Happy Sunday — and congratulations to the doctors trialling a potential cure for melanoma. Called the world’s first “personalized mRNA cancer vaccine for melanoma,” this could be a game changer. This neoantigen therapy is customized for each patient. It’s being tested for other cancers, too, like lung, bladder, and kidney. We wish the doctors and patients — and anyone else with skin in the game — good luck in this phase 3 trial!

Here are some fun stories to get you pumped for the weekend! (5 minute read)

1. Frontline tales (and fails)

2. US fertility rate is… low.

3. Pro-Palestine protests on US campuses

4. The end of paperwork, or is it?

5. Postcall Crossword and Image Challenge

Frontline Tales ⛺️

Have a riveting tale of your own? Send us your best story from the OR, ED, or clinic (anonymized, obviously).

Speed Reads 🚤 

1: 🍼 The US fertility rate has dropped to the lowest it’s been in more than a century. About 3.6 million babies were born in 2023, which is 54.4 live births for every 1,000 pregnant people. The previous low was in 2020, when there were 56 births for every 1,000 people. It’s interesting to note that 2023 was also the first full year the Supreme Court revoked the federal right to abortion.

2: 🫀 The first surgery to combine a heart pump and a gene-edited pig kidney was completed this past week. NYU Langone Health says the pig’s thymus gland was placed under the kidney to help the patient’s immune system accept the organ. Earlier this month, the world’s first recipient of a gene-edited pig kidney transplant was cleared to go home.

3:🚬 The Biden administration has indefinitely delayed plans to ban menthol cigarettes — a move that has been recommended by the FDA since 2021. Critics wonder if this move prioritizes politics over health concerns. According to the FDA, nearly 85% of Black smokers smoke menthol cigarettes, compared to just 30% of white smokers.

4:🖼️ Have you ever wanted to see the Mona Lisa rap? Well, want no more, thanks to Microsoft’s new AI, VASA-1. The tech was designed to recognize natural face and head movements, and works best when animating a still photo. Like most AI products, there’s a privacy concern surrounding the development, and Microsoft says they don’t plan on releasing VASA-1 to the public anytime soon.

5: 🪧 Across US college campuses, students are rallying to demand that universities divest from Israel and companies linked to its military activities. Echoing past movements against apartheid and fossil fuels, protesters aim to pressure universities into taking a moral stand. The complexities of endowment investments and mixed administrative responses suggest that achieving divestment is both challenging and controversial.

Paperwork may yet stay awhile 📝

“AI scribes are going to do all your paperwork and save you from burnout.” We’ve all heard it before, but now studies are emerging to test that claim.

What Happened? A study was conducted at Brigham and Women's Hospital to evaluate the impact of using large language models (LLMs) like GPT-4 in handling electronic patient portal messaging.

• Initially, six board-certified radiation oncologists manually responded to patient messages as they would in typical clinical practice.

• Following this, the same oncologists edited responses that had been pre-drafted by the GPT-4 model, ensuring these were clinically acceptable before sending.

• The responses were analyzed using a set of predefined content categories, and statistical significance was determined through pairwise comparisons. Both stages of responses were evaluated by the oncologists through surveys focusing on quality, safety, and helpfulness, with interphysician agreement also assessed to gauge consistency across responses.

So what happened?

The use of LLMs notably improved subjective efficiency in handling patient messages. An impressive "76.9% of 156 cases" reported enhanced efficiency when using LLM-drafted responses. However, this increased efficiency comes with potential dangers. Notably, the LLM drafts were found to pose a risk of severe harm in "11 (7.1%) of 156 survey responses" and resulted in a life-threatening error in one instance (0.6%). This underscores the critical need for oversight and careful editing by clinicians to prevent serious errors.

As well, the adoption of LLM drafts led to a higher interphysician agreement, with Cohen’s kappa (a statistical measure used to quantify the level of agreement between two people who each classify items into categories) increasing from 0.10 in manual responses to 0.52 in LLM-assisted responses. This improvement suggests that LLMs can help standardize responses across different clinicians. However, the content analysis showed a significant shift in the nature of the responses: LLM drafts were "less likely to include content on direct clinical action" and more likely to offer "extensive education, self-management recommendations, and a contingency plan." This shift indicates a possible over-reliance on educational content at the expense of immediate clinical directives.

This study follows other recent research papers by academic hospitals that are revealing limitations of large language models (LLMs) in medical settings, undercutting the AI hypetrain. Another study from University of California, San Diego found that use of an LLM to reply to patient messages did not save clinicians time, and another at Mount Sinai found that popular LLMs weren’t able to correctly map patients’ illnesses to diagnostic codes.

We may yet endure the drudgery of documentation, for now.

Source: Lancet Digital Health

Laugh of the Day

🕹️ Game ⛳️

Can you beat last week's average time of 4:24 👀

First question: What the drug does to the body…

Enjoyed our puzzle? Send it along to friends & co-workers!

Image Challenge 📸 

Source: Originally published on NEJM

Share Postcall, Get Merch.

🤘 PS. If you’re reading this, you know that our readers are America’s best and brightest physicians. Learn about partnering with Postcall if you’d like to reach them.