Daily Briefing for October 31, 2026

Show Notes

AI chatbots recommend care costing 3 times what physicians order

The voices on this episode are generated, and so is the script, from the day's post at Physician in the Loop. Every fact in the post links to its source.

Read the post and its sources

Get the free email briefing

Twenty general-purpose AI chatbots given 200 outpatient and emergency department notes recommended care costing an average of $314 per visit, more than three times the $97 the treating physicians ordered, a NEJM AI study found.

Most of the gap came from tests the chatbots suggested at visits where physicians had ordered none, the authors reported. A separate Annals of Internal Medicine benchmark found six commercial models usually held back on low-value care but recommended thyroid ultrasound for incidental subcentimeter nodules in 42.8% of responses.

In other news, Medline is removing or correcting all of its beds after reports of two deaths and 16 serious injuries, U.S. measles cases reached 4,852 for the year, STAT reported that Epic has used its market power against rival app makers, and the Food and Drug Administration approved the first plasmid DNA immunotherapy, for recurrent respiratory papillomatosis.

In this episode

• AI chatbots recommend care costing 3.2 times what physicians ordered, mostly by adding tests at visits where doctors ordered none
• Medline pulls all its beds from the market after two deaths and 16 serious injuries from fire and entrapment
• U.S. measles cases reach 4,852 in 2026 as Pennsylvania tells clinicians to treat every newborn of an infected mother as exposed
• Twelve app makers tell STAT that Epic cut off data access, spread rumors or charged high fees to undercut them
• Congo's Ebola outbreak passes 10,000 confirmed cases, with nearly 5,000 deaths
• FDA approves Voceryq, the first plasmid DNA immunotherapy, for adults with recurrent respiratory papillomatosis

What is Physician in the Loop?

How technological innovation is changing medicine for patients and the people who practice on the front lines.

Voice 1: This is Physician in the Loop, the daily briefing on AI in medicine. It's Saturday, October 31st, 2026.

Voice 1: Researchers fed 20 general-purpose AI chatbots the same 200 outpatient and emergency department notes that real physicians had already handled, and the chatbots recommended care that cost 3.2 times what those physicians actually ordered. The physicians' documented plans averaged $97 a visit. The AI plans ran $216 higher.

Voice 2: Where did that extra cost actually come from? More imaging, more specialists, what?

Voice 1: Mostly tests. The physicians recorded no testing at 66% of visits, and at every single one of those visits, every AI system still suggested testing, usually some stereotyped panel. That's the Technion, Stanford and Tel Aviv University team, writing in NEJM AI.

Voice 2: Did telling the chatbots to be cost-conscious help at all?

Voice 1: It shrank the gap but didn't close it. And four specialized medical AI systems the researchers tested ran even higher than the general-purpose ones, as much as three and a half times the physicians' cost.

Voice 2: So this isn't one rogue chatbot. It's the whole category defaulting to more.

Voice 1: A separate benchmark, in Annals of Internal Medicine, tested six commercial models on 225 Choosing Wisely scenarios and actually found the opposite instinct more often than not. The models endorsed overuse in about 5% of responses and withheld appropriate care in under 1%.

Voice 2: So where's the overuse concentrated, if it's that rare overall?

Voice 1: Two specific calls. Nearly 43% of responses recommended thyroid ultrasound for incidental, subcentimeter nodules, and more than 28% recommended routine electrocardiography in low-risk adults.

Voice 2: Those are both the kind of order a tired resident reaches for out of habit, not out of judgment.

Voice 1: Which is the pattern across both studies. These systems aren't randomly wrong. They lean toward testing, consistently, even when careful prompting pulls them back.

Voice 2: A different kind of safety story just landed, and this one isn't about software.

Voice 1: Medline is pulling or correcting every one of its branded beds, homecare, bariatric, long-term care, the pendants, the replacement parts, all of it. The FDA posted an early alert on it October 29th.

Voice 2: What's actually failing on these beds?

Voice 1: Electrical parts can overheat into a fire with excessive loading, jammed motors, or damaged pendants, extension cords and unapproved accessories. Separately, patients have been trapped or asphyxiated when accessories are used wrong, when someone lies on a pendant, or when rails are loose. As of October 20th, Medline had reported two deaths and 16 serious injuries.

Voice 2: Do hospitals have to pull these beds out of service right now?

Voice 1: No, the FDA says they can stay in use until repaired or replaced, but only unplugged except when repositioning, never on an extension cord, never self-repaired, and ideally with a second person helping. Some beds bought in the last five years qualify for a free replacement or refund.

Voice 2: So it's a watch-and-maintain alert, not an immediate evacuation order.

Voice 1: Right. Measles moved the other direction this week, upward.

Voice 2: The CDC counted 4,852 confirmed cases for the year as of October 29th, almost all of it, 96%, tied to outbreaks. Three deaths, and 43 new outbreaks recorded so far this year.

Voice 1: And Pennsylvania just issued specific guidance for pregnancy, given rising cases in its outbreak there. What's it actually telling clinicians to do?

Voice 2: Treat every newborn of a mother infected with measles in pregnancy as exposed. If the baby's asymptomatic and born anywhere from the mother's symptom onset through 14 days after her rash appears, give immune globulin and manage the infant in airborne isolation. Nonimmune pregnant patients should get IVIG within six days of exposure.

Voice 1: Does the advisory push toward early delivery in these cases?

Voice 2: No, it says plainly that measles is not on its own a reason to deliver. The guidance is about isolation and prophylaxis, not timing the birth.

Voice 1: A money story closes it out, and it's about who controls the data behind all of this.

Voice 2: Twelve companies in Epic's own app marketplace told STAT that Epic undercut their businesses, cutting off data access, spreading negative rumors, or charging high fees. That's from an October 28th report built on interviews with more than 50 people.

Voice 1: How much leverage does Epic actually have to do that?

Voice 2: A lot. It holds records on roughly 8 in 10 Americans and brings in more than $6.5 billion a year. Epic's side is that it offers more data connections than federal law even requires. Some hospital executives and developers told STAT the opposite, that its market power is what's keeping better software out.

Voice 1: A hospital picking an app from that marketplace is choosing from whatever Epic lets compete, not necessarily whatever works best.

Voice 2: That's the accusation, anyway, and it's just the opening piece of a series. Twelve companies willing to go on the record against the vendor that holds most of their customers' data is not a small thing.

Voice 1: Sources and corrections are at physicianintheloop.org, where you can sign up for the free email briefing. This briefing was read by AI voices.