AI が医療システムを改善し、誰もがマヨ・クリニック並のケアを受けられる可能性

My wife, Autumn, has spent nearly a quarter of the past four years in ERs and hospitals, untangling and battling three chronic conditions — and a shamefully broken U.S. medical system.

We live in Washington, D.C., with top-rated hospitals in our backyard. Yet her experience has been eye-opening and often horrifying, especially for a nation that spends twice as much on medical care as our rivals.

What it's like: ERs so hauntingly jammed she's often stacked like cargo in hallways, on a stretcher, waiting double-digit hours for care.

Well-intentioned and often talented nurses overwhelmed and overworked.

Doctors forbidden from, or uninterested in, working with experts beyond their domain to solve complex conditions. It's random luck if you get a specialist willing to even show up to your hospital room.

You need a forensic detective to hunt down medical records scattered across random portals and hospitals. And then a miracle to get anyone to piece them together for clues. "It feels like riding a stationary bicycle when you need to be somewhere," Autumn says.

And we're the lucky ones, with blue-chip insurance, a sophisticated understanding of the system and resources to attack problems. Most patients don't have those things.

All of this left us infuriated and hopeless — until we landed in a small Minnesota city and glimpsed medical nirvana: the Mayo Clinic in Rochester. It was like discovering a foreign land — or seeing a future that could and should be.

The best part: I soon discovered AI could easily help unlock the best of Mayo for everyone. Dr. Gianrico Farrugia, a gastroenterologist who's president and CEO of the Mayo Clinic, told me he's racing to show the federal government and other hospitals how to replicate the Mayo magic.

How it works: Mayo doctors work as teams. They're paid to heal, not to operate or medicate. Nurses handle four or so patients — fewer than we've experienced at even some of the best hospitals back home.

Everyone is kind and talented. Rooms are clean. Every appointment begins on time. Imaging and blood tests get scheduled immediately, not after months of anxious waiting. Results land instantly in your online portal.

But the real difference-maker: Doctors and patients leverage massive databases of medical information, sorted and analyzed by AI, then channeled into 500 different algorithms to help find cures and better serve patients.

It's a real-time, real-world fusion of man and machine, helping patients get healthier, faster by tapping into superior medical data about you and others with similar conditions.

Consider cases like Autumn's: She has several different things wrong requiring expertise across many domains. Few, if any, humans could piece together this puzzle alone. This is where AI could make a big impact — now.

Mayo recently signed a deal with Microsoft to expand and expedite its own frontier AI model to further scale and refine this work. Both Mayo and Microsoft are fine-tuning the AI with hopes of sharing it broadly.

"None of this works unless health data is arranged maximally for humans and AI agents," Farrugia told me. "What makes it work or scalable is organizing all the knowledge so doctors can use it optimally and so agents can use it optimally, so we get the best data to help patients."

The catch: "I could give this to every hospital — and they can't use it." That's because most hospitals lack the technology, money, systems and focus to use health data drawn from patients and broader pools.

The government needs to force hospitals to change, with "carrots and sticks," Farrugia said. A new data architecture needs to be formed ASAP. Done right, this would be a life-changer for the chronically ill like Autumn, who often get the worst, least coordinated attention.

Here's the kicker: A variety of studies show Mayo is cost-competitive, and works with private insurance as well as Medicare and Medicaid.

Scaling Mayo's system everywhere

image Illustration: Aïda Amer/Axios. Stock: Getty Images

It's a national disgrace that a nation so rich, so innovative, provides such crappy care to our sickest Americans. Most pay new Cadillac prices for used Ford Escort outcomes.

This isn't OK: We could simply admit our failure and rebuild the health care system, using Mayo as the base model.

Mayo is built on three core principles:

Doctors are paid flat salaries. No bonuses for more procedures, scans, tests or visits. They're paid to heal. American health care's original sin is fee-for-service: We pay most doctors for volume, so we get volume. Mayo removed the incentive to overtreat, and the entire culture reorganized around the question that should drive all medicine: What does this patient actually need?

Doctors work as teams, not individual know-it-alls. Mayo practices multidisciplinary medicine. So gastroenterologists, liver specialists and surgeons all work together to evaluate complex cases like Autumn's. The doctors do this in days, under one roof, with constant communication. Compare that to the standard American experience: months of siloed referrals, repeated tests and specialists who, maddeningly, never speak to each other.

Patient-first culture. This was the most striking thing to us: Every person from surgeons to desk clerks seemed caring and kind. They were obsessed with understanding Autumn and smartly diagnosing her before recommending any actions. As someone who started and has run two companies, I can tell you: Cultures are controllable and scalable. We should demand this of every medical institution in America.

And here's proof these principles scale: They aren't unique to Rochester.

Kaiser Permanente pays salaries to its physicians and coordinates care for 12 million members. Cleveland Clinic pays flat salaries with one-year contracts and annual reviews. Geisinger built integrated, team-based care in the hills of rural Pennsylvania, hardly a talent magnet on paper. Intermountain in Salt Lake City shows promise in spreading Mayo-like performance and outcomes elsewhere.

But it's the power of AI that could make the biggest difference and scale this further, the fastest, Farrugia argues.

Mayo's scarcest asset — coordinated diagnostic brilliance based on real data, cases, lab results and patients — is exactly what AI is getting good at. Mayo knows it, running more than 12,000 clinical studies and building its Mayo Clinic Platform to digitize its expertise.

Testing technology: Before we tried Mayo, Autumn's coordinating doctors asked to use AI-assisted recording and transcription. I did the same for the AI agent I built to tackle Autumn's case using Claude. Both of us fed this into systems to make better data-informed calls.

To this day, my personal AI agent has proven smarter than every doctor, other than Mayo's.

Think about what Mayo data changes: A community hospital in Oshkosh, Wisconsin, will never recruit 4,000 Mayo-caliber specialists. It doesn't need to. It could tap into Mayo's expertise and data to instantly help tens of millions of patients.

The harder-to-implement components of scaling Mayo — the talent, the brand, the destination economics — matter far less if the intelligence itself can be tapped by all doctors and patients.

A new health care goal could be Mayo-quality care for everyone. This moves us beyond government-run health care versus free markets.

The bottom line: American health care is the most expensive, bureaucratic, seemingly unfixable program on Earth. But smart policymakers, working with experts at Mayo and beyond, could spare millions of Americans Autumn's health care hell. Shame on us if we don't.

How Mayo's president and CEO sees AI

image Image via Mayo Clinic

I asked Dr. Farrugia to share his own lessons learned on AI and how he's implemented them at Mayo:

Organizations need to massively decentralize the responsibility and opportunities for finding uses for AI. The people closest to the work are best positioned to identify where AI can actually help because they know the workflow, the friction points and what patients or customers need. Governance, however, needs to be centralized to keep the organization out of trouble. That means the gate to full implementation comes much later in the process — but when it comes, it needs to be a very strong gate.

Avoid making definitive statements about future uses of AI and their implications. Those statements will be wrong in months, if not sooner.

Require each member of the C-suite to have built at least one agent on their own and to use more than one LLM at work several times a week. Only then will they be able to make the right decisions about AI for the institution.

America's medical future, imagined

image Illustration: Aïda Amer/Axios. Stock: Getty Images

The only way to pull or push people to a different place or worldview is to show them what better looks like. So I sketched this out, based on Autumn's experience and what Mayo is building with a multibillion-dollar bet on the hospital of 2030.

Imagine avoiding hospitals altogether, which would be the surest sign of a high-functioning medical system. How that could happen:

At a young age, Autumn's genetics, family history and current health are studied and stored in one safe, easy-to-read, easy-to-share place.

She's given recommendations on diet, supplements and general health maintenance, personalized based on her predispositions. Her potential issues are monitored and weighed against large data sets of similar people.

She decides how to act on this data. But everyone has equal visibility.

How it could/should work: Imagine the response once Autumn started getting sick. (Her earliest symptom was pericarditis, inflammation around her heart.)

Instead of going to a random doctor with scant understanding of her condition, she gets routed to someone with true expertise, who taps into a pool of data on other women with pericarditis.

She hops on a Zoom with this person — who could be anywhere worldwide with access to her health data. Her vitals are easily checked at home beforehand and fed into the data.

The doctor and Autumn decide together if an in-person examination is warranted. In many cases, a doctor on the screen and a nurse in her home could suffice.

Now, imagine Autumn is actively sick and needs to go to the hospital.

She never arrives a stranger. The workup happens before she travels. Her local doctor and the specialists see the same screen, with the plan set before she packs her bag. The one care leader is already clearly ID'd.

Her phone carries her medical history and directions to where she's headed. It provides photos of everyone she'll meet, along with their backgrounds.

She lands in what Mayo calls a neighborhood: imaging, labs, procedures, consults and recovery all co-located for complex patients, instead of a scavenger hunt across buildings and appointment desks. Her costs are clear and flagged.

The room itself feels healthy: controls for sound, lighting, temperature and TV. An electronic screen shows today's schedule and exactly what has to happen before she goes home. Family can easily Zoom in to chat and see her. Robots run supplies and meals, so nurses spend their hours on Autumn, not chores.

AI stays in the background: sequencing appointments across specialists, transcribing and summarizing conversations, catching conflicts in a brutal medication regimen. The machines absorb the paperwork. The humans focus on Autumn.

She goes home to connected devices tracking her vitals and her doses, alerts firing to her team when the numbers slide, virtual check-ins reading the trend line. The complication gets caught before she needs to return again.

The bottom line: None of this requires an invention. Every piece of it exists right now, somewhere. What's missing isn't technology. It's the will to rebuild care around the patient — the opposite of our existing mess.

📈 If you're a CEO or on a CEO's team: Ask to join Jim's new weekly Axios C-Suite newsletter.