Microsoft AI outdiagnoses generalists on complex cases in Science study; LLMs recommend low-value care in 5.4% of cases; FDA finalizes app guidance

• AI Oct 22, 2026
  Microsoft's MAI-DxO diagnoses 83.2% of complex NEJM cases versus 36.1% for generalist physicians in Science study
  Microsoft's MAI-DxO system correctly diagnosed 83.2% of complex New England Journal of Medicine cases, compared with 36.1% for generalist physicians using search and reference tools and 22.3% without them, according to a study published Oct. 22 in Science. The physician comparison used a held-out set of 56 cases. On 325 emergency department admissions from the MIMIC-IV-ED dataset, it reached 80.3% accuracy versus 46.8% for physicians. The authors wrote that the text-only comparisons "do not establish superiority over specialists, multidisciplinary teams, or real-world clinical practice." All authors are Microsoft employees with equity, and Microsoft funded the research, Becker's Hospital Review reported.
  Sources: Microsoft AI, Becker's Hospital Review
• AI Oct 19, 2026
  Six LLMs recommend low-value care in 5.4% of cases and withhold appropriate care in 0.6% in Choosing Wisely test
  A team from Clalit Health Services in Tel Aviv tested six commercial large language models (LLMs), which generated 4,050 responses to 225 scenarios built from 15 Choosing Wisely recommendations, in a study published Oct. 19 in Annals of Internal Medicine. Overuse occurred in 5.4% of cases, and eight of the 15 recommendations drew no overuse. Most overuse came from thyroid ultrasound for an incidental subcentimeter nodule (42.8% of responses) and routine electrocardiography in low-risk adults (28.3%), according to AuntMinnie.
  Sources: Annals of Internal Medicine, AuntMinnie
• FDA Oct 20, 2026
  FDA finalizes guidance on when drug makers' companion apps belong on a drug's prescribing label
  The Food and Drug Administration (FDA) issued final guidance Oct. 20 on how drug companies can develop, distribute and label apps. Apps connected to a medical device, such as an app paired with an asthma inhaler, generally should be described on the drug's prescribing label, while a standalone app such as a symptom tracker or dose reminder should appear there if the company can show it is essential for safe and effective use, the FDA said. "Companion apps are quickly moving beyond simple dose reminders toward tools that adapt to each patient," said Jared Seehafer, FDA deputy commissioner for technology.
  Source: FDA

The weekly innovation roundup for internal medicine, published every Sunday for the week before; every item links to its sources. Latest internal medicine news · All specialties