Skip to main content
Back to SentryData study · June 2026

How often do recent papers cite retracted research?

We checked the reference lists of 12,402 recent (2025) papers across 54 major biomedical journals — 744,770 individual citations in all — against the Retraction Watch database and Crossref’s retraction notices, the same data behind our free reference-integrity checker. Here is what we found, with the method and named examples so you can check it yourself.

7.1%
of papers cite at least one retracted, expression-of-concern, or corrected reference (875/12,402; 95% CI 6.6–7.5%).
1.4%
cite a retracted or expression-of-concern reference specifically — the higher-severity subset (170/12,402; 95% CI 1.2–1.6%). About 1 in 73 recent papers.

What this means (and what it doesn’t)

A paper isn’t wrong because one of its references was later retracted. Retractions often land after a citing paper is written, and the citing authors usually had no way to know. The point is narrower and more useful: at submission time, nobody ran a check that would have caught it — and that check now takes seconds. When the cited work has been retracted for data manipulation or unreliable results, a citation that still leans on its original finding is a real problem a reviewer can flag.

We lead with the 1.4% retracted-or-EoC number rather than the headline 7.1%, because “correction” is the lowest-severity flag (an erratum is not misconduct) and the most common — 783 of the flagged references were corrections, versus 162 retractions and 25 expressions of concern. Both numbers are a conservative lower bound: this run did not include the production engine’s hijacked-journal clone-domain check.

The problem is concentrated in lab-based fields

The corpus average hides a sharp gradient by field. Cell and cancer biology run far above it — Cancer Cell 16.4%, Cell 15.4%, Cell Reports 14.7%, Cancer Research 14.4%, with immunology (Immunity 12.4%, Nature Immunology 10.9%) close behind. Clinical medicine sits near the floor: the New England Journal of Medicine 2.3%, JAMA 2.8%, Annals of Internal Medicine 2.5%, The Lancet Oncology 1.3%. The fields where data integrity has been most contested are the same fields whose recent papers most often carry a flagged reference. If you work in cell biology, oncology, or immunology, your bibliography’s base rate is roughly five times a clinical journal’s.

By journal

JournalSampledPapers w/ flagRate
Nature Reviews Immunology42921.4%
Nature Reviews Drug Discovery30516.7%
Cancer Cell1282116.4%
Cell3054715.4%
Cell Reports3064514.7%
Cancer Research2914214.4%
Journal of Experimental Medicine1582012.7%
Diabetes1822312.6%
Immunity1782212.4%
American Journal of Human Genetics1661911.4%
Nature Immunology1832010.9%
Neuron2322510.8%
Molecular Cell274279.9%
Developmental Cell205209.8%
Science Translational Medicine255249.4%
Nature Biotechnology163159.2%
eLife289269.0%
Nature Genetics264228.3%
Nature294248.2%
Biochemical and Biophysical Research Communications297248.1%
Journal of Immunology306237.5%
Journal of Cell Biology209157.2%
Nature Communications303216.9%
Journal of Biological Chemistry306216.9%
PLOS Genetics295206.8%
Nature Neuroscience208146.7%
PLOS Biology308196.2%
Nature Methods227146.2%
Proceedings of the National Academy of Sciences299186.0%
Nucleic Acids Research302186.0%
BMC Bioinformatics270165.9%
Nature Cell Biology172105.8%
Journal of Virology297165.4%
BMJ14874.7%
Science Advances307144.6%
Journal of Bacteriology19894.5%
Gastroenterology17984.5%
Blood296134.4%
Brain302134.3%
PLOS Medicine14064.3%
Nature Medicine306134.2%
Science286124.2%
Molecular Biology and Evolution268114.1%
The Lancet244104.1%
Circulation253104.0%
Bioinformatics17263.5%
The Journal of Neuroscience300103.3%
PLOS ONE30093.0%
JAMA18052.8%
Annals of Internal Medicine16242.5%
New England Journal of Medicine29872.3%
The Lancet Oncology15821.3%
Journal of the American Society of Echocardiography8311.2%
The Lancet Neurology7800.0%

All 54 journals shown. Most have ~150–300 papers sampled, so individual rates are meaningful; the few small-sample journals (the review titles, JASE, Lancet Neurology) should be read as directional. The corpus-level rate is the most reliable figure.

Real examples

Each of these is a 2025 paper in a major journal whose reference list includes a paper that has since been retracted. In several cases the retraction landed after the citing paper was written — the point is that a pre-submission check now catches it in seconds. Links go to the DOIs so you can verify both ends.

Method

  • Sample: 12,402 recent (2025) journal-articles drawn across 54 major biomed journals via the Crossref /works API by ISSN, with reference lists of at least 10 resolvable DOIs (4,969 thinner candidates excluded).
  • Check: all 744,770 reference citations (435,108 unique cited DOIs) matched against the live Retraction Watch dataset (over 61,000 notices) and Crossref’s update-to[] retraction notices — the identical code path behind the free checker. Severity precedence: retracted > expression of concern > correction.
  • Confidence intervals are Wilson 95%. Lower bound: the hijacked-journal clone-domain check was not run in this batch; production applies it too.
  • Known biases (stated plainly): flag probability scales with reference-list length, so excluding thin-reference papers pushes the rate up; read 7.1% as the rate among papers with substantive bibliographies, not all 2025 papers. The 54-journal panel spans clinical medicine, oncology, cell and molecular biology, immunology, microbiology, genetics, neuroscience, and methods, but it is a sample of major journals, not a random draw of all biomedical literature; the per-field gradient above is the honest way to read it.

Data sources are credited on the Sentry about page; if you cite this study, please also cite the Retraction Watch Database (ISSN 2692-4579) and Crossref. Need the full row-level data (every citing/cited DOI pair)? Email erik@manusights.com.

Download the data

The per-journal table and study summary, free to reuse under CC BY 4.0 with attribution. Please credit Manusights, the Retraction Watch Database (ISSN 2692-4579), and Crossref.

How to cite this study

APA

Manusights. (2026). How often do recent papers cite retracted research? A 12,402-paper study. https://manusights.com/tools/reference-integrity/retracted-citation-study

BibTeX
@misc{manusights2026retracted,
  title        = {How Often Do Recent Papers Cite Retracted Research? A 12,402-Paper Study},
  author       = {{Manusights}},
  year         = {2026},
  howpublished = {\url{https://manusights.com/tools/reference-integrity/retracted-citation-study}},
  note         = {Data study. 12,402 recent biomedical papers across 54 journals; built on Crossref and the Retraction Watch Database (ISSN 2692-4579).}
}

Embed this badge

Cite the finding on your site

Writing about citation integrity, retractions, or pre-submission checks? Drop this badge into your post or library guide — it links back to the study and its method.

Manusights: 1 in 73 recent papers cite a retracted or expression-of-concern reference
Markdown
[![Manusights: 1 in 73 recent papers cite a retracted or expression-of-concern reference](https://manusights.com/api/tools/reference-integrity/retracted-citation-study/badge)](https://manusights.com/tools/reference-integrity/retracted-citation-study)
HTML
<a href="https://manusights.com/tools/reference-integrity/retracted-citation-study"><img src="https://manusights.com/api/tools/reference-integrity/retracted-citation-study/badge" alt="Manusights: 1 in 73 recent papers cite a retracted or expression-of-concern reference" /></a>

Check your own bibliography

Paste your reference list into the free Manusights reference-integrity checker — no login, no email gate — and it flags retracted, expression-of-concern, corrected, and hijacked-journal references in one pass.

Preparing a manuscript for submission? Our $39 manuscript readiness scan runs the full citation-integrity suite across your entire reference list alongside methodology, statistics, and journal-fit checks. The preview is free.