Retraction Watch doi checker

ByYuki Furukawa

This tool checks a list of DOIs against Retraction Watch data. Paste the DOIs of the records you have screened or included, and it returns a table flagging which of them carry a retraction or a correction, together with the stated reasons.

How to use

Input is a tab-separated block pasted into the text area, for example a column copied out of Excel or a reference manager export. There is no file upload. The first line must be a header row and it must contain a column named doi.

  • Three DOI formats are recognised and normalised: a bare DOI (10.1016/...), a DOI with a DOI: prefix, and the full URL form (https://doi.org/10.1016/...). You do not need to clean the column first.
  • Any additional columns you paste — a study ID, a first author, your own screening decision — are preserved and carried through to the result table, so the output can be joined back to your screening spreadsheet.

The result table has the columns doi (as you entered it), doi_norm (the normalised form actually used for matching), RetractionNature, Reason and Notes, followed by your own columns. RetractionNature takes one of three values: Retraction, Correction, or No flag. Reason is a semicolon-separated list, so a single record can carry several reasons. The table can be downloaded as a CSV file named retraction_check_YYYY-MM-DD.csv, which is worth keeping as documentation of when the check was run.

Two points about the output shape are easy to miss:

  • One DOI can produce more than one row. If several records match the same DOI, each becomes its own row. In one check, two pasted DOIs produced three output rows: 10.1016/s0140-6736(97)11096-0 returned both a Correction row and a Retraction row, with reasons including falsification or fabrication of data, while 10.1136/bmj.n71 returned No flag. Deduplicate by doi_norm before you count how many of your included studies are affected.
  • Retraction dates are not returned. The output tells you that a flag exists, not when it was issued, so it cannot by itself tell you whether a retraction predated your search.

Method and primary sources

Matching is done on the normalised DOI string, server-side; nothing about the content of the paper is assessed. A flag is therefore a prompt to go and read the publisher’s notice, not a verdict, and a No flag result means only that no matching record was found at the moment of the query.

Checking for retractions is one narrow component of a much larger question, which is whether the trials in a review can be trusted at all. For a structured approach to that question, see the INSPECT-SR tool [1], with its user resources at [2]. Retraction status is the easiest signal to automate and the last one to appear; problems with a trial are usually visible in its own report long before a notice is issued.

How to cite

Record the check as a dated step in the review process rather than as an analytic method:

The DOIs of all included records were checked against the Retraction Watch database using the Retraction Watch DOI Checker (Furukawa Y, https://yukifurukawa.jp/retraction_watch_doi_checker/) on [date]. Flagged records were verified against the publisher’s retraction or correction notice before any decision was made about their inclusion.

Stating the date matters: a check is only valid as of the day it was run, and for a living review it needs repeating at every update.

Limitations

  • A doi column is required. Records without a DOI — older trials, conference abstracts, some non-English journals, most grey literature — cannot be checked here and need to be handled by hand.
  • Matching is exact on the normalised DOI. A typo in the DOI produces No flag, which looks the same as a clean result.
  • No dates in the output, and no link to the notice itself; verification against the journal record is a manual step.
  • Duplicate rows per DOI mean the row count of the output is not the number of affected studies.
  • The app does not show the version or update date of the underlying data. The result reflects whatever state that data was in at the moment of the query, and there is no way to tell from the screen how recent it is. Record the date on which you ran the check, re-run it before submission, and re-run it again at each update of a living review.
  • Absence of a flag is not evidence of trustworthiness. Retraction is a slow, incomplete signal, and this tool does not assess integrity, plausibility or risk of bias.
  • The decision of what to do with a flagged study — exclude it, retain it in a sensitivity analysis, or report it separately — is a protocol decision, ideally made before the check is run.

References

  1. Wilkinson J, Heal C, Flemyng E, et al. INSPECT-SR: a tool for assessing trustworthiness of randomised controlled trials. Preprint. medRxiv. 2025;2025.09.03.25334905. Published 2025 Oct 21. doi:10.1101/2025.09.03.25334905
  2. INSPECT-SR: resources for users. https://osf.io/b74wj/overview

Data source: the flags and reasons returned here come from Retraction Watch. The Retraction Watch dataset is distributed publicly by Crossref at https://gitlab.com/crossref/retraction-watch-data/. The app does not display which snapshot of that data it is querying, or when the snapshot was taken.