A new machine learning system flagged more than 250,000 cancer research papers that could be linked to so-called “paper mills.”
This research BMJexamined 2.6 million cancer research papers published between 1999 and 2024. The research was led by QUT researcher Professor Adrian Barnett from the School of Public Health and Social Work and the Australian Center for Health Services Innovation (AusHSI), along with a group of international collaborators.
Researchers found that more than 250,000 papers contained descriptive patterns similar to those seen in studies that had already been retracted due to fabrication.
“Paper mills are companies that sell fake or low-quality scientific research. They produce ‘research’ on an industrial scale, and our findings suggest that the problem in cancer research is much bigger than most people realize,” Professor Barnett said.
How paper mills produce fake research results
Paper mills sell authorship and, in some cases, completed, off-the-shelf scientific papers. These studies may include recycled text, unusual or difficult language, and fabricated data and images.
“Perhaps they rely on fixed templates that can be detected by large-scale language models that analyze patterns in the text,” Professor Barnett says.
To search for these patterns, Barnett and his colleagues trained a language model called BERT. The system was designed to identify subtle textual “fingerprints” that repeatedly appear on products from known paper mills.
When evaluated using a validated example, the model correctly detected suspicious documents 91% of the time.
“We’ve basically built a scientific spam filter,” says Professor Barnett.
“Just as email systems can detect unwanted messages, our tool flags papers that match the style and structure found in retracted fraudulent works.”
Suspicious papers on cancer are rapidly increasing
Extensive analysis revealed several key trends.
The number of flagged papers has increased rapidly over the past two decades, from about 1% in the early 2000s to a peak of more than 16% in 2022.
This suspected issue occurs across thousands of journals published by major companies, including many reputable and influential journals.
Questionable papers were particularly prevalent in areas such as molecular cancer biology and early-stage laboratory research.
Certain cancer types, such as stomach, liver, bone, and lung cancers, had particularly high rates of flagged studies.
Journal begins testing AI tools
Three scientific journals are already testing the system as part of their editorial review process. The purpose is to help editors identify potentially fabricated manuscripts before they are sent to external experts for peer review.
The researchers also plan to make the tool available for use in other scientific fields. They expect that accuracy to improve as more confirmed examples of paper mill activity become available.
However, the team stressed that papers identified by the system should not be automatically treated as fraudulent. The results are warning signals, not confirmed findings of wrongdoing, and each case still requires scrutiny by human experts.
Why fake research can harm patients
“Cancer research impacts clinical trials, drug development and patient care,” Professor Barnett said.
“When fabricated studies enter the evidence base, they can mislead real scientists and ultimately slow progress for patients. That’s why it’s important that we get ahead of this problem.”
