US government has altered more than 100 scientific databases since 20 January

On 3 July 2025, details of changes in numerous US public health datasets were reported in a letter to The Lancet. [1]

The researchers compared changes in words (not numbers) in 232 selected databases between 20 January to 25 March 2025 to archived versions, and found substantial alterations in almost half (114/232). These commonly involved changing gender to sex including in historical datasets and most of these changes were not formally recorded.

The datasets included US Veterans Affairs, the US CDC datasets tracking tobacco use, stroke mortality data, and a survey of nutrition, physical activity and obesity.

The authors noted limitations in their study that did not allow for analysing the numerical data but still showed that US public health agencies have been “altering the contents of those datasets in ways that might be politically motivated and not transparent”.

They also note that changing gender to sex is not a trivial alteration because it “degrades the quality of the information” and “changes the accuracy of the dataset and the conclusions that can be drawn”.

The US government hosts more than 300,000 datasets including 173 related to HIV and unrecorded data manipulation risks making these sources untrustworthy and unusable. [2]

comment

Many of the altered records were datasets that had been sealed and capped years ago, representing decades of work by thousands of researchers.

It is a signal that the current US administration is not a fit custodian for any scientific archives. Bedrock institutions including the National Library of Medicine and PubMed rapidly need to be duplicated on mirror sites.

Although the US Department of Health and Human Sciences has a $94 billion budget, this was recently cut from $126, representing a 25% reduction, and the new Secretary has already cancelled thousands of ongoing studies.

The most prominent and established HHS HIV guidelines have been recently identified as a project that could easily be cut.