Abstract
This cross-sectional study describes trends and characteristics of retracted articles in the biomedical literature from 1971 to August 2020.
Retracting problematic articles helps to maintain the accuracy and integrity of the biomedical literature. However, retractions rarely receive the same attention as the initial publication; they may be quoted and included in reviews and meta-analyses.1 We describe trends and characteristics of retracted articles in the biomedical literature from 1971 to August 2020.
Methods
On August 12, 2020, we queried the Retraction Watch database, a comprehensive database of retracted articles, for items with the term medicine in the subject code.2 We excluded retractions of abstracts, autobiographies, bulletins, committee opinions, and non-English articles. We used the Scopus database3 to determine the number of publications in the biomedical literature during the corresponding time period (eMethods and eFigure in the Supplement). We determined the percentage of retractions using the number of retractions in a given year as the numerator and the number of publications in the same year as the denominator. For each retraction, we recorded the journal name, impact factor, article type, year of publication and retraction, reasons for retraction (no more than 10 reasons were listed in the database), field, paywall presence, and number of citations. Articles with authors from more than 1 country were classified as “multiple countries.” We classified 8 reasons for retraction as scientific misconduct: data fabrication, plagiarism, duplication, ethical violation, conflict of interest, authorship issues, missing informed consent, and compromised peer review.4 We report continuous variables as medians and interquartile ranges, and categorical variables as counts and percentages. We used segmented piecewise regression to estimate change-points. The Weill Cornell Institutional Review Board exempted the study from review because it does not constitute human subjects research.
Results
We identified 5209 articles published from January 1923 through July 2020 and retracted from January 1971 through August 2020. Of the retractions, 4364 (83.8%) were for original articles in clinical medicine, and 3245 (62.3%) were for scientific misconduct; 3608 articles (69.3%) had more than 1 reason for retraction (Table). As only 4 retractions were before 1980, we calculated trends for 1980 to 2020. The number and percentages of retractions by year increased from 1980 to 2014 (particularly after 2004) and declined after 2015 (Figure, A and B). Retractions for scientific misconduct also increased from 1980 to 2014 and subsequently declined (Figure, C).
Table. Characteristics of the Retracted Articles.
| Characteristic | No. (%) |
|---|---|
| No. of articles | 5209 |
| Article type | |
| Original article | 4364 (83.8) |
| Meta-analysis/review | 446 (8.6) |
| Case report | 254 (4.8) |
| Commentary/editorial/letter | 111 (2.1) |
| Guidelines | 14 (0.3) |
| Field of interest | |
| Clinical medicine | 1868 (35.9) |
| Basic science | 1333 (25.6) |
| Surgery | 839 (16.1) |
| Preclinical studies | 513 (9.8) |
| Health-associated disciplines | 329 (6.3) |
| Multiple fields | 324 (6.2) |
| Continent | |
| Asia | 2345 (45.0) |
| North America | 1041 (20.1) |
| Europe | 810 (15.6) |
| Africa | 113 (2.1) |
| South America | 65 (1.2) |
| Australia | 56 (1.1) |
| Multiple continents | 756 (14.5) |
| Country, No./No. (%)a | |
| China | 1025/1 301 086 (0.08) |
| US | 973/7 925 359 (0.01) |
| Japan | 495/1 518 911 (0.03) |
| India | 296/702 740 (0.04) |
| Germany | 210/1 252 891 (0.02) |
| Italy | 139/895 359 (0.02) |
| United Kingdom | 136/2 045 725 (0.01) |
| Canada | 61/1 006 743 (0.01) |
| Australia | 51/733 657 (0.01) |
| France | 41/835 361 (0.01) |
| Reason for retractionb | |
| Scientific misconduct | 3245 (62.2) |
| Duplication | 1225 (23.5) |
| Plagiarism | 724 (13.9) |
| Fabrication of data | 631 (12.1) |
| Issues about authorship | 378 (7.3) |
| Fake peer review | 313 (6.0) |
| Ethical violations | 232 (4.5) |
| Lack or withdrawn of informed consent | 85 (1.6) |
| Conflict of interest | 43 (0.8) |
| Unspecified misconduct | 781 (15.0) |
| Error(s) in the manuscript | 1949 (37.4) |
| Issues with the journal or publisher | 1010 (19.4) |
| Retractions for scientific misconduct by country, No./No. (%)c | |
| China | 699/1025 (68.2) |
| US | 474/973 (48.7) |
| Japan | 377/495 (76.1) |
| India | 229/296 (77.4) |
| Germany | 154/210 (73.3) |
| Italy | 98/139 (70.5) |
| United Kingdom | 58/136 (42.6) |
| Australia | 29/51 (56.9) |
| Canada | 27/61 (44.2) |
| France | 24/41 (58.5) |
| No. of journals with ≥1 retraction | 1828 |
| Impact factor of the journal of publication, median (IQR) | 3.17 (1.9-4.7) |
| Years from publication to retraction, median (IQR) | 1.8 (0.6-4.7) |
| Retractions due to scientific misconduct, mediand | 2.4 |
| Retractions for other reasons, median | 1.1 |
| No. of authors, median (IQR) | 5.0 (3.0-7.0) |
| No. of citations, median (IQR) | 9.0 (3.0-25.0) |
Abbreviation: IQR, interquartile range.
For country-level data, the numerator is the total number of retractions over the study period, and the denominator is the total number of publications over the study period. Percentages by countries were calculated for the 10 countries with the highest numbers of retractions. Retractions with authors from multiple countries were excluded from these calculations.
Only the top 3 reasons for retraction are reported; 3608 retracted articles had more than 1 reason for retraction.
The numerator is the total number of retractions owing to misconduct by country, and the denominator is the total number of retractions by country over the study period. Retractions with authors from multiple countries were excluded from these calculations.
The median time from publication to retraction is significantly higher for retractions owing to misconduct (P < .001).
Figure. Time-Segmented Analysis Showing Number of Retractions, Percentage of Retractions, and Percentage of Retractions Owing to Misconduct by Year.
The 2020 data point accounts for retractions from January through August.
The median (interquartile range) time from publication to retraction was 1.8 (0.6-4.7) years: 2.4 years for retractions owing to scientific misconduct and 1.1 years for other reasons (Table). Over the study period, the median time from publication to retraction significantly decreased (start of study period, 0.83 years, vs end of study period, 0.14 years; P < .001). For retractions owing to scientific misconduct, the most common reasons were duplication, plagiarism, and fabrication of data (Table). For retractions for other reasons, the most common explanations were errors in the article and issues with the journal or publisher, such as inadvertently moving an article to production before final acceptance.
Most authors (3729 of 4138 [90.1%] first authors and 3525 of 3896 [90.5%] last authors) had only 1 retraction. Two authors had more than 100 retractions. Scientific misconduct was the primary reason—164 of 166 retractions (98.8%) for 1 author and 82 of 102 retractions (80.4%) for the other. Three other authors had between 50 and 100 retractions; these 5 authors accounted for 7.3% (381 of 5209) of all the retractions.
The median number of citations of retracted articles was 9.0 (interquartile range, 3.0-25.0). Of the retracted articles, 166 (3.2%) were cited more than 100 times, and 78 (1.5%) were cited more than 200 times. China, India, and Japan had the highest rates of retraction overall, and India, Japan, and Germany had the highest percentage of retractions for scientific misconduct.
Discussion
The annual rate of retracted articles in the biomedical literature increased from 1980 to 2014 but decreased after 2015; scientific misconduct was the most common reason for retraction. Although the declining rate of retractions, overall and for scientific misconduct, is encouraging, we did not study the explanations for this trend. Other limitations include the possibility that we missed some retractions despite the use of a frequently updated database, as well as retractions of recently published articles.
eMethods. Scopus search strategy
eFigure. Flowchart of the articles selection process
References
- 1.Couzin J, Unger K. Scientific misconduct. cleaning up the paper trail. Science. 2006;312(5770):38-43. doi: 10.1126/science.312.5770.38 [DOI] [PubMed] [Google Scholar]
- 2.Retraction Watch Database . Accessed August 12, 2020. http://retractiondatabase.org/RetractionSearch.aspx
- 3.Scopus Database . Accessed November 18, 2020. https://www.scopus.com/home.uri
- 4.Resnik DB, Rasmussen LM, Kissling GE. An international study of research misconduct policies. Account Res. 2015;22(5):249-266. doi: 10.1080/08989621.2014.958218 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
eMethods. Scopus search strategy
eFigure. Flowchart of the articles selection process