A paper can change after publication. A correction amends the record; an expression of concern raises an unresolved question; a retraction formally changes the article’s status. Read the notice before deciding which findings remain usable. A warning label alone does not establish an author’s intentions.
Identify the notice before interpreting it
| Record | What to check |
|---|---|
| Correction or erratum | Which text, data or analysis changed, and whether your summary uses the affected information. |
| Expression of concern | What remains unresolved and whether the journal subsequently reported an outcome. |
| Retraction notice | The stated reason, the identified article and any replacement publication. |
| Comment or letter | The argument being made and whether a separate formal status notice exists. |
NLM links these records in PubMed. The original retracted article receives the publication type Retracted Publication; the notice receives Retraction Notice. NLM does not use those labels to distinguish honest error from misconduct. A linked comment is not automatically a retraction. NLM linked-citation policy.
A correction can change one detail or the main conclusion
ICMJE recommends dated corrections and traceable article versions. Serious errors that invalidate results or conclusions may require retraction. Where honest error substantially changes findings but the underlying science remains valid, retraction with republication may be considered after further review. Ordinary scientific disagreement is not itself an error requiring correction. ICMJE corrections and version control.
Fictional example: correcting a misspelled institution would not, by itself, change an effect estimate. Correcting a spreadsheet formula used for every participant might change the estimate, its uncertainty and the conclusion. Neither the word “correction” nor the length of the notice tells you which situation applies. Trace the stated change into the specific claim.
Do not turn uncertainty into an accusation
ICMJE permits an expression of concern while procedures addressing concerns are pending. It calls for retraction notices to explain their reasons and for retracted articles to remain accessible with clear labels. Continuing access therefore does not restore an article’s previous status. ICMJE guidance on concerns and retractions.
For a notice labeled “withdrawn” or “reinstated,” report the publisher’s explanation rather than guessing what happened. If the notice does not resolve a question, say so. Avoid describing a retraction as fraud unless the authoritative record establishes that reason.
A status check you can repeat
- Match the record. Save the article title, DOI, publisher link and date checked. Similar titles are not sufficient.
- Open the publication history. Read the current article page and each linked notice, including its date and any replacement version.
- Cross-check the links. Compare PubMed’s related records with the publisher history. Use Crossref or Retraction Watch as an additional route to updates.
- Map the consequences. List the sentences, tables and conclusions in your summary that depend on the affected information.
- Revise the claim. Change or remove unsupported statements, cite the notice and identify the version now used.
This is an editorial workflow, not a certification that a paper is free of problems. Crossref provides publisher and Retraction Watch update metadata, but states that the latter’s corrections and expressions of concern are less comprehensive than its retraction records. An empty metadata search is not proof that no notice exists. Crossref Retraction Watch documentation.
Make the change visible in your summary
A useful note identifies the check date, notice, affected result and replacement source. For example: “The publisher corrected the analysis on [date]. This summary now uses [identified version] and no longer repeats [affected claim].” Fill in only details that the records support. Adding a warning while leaving the unsupported headline unchanged does not repair the summary.
If the article, notice and database disagree, describe that discrepancy and avoid presenting the status as settled. Broader assessment of study design belongs in How to Read a Research Study. Browse related topics in Research Evidence Guides.
Review and corrections
Sources checked October 1, 2026. This AI-assisted educational summary was checked against the linked official guidance. It does not adjudicate any author’s conduct or claim independent scientific specialist review. Report a source discrepancy through Contact Us.