Medical Maximalism: Physicians on OpenEvidence, Over-Testing, and When to Stop

On November 3, 2026, nephrologist Brent A. Williams, MD posted on X that OpenEvidence “definitely recommends over-testing and overtreating.” The next day he called the pattern “medical maximalism.” A heart rhythm doctor and a lung cancer oncologist replied with related criticisms. OpenEvidence and its Chief Medical Officer answered in the same thread. And the original post that started the discussion was deleted.

Who said what

• Brent A. Williams, MD (nephrologist): OpenEvidence recommends over-testing and overtreating.
• Giuliano Palumbo, M.D. (lung cancer medical oncologist, IRCCS Pascale, Naples): guidelines rarely say when to stop.
• John Mandrola, MD (heart rhythm doctor, editor of Sensible Medicine): it stays “too close to the party line.”
• PulseMed (medical AI company): tools are “rewarded for completeness rather than appropriateness.”
• Travis Zack MD (Chief Medical Officer, OpenEvidence): “I agree!!!” A fix is in progress.
• Aakash Desai (thoracic and phase 1 oncologist, O’Neal Cancer Center at UAB): build a “#ChoosingWisely version of models.”

What is medical maximalism?

It is Williams’s term for clinical AI that recommends more testing and more treatment. His first post (since deleted) (@BrentAWilliams2):

“I do like Open Evidence, but the main issue that I have with it is that it definitely recommends over-testing and overtreating.” — Brent A. Williams, MD

He used the phrase the next day, in a reply to Giuliano Palumbo, M.D.:

Yes I think that the are trained on a model of medical maximalism.

— Brent A. Williams, MD (@BrentAWilliams2) November 4, 2026

Why would a clinical AI tool over-test?

The thread gave two explanations.

The guidelines. Palumbo (@GiulyPalumbo) is a lung cancer medical oncologist at IRCCS Pascale in Naples, Italy. He sees the same issue across large language models:

This an issue I found also with LLMs; I Guess It derives from the fact that guidelines, which LLMs rely Upon, tell you what to do, but rarely there are indications on when to stop or avoid treatments.

— Giuliano Palumbo, M.D. (@GiulyPalumbo) November 4, 2026

The incentive. PulseMed (@PulseMedAI), a medical AI company, replied:

Over-testing is the predictable failure mode when a tool is rewarded for completeness rather than appropriateness. Every recommendation should carry its evidence threshold and its cost, and the clinician who orders remains accountable for whether it changes management.

— PulseMed (@PulseMedAI) November 4, 2026

Is OpenEvidence good at critical appraisal?

John Mandrola, MD (@drjohnm) says no. Mandrola is a heart rhythm doctor, a writer for Medscape and editor of Sensible Medicine. His example was Watchman:

It stays too close to the party line, IMO. Ask it about Watchman, for instance. Not good at critical appraisal.

— John Mandrola, MD (@drjohnm) November 4, 2026

How did OpenEvidence respond?

Twice, both within hours. The company account (@OpenEvidence) asked Williams for examples:

Appreciate your feedback. Would it be possible to DM us examples or queries where that happens?

— OpenEvidence (@OpenEvidence) November 4, 2026

Travis Zack MD (@TravisZack2), Chief Medical Officer at OpenEvidence, quoted Williams’s post and agreed. He said past versions of the differential diagnosis model were not trained to prioritize, and named the team member working on it:

I agree!!! Especially past iterations of our ddx model wasnt trained to help prioritize, and instead just lists "to rule out x, do y" @MondiraRay is a physician scientist on our team actively working on this. More to come that will both narrow to the evidence based next steps and even request the most pertinent pieces of missing information to help us do that

— Travis Zack MD (@TravisZack2) November 4, 2026

What did an oncologist ask OpenEvidence to build?

A Choosing Wisely version. Aakash Desai (@ADesaiMD), a thoracic and phase 1 oncologist at the O’Neal Cancer Center at UAB, replied to Zack:

That is great to hear, would be great to build a #ChoosingWisely version of models where #AI can actually help in reducing cost of care while maintaining high level of diagnostic/testing accuracy!

— Aakash Desai (@ADesaiMD) November 4, 2026

Did physicians say to stop using OpenEvidence?

No. Williams’s post opened with “I do like Open Evidence.” The criticism was specific: over-testing, overtreating and weak critical appraisal.

Zack described two changes: answers that “narrow to the evidence based next steps” and a model that will “request the most pertinent pieces of missing information.”

Why does this matter for oncology?

Two physicians in the thread, Palumbo and Desai, treat lung cancer. The same questions come up in the oncology AI posts KOL Pulse tracks:

• The Curator Era: 6 Ways AI Is Already Rewiring the Oncology Community covers OpenEvidence as an AI knowledge layer, the verification paradox and automation bias.
• ESMO 2026: Oncology is the Tip of the Spear for AI in Healthcare covers the JAMA analysis of OpenEvidence use by specialty.
• The KOL Pulse AI in Oncology Digest collects what physicians post about clinical AI tools.

Frequently asked questions

What is medical maximalism?

It is the term nephrologist Brent A. Williams, MD used on X on November 4, 2026 for clinical AI tools that recommend more testing and more treatment. He wrote that the tools are trained on a model of medical maximalism.

What did physicians say about OpenEvidence and over-testing?

Brent A. Williams, MD wrote that he likes OpenEvidence but that it definitely recommends over-testing and overtreating. John Mandrola, MD wrote that it stays too close to the party line and is not good at critical appraisal. Giuliano Palumbo, M.D. wrote that guidelines tell you what to do but rarely say when to stop or avoid treatments.

How did OpenEvidence respond?

The OpenEvidence account asked Williams to send examples by direct message. Travis Zack MD, Chief Medical Officer at OpenEvidence, replied that he agreed, said past versions of the company's differential diagnosis model were not trained to help prioritize, and said a physician scientist on the team is working on it. Thoracic oncologist Aakash Desai replied that it would be great to build a #ChoosingWisely version of models.

Did the physicians say to stop using OpenEvidence?

No. Williams opened his post with the words "I do like Open Evidence." The criticism in the thread was about over-testing, overtreating and critical appraisal.

Why would a clinical AI tool recommend too much testing?

Two explanations were given in the thread. Giuliano Palumbo, M.D. pointed to guidelines, which say what to do but rarely when to stop. PulseMed wrote that over-testing is the predictable failure mode when a tool is rewarded for completeness rather than appropriateness.

All quotes are verbatim from public X posts made November 3–4, 2026.