AI Prescribers, Mushroom Hype, mRNA's Next Targets

Editor's note: This is a STAT+ paywalled story. Only the first few paragraphs are free, so this summary is short. The free part has no effect sizes, trial names beyond one drug, or details on the "divergent results" in the headline.

What the free text says

In August, Merck and Moderna reported that their mRNA cancer vaccine, intismeran autogene, slowed the return and spread of melanoma when added to an existing treatment. The trial was late-stage. The article doesn't name the existing therapy, how big the benefit was, or how long patients were followed.

The vaccine uses messenger RNA to train the immune system to recognize and attack cancer cells. The approach was first devised decades ago. Karen Knudsen, CEO of the Parker Institute for Cancer Immunotherapy, said the field had been waiting "with bated breath" for the result. Reporter Allison DeAngelis writes that it suggests the approach is ready for more drug development.

STAT frames the open questions this way: which other cancers the approach could work in, and whether mRNA is the right platform for the job. The headline says vaccine trials have produced conflicting results, but those trials and their numbers sit behind the paywall.

How to read it

What's missing matters. Without the hazard ratios, the size of the trials, or which trials failed, a reader can't judge how far the melanoma result carries over. For now, the evidence here is one positive late-stage melanoma readout from a company announcement, plus a report that results elsewhere are mixed.

This is an opinion piece. It has no trial, no data and no effect sizes. Adam Cifu, an internist and one of the three Sensible Medicine authors, is answering his colleague Vinay Prasad's recent post, "Things you should have stopped doing already because of AI." A reader called Prasad's list "first-level thinking" and asked the obvious follow-up: if doctors stop doing those things, what should they do instead? Cifu gives seven answers.

Where Cifu thinks the freed-up time should go

Prepare for visits. Cifu says AI built into the electronic health record (EHR) can already give a decent summary of a patient's history with little prompting. That leaves the doctor to review test results and colleagues' assessments, and to walk in with an agenda they are willing to drop if the patient has a different one.

Spend time with patients. AI can help with diagnosis and treatment options. Cifu argues that working out who the patient is and what matters to them is still the doctor's job.

Document your thinking. He argues that billing rules have filled medical records with checkboxes and copy-pasted bloat, and that AI chart summaries make this padding even less useful. His suggestion: let AI draft the note formatted for billing. Then spend about five minutes writing your own paragraph or bullet list on the plan and the reasons for it, for yourself and for colleagues.

Talk to patients between visits. Cifu wants AI to take over routine messages to insurers, home care agencies and prior-authorization reviewers. With supervision, it could also run check-ins on blood sugar, blood pressure and weight. Doctors could then make the calls that matter, such as explaining an abnormal calcium result or a frightening EKG report.

K21 is a quaternary ammonium silane that kills viruses, bacteria and fungi by disrupting their membranes. It is also known to speed wound healing, which its antimicrobial action alone doesn't easily explain. A new paper in Life Science Alliance (doi:10.26508/lsa.202603852) looks for that second mechanism in macrophages. These immune cells drive wound repair. They first act as pro-inflammatory, infection-fighting cells, then switch to a pro-regenerative state that recruits fibroblasts and keratinocytes to rebuild the epithelium.

What the researchers did

The team grew macrophages from human peripheral blood monocytes. They chose these cells because they are easy to obtain, can be isolated in a fairly naïve state, and can be pushed toward M1 (inflammatory) or M2 (regenerative) states in a controlled way. They then combined cytokine and cell-surface marker profiling with single-cell RNA sequencing to see how K21 changes the mix of macrophage subtypes.

The authors report that K21 reshapes the macrophage population. They identify one key effect: it induces mitochondrial fission and mitophagy, the process that splits off damaged mitochondria and breaks them down. Better mitochondrial quality control leads to better mitochondrial function.

To check whether this happens across a whole organism and is conserved across species, they gave K21 to the nematode C. elegans. In the worms, K21 caused mitochondrial fission that depended on DRP-1, the worm's main fission protein, along with mitophagy. It did not harm viability, development or fertility. It also changed which metabolic genes were switched on and , which the authors say matches what they saw in human cells.

A note first: this is a STAT+ article and only its first few paragraphs are free. The headline mentions pilots in women's health, and the URL says Utah picked third-party auditors, but none of those details are in the free text. This summary covers only what is visible.

What Utah approved

On Monday, Utah regulators announced that a startup, Nolla Health, can launch an AI service in the state that assesses patients for mild-to-moderate acne and writes prescriptions. It works from two inputs: a structured intake questionnaire and photos of the patient's skin. From these, the AI decides whether one drug from a fixed list of low-risk acne medications is appropriate.

There are two stages:

• Stage one: a licensed clinician reviews each AI decision and signs off on every prescription.
• Later stages: Nolla may be allowed to prescribe with no human review at all.

The free text doesn't say what Nolla would have to show to reach that second stage. It also doesn't say which drugs are on the list or how errors will be tracked.

The sandbox behind it

The pilot is part of a large expansion of Utah's "AI sandbox." The program is run by the state's Office of Artificial Intelligence Policy (OAIP), which can waive regulations for companies testing AI products under state supervision. That includes the rules on who may practice medicine. Normally a licensed professional must make the prescribing decision. In the sandbox, the state can suspend that requirement while it watches how the product performs.

The company