"Hidden" Changes To US Health Data Swapping "Gender" For "Sex" Spark Fears For Public Trust

A new analysis has found that over 100 health datasets from US government agencies were “substantially altered” in the two months after the beginning of the new administration earlier this year, in many cases switching the word “gender” to “sex”. Only a small number of the changes were logged, something the authors behind the analysis say could undermine the accuracy of and trust in science.

The duo behind the analysis is Janet Freilich, a professor at the Boston University School of Law, and Aaron S Kesselheim, a professor at Harvard Medical School. They looked at a total of 232 datasets from the US Department of Health and Human Services, Centers for Disease Control and Prevention, and Veterans Affairs, all of which had been modified between January 20 and March 25, 2025 and weren’t otherwise routinely updated monthly or more frequently.

Using archived versions of the datasets hosted by Internet Archive, they manually tracked where changes had been made to words in the datasets – and they found that 114 (49 percent) of them had been “substantially altered”. Out of those 114 changed datasets, only 15 of them had indications that alterations had been made.

Freilich and Kesselheim also found that in 106 of the 114 datasets (93 percent) had been changed to switch the word “gender” to “sex”. While the motivations for such a change hasn’t been explained by those responsible for hosting the datasets, the authors note that they “could be intended to comply with a Presidential directive for agencies to remove ‘messages that promote or otherwise inculcate gender ideology’”. That’s the same directive that ended up technically making every American female.

Regardless of the reasoning behind the change, the authors write that “this is not a trivial alteration,” for two main reasons – accuracy and trust.

“Because some respondents will answer questions about gender differently from questions about sex, changing these terms changes the accuracy of the dataset and the conclusions that can be drawn,” Freilich and Kesselheim explain. “These data are currently used to study health interventions and outcomes, so secretly changing terms degrades the quality of the underlying information and can undermine the interpretation of the results of these studies – or even invalidate the results themselves.”

Misinterpretation of data is no small matter when it comes to health data – it’s the sort of information that can be used as a basis for healthcare recommendations, which could be rendered “ineffective or counterproductive” by altering the data, the researchers argue.

Not only might this have consequences for public health, but also the public’s faith in science. “Such changes, when discovered, reduce trust in the data that underly public health and, consequently, health interventions,” Freilich and Kesselheim write. “This reduction in trust hinders the progress of science, medicine, and public health, and reduces individual willingness to rely on expert recommendations.”

The duo also suggest that it could affect trust amongst the international research community too, with researchers from abroad less willing to put their own data in repositories hosted by the US Government if they think it could be altered in some way.

As the authors point out, there are some limitations to their analysis. In this case, they didn’t carry out something called inter-rater reliability testing, which is a measure of how much people agree about something subjective that they’re assessing – in this case, whether a change made to the data was clerical or routine, or potentially substantial.

Nonetheless, Freilich and Kesselheim are clear that the issue at play here is transparency. “Despite Secretary Robert F Kennedy Jr's (Department of Health and Human Services) calls for 'radical transparency', unlogged data manipulation moves away from meaningful transparency,” they write. “It is inevitable that some words applied to data collection will be politically controversial or the result of politicised choices and lack universal consensus. However, transparency can ensure that these datasets are still trusted and useful.”

The solution “to best facilitate public health and scientific research,” they conclude, is that “databases should use terms that accurately describe the data collected and, if changes must be made, they should be clearly logged.”

The paper is published in The Lancet.