Bettmann / Getty ImagesAt an online conference held on May 24, experts in academic integrity discussed the challenges faced by journal publishers in dealing with articles generated by paper mills — illegal services that fabricate studies, data, and images and sell authorship of fraudulent articles to researchers, even helping submit them for publication. Organized by the UK-based Committee on Publications Ethics (COPE) and the association of scientific communication companies STM, based in The Hague in the Netherlands, the United2Act virtual meeting brought together researchers, journals, and representatives of research funding agencies. “It was the first time we had a group of people come together and co-create a set of actions which we’re going to take forward,” mathematician Deborah Kahn, a member of COPE’s board of trustees, told the journal Nature.
The group plans to publish a plan of action in the near future. The ideas under consideration include a proposal that scientific journals start requiring authors submitting manuscripts for publication to provide primary data from their research with digital watermarks. This feature could include references to the data’s origin in the form of unique identification codes, allowing editors and reviewers to assess whether the information behind the study is genuine. According to the participants of the meeting, there is currently no uniform set of requirements for submitting primary data, which can hinder attempts to determine their veracity. One obstacle to such a change is that it would require investment in data management by scientific journals.
Another strategy being considered is the sharing of data between scientific publishers when there is evidence that a paper mill is involved — it is common for large volumes of fraudulent papers to be submitted to various titles, and editors often do not realize that the problem is bigger or more serious than it seems. The challenge is to share information without violating data protection rules, since maintaining the confidentiality of submitted articles is a key principle of scientific communication. COPE offers a confidential forum where affiliated publishers can consult with other journals or companies that have faced a particular case of misconduct when they suspect they are victims of the same type of fraud.
It is not only to identify articles produced by paper mills. According to a study by COPE that served as a basis for discussions at the conference, fraudulent articles are commonly written in a way that is designed to avoid being detected by plagiarism software while respecting the publication rules of the journals to which they are submitted, so as not to raise suspicion and to increase the chance of being accepted. But there are characteristics that can help journal publishers identify them, for example, they are more common in the fields of cellular and molecular biology; contain very clean western blot images, which may be an indication of manipulation; use the private email addresses of authors, instead of emails linked to an institution; make changes to the list of authors while the article is being reviewed; and omit ethics committee approval of animal studies.
A study published in The British Medical Journal (BMJ) in November found that of 33,700 articles retracted between 2004 and 2021, 1,182 came from paper mills — the authors of 96.8% of this subset were from institutions in China and 76.9% were primarily affiliated with a hospital. It seems that while paper mills are increasingly common, the methods of tackling the problem are also improving. One of the conference attendees, Sabina Alam, director of ethics and editorial integrity at Taylor & Francis, told Nature that the number of potential cases of misconduct investigated by her team has increased tenfold between 2019 and 2022 — with about half relating to articles produced by paper mills. In 2023, the problem has been even worse: by May, there had already been more cases than in the entire year of 2022.
According to Alam, the publishing company’s ability to detect fraudulent articles was boosted by software made available last year on the STM Integrity Hub, designed to help publishers scrutinize manuscripts submitted for publication. The cloud-based platform helps editors from 15 scientific communication companies to identify cases of misconduct. They can also share their experiences without violating data privacy and antitrust laws, as well as jointly developing new tools.
One example is an STM-sponsored collaboration involving 24 publishers and scientific data analysis companies announced at the end of 2022, which is developing automated systems capable of analyzing more than 70 possible signs of manipulation frequently seen in articles produced by paper mills, such as doctored images, linguistic errors, and suspicious email addresses. The group shared one of its first tools in April: a website where manuscripts can be uploaded and evaluated for signs of fraud. If anything suspicious is detected, the system sends an alert to the editors so that they investigate the case in more depth.
The conference was held against the backdrop of a more recent threat: the emergence of artificial intelligence (AI) that could potentially make fraud more sophisticated and harder to detect. There is still no evidence that AI tools that generate text, such as ChatGPT, or produce realistic images, such as MidJourney, are compromising the scientific literature, but for Russian sociologist Anna Abalkina, a researcher at the Free University of Berlin, it could be simply a matter of time. Because the peer-review process generally takes a few months to complete, new problems can take some time to become apparent.
Republish