MEC Veradigm 8.26
News|Articles|August 21, 2026

Trump picks Heidi Overton to lead the FDA; CMS tightens prior auth disclosure rules; JAMA authors say AI alone will beat doctors — Morning Medical Update Weekly Recap

Fact checked by: Keith A. Reynolds

Key Takeaways

  • President Trump nominated Heidi Overton to be FDA commissioner amid concerns about limited large-agency management experience, involvement in vaccine scheduling policy, and potential influence over mifepristone regulation.
  • Senate HELP Chair Bill Cassidy signaled substantial reservations, citing FDA staffing and morale challenges and calling the vaccine-order role “almost disqualifying,” while the White House emphasized her interagency coordination experience.
SHOW MORE

The top news stories in medicine this week.

Trump picks Heidi Overton to lead the FDA

The Senate health committee chairman says her role in the Aug. 10 childhood vaccine order is almost disqualifying on its own.

President Donald Trump announced Aug. 19 on Truth Social that he is nominating Heidi Overton, M.D., Ph.D., his deputy assistant for domestic policy, as commissioner of the U.S. Food and Drug Administration (FDA).

Overton earned her medical degree from the University of New Mexico and a doctorate in clinical investigation from Johns Hopkins University, and she was chief policy officer at the America First Policy Institute, a conservative think tank, before joining the administration. If confirmed, she would run an agency of roughly 18,000 employees with authority over prescription drugs and medical devices, the food supply, tobacco, supplements and cosmetics. The FDA has been run by acting commissioner Kyle Diamantas since Marty Makary, M.D., resigned in May.

Two parts of her record are drawing scrutiny. Overton appeared with Trump at the Aug. 10 signing of an executive order calling for childhood vaccines to be spaced out and for the combined measles, mumps and rubella (MMR) shot to be split into three single-disease vaccines. At that event she said the administration still recommends the measles vaccine for all children but wants parents informed about timing and given the option to separate the doses.

She has also long opposed abortion. In papers written at the America First Policy Institute, she argued that mifepristone is dangerous to women and called on the federal government to ban its distribution by mail. As commissioner, Overton would oversee the rules governing it.

Bill Cassidy, M.D. (R-Louisiana), a physician who chairs the Senate Health, Education, Labor and Pensions (HELP) Committee, wrote on X, formerly Twitter, that while he respects Overton's experience as a physician, he has strong concerns about the nomination, citing her lack of managerial experience at an agency already dealing with staffing and morale problems, and her role in the vaccine order, which he called "almost disqualifying." He has not said how he will vote. White House spokesman Kush Desai said Overton has spent the past year and a half coordinating an interagency policymaking process with senior officials across U.S. health agencies, and will play a similar role at the FDA.

CMS tightens prior authorization disclosure rules after AMA review

The AMA found one plan posting an 832-page list of billing codes without a word of plain English.

CMS has updated its guidance on federal prior authorization transparency requirements, the American Medical Association (AMA) announced Aug. 17, strengthening enforcement of a 2024 rule that requires payers to publicly post which services need prior authorization and how those requests turn out. Earlier this year the AMA examined how 15 Medicare Advantage contracts were implementing the transparency provisions of the 2024 Interoperability and Prior Authorization final rule and found what it called a consistent pattern: plans that appeared to comply on the surface while presenting disclosures in places and formats that made them difficult or impossible to find and use.

One plan posted an 832-page list of billing codes without a word of plain English, AMA President Willie Underwood III, M.D., M.Sc., M.P.H., said in a statement. Others buried required information behind physician or member portals and deep within plan websites, reported turnaround times without units, published statistics that did not add up, and omitted entire categories of care including behavioral health and post-acute services.

The AMA documented the problems and recommended corrective actions in a May 22 letter to CMS, followed by additional comments.

CMS incorporated several of those recommendations. Disclosures are not publicly accessible, the agency says, if they are available only through password-protected portals or cannot be reached through ordinary navigation from a payer's public-facing website. Plans must publicly identify all medical items and services requiring prior authorization.

Lists of procedure codes without plain-language descriptions do not satisfy the requirement, and CMS recommends a single comprehensive list organized by uniform service categories, with CPT codes, plain-language descriptions and a machine-readable format.

Every turnaround-time metric must include a unit of time, and median turnaround times of less than one day must be reported in hours rather than rounded to zero days.

The AMA is urging CMS to go further by defining prior authorization based on how the process functions regardless of whether a payer calls it precertification or advance notice or delegates it to a third-party utilization management vendor, by linking directly to plan prior authorization requirements from Medicare Plan Finder and HealthCare.gov, and by requiring rather than recommending standardized reporting templates.

Underwood called the guidance an important step but said "the work is not finished."

Related content: 10 numbers that show prior authorization is still broken and A year after the prior authorization pledge, what changed? with Colin Banas, M.D., of DrFirst

JAMA authors say AI alone will outperform physicians and physician-AI teams by 2030

Once AI alone beats physicians at a task, the authors argue, putting a physician back in to check the work makes results worse.

A perspective published Aug. 17 in JAMA argues that artificial intelligence (AI) alone already matches or exceeds physicians at five core cognitive medical tasks, and that adding a physician to review the AI's work degrades performance rather than catching errors. The five tasks are eliciting medically relevant information, establishing a differential diagnosis, specifying diagnostic testing, prescribing guideline-concordant treatment and managing chronic disease. The authors are Ezekiel J. Emanuel, M.D., Ph.D., and Abe Baker-Butler, of the University of Pennsylvania; Neal Khosla, M.S., CEO of Curai Health; and Vinod Khosla, M.S., MBA, of Khosla Ventures.

Among the studies cited: OpenAI's o3 model ranked the correct final diagnosis first in 60% of 377 real-world complex cases, while 20 internal medicine physicians did so in 15.9% of a 302-case subset. Microsoft AI Diagnostic Orchestrator, held to an $8,000 testing budget on 56 complex cases, reached the correct diagnosis 4.02 times more often than physicians working without access to colleagues, textbooks or the internet, at 19.1% lower cost per case. A 2024 meta-analysis of 106 human-AI experiments found that pairing a human with AI improved results when the human was already the better performer, but significantly worsened performance when AI alone was better.

Neal Khosla is CEO of Curai Health, an AI primary care company, and Vinod Khosla has invested in OpenAI, Curai Health and Limbic. Their position runs counter to the AMA and the American College of Physicians, both of which hold that AI should support rather than replace physician decision-making. AMA CEO John Whyte wrote on LinkedIn that medicine "is not a collection of tasks," and compared AI-only care to boarding a plane with autopilot but no pilot in the cockpit.

The authors list their own caveats. Most of the supporting evidence comes from simulations of discrete cognitive tasks rather than real clinical encounters. Liability, regulatory and reimbursement barriers limit study of AI alone in live settings. Adversarial stress testing shows large language models remain brittle in some cases. And many cognitive tasks are tied to physical procedures such as surgery, deliveries, interventional radiology and colonoscopies that robotics cannot yet perform autonomously.

Even so, the authors argue, superior autonomous AI will likely be ready to deploy in some, maybe many, workflows by 2030.