Prompt Engineering in Med School; Messages From Dr. ChatGPT; AI Scientist’s Papers

Welcome to MedAI Roundup, highlighting the latest news and research in healthcare-related artificial intelligence each month.

Doctors — especially medical school students — should get training in “prompt engineering” to optimize generative AI in medicine, researchers argue in a letter in Academic Medicine.

AI chatbots are playing more of a role in generating patient-facing medical information, from answering direct medical questions to drafting messages from physicians. (New York Times)

In the U.K., 1 in 5 primary care physicians reported using AI chatbots in their practice, according to a study in BMJ Health & Care Informatics.

A new “AI scientist” can write and review its own research papers, but its usefulness is still in question. (Nature)

ChatGPT version 3.5 outperformed trainee doctors at assessing complex respiratory illness in children, according to a study presented at the European Respiratory Society Congress.

Scientists are using AI to crack “the olfactory code,” helping to unravel the mysterious biology of smell. (Nature)

About half of clinicians in a small trial said AI-powered clinical documentation tools offered time management benefits, but a substantial proportion said it had no impact on their overall EHR experience, according to a paper published in JAMA Network Open.

Another study in JAMA Network Open found that machine learning improved electronic detection of potential diagnostic errors.

Researchers writing in Nature Medicine say they’ve developed a much-improved model for using AI in drug repurposing.

Using an AI-based early warning system to detect risk of adverse events among high-risk patients led to fewer non-palliative hospital deaths than before the system was implemented, according to a study in the Canadian Medical Association Journal.