# How to Set Up a Shared Reference Library for a Systematic Review Team By the end of this guide you will have a working shared reference library that every member of your systematic review team can import into, deduplicate, tag with screening decisions, and export to a screening platform or PRISMA flow count. You will need a reference management program installed on each team computer (this walkthrough uses Zotero because it is free and supports group libraries), exported search results from your databases in RIS format, a screening tool account such as Rayyan or Covidence if your protocol uses one, and a short written agreement among coauthors about who holds each role. A systematic review lives or dies on the integrity of its reference records. Published reviews routinely pull in enormous volumes of records. A review of teamwork and implementation in healthcare settings identified 10,489 results and reduced them to 58 included articles [1]. A review of patient perspectives on robotic surgery screened 7,889 references down to 18 included studies [2]. A meta-epidemiological review of kidney transplant biomarkers critically appraised 804 studies with expert reviewers [3]. At that scale, an email attachment of PDFs and a spreadsheet of citations will fail within days. A shared reference library is the infrastructure that prevents that failure. ## What a Shared Reference Library Actually Does ![Flowchart showing database exports and grey sources merging into the group library, then deduplication, screening export and a frozen snapshot.](/article-images/illustrations/shared-reference-library-systematic-review-record-intake-path.webp) A shared reference library is a single structured database of citations, abstracts, PDFs, and tags that every reviewer reads from and writes to. It is distinct from three other tools people often confuse with it. A **reference manager** (Zotero, Mendeley, EndNote, Paperpile) stores and organizes references and lets teams share them. A **screening platform** (Rayyan, Covidence) is built for blinded title and abstract screening by two independent reviewers and imports records from a reference manager. A **search log spreadsheet** records dates, database strings, and result counts for the methods section. All three connect, but none replaces the others. The reference manager is the source of truth for every record the team has ever retrieved. The connection works in both directions. Reviews export records to screening platforms, and they also receive records back. Many published reviews supplement database searches with citation chasing. The renal fat imaging review screened reference lists during its search [4]. The German emergency care review conducted reference- and cited-by-searches alongside database searches [5]. The digital health economic evaluation protocol plans both backward citation searching of reference lists and forward citation tracking, plus contacting study authors for missing data [6]. Every one of those retrieved records has to land somewhere structured. That somewhere is the shared library. ## Reference Management Software: A Citation Manager Comparison Choose one program and use it for the whole review. Switching mid-review costs days of cleanup. The table below summarizes the mainstream options for review teams. | Software | Cost model | Sharing model | Strengths for review teams | Watch out for | |--|--|--|--|--| | Zotero | Free, open source | Group libraries with owner, admin, read/write, and read-only roles | Unlimited group members in principle, strong browser connector, easy RIS import and export, duplicate detection pane | Storage quota applies to attached PDFs on the free tier, so link PDFs in a shared drive instead | | Mendeley Reference Manager | Free with an Elsevier account | Private groups with invited members | Good PDF annotation, integrates with Elsevier tools | Group features have been pared back over the years, verify current limits before committing | | EndNote | Paid desktop license plus online component | Shared desktop libraries and EndNote Web libraries | Deep Word integration, robust deduplication settings, handles very large libraries | Steeper learning curve, sharing requires everyone on compatible versions | | Paperpile | Paid subscription | Shared folders tied to Google accounts | Clean Google Docs integration, simple interface | Requires Google ecosystem, fewer structured deduplication controls | For a team with no budget, Zotero is the default recommendation because group libraries are free, roles are granular, and the RIS round trip to screening tools is reliable. Teams whose institutions already license EndNote often stay with it because the deduplication options are more configurable, which matters when database exports overlap heavily. ## Setting Up a Shared Reference Library for a Systematic Review, Step by Step The walkthrough below uses Zotero with concrete commands and expected outputs. The logic transfers directly to EndNote or Mendeley. ### Step 1. Register the review and lock the search plan before importing anything Create the protocol first, not after the library. The cancer and incarceration protocol specifies its databases (MEDLINE, EMBASE, Cochrane Library, CINAHL, Web of Science, and Sociological Abstracts) and the planned search date in the protocol itself, before any records exist [7]. The emergency care review registered in PROSPERO in 2023 and documented that the final search ran in April 2025 [5]. Your shared library inherits this discipline: every record that enters it must be traceable to a search described in your methods. Write down, before Step 2, the database list, the date each search will run, and the export format. Registration in PROSPERO or OSF gives you a citable anchor for these decisions. ### Step 2. Create the group library In Zotero: 1. Create an account at zotero.org with an institutional email. 2. Go to **Groups**, then **Create a New Group**. 3. Name it with the review registration prefix plus a short title, for example `CRD420251035771-ectopic-breast-ca`. Published reviews cite their PROSPERO registration number in the methods section, and naming the library after it removes all ambiguity [8]. 4. Under **Group Type**, choose **Private**. Public group libraries are visible to anyone on the web, which is inappropriate for unpublished screening data. 5. Under **Library Reading**, set members to read-only in the group settings and grant write access individually. Expected result: a group URL of the form `zotero.org/groups/XXXXXX/review-name`, visible on your account page and shareable by invitation email. Now assign roles. Zotero group roles are Owner, Admin, and Member, with Member permissions set to read-only or read/write. A sensible split for a review team is one admin (the information specialist or lead author), read-write for the two or three people who import and tag records, and read-only for senior investigators who only need to check work. ### Step 3. Export records from every database in one consistent format Run each search in each database and export in RIS format. RIS is the tagged plain text format that every reference manager and screening platform reads. In PubMed, run the search, select the results, choose **Send to**, then **Citation manager**, then **Create file**. In Embase, Ovid, Scopus, and CINAHL the export dialog offers RIS explicitly. Export in batches of 500 to 2,000 records rather than one giant file, because large single exports time out. A single RIS record looks like this: ``` TY - JOUR AU - Oh, E AU - Mueller-Alcazar, A TI - Effects of digital communication tools on patients, family members and health care professionals in adult ICUs T2 - Crit Care PY - 2026 DO - 10.xxxx/example ER - ``` Two export settings matter. First, include abstracts in the export, because title and abstract screening runs on abstracts. Second, keep each database export in its own file. You will tag by source in Step 5, and merged source files destroy that possibility. ### Step 4. Merge batch files from the same database and count everything If a database produced several batch files, concatenate them on the command line before import: ```bash cat medline_batch1.ris medline_batch2.ris medline_batch3.ris > medline_all.ris grep -c "^TY - " medline_all.ris ``` The `grep -c` command counts records by counting the `TY` tags, since every RIS record begins with one. Expected output: ``` 3487 ``` That number, 3,487 records from MEDLINE, is the exact figure that goes into your PRISMA flow diagram for that database. Record it in your search log now. Teams that skip this step later reconstruct counts by hand, which is where flow diagram discrepancies come from. ### Step 5. Import into the group library and tag by source In Zotero, open the group library, create a collection named for the database and date (for example `MEDLINE 2026-01-15`), then use **File, Import** on `medline_all.ris`. In the import options, place imported items into the new collection. Before importing, add a line to your Zotero import preferences so that automatic tags are saved. Better still, rename each export file with a short prefix and use a tag reflecting it: - `_medline` for MEDLINE records - `_embase` for Embase records - `_cinahl` for CINAHL records - `_scopus` for Scopus records - `_grey` for registry hits, reference list items, and other non-database sources The source tag survives deduplication and answers a question peer reviewers ask: where did this record come from? Real reviews draw from many places. The digital communication tools review searched MEDLINE, CINAHL, PsycINFO, PSYNDEX, the Cochrane Library, PROSPERO, plus reference lists and trial registries [9]. The Ethiopian informed consent review added national university databases and Google Scholar to the major databases [10]. The malaria schistosomiasis review included ProQuest and publisher platforms such as Wiley Online Library and SpringerLink [11]. Each of those sources deserves its own tag. ### Step 6. Deduplicate before screening Overlapping databases mean overlapping records. A MEDLINE and Embase combination typically produces substantial overlap, and every duplicate a reviewer screens is wasted effort. Zotero ships with a duplicate detector. In the group library, select **Duplicate Items** in the left pane, review each candidate pair, and merge. Zotero merges the records and preserves tags from both. For a first-pass count of duplicate DOIs across a merged file, use this: ```bash grep -i "^DO - " medline_all.ris | tr 'A-Z' 'a-z' | sed 's/ *$//' | sort | uniq -d ``` Expected output is a list of DOIs appearing more than once: ``` 10.1186/s12913-024-11223-4 10.1016/j.jclinepi.2023.09.002 ``` Each line is a candidate duplicate. Not every matching DOI pair is a true duplicate (correction notices and reprints share DOIs with quirks), so treat the list as a work queue, not a verdict. Two rules make deduplication safe. Back up the library immediately before the deduplication pass, because merges are hard to reverse in bulk. And deduplicate once, at a defined point, rather than continuously, because you need the pre-deduplication count for the PRISMA flow diagram. Screening platforms also run their own deduplication, so document which pass produced your official figures. ### Step 7. Define a screening tag vocabulary in writing Decide the tag set before the first reviewer opens a record, and write it in the team handbook. A workable minimum set: - `include` for records passing title and abstract screening - `exclude` for records failing screening - `maybe` for records requiring a second reviewer - `excluded-population`, `excluded-design`, `excluded-outcome`, `excluded-language` for reason codes - `fulltext-pending`, `fulltext-included`, `fulltext-excluded` for the full text stage - `data-extracted` once a study has been extracted Reason codes map directly onto the PRISMA 2020 flow diagram, which requires counts of records excluded at screening and full text stages with reasons. One shared vocabulary means one person can count exclusions by filter and produce the flow numbers in minutes. ### Step 8. Build in citation searching, registries, and alerts Database searches alone do not satisfy modern methods sections. Build these retrieval channels into the same library. **Backward citation searching** means screening the reference lists of included studies. **Forward citation tracking** means finding newer papers that cite your included studies. Both are standard components of published and planned reviews: the economic evaluation protocol specifies both explicitly, with author contact for missing data [6], and the emergency care review performed reference- and cited-by-searches [5]. Import every retrieved record into the shared library with the `_grey` or `_citsearch` tag so they are countable in the flow diagram. **Trial registries and grey literature.** The digital communication tools review screened reference lists and trial registries for additional and ongoing studies, and updated its search from 2023 through July 2025 [9]. Registry records are not journal articles, so import them as report or document item types rather than journal articles, and tag them `_registry`. **Search updates.** Long reviews need a rerun strategy. The gestational diabetes AI prediction review supplemented its seven-database search with backward and forward reference screening and biweekly alerts to capture newly published studies [12]. Configure the same alerts inside each database (PubMed saves searches with email or RSS alerts), and import alert hits into a dated collection such as `ALERTS 2026-03-01`. Your protocol should state the original search date and the update date, as the ICU communication review did with searches updated to July 2025 [9]. ### Step 9. Export to the screening platform When the deduplicated library is ready, select all records, right-click, choose **Export Collection**, pick RIS format, and include abstracts. Import that file into Rayyan or Covidence. Screening then happens outside the reference manager, with blinded independent reviewers. The emergency care review screened titles and abstracts and assessed full texts with two independent reviewers, then had a second reviewer check data extraction [5]. The reference manager's job during screening is passive: it stays intact while the screening platform tracks decisions. When full text screening finishes, mark final decisions back in the reference manager using the Step 7 tags so the library reflects the final included set. ### Step 10. Freeze the library and document the flow Once all records are screened, take a snapshot. In Zotero, export the full group library as RIS and as BibTeX, and store both with the review's data package. Record four counts for the PRISMA flow diagram: records identified from databases, records identified from other methods, duplicates removed, and records screened. The robotic surgery review's methods section reports exactly this kind of chain, from 7,889 identified references through design-based evaluation to 18 included studies [2]. Your library, tagged and countable, is what makes those numbers reproducible. ## Shared Reference Library Tips That Prevent Rework ![Four cards listing one library per paper, collections for structure, linked PDFs in a shared drive, and weekly backup exports.](/article-images/illustrations/shared-reference-library-systematic-review-rework-prevention.webp) - **One library, one record per paper.** If two reviewers need the same PDF, it should already be there. Duplicated PDFs in a shared drive are how version chaos starts. - **Never edit master records during data extraction.** Fixing a typo in an author name after extraction begins can orphan the extracted data mapping. Log corrections and apply them once, deliberately. - **Use collections for structure, tags for status.** A collection answers "which database and date produced this record." A tag answers "what happened to it in screening." Mixing the two produces libraries nobody can query. - **Keep PDFs in a linked shared drive.** Zotero's free storage quota fills quickly with hundreds of full texts. Store PDFs in the team's institutional drive and link the folder in Zotero settings, so the reference record and the PDF stay associated without consuming group storage. - **Run a weekly backup export.** A 30 second RIS export each week means a sync failure or accidental bulk deletion costs you days, not the whole review. - **Agree on author name formats at the start.** Duplicate detection matches on author, title, and year. If one reviewer's imports carry initials and another carries full names, real duplicates evade detection and you screen the same paper twice. ## Common Mistakes and Limitations The most common mistakes in a shared reference library for a systematic review are predictable, and each has a direct fix. **Importing into personal libraries first.** A reviewer imports 2,000 records into a personal account, screens them there, and plans to migrate later. Migration never happens cleanly. The fix is procedural: import directly into the group library, always. **Skipping the pre-deduplication count.** Teams deduplicate immediately and cannot reconstruct how many records the searches identified. Fix: the `grep -c` count in Step 4 happens before any deduplication, and the number goes into the search log the same day. **Merging duplicates in the wrong place.** Screening platforms deduplicate internally, and so does the reference manager. If both run, the team reports two different duplicate counts. Fix: decide in the protocol which pass is authoritative and document it. **Treating the reference manager as a screening tool.** Reference managers store decisions as tags but do not support blinded dual screening or conflict resolution. Reviews that screen inside the reference manager lose the independent-reviewer structure that published methods describe [5]. Fix: screen in a platform built for it. **No source tagging.** Without source tags, the team cannot answer "how many records came from CINAHL," which many journals and peer reviewers ask. Fix: the underscore-prefixed tag convention in Step 5. **Uncontrolled write access.** Everyone with write access can bulk-delete collections. A single misclick on a 5,000 record library is a disaster. Fix: read-only roles for most members, write access for two named people, weekly exports as the safety net. **Letting the library drift from the protocol.** A protocol commits to specific databases and dates [7]. If a team silently adds a database mid-review, the methods section and the flow diagram stop matching. Fix: any retrieval channel added later gets documented, dated, and tagged, then reported as an update in the methods. A final limitation applies to any tool: the reference manager guarantees traceability, not quality. A complete, deduplicated, perfectly tagged library can still contain flawed primary studies, which is what critical appraisal tools are for. The reference library's job is to ensure no eligible study goes missing and every decision is auditable. ## Frequently Asked Questions ### Which citation manager should a systematic review team use? Zotero for budget-limited teams, EndNote when the institution already licenses it and the team wants configurable deduplication, Mendeley when the team lives in the Elsevier ecosystem, and Paperpile for Google Docs based writing. The comparison table above summarizes the tradeoffs. Pick one before the first search runs and commit for the whole review. ### How do we get references from PubMed into a shared reference library? Run the search in PubMed, use **Send to, Citation manager** to download an RIS file, then import that file into the group library collection named for the database and date. Export in batches and count records per batch with a command like `grep -c "^TY - " file.ris` so your PRISMA counts exist before deduplication. ### When should we remove duplicates? Run a first deduplication pass in the reference manager after all initial database imports are complete, and record the pre-deduplication and post-deduplication counts for the PRISMA flow diagram. Screening platforms may run a second pass, so document which pass produced the official figures in your methods. ### Can screening decisions be made directly in the shared reference manager? Basic inclusion and exclusion can be tracked with tags, but blinded dual screening, conflict resolution, and reason-code reporting belong in a dedicated screening platform such as Rayyan or Covidence. Use the reference manager as the record store and the screening platform for decisions, then mirror final decisions back into the library tags. ### How do we handle citation searching and grey literature? Import every record from reference list screening, cited-by searches, trial registries, and author contact into the same shared library, tagged with a source code such as `_citsearch` or `_registry`. Backward and forward citation searching are documented components of contemporary review methods, and registry screening is common in ICU and intervention reviews. ### How do we keep the library current during a long review? Set saved search alerts in each database, run them on a fixed schedule such as biweekly, and import hits into dated alert collections. State the original search date and the update date in the protocol and the final manuscript, as published reviews do when reporting updated searches. ### What permissions should team members have in the group library? Give the information specialist or lead author admin rights, give the two or three people responsible for importing and tagging read-write access, and give everyone else read-only access. Restricting write access prevents accidental bulk deletions and keeps the record set stable during screening. ### How does the shared library produce the PRISMA flow diagram numbers? Every count in the flow diagram comes from the library and its logs. Records identified equals the sum of the `grep` counts per database plus tagged records from other sources. Duplicates removed equals the pre-deduplication count minus the post-deduplication count. Exclusions at each stage come from the screening reason tags. Because each number has a defined origin, a peer reviewer or journal can verify the flow without trusting anyone's memory. ## Sources 1. [Teamwork and implementation of innovations in healthcare and human service settings: a systematic review.](https://pubmed.ncbi.nlm.nih.gov/39010100/) 2. [From expectations to experiences: a systematic review of patient and public perspectives on robotic surgery.](https://pubmed.ncbi.nlm.nih.gov/40810856/) 3. [Prognostic Biomarkers in Kidney Transplantation: A Systematic Review and Critical Appraisal.](https://pubmed.ncbi.nlm.nih.gov/38053242/) 4. [Renal Fat Accumulation Assessed by MRI or CT and Its Association with Clinical and Metabolic Disorders: A Systematic Imaging Review.](https://pubmed.ncbi.nlm.nih.gov/40566049/) 5. [How do interventions in primary care or in emergency departments impact less urgent patients' emergency care utilization? A systematic review.](https://pubmed.ncbi.nlm.nih.gov/41645205/) 6. [Economic evaluations of digital health interventions on maternal, newborn and child health in low-income and middle-income countries: a systematic review protocol.](https://pubmed.ncbi.nlm.nih.gov/42097650/) 7. [Cancer prevalence, incidence, mortality and survival among people with past or current incarceration globally: a systematic review and meta-analysis protocol.](https://pubmed.ncbi.nlm.nih.gov/42790903/) 8. [Ectopic breast cancer in vulva and other non-axillary sites: a systematic review.](https://pubmed.ncbi.nlm.nih.gov/42015262/) 9. [Effects of digital communication tools on patients, family members and health care professionals in adult ICUs: a mixed-methods systematic review.](https://pubmed.ncbi.nlm.nih.gov/41776658/) 10. [Patient knowledge of surgical informed consent and shared decision-making process among surgical patients in Ethiopia: a systematic review and meta-analysis.](https://pubmed.ncbi.nlm.nih.gov/38218929/) 11. [Prevalence of Plasmodium falciparum coinfection with Schistosoma haematobium and Schistosoma mansoni among children in sub-Saharan Africa: a systematic review and meta-analysis.](https://pubmed.ncbi.nlm.nih.gov/41935296/) 12. [Performance of AI in Predicting the Progression of Gestational Diabetes to Type 2 Diabetes: Systematic Review and Meta-Analysis.](https://pubmed.ncbi.nlm.nih.gov/42424585/)