AI-Generated Biomedical Papers Flood Journals, Raising Quality Concerns

The scientific literature, particularly in the biomedical field, is at risk of being inundated with low-quality research papers making misleading health claims based on easily accessible public data sets, warn researchers. A significant concern is the use of artificial intelligence (AI) tools to generate these papers without proper scrutiny of the underlying data or methodology. In a study published in PLoS Biology on May 8, researchers analyzed over 300 papers that utilized data from the US National Health and Nutrition Examination Survey (NHANES), a extensive and publicly available database of health records. These papers often followed a predictable pattern, linking a single variable, such as vitamin D levels or sleep quality, to complex disorders like depression or heart disease. This approach ignores the multifaceted nature of these conditions, which are influenced by numerous factors. "We are witnessing a rapid increase in the number of publications that are extremely formulaic and could easily be generated by large language models," notes Matt Spick, a biomedical scientist at the University of Surrey in Guildford, UK. Spick and his team discovered that many of these associations lacked robust statistical support and appeared to be the result of data cherry-picking. Charlie Harrison, a computational biologist at Aberystwyth University in Ceredigion, UK, illustrates the issue with an analogy: "Imagine you’re taking an exam where you can keep adding questions until you get enough right to pass. You then only submit the correct answers and discard the wrong ones. This is essentially what some of these studies are doing." Ioana Alina Cristea, a clinical psychologist and meta-researcher at the University of Padua, Italy, concurs that the papers "seem to be written with a recipe." She emphasizes the importance of systematic evaluations to understand the scope of the problem. "We need these assessments to gauge how widespread this issue is and to take appropriate action," Cristea adds. The NHANES database, which has been collecting data on health, diet, and lifestyle from thousands of Americans since the 1960s, has seen a surge in usage over the past two years. According to the PubMed index of biomedical literature, more than 2,200 association studies using NHANES data were published in 2024, and over 1,200 have already been published this year. Harrison, Spick, and their colleagues focused on a sample of 341 NHANES-based studies published between 2014 and 2024. These papers appeared in 147 different journals, including those from major publishers like Frontiers Media, Elsevier, and Springer Nature. The study highlights a broader debate within the scientific community: whether it is ethical for AI to write science papers. While AI can facilitate the generation of research, there is significant concern about the lack of transparency and the potential for producing misleading results. A recent Nature survey revealed that researchers are divided on this issue, reflecting the complexity and urgency of addressing the integrity of scientific publications in the age of advanced AI. The findings from Spick and his colleagues underscore the need for stringent peer review and transparent methodologies to ensure that research using public data sets, such as NHANES, remains rigorous and reliable. As AI continues to evolve and become more integrated into research processes, the scientific community must remain vigilant to maintain the high standards and credibility of their work.