AI detects conflicts of interest in medical journals, spots unexpected trend

Medical journals accepting reprint fees are much more likely to publish articles written by authors who received industry payments, according to a new analysis published in PLOS ONE.

The researchers tracked more than 128,000 articles from 159 different journals, developing a machine learning-based AI model capable of tracking individual conflicts of interest in English-language disclosure statements. Overall, the model found that articles written in journals accepting reprint fees were 2.81 times more likely to include potential conflicts of interest.

“I was honestly surprised by the findings here,” lead author S. Scott Graham, an assistant professor of rhetoric at the University of Texas at Austin, said in a statement. “There’s a famous story about one company pulling a multimillion-dollar contract from the Annals of Internal Medicine because they didn’t like an article published in the journal. All the available literature suggests that ad revenue should be the real concern, but that’s not what we found.”

When journals make advertising space available but don’t accept reprint fees, it is associated with fewer conflicts of interest. Also, the team noted, when journals are owned by “large publishing companies,” articles are expected to have an average of 3.2 more conflicts of interest.

“If we’re going to make sure that medical journals are publishing the best science available, we need to focus on the commercial relationships that actually have an effect,” Graham said. “The issue with reprints also suggests that academics may need to take open access publishing even more seriously.”

The study is available here.