already been done.

Beyond these, the choice is driven by the question: regional databases where the relevant literature is regionally published, subject databases for nursing, psychology, allied health, rehabilitation or veterinary questions, and citation indexes for the forward and backward citation chasing described below. At this point, consultation with a specialist librarian is more valuable than additional unassisted searching; see literature search support and search strategy peer review 19,20,21.

What the three core databases contain

DatabaseSizeCoverage fromJournals and scopeWhy you would search it
MEDLINE~30 million references1946More than 5,000 journals indexed, in about 40 languagesThe mature controlled vocabulary makes a precise, documentable strategy possible
EmbaseMore than 40 million records1947More than 8,000 current journals from approximately 100 countries, including more than 10 million records from approximately 3,000 journals not in MEDLINEAdds journals MEDLINE does not index, plus conference abstract coverage
CENTRALMore than 2,000,000 records as of April 2024Assembled from searches, hand-searching and other sourcesA curated register of reports of controlled trials rather than a general databaseHighest yield per record screened for a review of randomised evidence

Figures from the Cochrane Handbook for Systematic Reviews of Interventions, Chapter 4.

~30 millionreferences in MEDLINE, from 1946 onwards
40+ millionrecords in Embase, from 1947 onwards
2,000,000+records in CENTRAL as of April 2024
10+ millionEmbase records from journals MEDLINE does not index

Database sizes as reported in the Cochrane Handbook for Systematic Reviews of Interventions, Chapter 4, on searching for and selecting studies.

Adequacy of single-database searching

This question warrants a direct answer rather than a reflexive one. The evidence cited in the Cochrane Handbook indicates that a single-database search finds most, but not all, of what a multi-database search finds 1. Coverage studies found the majority of included studies could be identified via PubMed, with a range of 75% to 92% reported in one analysis, and relatively high coverage of 68% to 73% reported across topics in another 1. These fractions are substantial but incomplete.

Two consequences follow. First, the answer depends on what the missing fraction contains. If the studies only findable in Embase are systematically different — non-English, European, pharmaceutical, reported only as conference abstracts — then missing them biases the review rather than merely shrinking it. Since Embase holds more than 10 million records from approximately 3,000 journals that MEDLINE does not index, that is a real possibility rather than a theoretical one 1.

Second, the decision is a trade-off with screening workload and should be made explicitly. Each additional database adds records to screen at a rate the team must absorb; the screening page sets out the indicative rates to plan against. The defensible positions are: search multiple databases, or search one and state plainly which database and what is therefore outside the review’s reach 3,10. The indefensible position is to search one and describe the search as comprehensive.

For rapid reviews and other deliberately abbreviated designs, a restricted search is part of the design rather than a shortcut — but it has to be declared as such. See rapid review and scoping review for designs where this is explicit.

Trial registries and regulatory documents

Registries serve two purposes in a review, and the second is more useful than the first. The obvious purpose is finding studies: a registry entry identifies a trial that was conducted whether or not it was ever published, and some registries carry summary results 25. The less obvious and more valuable purpose is comparison: a registered protocol records the outcomes the investigators stated they would measure, which is the only means of detecting that a publication reports a different set 15,18.

Regulatory documents are the other major non-bibliographic source. Drug approval packages from regulators contain study reports, statistical reviews and analyses that frequently differ in detail from the published papers of the same trials 24. FDA drug approval packages are available online for drugs approved since 1997; earlier approvals require a freedom of information request, which is slower but not impossible 24. For devices, the submission and surveillance record is the equivalent source — see our page on reviews of implantable devices and on post-market surveillance.

Registries should be searched as a named step with a documented strategy, not as an afterthought. The registration identifier should be recorded for every included study for which one can be found, and the review should report when an included study has no retrievable registration, since that absence is itself informative. Where a registered trial appears never to have been published, the review should state this; an unpublished registered trial is a gap in the evidence base of which readers need to be aware.

Contacting investigators: low yield and resource planning

Contacting investigators, manufacturers and trial groups is the part of the search most often promised in a protocol and least often performed, and its yield should be estimated realistically. In one review that attempted it, contacting trialists yielded data for six of the 38 randomised trials included 1.

This low rate is a reason to plan for contact, not a reason to omit it. Three practices distinguish a yield of six of 38 from none. Requests should be specific: a named outcome at a named timepoint, not “any additional data”. Requests should be made early, because correspondence takes months and may otherwise coincide with the submission date. Every attempt should be recorded, with the date, the recipient, the request and the response, so that the review can report the attempt even where it failed. An unanswered request that is reported constitutes evidence about the evidence base; an unanswered request that is not reported is simply missing data.

The same applies to manufacturers for device and drug questions, and to the authors of conference abstracts whose full study was never published. Most should be expected not to reply, and the denominator should be reported.

Grey literature, theses and conference abstracts

Grey literature is material not issued by a commercial publisher: dissertations, institutional and government reports, working papers, conference proceedings and abstracts, and technical reports. The category is defined by how something was distributed, not by how well it was done, and conflating the two is one of the more consequential errors in search design, because the literature that fails to get published is not a random sample of the literature that was conducted 14.

Conference abstracts present a particular difficulty. They identify studies that may never appear in full, which is precisely why they matter, but they carry too little methodological and result detail for confident extraction, and the results reported in an abstract do not always match the eventual paper. The workable approach is to treat an abstract as a signal that a study exists, use it to trigger a search for a full report and a registry entry, attempt author contact where neither exists, and decide in the protocol, not afterwards, whether an abstract-only study can contribute to the synthesis and with what limitations noted.

Theses and dissertations warrant searching for the same reason: each may contain a completed study with a full methods chapter that was never submitted to a journal. Embase’s conference coverage and dedicated thesis repositories cover much of this ground; a librarian can identify the appropriate repositories for the field.

Reference lists, citation tracking and hand-searching

These methods achieve something that database searching cannot: they identify studies that the search terms did not describe. A trial with an idiosyncratic title, an unusual intervention name or indexing that predates the relevant vocabulary can be invisible to a well-built strategy and obvious in the reference list of a relevant paper 8.

Three inexpensive techniques are available. Backward chasing: screen the reference lists of every included study and of existing reviews on the topic 8,26. Forward chasing: identify all records that cite the included studies, using a citation index. Related-record searching: apply the database’s own similarity function to the most central included studies, which sometimes surfaces records whose indexing differs from the search terms.

Hand-searching (reading the tables of contents of key journals across a date range) is laborious but still occasionally decisive for older literature and for fields whose indexing is incomplete. It is also how much of CENTRAL was built. Whether it justifies the effort depends on how much of the evidence base predates adequate indexing. The methods used should be reported in either case: the number of additional records identified by each supplementary method is a line in the flow diagram that reviewers read closely, and it is covered by the reporting requirements described on our PRISMA page 3,10.

Recording sources for search reproducibility

A search is reproducible when another team can re-run it and obtain the same result. That requires, for each database: the database name and the platform or interface used, the date the search was run, the full strategy exactly as entered including all line numbers and operators, any limits or filters applied, and the number of records retrieved 3,11. Per database, not aggregated — the combined total tells a reader nothing about where the evidence came from.

For non-bibliographic sources, record the registry or repository, the search terms or browse path, the date, and the yield. For contact attempts, record the recipient, date, request and outcome. For supplementary methods, record which method produced how many additional included studies.

Two habits reduce this burden considerably. The search log should be kept contemporaneously, in one file, rather than reconstructed at submission; a reconstructed search log requires more work and is less accurate. The raw export from each database should also be saved alongside the strategy, so that the deduplication is auditable 16. Reporting then becomes a matter of transcription from the log. Where a search requires independent scrutiny before it is run, search strategy peer review catches the structural errors — missing synonyms, a misplaced Boolean operator, a filter that silently excludes a decade — that are expensive to discover after screening has started.

The sources paragraph in a methods section

Describes a search nobody can repeat

A comprehensive literature search was performed in PubMed, Google Scholar and the Cochrane Library to identify all relevant studies published up to 2024. Additional studies were identified from the reference lists of relevant articles. Grey literature was also searched.

“Comprehensive” is a claim, not a method. No platform, no date, no strategy, no per-database counts — nothing here can be re-run or checked.

Google Scholar is named as a database. It cannot be searched with a stable, documented strategy, so its inclusion makes the search less reproducible rather than more complete.

“Up to 2024” is not a search date. A search run in January and one run in December return different sets; the reader needs the exact date.

“Grey literature was also searched” names nothing. Which registries, which repositories, which terms, what yield — all absent, so the claim is unverifiable.

No registry or regulatory search. Studies conducted but not published, and outcomes reported differently in the regulatory record, are structurally outside this review’s reach and the reader is not told.

No record of contact attempts. Unpublished data was either not sought or sought and not reported; either way the reader cannot tell which.

Reports a search that can be re-run

We searched MEDLINE via Ovid (1946 to 14 March 2024), Embase via Ovid (1947 to 14 March 2024) and CENTRAL via the Cochrane Library (Issue 3, 2024). The full strategies, exactly as run, with line numbers and record counts for each database, are in Supplementary File 1. No language or publication-date limits were applied; the MEDLINE strategy used the Cochrane highly sensitive search strategy for identifying randomised trials. Records retrieved are reported for each database separately in Supplementary File 1 and in the flow diagram, together with the number remaining after deduplication and the deduplication method. We searched ClinicalTrials.gov and the WHO ICTRP on 15 March 2024 using the intervention and condition terms listed in Supplementary File 1, and screened the Drugs@FDA approval packages for every in-scope agent approved since 1997. We screened the reference lists of all included studies and of the previously published reviews listed in Supplementary File 1, and ran forward citation searches on all included studies in Web of Science on 18 March 2024. We emailed the corresponding authors of every study with a missing outcome at the 12-week timepoint, requesting that outcome specifically; the number asked, the number replying and the number supplying usable data are reported in the results, and all contact attempts with dates and outcomes are listed in Supplementary File 2.

Each database is named with its platform, its coverage span and the exact date of the search, which is the minimum for re-running it.

Per-database record counts are given, so the reader can see where the evidence came from and the deduplication is auditable.

Registry and regulatory searching appear as named steps with dates and terms, so unpublished and differently reported studies were actually looked for.

The supplementary methods are separated and quantified, so their contribution to the included set is visible in the flow diagram.

Author contact is reported with a denominator — how many were asked, how many replied, how many supplied usable data — which tells the reader how much of the evidence base stayed out of reach.

Every claim in the paragraph points at a supplementary file, so nothing rests on the word “comprehensive”.

A methods reviewer cannot assess a search that cannot be reconstructed, so the first version obliges the reviewer either to accept the search on trust or to reject it. The second version takes longer to read but no longer to write, because it is a copy of the search log kept while the search was being run.

Key points

Key points

A systematic review draws on three tiers of source. Bibliographic databases carry the published journal literature: MEDLINE holds approximately 30 million references from 1946 onwards across more than 5,000 journals in about 40 languages, Embase more than 40 million records from 1947 including more than 10 million from approximately 3,000 journals MEDLINE does not index, and CENTRAL more than 2,000,000 curated reports of controlled trials as of April 2024. Non-bibliographic sources — registries, regulatory packages such as the FDA approval packages online since 1997, clinical study reports — hold studies that were conducted without being published and the protocol record that exposes selective outcome reporting. The third tier is people, and its yield is low: one review obtained data from trialists for six of its 38 included trials. A search of PubMed alone identifies most of what a wider search finds, with coverage reported between 75% and 92% in one analysis and 68% to 73% across topics in another — which makes a single-database search a stated limitation rather than a comprehensive search. When each database, platform, date, strategy and yield is recorded during the search, the reporting can be completed directly from the log.

Application to your study

Application to your study

  1. A search log should be started at the outset of the review. One file, one block per source: name, platform, coverage span, date run, full strategy as entered, limits, records retrieved.
  2. The database set should be decided explicitly and the rationale recorded. If only one database is searched, it should be named, the material consequently outside the review’s reach should be stated, and the search should not be described as comprehensive.
  3. Registry searching should be included as a named step. Registries should be searched with recorded terms and dates, and the registration identifier recorded for every included study, noting those without one.
  4. The regulatory record should be checked for drugs and devices within scope. Approval packages are online for drugs approved since 1997, and they frequently contain outcome detail the papers do not.
  5. Requests to investigators should be specific, made early and counted. A named outcome at a named timepoint, months before submission, with every attempt and reply logged.
  6. Backward and forward citation chasing should be conducted for every included study. The number of additional included studies produced by each supplementary method should be recorded.
  7. The strategy should be peer reviewed before it is run at scale. A misplaced operator or a filter that excludes a decade is inexpensive to correct at this stage and costly after screening.

Frequently asked questions

It finds most of the evidence, not all of it, and the appropriate answer depends on what the remainder contains. Coverage studies cited in the Cochrane Handbook found the majority of included studies could be identified via PubMed, with a reported range of 75% to 92% in one analysis and relatively high coverage of 68% to 73% across topics in another. The difficulty is that the missing fraction is unlikely to be random: Embase alone holds more than 10 million records from approximately 3,000 journals that MEDLINE does not index, which skews towards non-English, European and conference-reported work. A single-database search can be a legitimate design choice for a rapid review, provided it is declared as a restriction rather than described as comprehensive.

MEDLINE is the indexed bibliographic database: approximately 30 million references from 1946 onwards, with more than 5,000 journals in about 40 languages indexed for it, and a mature controlled vocabulary. PubMed is a free interface that searches MEDLINE plus additional records, so the two overlap heavily but are not identical, and a strategy written for one platform does not transfer unchanged to another. Embase is a separate database: more than 40 million records from 1947 onwards, more than 8,000 current journals from approximately 100 countries, and more than 10 million records from approximately 3,000 journals that MEDLINE does not index. For review purposes, the practical consequence is that MEDLINE and Embase are complementary, and that the interface used must be reported as well as the database.

CENTRAL is the Cochrane Central Register of Controlled Trials, a curated register of reports of controlled trials rather than a general bibliographic database. It is assembled from database searching, hand-searching and other sources, and as of April 2024 it contained more than 2,000,000 records. For reviews that include randomised or controlled trials, it is the highest-yield single source relative to the number of records to be screened, because the filtering for trial reports has already been performed. It is not a substitute for MEDLINE and Embase, because it is restricted by design to trial reports and will not find non-randomised evidence or the wider literature.

Three routes are available, in descending order of reliability. Trial registries identify studies that were conducted whether or not they were published, and some carry summary results; they also provide the registered outcome list, which is the means of detecting that a publication reports a different set. Regulatory records are the second route: FDA drug approval packages are available online for drugs approved since 1997, and earlier approvals can be requested under freedom of information. The third route is contacting the investigators, which is occasionally successful: one review obtained data from trialists for six of its 38 included trials. Several months should be allowed for correspondence, requests should specify a named outcome at a named timepoint, and every attempt should be reported with its outcome.

Grey literature means material that was not issued by a commercial publisher: theses, institutional and government reports, conference proceedings and abstracts, and technical reports. It is a distribution category, not a quality category. A completed trial reported only in a dissertation is not a weaker trial than one in a journal; it is a trial the journal literature does not contain, and excluding it leaves the review vulnerable to the fact that what is published is not a random sample of what is conducted. The practical caution concerns conference abstracts specifically: they carry too little methodological detail for confident extraction, and their results do not always match the eventual full report. An abstract should be treated as evidence that a study exists, prompting a search for the full report, the registry entry or the author.

Not as a database of record. It cannot be searched with a documented, reproducible strategy, the result set is not stable between runs, there is no controlled vocabulary, and the number of retrievable results is capped in practice. These limitations make it impossible to report in the form required for a systematic review search: database, platform, date, exact strategy and record count. It is useful as a supplementary tool, particularly for forward citation chasing and for finding theses and reports that indexed databases miss. It should be reported as a supplementary method, with the date and its contribution, not as one of the review’s databases.

For every bibliographic database: the database name, the platform or interface, the coverage span, the date the search was run, the complete strategy exactly as entered with line numbers, any limits or filters, and the number of records retrieved from that database alone. For registries and repositories: the source, the terms or browse path, the date and the yield. For contact attempts: recipient, date, what was requested and what was received. For supplementary methods: how many additional included studies each one produced. This information should be kept in a single log during the work rather than reconstructed at submission; a reconstructed log takes longer to produce and is less accurate, and its inconsistencies are precisely what a methods reviewer notices.

No single number is defensible. The relevant question is instead which sources could plausibly contain studies meeting the eligibility criteria, and that depends on the field, the languages in which the relevant work is published, the period from which the evidence comes, and whether the intervention is pharmaceutical, surgical, behavioural or organisational. For most health intervention reviews, the working core is MEDLINE, Embase and CENTRAL, plus subject databases where the question touches nursing, psychology, allied health or rehabilitation, plus regional databases where the literature is regionally published, plus registries. The constraint on the other side is screening capacity, which is real and should be planned for rather than discovered.

References

  1. Chapter 4: Searching for and selecting studies. Cochrane Handbook for Systematic Reviews of Interventions, version 6.5. Cochrane. 2024;6.5. read 9 October 2026. cochrane.org
  2. Higgins JPT, Thomas J, Chandler J, Cumpston M, Li T, Page MJ, Welch VA (editors). Cochrane Handbook for Systematic Reviews of Interventions, version 6.5. Cochrane. 2024;6.5. read 9 October 2026. cochrane.org
  3. Rethlefsen ML, Kirtley S, Waffenschmidt S, Ayala AP, Moher D, Page MJ, Koffel JB; PRISMA-S Group. PRISMA-S: an extension to the PRISMA Statement for Reporting Literature Searches in Systematic Reviews. Systematic Reviews. 2021;10(1):1-19. read 9 October 2026. doaj.org
  4. Reddy SM, Patel S, Weyrich M, Fenton J, Viswanathan M. Comparison of a traditional systematic review approach with review-of-reviews and semi-automation as strategies to update the evidence. Systematic Reviews. 2020;9(1):243. doi:10.1186/s13643-020-01450-2 read 9 October 2026. doi.org/10.1186/s13643-020-01450-2
  5. Jonnalagadda SR, Goyal P, Huffman MD. Automating data extraction in systematic reviews: a systematic review. Systematic Reviews. 2015;4:78. read 9 October 2026. cochrane.org
  6. Moher D, Shamseer L, Clarke M, Ghersi D, Liberati A, Petticrew M, et al. Preferred reporting items for systematic review and meta-analysis protocols (PRISMA-P) 2015 statement. Systematic Reviews. 2015;4(1):1. doi:10.1186/2046-4053-4-1 read 9 October 2026. doi.org/10.1186/2046-4053-4-1
  7. Tsafnat G, Glasziou P, Choong MK, Dunn A, Galgani F, Coiera E. Systematic review automation technologies. Systematic Reviews. 2014;3(1):74. doi:10.1186/2046-4053-3-74 read 9 October 2026. doi.org/10.1186/2046-4053-3-74
  8. Horsley T, Dingwall O, Sampson M. Checking reference lists to find additional studies for systematic reviews. Cochrane Database of Systematic Reviews. 2011;2011(8):MR000026. doi:10.1002/14651858.MR000026.pub2 read 9 October 2026. doi.org/10.1002/14651858.MR000026.pub2
  9. Bolanos F, Salatino A, Osborne F, Motta E. Artificial intelligence for literature reviews: opportunities and challenges. Artificial Intelligence Review. 2024;57(10):259. doi:10.1007/s10462-024-10902-3 read 9 October 2026. doi.org/10.1007/s10462-024-10902-3
  10. Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021;372:n71. doi:10.1136/bmj.n71 read 9 October 2026. doi.org/10.1136/bmj.n71
  11. Bramer WM, de Jonge GB, Rethlefsen ML, Mast F, Kleijnen J. A systematic approach to searching: an efficient and complete method to develop literature searches. Journal of the Medical Library Association. 2018;106(4):531-541. doi:10.5195/jmla.2018.283 read 9 October 2026. doi.org/10.5195/jmla.2018.283
  12. Mayo-Wilson E, Li T, Fusco N, Dickersin K. Practical guidance for using multiple data sources in systematic reviews and meta-analyses (with examples from the MUDS study). Research Synthesis Methods. 2018;9:2-12. read 9 October 2026. cochrane.org
  13. Munn Z, Peters MD, Stern C, Tufanaru C, McArthur A, Aromataris E. Systematic review or scoping review? Guidance for authors when choosing between a systematic or scoping review approach. BMC Medical Research Methodology. 2018;18(1):143. doi:10.1186/s12874-018-0611-x read 9 October 2026. doi.org/10.1186/s12874-018-0611-x
  14. Paez A. Gray literature: An important resource in systematic reviews. Journal of Evidence-Based Medicine. 2017;10(3):233-240. doi:10.1111/jebm.12266 read 9 October 2026. doi.org/10.1111/jebm.12266
  15. Baudard M, Yavchitz A, Ravaud P, Perrodeau E, Boutron I. Impact of searching clinical trial registries in systematic reviews of pharmaceutical treatments: methodological systematic review and reanalysis of meta-analyses. BMJ. 2017;356:j448. read 9 October 2026. cochrane.org
  16. Bramer WM, Giustini D, de Jonge GB, Holland L, Bekhuis T. De-duplication of database search results for systematic reviews in EndNote. Journal of the Medical Library Association. 2016;104(3):240-243. doi:10.3163/1536-5050.104.3.014 read 9 October 2026. doi.org/10.3163/1536-5050.104.3.014
  17. Marshall IJ, Kuiper J, Wallace BC. RobotReviewer: evaluation of a system for automatically assessing bias in clinical trials. Journal of the American Medical Informatics Association. 2016;23:193-201. read 9 October 2026. cochrane.org
  18. Jones CW, Keil LG, Holland WC, Caughey MC, Platts-Mills TF. Comparison of registered and published outcomes in randomized controlled trials: a systematic review. BMC Medicine. 2015;13:282. read 9 October 2026. cochrane.org
  19. Rethlefsen ML, Farrell AM, Osterhaus Trzasko LC, Brigham TJ. Librarian co-authors correlated with higher quality reported search strategies in general internal medicine systematic reviews. Journal of Clinical Epidemiology. 2015;68(6):617-626. doi:10.1016/j.jclinepi.2014.11.025 read 9 October 2026. doi.org/10.1016/j.jclinepi.2014.11.025
  20. Koffel JB. Use of recommended search strategies in systematic reviews and the impact of librarian involvement: a cross-sectional survey of recent authors. PLOS ONE. 2015;10(5):e0125931. doi:10.1371/journal.pone.0125931 read 9 October 2026. doi.org/10.1371/journal.pone.0125931
  21. Rethlefsen ML, Murad MH, Livingston EH. Engaging medical librarians to improve the quality of review articles. JAMA. 2014;312(10):999-1000. doi:10.1001/jama.2014.9263 read 9 October 2026. doi.org/10.1001/jama.2014.9263
  22. Elliott JH, Turner T, Clavisi O, Thomas J, Higgins JP, Mavergames C, Gruen RL. Living systematic reviews: an emerging opportunity to narrow the evidence-practice gap. PLOS Medicine. 2014;11(2):e1001603. doi:10.1371/journal.pmed.1001603 read 9 October 2026. doi.org/10.1371/journal.pmed.1001603
  23. Gross A, Schirm S, Scholz M. Ycasd - a tool for capturing and scaling data from graphical representations. BMC Bioinformatics. 2014;15(1):219. doi:10.1186/1471-2105-15-219 read 9 October 2026. doi.org/10.1186/1471-2105-15-219
  24. Turner EH. How to access and process FDA drug approval packages for use in research. BMJ. 2013;347. read 9 October 2026. cochrane.org
  25. Zarin DA, Tse T, Williams RJ, Califf RM, Ide NC. The ClinicalTrials.gov results database: update and key issues. New England Journal of Medicine. 2011;364:852-860. read 9 October 2026. cochrane.org
  26. Papaioannou D, Sutton A, Carroll C, Booth A, Wong R. Literature searching for social science systematic reviews: consideration of a range of search techniques. Health Information and Libraries Journal. 2010;27(2):114-122. doi:10.1111/j.1471-1842.2009.00863.x read 9 October 2026. doi.org/10.1111/j.1471-1842.2009.00863.x
  27. Grant MJ, Booth A. A typology of reviews: an analysis of 14 review types and associated methodologies. Health Information and Libraries Journal. 2009;26(2):91-108. doi:10.1111/j.1471-1842.2009.00848.x read 9 October 2026. doi.org/10.1111/j.1471-1842.2009.00848.x
  28. Gotzsche PC, Hrobjartsson A, Maric K, Tendal B. Data extraction errors in meta-analyses that use standardized mean differences. JAMA. 2007;298:430-437. read 9 October 2026. cochrane.org
  29. Jones AP, Remmington T, Williamson PR, Ashby D, Smyth RL. High prevalence but low impact of data extraction and reporting errors were found in Cochrane systematic reviews. Journal of Clinical Epidemiology. 2005;58:741-742. read 9 October 2026. cochrane.org
  30. Richardson WS, Wilson MC, Nishikawa J, Hayward RS. The well-built clinical question: a key to evidence-based decisions. ACP Journal Club. 1995;123(3):A12-A13. doi:10.7326/ACPJC-1995-123-3-A12 read 9 October 2026. doi.org/10.7326/ACPJC-1995-123-3-A12

Every source above is a peer-reviewed article, a publisher’s record of one, or a standard issued by the body that owns it. Each carries the date it was last read. Where a standard moves, this article is revised rather than left up with a note.

How to cite this article

Pick your referencing style and copy it. Your university or target journal may vary the punctuation — check its guide if it has one.

Pubrica Academy. (2026, October 9). Sources of Data for a Systematic Review: Databases, Registries, Regulatory Records and Investigators. Pubrica. https://pubrica.com/academy/systematic-review/what-are-the-different-sources-of-data-used-to-extract-for-writing-a-systematic-review/

This article is written to be learned from. Cite it if you are referring to what it says; if you are using the evidence behind it, cite the sources listed above instead.

Search log and source inventory

This log should be opened on the first day of searching, not in the week of submission. Once completed, it serves both as the audit trail and as the supplementary file that the reporting checklist requires.

SEARCH LOG — REVIEW: ______________________________
Information specialist / librarian involved: ______________
Strategy peer-reviewed before running?  [ ] yes, by ______  [ ] no

=== TIER 1: BIBLIOGRAPHIC DATABASES ===
Repeat this block for every database.

  Database: ____________________________________________
  Platform / interface: ________________________________
  Coverage searched: from __________ to __________
  DATE RUN (day/month/year): ___________________________
  Limits or filters applied: ____________________________
  Published search filter used (name it): _______________
  Records retrieved: ___________________________________
  Strategy, exactly as entered (paste with line numbers):
    1. ______________________________________________
    2. ______________________________________________
    3. ______________________________________________
  Raw export saved as: ________________________________
  Re-run before submission?  [ ] yes, date: ______  [ ] no

  TOTAL RECORDS, ALL DATABASES: __________
  AFTER DEDUPLICATION: __________   Method / software: __________

=== TIER 2: NON-BIBLIOGRAPHIC SOURCES ===
  Trial registries searched:
    Registry: ____________  Terms: ____________  Date: ______  Yield: ____
    Registry: ____________  Terms: ____________  Date: ______  Yield: ____
  Regulatory records:
    Agency / portal: ____________  Products checked: ______________
    Approval packages retrieved (online from 1997 for FDA drugs): ____
    Freedom of information request needed?  [ ] yes, filed ______  [ ] no
  HTA reports / repositories / clinical study reports:
    Source: ____________  Date: ______  Yield: ____
  Registered studies found with NO publication located: ____
    (list them — this belongs in the review)

=== TIER 3: PEOPLE ===
  Contact attempts (one line each):
    Date | Recipient | Study | Exactly what was requested | Reply? | Data supplied?
    ____ | ________ | _____ | ________________________ | _____ | _____
  TOTALS: asked ____  replied ____  supplied usable data ____

=== SUPPLEMENTARY METHODS ===
  Reference lists of included studies screened?  [ ] yes  Additional includes: ____
  Reference lists of previous reviews screened?  [ ] yes  How many reviews: ____
  Forward citation search: source __________  date ______  Additional includes: ____
  Related-record / similar-article search: date ______  Additional includes: ____
  Hand-searching: journals __________  years ______  Additional includes: ____

=== WHAT IS OUTSIDE THIS REVIEW'S REACH ===
  State plainly, for the limitations section:
    Databases not searched and why: ______________________
    Languages excluded, if any: __________________________
    Date range not covered: ______________________________
    Source types not searched: ___________________________

=== REPORTING CHECK ===
  [ ] Every database reported with platform, coverage span and exact date
  [ ] Full strategies in a supplementary file, as run, not summarised
  [ ] Per-database record counts given, not only the combined total
  [ ] Registry and regulatory searching described as named steps
  [ ] Contact attempts reported with a denominator
  [ ] Supplementary methods quantified in the flow diagram
  [ ] The word "comprehensive" does not appear unsupported

Pubrica AcademyWritten and maintained by Pubrica’s subject matter experts. Articles are revised when the underlying standards change, and each article states the date on which it was last reviewed.

How Pubrica can help

Pubrica builds and documents the search with an information specialist rather than a keyword list: database-by-database strategies with controlled vocabulary, peer review of the strategy before it is run at scale, and registry and regulatory searching as named steps. The deliverables comprise the completed search log, the raw exports and a clear statement of what the search could not reach, which is the element a methods reviewer typically examines first.

Find something else

Search the Pubrica Academy

Access is free and requires no account. Enter a topic, or open the full library to filter by category, sub-topic or resource type.

On receipt of a review question, Pubrica provides a database-by-database search strategy with the search log already established.

A scoping assessment is free and takes two business days. It provides a written assessment of what the work requires and which specialist would undertake it.

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.