An MIT study finds non-clinical information in patient messages, like typos, extra whitespace, or colorful language, can reduce the accuracy of a large language model deployed to make treatment recommendations. The LLMs were consistently less accurate for female patients, even when all gender markers were removed from the text.

The COVID-19 pandemic yielded important advances in testing for respiratory viruses, but it also exposed important unmet needs in screening to prevent the spread of infections in high-risk settings.

While PCR (polymerase chain reaction) tests are the gold standard for detecting viral infections, they remain a challenge for screening large numbers of people in places vulnerable to outbreak — such as health care centers and nursing homes — due to high costs and the fact that different tests are required for each virus.

A new Yale study, however, finds that an alternate strategy — using a nasal swab to screen for an antiviral protein produced by the body as a defense against infection — can be an effective method for ruling out respiratory infections, limiting PCR testing only to those most likely to be infected, at a fraction of the cost. 

The study was published online on June 20 in The Lancet eBiomedicine.