How to verify medical citations: PMID, DOI, retractions and claim support

A concrete task with RigorMed

1. Connect a sentence to the paper it cites

   These are suggested task inputs and outputs to request, not a new execution, a customer outcome or a guarantee of completion.

   Materials to provide: The sentence to cite, PMIDs/DOIs and bibliographic records, plus lawfully readable original text or relevant passages.

   Requested outputs / delivery goal: Citation-checking records, a claim-to-source mapping table and a list of missing evidence.

2. An example request

   Using the bibliographic records I provide, match each title and identifier, then map manuscript claims to locations in the supplied original text. Record claim, source, evidence location and whether support is sufficient. Without original text, organise bibliographic records only. Mark unknown correction, retraction or version status for further checking.

3. What still needs review

   A resolving identifier does not prove a claim. Corrections, retractions and version status still require verification with the publisher.

Four steps back to original evidence

1. Check the work’s identity

   Locate the PubMed record by PMID, DOI or title and open the publisher page. Compare title, authors, journal, year, volume, issue, pages or article number, and DOI/PMID. Distinguish preprints, conference abstracts, published articles and correction notices. A similar title or syntactically valid DOI is not a match; do not invent missing fields.

2. Check versions, corrections and retractions

   Read linked PubMed notices, publisher version information and available Crossmark updates; record the date and status actually checked. ICMJE requires checking for retractions and avoiding retracted articles except when discussing the retraction itself; for MEDLINE-indexed journals it treats PubMed as authoritative for retraction information. Read corrections to assess their effect on the result you intend to cite.

3. Match each claim to the original text

   Write down the proposed sentence, then examine relevant methods, results, tables, figures and limitations. Record section, page or table. Check population, design, exposure or intervention, outcome, direction and uncertainty. Do not convert association into causation, generalize a subgroup to everyone, or substitute a review for accessible original research.

4. Resolve conflicts before exporting references

   Crossref metadata or DOI content negotiation can help retrieve and export bibliographic records, but check field mappings, journal rules and the original. If identifiers conflict, status is unclear or support is absent, pause the claim and correct, replace or remove the reference. Changing citation style is not a remedy. Preserve the author’s decision and date.

Conceptual example: a genuine paper can still fail to support a claim

Assume a candidate observational cohort solely to illustrate checking. This is not an actual paper or result; no fabricated PMID, DOI or effect estimate is supplied.

If only the abstract is available, label the check as abstract-level and retain only statements it supports without overinterpretation. When a key claim needs unavailable full text or supplements, retain the gap rather than claiming full-text verification.

Keep a reviewable citation record

For important references, keep at least these items as a practical recommendation, not a universal required format:

• The supported sentence and its manuscript location.
• Verified title, authors, year, available PMID/DOI and source URL.
• Version read, abstract/full-text/supplement level and access date.
• Sources and status for correction/retraction checks, including unresolved items.
• Original evidence location, scope, reviewer and decision.

Retain records and necessary excerpts within the material’s licence; do not publicly share restricted full text or private data. Correct citation style does not establish sound design, evidence quality or clinical applicability.

Frequently asked questions