Weekly health-care AI brief

Issue 02 · Unsent dry run

Good morning. The medical record finally stopped acting like a filing cabinet.

The interesting part is not that ChatGPT can read a chart. It is that the systems holding the chart are beginning to open doors AI can walk through. This week, that same shift shows up in an ECG and a protein map, while a familiar drug starts a very different conversation about aging.

Four stories worth your time

Start with the one that matters to you

Lead story · From record to workflow

Epic’s walls are getting a new door →

Second look · Detection

A century-old heart test learned a second language →

Deep signal · Discovery

One hundred autism genes start looking like fewer problems →

Early signal · Longevity

The GLP-1 story just acquired a second clock →

A dark digital anatomical heart connected by neural pathways

From record to workflow · 5 minute read

Epic’s walls are getting a new door

OpenAI says health-care organizations can now bring authorized patient context from supported Epic environments into ChatGPT for Healthcare—or place ChatGPT inside the EHR workflow. The same release adds structured access to nine public sources, including PubMed, DailyMed and ClinicalTrials.gov.

That is a bigger move than another medical chatbot. Appointment notes, lab results, medications and specialist documentation can be reviewed in one governed workspace, with answers pointing back to the chart. OpenAI reports that physicians rated 99.1% of responses safe across 4,363 ratings and 27 EHR use cases. Useful evidence—but still vendor-reported evidence, not an independent outcome study.

Why this matters

The model is no longer the whole product. Governed access to the record—and a clean handoff back into the clinician’s existing workflow—is becoming the real distribution advantage.

Open the primary record →

Three more major stories

The rest of the briefing

Second look · 3 minute read

A century-old heart test learned a second language

A British Heart Foundation-backed model trained on 10.6 million ECGs found signals the human eye cannot see in under two seconds. In two U.S. test groups it identified up to 81% of reduced heart-pumping function and up to 90% of aortic stenosis. It is a triage flag—not a diagnosis—and is now being tested on 590 NHS patients. Open the record →

Deep signal · 3 minute read

One hundred autism genes start looking like fewer problems

UCSF mapped more than 1,800 protein interactions across 100 high-confidence autism-risk genes; 87% had not been reported before. AlphaFold helped locate where mutations disrupt protein interfaces, and organoids showed distinct mutations converging on a smaller set of shared machinery. The result is a therapeutic map, not a treatment—but it may turn hundreds of genetic starting points into fewer drug targets. Open the record →

Early signal · 2 minute read

The GLP-1 story just acquired a second clock

Researchers started semaglutide in 20-month-old female mice and continued treatment through the lifespan. Median survival rose from 742 to 834 days, alongside improved function and fewer markers of age-related decline. The caveat is the story: this is one mouse strain, one sex and one regimen. Whether GLP-1 treatment changes human aging will require long-term clinical studies. Open the paper →

One more signal

Quick hit

The lab bench

Anthropic’s Model Hardware Standard gives AI agents one interface for microscopes, liquid handlers and robot arms. The first drug-discovery examples are proofs of concept—not autonomous science. →

What we’re watching

Can the two-second ECG flag survive real clinical use?

The retrospective numbers are strong. The useful number will come from the 590-patient NHS test: whether the flag finds the right people early enough to change who gets an echocardiogram—and what happens next.

Reader pulse

Which signal should we follow first?

AI inside the medical record →

AI finding disease in routine tests →

One click. We’ll share the results next week.

The Next Patient

A weekly briefing on the AI developments that could change health care.

View online · Forward · Unsubscribe