Managing References and Search Records

Overview

A literature search produces records from multiple databases and other sources. These records must be imported, labelled, deduplicated, screened, linked and retained in a way that preserves where they came from and what happened to them. Reference management is therefore part of methodological transparency.

A reliable evidence library separates raw search exports from working files, preserves source tags, documents deduplication and distinguishes one study from its multiple reports. Reference-management software such as Zotero or EndNote can support these tasks, but no software removes the need for planned procedures and manual verification.

Why this matters

Poor record management can produce incorrect PRISMA counts, repeated screening, accidental deletion of unique records, loss of full texts and uncertainty about which search version was used. It can also make an update impossible because the team cannot reconstruct the original evidence base.

Deduplication is particularly sensitive. Database records may differ in punctuation, author order, title wording, DOI format or publication stage. Automated matching can miss duplicates or wrongly merge distinct records. Published evaluations show that deduplication algorithms vary in sensitivity and specificity [1,2].

Records, reports and studies

These terms should not be used interchangeably.

• A record is a citation or search result imported from a source.
• A report is a document or data source describing a study, such as an article, abstract, thesis or registry entry.
• A study is the underlying investigation involving a defined set of participants, records, specimens or settings.

One record may be duplicated across databases. One study may have several non-duplicate reports. Deduplication operates mainly at the record level; study linking occurs after the team has enough information to recognise related reports.

Building the evidence library

A practical library usually contains layers.

Raw search exports

Save each database export exactly as received before editing. Use a stable naming convention, for example:

2026-07-14_MEDLINE_PubMed_postpartum-followup_1284.nbib
2026-07-14_Embase_postpartum-followup_1941.ris

Raw files allow recovery if later processing is incorrect.

Master reference library

Import records into a designated project library. Avoid mixing the review with personal citations. Create groups or collections for raw imports, deduplicated records, screening stages, included studies and linked reports.

Full-text and supplementary files

Store PDFs and other documents using consistent filenames linked to the reference record. Retain supplementary files, protocols, registry pages and correction notices where relevant.

Structured review data

Screening decisions, exclusion reasons, extraction data and appraisal results may be managed in specialised software or structured spreadsheets. These files should share stable record or study identifiers with the reference library.

Preserving provenance with source tags

Before deduplication, apply a source tag or custom field to every import. Useful information includes:

• database or source;
• platform;
• search date;
• search version;
• import batch; and
• supplementary-search method.

When records are merged, retain all source tags. This allows the team to calculate retrieval by source, investigate missing batches and reconstruct the flow of records.

Minimum record-management fields

illustrative

Table 1. Core fields for an evidence library

Field	Example	Purpose
record_id	R000184	Stable citation-level identifier
source_tags	MEDLINE; Embase	Provenance before deduplication
import_batch	2026-07-14-v1	Links record to search export
screening_status	Full text	Current workflow stage
study_id	S0027	Links several reports of one study
report_role	Main results article	Describes relationship within study
file_status	PDF obtained	Tracks document retrieval
notes	Correction published	Preserves important context

Note: Software implementation may differ, but the underlying functions should remain available.

Deduplication workflow

Deduplication should occur after all planned batches for that stage are imported and source-labelled.

Step 1: Normalise where appropriate

Software may standardise DOI formatting, page ranges or title punctuation. Do not overwrite raw files. Normalisation improves matching but can also conceal differences if applied carelessly.

Step 2: Run automated or rule-based matching

Common matching fields include DOI, title, author, year, journal and page number. A DOI match is strong but not infallible; metadata errors and corrections occur.

Step 3: Review probable matches

Inspect records that differ slightly. Decide which citation retains the most complete metadata and all source tags. Keep both when uncertainty remains.

Step 4: Search for residual duplicates

Sort by title, first author, year or DOI and inspect near matches. Early-online and final issue versions may have different dates or pagination.

Step 5: Record counts and method

Document the software, version, rules, number removed automatically, number removed manually and number remaining.

DECISION GUIDE

Are these records duplicates?

Consider them likely duplicate records when they describe the same report and match on several strong fields.

Do not merge automatically when:

• titles are similar but populations differ;
• one item is a protocol and another is a results article;
• one is a conference abstract and another is a full report;
• one is a correction, commentary or secondary analysis;
• author and year match but the report title differs; or
• the metadata are too incomplete to establish identity.

Linking multiple reports of one study

Study linkage usually becomes clearer during full-text screening and extraction. Assign one study identifier and retain separate report identifiers. Identify the role of each report:

• protocol;
• registry entry;
• primary results;
• follow-up results;
• subgroup analysis;
• process evaluation;
• qualitative component;
• correction or retraction notice; and
• secondary analysis.

Use all linked reports to understand methods and outcomes, but avoid counting the same participants more than once in synthesis.

Search logs and version control

A search log should include:

• review question and protocol version;
• database, platform and coverage;
• complete search strategy;
• search date and result count;
• export filename and format;
• searcher;
• peer-review status;
• deduplication batch and count;
• updates or amendments; and
• notes on errors or reruns.

Search strategies change during development. Label drafts clearly and identify the version used for the reported review. PRISMA-S and PRISMA 2020 require transparent reporting of information sources, search dates, record handling and study-selection flow [3,4]. When a search is rerun, preserve both the original and update files and record the date range or method used to identify new records.

PROTOCOL CHECKLIST

Evidence-library setup checklist

• Create a dedicated project library and folder structure.
• Save untouched raw exports.
• Tag every import by source and date before deduplication.
• Define record, report and study identifiers.
• Specify deduplication rules and manual verification.
• Define file-naming and PDF-storage conventions.
• Create a search and amendment log.
• Establish backup and access permissions.
• Test export to the selected screening platform.
• Document responsibility for updates and final archiving.

Backups, access and security

Maintain at least one independent backup and avoid relying on a single local device. Cloud synchronisation is not identical to backup because accidental deletion may synchronise across devices. Define who can edit, merge or delete records.

Most bibliographic metadata are not sensitive, but author correspondence, confidential reports or unpublished data may require controlled access. Do not upload restricted documents to third-party platforms without permission.

Reproducible evidence objectsMMed extension

For a durable review, treat search strategies, raw exports, deduplication logs, screening data and study links as research objects. Where permitted, deposit non-copyrighted materials such as strategies, blank forms, code and de-identified decision logs in a stable repository. FAIR principles encourage data and metadata to be findable, accessible, interoperable and reusable, while recognising legitimate access restrictions [5].

Worked example

WORKED EXAMPLE

Organising the postpartum follow-up review

Import: The team imports MEDLINE, Embase, CINAHL, registry and conference records into separate source-labelled collections.

Raw preservation: All original .nbib and .ris files are stored read-only.

Deduplication: Exact DOI and high-confidence title matches are merged first. Near matches are reviewed manually. The accepted manuscript and final article are treated as duplicate records of the same report, retaining the final citation and all source tags.

Study linkage: The registry, conference abstract, main article and correction remain separate reports linked to study S0012.

Audit: The log records 3,482 imported records, 1,104 duplicate citations removed and 2,378 unique records sent to screening. Counts are illustrative.

Update: Six months later, the update search is imported as a new batch and compared with the existing library without overwriting the original search.

Appraising evidence management

APPRAISAL CHECKLIST

Reference-management appraisal questions

• Were raw exports retained?
• Was every import tagged before deduplication?
• Is the deduplication method documented?
• Were uncertain matches manually verified?
• Are records, reports and studies distinguished?
• Can multiple reports of one study be identified?
• Are search versions, updates and amendments traceable?
• Are backups, permissions and file storage adequate?
• Can flow-diagram counts be reconstructed from the records?

Suggested protocol or reporting wording

Common mistakes

Knowledge check

What is the difference between a duplicate record and a multiple report?

Answer and reasoning

Duplicate records are repeated citations to the same document, often retrieved from different databases. Multiple reports are distinct documents—such as a registry entry, abstract and article—that describe the same underlying study. Duplicate records may be merged; distinct reports should be retained and linked.

Why should source tags be added before deduplication?

Answer and reasoning

Once duplicate citations are merged, the team still needs to know every source that retrieved the record. Source tags preserve provenance and support source-level counts, troubleshooting and search evaluation.

Why is conservative deduplication preferable when two records are uncertain?

Answer and reasoning

Deleting a unique eligible record is usually more harmful than temporarily retaining a duplicate. The uncertain pair can be resolved during later screening when more information is available.

What should be preserved when an updated search is run?

Answer and reasoning

Preserve the original search files, strategy, date and library; import the update as a separately labelled batch; document how new records were identified and deduplicated against the existing evidence base.

Key take-home messages

SUMMARY

Key take-home messages

• Evidence management is part of methodological transparency.
• Preserve raw exports and tag every source before deduplication.
• Distinguish records, reports and underlying studies.
• Automated deduplication requires manual verification and an audit trail.
• Stable identifiers, search logs, backups and version control make a review reproducible and updateable.

References

1. Rathbone J, Carter M, Hoffmann T, Glasziou P. Better duplicate detection for systematic reviewers: evaluation of Systematic Review Assistant-Deduplication Module. Syst Rev. 2015;4:6. https://doi.org/10.1186/2046-4053-4-6
2. Bramer WM, Giustini D, de Jonge GB, Holland L, Bekhuis T. De-duplication of database search results for systematic reviews in EndNote. J Med Libr Assoc. 2016;104(3):240–243. https://doi.org/10.3163/1536-5050.104.3.014
3. Rethlefsen ML, Kirtley S, Waffenschmidt S, Ayala AP, Moher D, Page MJ, et al. PRISMA-S: an extension to the PRISMA Statement for Reporting Literature Searches in Systematic Reviews. Syst Rev. 2021;10:39. https://doi.org/10.1186/s13643-020-01542-z
4. 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. https://doi.org/10.1136/bmj.n71
5. Wilkinson MD, Dumontier M, Aalbersberg IJ, Appleton G, Axton M, Baak A, et al. The FAIR Guiding Principles for scientific data management and stewardship. Sci Data. 2016;3:160018. https://doi.org/10.1038/sdata.2016.18