Retraction of publications is an increasing problem in science and one of the most troublesome issues for biomedical journals.1–3 The impact of retraction is negative whether it is due to misconduct or not, and it is believed that fraud is more frequent than assumed.4,5 Retracted publications usually cause lack of trust in medical science and may affect the credibility of scientific journals, even in the case that some journals are proactive in retraction policies.

Detection of plagiarism has become a concern for editors, and measures such as the implementation of plagiarism detection software have been taken.6–8 The literature regarding the use of surveillance tools to detect plagiarism and its results is scarce. In fact we have not found any paper analyzing the use of such tools comparing biomedical journals’ characteristics such as the impact factor. The use of these tools has increased plagiarism detection, but these detected manuscripts do not become public, as when detected during the editorial process authors are warned and therefore the manuscript is not formally retracted because it is directly rejected. Therefore, the increase in retractions due to plagiarism is a result of the manuscripts that have escaped from the journals detection system, which are now increasingly used by high-impact journals.9

Literature focusing on misconduct has been centered more on authors instead of journal's characteristics1,10–15 involved in retraction or misconduct. Misconduct rates by journal type do not necessarily mean a weak peer-review policy by a specific journal but a high ethical standard or a higher surveillance or sensitivity on claims of what a particular journal has published. The surveillance of what is published is expected to be higher in a high impact journal. We aimed to describe the journal characteristics, mainly the impact factor and journal categories, involved in the retraction process to have a better picture of retraction frequency, particularly that related with misconduct. This paper is part of a previously published study,15 and is exploiting additional information from the same database.

We present here a description of retractions broken down by journals’ IF. We have found that, of the retracted manuscripts, 51.4% belong to high-impact journals and that it seems that misconduct among retracted papers has apparently a higher presence in low-quartile journals. The most frequent cause of misconduct is plagiarism, accounting for 38% of all retractions in 2016. Finally, only 1.5% of retractions were marked as having triggered an action by journals, a very low percentage. We do not have data on the percentage of journals contacting authors when there is a retraction and if there are differences on the frequency of this contact regarding the journal's IF. However, and despite no clear data were available, different implications for authors have been previously described following retractions.20 Similarly, some journals decide to penalize authors by preventing them from publishing in the journal for a period of time.8,21

Fang et al.1 have shown in a study including papers retracted until 2012 that the mean time to retraction was 32.9 months, which is more than three times longer that the time observed in the present study. We might speculate that this finding served editors to speed up fraud investigation when there is a suspicion and, as a consequence, the time to retraction has been shortened. The same group has also shown that time to retraction because of fraud is longer than when retractions were due to other reasons. This result is in agreement with our finding of a longer time to retraction because of misconduct. This might be related to the difficulty to ensure the facts and to confirm the fraudulent conduct. Also, when retraction is due to reasons other than misconduct, authors might be involved in the process of retraction, and that seems to be a factor related with reduced time to retraction.

Journal IF also has an influence in retraction with a higher proportion of misconduct among retracted papers published in low-impact journals. Previously, the contrary has been shown, suggesting that the greater visibility of high-impact journals promotes higher inspection and a quicker retraction. We have also found that 51.4% of the retracted papers were published in high-quartile journals. It is noteworthy that in high-impact journals, a significant proportion of the misconduct is related to data manipulation by authors, suggesting that high IF might be more attractive as these publications are generally linked to improvements in CVs, grant success, and better job opportunities. This might provide some incentive to commit misconduct to publish in a high-ranking journal even though IF and quantitative measurement of publications has been largely questioned.16,22,23

Plagiarism accounts for the highest number of retractions for the study period. These data are striking and deserve to be commented given that we have reasons to think that the real number of authors plagiarizing is higher. The use of plagiarism detection software has become more widespread. Despite the fact that the use of these systems prevents having plagiarized published manuscripts, retractions due to plagiarism are increasing by year. It is worth to mention that plagiarism was the main reason for retraction independently of the journal impact, indicating that the implementation of plagiarism detection systems are in great need, independently of the journal. However, these systems have probably prevented finding a significant increase. Hence, and despite awareness of editors, reviewers and readers who undoubtedly are contributing to this increase in detection and retraction, more action has to be done.8,24,25 However, efforts to limit this phenomenon have to start from the very moment when a manuscript arrives to a medical journal and is assigned to an editor. Detection of plagiarism can be done in three ways. First, detection can rely on the capacity of reviewers and editors to identify previously published papers. Second, it can use plagiarism detection software including that for image fraud detection.24,25 And finally, detection can occur when a reader or the author is capable of identifying the plagiarism from an already published paper. This method seems to be the only way to identify plagiarism when the original publication is in a language other than English and is not published in a widely visible journal.26 Actions other than warning authors are needed, such as those in based in education. It has already been tested that warnings against plagiarism are insufficient.27,28 There is also a lack of standardization and clarity on retraction management, and databases should provide tools and procedures to trace the retractions and authors committing misconduct.29

It is worth to mention, that the entire editorial process is not paid, nor valued, and that in general there is not any specific training for the work. Also, the editorial process is based on good faith and goodwill. Consequently, to ask Editors and reviewers an excessive amount of work and to be fully aware of new ways of committing fraud might not be realistic. Some tasks using plagiarism detection using commercial software should be done by biomedical journals’ administrative staff before allowing the manuscript to enter in the editorial process.

Biochemistry and molecular biology and pathology journals are, respectively, the JCR categories with a greater number of retractions, and with the highest proportion of misconduct among retracted papers, though we do not know the total number of published papers within those categories. Despite retractions and misconduct are widespread, this basic discipline accounts for the greater proportion of retractions. Before, non-association was found between a general medicine journal (vs. other) and misconduct.13 In general terms, we believe that the threat of retraction and misconduct is everywhere, and all journals have to be aware of it and implement measures to limit it.

The main limitation of our study results from the fact that we are only evaluating retractions that have been identified after being published, and we do not address all the plagiarism detected by the journals including any other fraud avoided by early identification during the editorial process affecting rejected papers. Also, as has been previously acknowledged,15 the ability to classify publications as misconduct or not is limited and some data are not accessible, such as quartile of all the publications of the study period. Also, we cannot evaluate the percentage of journals contacting authors and/or institutions when there is a retraction, and if this is related to the journal's IF. The dynamic nature of data obliges us to be cautious. An additional limitation is that PubMed was the only database consulted to gather the reported data, though we need to take into account the fact that, for medical literature, Medline is the main database, and where the vast majority of journals are indexed. PubMed is also the database where a reader would notice if an article has been retracted. Finally, and perhaps most important, we do not know the total number of papers published in the analysed journals and therefore we cannot infer that the incidence of retractions is higher in low-impact biomedical journals compared to high impact journals. We only can conclude that the crude number of retractions (and also retractions due to misconduct) is higher in low impact journals

To conclude, retraction of scientific publications is associated with both high and low-impact journals, with plagiarism as the main reason of misconduct, independently of the journal IF. It would be recommendable for journals to implement the use of plagiarism detection systems and to standardize measures both before and after the publication of the manuscript. We encourage journal editors detecting misconduct to promote actions beyond those limited to the journal.

Carlos Álvarez-Dardet.

The corresponding author on behalf of the other authors guarantee the accuracy, transparency and honesty of the data and information contained in the study, that no relevant information has been omitted and that all discrepancies between authors have been adequately resolved and described.

I. Campos-Varela and A. Ruano-Raviña wrote the manuscript and R. Villaverde-Castañeda critically edited and reviewed the manuscript. All authors provided critical input to the manuscript's content and have approved the final version. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted.

None.

None.