Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

Physician in the Loop

Physician in the Loop artwork

Daily Briefing for October 10, 2026

Oct 10, 20265 minAudio

FDA clears Epic deterioration model; Bayer sells Blackford radiology AI; chest X-ray AI flags osteoporosis; Houston Methodist ties OR AI to more cases 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 The Food and Drug Administration has cleared Epic Systems' Deterioration Index Version 2, a model in Epic's record that predicts which hospitalized adults will decline, in a 510(k) decided Sept. 18, agency records show. The model is Epic's first AI clearance from the agency's device center, according to a review of September clearances by the consultancy Innolitics. Its validation showed lower sensitivity for Black and Hispanic patients than for white patients at the medium-risk threshold, Becker's Hospital Review reported. In radiology, Sopris Capital closed its purchase of Blackford Analysis from Bayer, and Korean researchers reported a chest X-ray model that identified osteoporosis in three outside cohorts. In surgery, Houston Methodist's innovation chief described an operating-room camera system linked to 7% more cardiothoracic cases a month, and a Northwell Health team published a skills score built from robotic console data. In this episode FDA clears Epic's Deterioration Index Version 2, an inpatient decline model hospitals have run inside Epic for years Sopris Capital buys Blackford Analysis, a platform carrying more than 130 radiology AI apps, from Bayer and plans to merge it with Azra AI Chest X-ray model identifies osteoporosis with AUROCs of 0.88 to 0.93 in three Korean validation cohorts of more than 150,000 exams Houston Methodist links operating-room camera AI to 7% more cardiothoracic cases a month; the hospital holds equity in the vendor Robotic console data separate attending surgeons from trainees in inguinal hernia repair with an outside-cohort AUC of 0.84

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show.

Voice 1: This is Physician in the Loop, the daily briefing on AI in medicine. It's Saturday, October 10th, 2026.

Voice 1: The FDA just gave Epic its first AI clearance ever from the device center, and the validation data show the model catches far fewer at-risk Black and Hispanic patients than white ones.

Voice 2: What is this thing actually supposed to catch?

Voice 1: It is the Deterioration Index Version 2, built into the Epic record, scoring adult inpatients zero to 100 for risk of an ICU transfer, a rapid response call, a code within 24 hours, or death within 72 hours. It uses 195 inputs from the chart, and each hospital sets its own alert threshold.

Voice 2: And the gap in who it flags?

Voice 1: At the medium-risk threshold, sensitivity was 26% for Black patients, 32% for Hispanic patients, and 51% for white patients. That came out of retrospective testing on about 10.7 million scores across three U.S. sites, where the overall area under the curve was 0.83.

Voice 2: Is this a brand-new model, or just a fresh stamp on something hospitals already run?

Voice 1: Hospitals have run earlier versions for years. The FDA cleared this one September 18th as substantially equivalent to AgileMD's eCARTv5, about eight months after Epic filed. It is locked, with no predetermined change control plan, so a retrained version would likely need a new submission.

Voice 2: Locked models age. Is there any sense of how the earlier version actually performed?

Voice 1: A 2024 Yale New Haven Health study ranked that earlier version near the bottom of six early warning scores.

Voice 2: Different kind of deal in radiology. Sopris Capital just closed its purchase of Blackford Analysis from Bayer.

Voice 1: What does Blackford actually give an imaging department?

Voice 2: Access to more than 130 AI applications from more than 45 vendors, all through one platform. The price was not disclosed. Bayer bought Blackford in 2023 and then moved away from its AI platform business the following year.

Voice 1: So what happens to it now?

Voice 2: Sopris plans to fold it into Azra AI, a company it already owns, in the coming weeks. Blackford stays a separate legal entity, with operations in Scotland and the United States.

Voice 1: A platform that size changing hands twice in a few years says something about how unsettled this corner of radiology AI still is.

Voice 2: Speaking of radiology, there is a chest X-ray model out of Korea that reads bone density without a DXA scan.

Voice 1: How well does it actually work?

Voice 2: Researchers at Seoul National University Bundang Hospital trained it on 46,478 chest radiographs paired with DXA scans. In a later hospital cohort of 6,391 exams, the area under the curve for osteoporosis was 0.88. In two health-checkup cohorts of 24,132 and 122,535 exams, it hit 0.93 and 0.92.

Voice 1: Does that translate into fewer scans, or just a better reading of the same ones?

Voice 2: In a simulation, using the model to decide who gets a DXA scan cut the number of scans needed by about half, while still catching 98% to 99% of osteoporosis cases.

Voice 1: Any caveats worth flagging?

Voice 2: It is observational and retrospective, and several authors reported ties to imaging AI companies, which they said were unrelated to this work.

Voice 1: A different kind of imaging story out of Houston Methodist, this one about operating rooms, not pictures.

Voice 2: What is actually being measured there?

Voice 1: Roberta Schwartz, the hospital's chief innovation officer, described an operating-room camera system from Apella tied to 7% more cardiothoracic cases a month. Volume rose by about 25 cases over a baseline of about 339, measured against a synthetic control built from 11 sites that did not have the system.

Voice 2: Did that hold up statistically, or is this one of those gains that looks better in a press release than a methods section?

Voice 1: A placebo permutation test of that gain did not reach conventional significance, and a 2.7-minute drop in median late-start minutes was described by the authors as exploratory.

Voice 2: And the hospital has a financial stake in the company making this claim?

Voice 1: It does. Houston Methodist holds a minority equity interest in Apella, and an Apella employee ran the original analyses, which the hospital's own team then replicated.

Voice 2: A hospital replicating its own vendor's numbers is better than nothing, but it is not the same as an outside check.

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.

Open the transcript file

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. Special Topic: What changes on Jan. 1

    Nov 20, 20267 minAudio

    Laws in at least eight states that touch AI in health care take effect on Jan. 1. Most bind insurers and chatbot companies. A few, in California and Colorado, reach the practice itself. Read the special topic and its...

    Credit the people on this episode

  2. Friday Letter: On a short leash

    Nov 20, 20267 minAudio

    Software has begun to act for doctors, not just advise them. It should be kept on a short leash, because the doctor watching it rarely notices when it slips. Read the letter and its sources:...

    Credit the people on this episode

  3. Daily Briefing for November 20, 2026

    Nov 20, 20266 minAudio

    Hospitals race to patch a Java flaw found with AI in an hour 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...

    Credit the people on this episode

  4. Special Topic: Not the doctor's to sign

    Nov 19, 20267 minAudio

    Free AI tools for doctors come with a HIPAA contract one doctor can accept alone. A hospital's records need the hospital's signature, and the companies should make the safe path the default. Read the special topic and...

    Credit the people on this episode

  5. Daily Briefing for November 19, 2026

    Nov 19, 20267 minAudio

    AI hacking agents seize some networks in under 3 hours, FBI warns 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....

    Credit the people on this episode