How AI Could Save a Collapsing Healthcare System w/ Dr. Robert Pearl, Author, "ChatGPT, MD"

American employers now spend over $25,000 a year to cover a single family, and chronic disease is driving the system toward collapse. Yet medicine is still built around a doctor's office visit every three to four months.

Dr. Robert Pearl, former CEO of the Permanente Medical Group, Stanford professor, and author of ChatGPT MD, joins host David E. Williams to make the case that generative AI is the only tool that can shift medicine from episodic to continuous care, and why without it, the chronic disease crisis will break American healthcare entirely.

🎙️⚕️ABOUT DR. ROBERT PEARL
Robert Pearl, MD served as CEO of The Permanente Medical Group (Kaiser Permanente) for 18 years and former president of The Mid-Atlantic Permanente Medical Group, leading 10,000 physicians and 38,000 staff responsible for the care of 5 million Kaiser Permanente members on the west and east coasts. He is a clinical professor of plastic surgery at Stanford University School of Medicine and faculty at the Stanford Graduate School of Business, teaching healthcare strategy, technology, and leadership. Pearl is board certified in plastic and reconstructive surgery, with his medical degree from Yale and residency at Stanford.

He is the author of three books: Mistreated: Why We Think We're Getting Good Healthcare—And Why We're Usually Wrong, a Washington Post bestseller (2017); Uncaring: How the Culture of Medicine Kills Doctors & Patients, a Kirkus star recipient (2021); and ChatGPT, MD: How AI-Empowered Patients & Doctors Can Take Back Control of American Medicine (April 2024). All profits go to Doctors Without Borders.

Pearl is a LinkedIn Top Voice in healthcare and host of the podcasts Fixing Healthcare and Medicine: The Truth, with two monthly newsletters reaching 50,000+ combined subscribers. He has presented at The World Healthcare Congress, the Commonwealth Club, TEDx, HLTH, NCQA Quality Talks, and international conferences in Brazil, Australia, and India. His insights on generative AI have been featured in Associated Press, USA Today, Forbes, Fast Company, WIRED, Modern Healthcare, Medscape, Becker's Hospital Review, the Advisory Board, and the Journal of AHIMA.

🎙️⚕️ABOUT HEALTH BIZ PODCAST
HealthBiz is a CareTalk podcast that delivers in-depth interviews on healthcare business, technology, and policy with entrepreneurs and CEOs. Host David E. Williams — president of the healthcare strategy consulting boutique Health Business Group — is also a board member, investor in private healthcare companies, and author of the Health Business Blog. Known for his strategic insights and sharp humor, David offers a refreshing break from the usual healthcare industry BS.

Need a Strategy Partner?
For over 20 years, Health Business Group has helped healthcare software companies, tech-enabled services businesses, life sciences companies, and payers make smarter strategic decisions. Led by podcast host David Williams, the firm advises clients on sharpening AI positioning, entering new segments, and building commercial strategies for value-based care. See examples of our work here

GET IN TOUCH
Follow CareTalk on LinkedIn
Become a CareTalk sponsor
Guest appearance requests
Visit us on the web
Subscribe to the CareTalk Newsletter

Transcript

David: 0:00

Nearly three quarters of physicians use AI in their daily practice. Most of them are still terrified about it, and today's guest argues that we're at a tipping point. Either we use AI to solve the chronic disease crisis, or the entire system collapses under its own weight. Dr. Robert Pearl is former CEO of the Permanente Medical Group with 10,000 doctors, so he knows exactly how hard it is to change a physician's mind. But as a Stanford professor, author of Chat, CPT MD and host of two podcasts, he knows that physician practice has changed forever. Hi everyone. I'm David Williams, president of Strategy consulting firm, health Business Group, and host of the Health Biz Podcast, where I interview top healthcare leaders about their lives and careers. Dr. Pearl, welcome.

Dr. Pearl: 0:51

David, it's a privilege and a pleasure to be with you today.

David: 0:54

Outstanding. So your latest book, which is called Chat, GPT, md, uh, you wrote, you co-authored it in collaboration with chat GPT itself. Uh, why'd you choose to write the book that way? And what did you learn from that process about working alongside AI as your co author?

Dr. Pearl: 1:11

I wrote the book three years ago, and so much has changed. I would've written a totally different book today, but I knew that at the time, and so I wanted to figure out, number one, how could I write the book as quickly as possible? And number two, I just wanted to get more experience. Remember I wrote this book months after the first version of Chatri PT was released, and I could just see the trajectory. In fact, the book I wrote about, I said. It would double power every year, and it's done more than that in terms of the speed of advance. So to a large extent, I actually. Wrote it with chat BT, the way to write it with a resident, but where I wrote almost all of it, there's six pages that I let the readers see what chat GBT would have done, but it's just really to expose them again at the time now it would be totally unnecessary with 40% of Americans using chat GPT already or one of the other large language models for medical questions.

David: 2:13

Yeah, it's interesting. I, I had seen when you were, you were writing it, you hadn't really waited for it to mature and you said, I'm gonna jump in now, uh, and do it. And now you already anticipated. Of course. My, my next question is about how it'd be different in, uh, in 2026. I read a couple books around that same time, and. Where there's some similar themes, which is to say like, just remember the version of whatever you're using now is the worst AI you're ever gonna use. So don't think in terms of its limitations, but in terms of its possibilities. So I appreciate that

Dr. Pearl: 2:41

today would be just remarkably different. I mean, number one, it's multimodal. So when I rewriting about it, at the time it was only words. And now what we know is that it can analyze physical appearance of patients, it can use video, it can test motion. And remember also that today, uh, it's able to have vibe coding, which I think will be a major advance. Remember, vibe coding is a way for an individual to create a computer application without knowing anything about writing a computer code. And what that to me is going to do is allow doctors to create applications for patients specific to their practice, even specific to the patient.'cause the tools are so efficient and effective today. No, I'm writing completely different stuff now than I did then, but of course. That was three years ago, which is a general AI world is an entire three decades.

David: 3:40

So you've argued that, you know, AI empowered patients and doctors can take back control of American medicine. And I got a little bit of a sense of it right there when you're describing it. But what does that look like, um, in practice? What, what is a, what should a patient's experience be like in an AI enabled healthcare system?

Dr. Pearl: 3:58

So I would, I'd say two things we should be talking about. One is just what's the experience we like for the patients, and number two, how does it take over American medicine for the patient? I think it's going to empower them to know about their diseases. So before they even come to the doctor, they will have had a conversation. You know, I think back to a podcast host that I spoke with actually right after the book came out, who had never used generative ai and her husband had fallen, she slid, he slid down the mountain about a hundred feet skiing, uh, and his, uh, shoulder still hurt him three months later. And she said, Dr. Pearl, you're a skier, you're a doctor. What's going on? I said, I think I know exactly what's going on with him, but why don't you use Chachi pt? See what it says. So five days later, she calls me back, said I did exactly what you said. It said he probably had a rotator cuff tear. I knew what the rotator cuff was. When I saw the doctor, he needed an MRI to establish the diagnosis. I knew why I didn't have to question whether it was overtreatment or not. And the clinician said to me after the surgery, oh, and the application also said, see an orthopedic surgeon because you almost definitely need surgery. And when I did, when I saw the uh, surgeon, he did the procedure. He said, if I had waited a couple more months, I couldn't have reattached the tendon to the bone.'cause it would've been, the muscle would've shrunk. That's empowerment. The issue to me really though, is how do you change and transform the American healthcare system? This system is breaking. We're seeing already the changes happening to Medicaid. We're seeing it to the individuals on the exchanges. We're starting to see Medicare. We're seeing every aspect of this. Being impacted, businesses are spending over $25,000 a year to provide medical care to a family of four. This is just not viable and the US economy keeps adding. Uh, over a hundred thousand jobs in healthcare while everyone else is using these tools to be able to be able to become more productive and more efficient. But the problem has been that clinicians have never had a good way to be able to take back control of American medicine. And what I mean by that is to be the drivers of the care for the patients. They're responsible. Insurers have done it. We've seen, uh, elected officials do it. We've seen hospital administrators. Generative AI is the tool to so solidify that bond between the doctor and the patient to be able to have the doctor in the patient's home rather than managing hypertension, the number one cause of strokes every three to four months in an office. Have an application there, often programmed by a vibe, uh, coding tool to be able to have the patient enter a couple of readings every single day for a month after a month. The doctor could figure out how is the patient doing? You don't have to wait three or four months. You could actually make the changes more rapidly, improve the health and be able to take, not pay for volume. The more you do, the more you get paid. Even if it doesn't do any good, but pay for value and now you have something that the clinician can do that no one else in the healthcare world can do it. The insurer can't do it. The elected official can't do it, and the possible administrator can't.

David: 7:16

At what level does this need to be done? Is this the individual, uh, physician working with a patient? Does it have to be done at a departmental level, whole hospital health system, or are we talking about at a national or global level?

Dr. Pearl: 7:28

It can't be done by an individual physician because taking what's called capitation risk for the total healthcare of a population, a single individual has too much variation, uh, on a day-to-day basis, uh, or a month to month basis for the individual to take that kind of accountability. Moreover, the individual doesn't have the. Ability to really leverage these tools in a way that's gonna be best for the patient. And what do I mean by that? So think about groups of doctors coming together, uh, maybe as few as 50, hopefully a hundred or more. And they can now use tools, they can have economies of scale in terms of some of the administrative tasks that can be able to collaborate and coordinate. When I was the CEO in Kaiser Permanente. What we did is when a patient was seeing a primary care physician, we had available'cause we could leverage our scale. We had, uh, almost 5 million members, in example, is just Northern California. And what would happen is that the primary care doctor could talk to the specialist. And in 40% of the cases could solve the patient's problem that otherwise would require to return visit to a specialist, and in another 30% was able to order the test up so when the patient saw the specialist, something could be done rather than the test to be ordered. And as soon as you do 70 or percent of the care or start the care in the primary care physician's office. Now you can have same day access to the specialty care. Try to imagine that happening to the most of the communities in this country. This waits that are weeks and often months to get into a specialist. And here we are offering same day access, same day information. That's what happens when you bring. Groups of physicians together when you have leadership sitting in place.

David: 9:17

Let's go back to your hypertension example 'cause I've heard you give it before and mention that you know, someone goes and gets their medication and they come back in four months and they check their blood pressure and all you really diagnose they have white coat syndrome. Uh, and you have to kinda do it again and you could, you could, you could reduce that time. I wanna go beyond that and ask you about the role of the family in chronic disease. Does AI play a role? I think about a lot of people with hypertension may not. Take care of it properly, get it treated, and often the spouses involved and trying to encourage them. What's the broader family units? Um, how does that change, uh, with ai?

Dr. Pearl: 9:50

Well, as the expression goes, seeing one family, you've seen one family, you haven't seen them all. So it's a tool that has broad application as an example. Uh, my wife and I, every Sunday evening we say, what do we want to eat this coming week? Uh, maybe we had a weekend. We were at a lot of events and we'd like to have calories cut back. Maybe we have a particular kind of food we want to be able to cook. We provide that information to a generative AI tool, and it gives us menus, it gives us shopping lists. Someone wants in a different language, someone wants to, with a different, uh, price point. All of that is possible. I'm a big runner. I run every day. But if someone's not wants a plan to be able to improve and put their information into the healthcare, into the application, it can accomplish all of that. It can ensure patients are taking their medication. So, as you noted. Tremendous opportunities for everyone in the family. Maybe the family members aren't quite sure. What did the clinician say? What does the medication actually do? What are the risks? Why is this laboratory test being done? What do the results mean? These are all things that can be done with a large language model, and the interesting part is that when you compare Cedar-Sinai Hospital did this, they compared clinicians. Against this technology, the new technology, not the technology. I wrote about three years ago, the technology of today, the technology was 10% better and more accurate. Interestingly enough, when the University of Arizona did a similar study, it found that it was four times more empathetic.

David: 11:30

So, you know, you've also cited surveys that show that something like 70% of patients are comfortable using generative AI with medical issues, but really only if a trusted clinician is guiding them. And without that guidance, you know, the comfort drops a lot. And at the same time, uh, you've got these provocative studies and you know, they have all different things you can critique about them that often a doctor does worse than the ai or they do worse using the AI than the AI alone or themselves alone or whatever. So what does this all mean about the, about the evolving role of the, of the physician? Assuming you see a role for a physician going forward.

Dr. Pearl: 12:05

Absolutely. I, I think you said it very, very well. It's gonna be this partnership. It's gonna be the dedicated clinicians working with empowered patients and generative AI tools, and they're gonna create outcomes that are 10 times better than any of them alone. You pointed out earlier when you talked about the three to four month gap between seeing the physician for the visit and seeing the return, what happens in the middle? We have no idea. Why not fill it in? As though, I'm not saying it's gonna be as good as the clinician go into the home every day, but that of course is an absurdity. It could never happen. How do we have that ability? One of my areas that I look at a lot is heart failure. So after someone has a heart attack or someone has an infection of the heart. The heart is weakened and medication is given, and the medication allows it to be functional, but the amount of decompensation that can be tolerated is actually very small. So when the individual forgets to take their medication or so the individual, let's say, has uh, uh, the flu, what you see is the patient now has an acute. Heart failure has to go to the ER and be admitted to the ICU and often dies. And as they live, they generate massive amounts of bills. What we can do basically is, and what doctors know I should say, is that it happens not immediately, like in an hour. It happens over about three days. And what do you start to see if you're the doctor? You see they gained some weight because they, um. Retain water. What you see is their ankles swell because gravity pulls the fluid down to the ankle level. They can't lie flat because the lungs fill up with fluid. They can't walk as many steps because they don't have as much pulmonary lung capability. You could assess all of that with a current tool. That's why I said it's so different than before the multimodal nature of it. You could use Bluetooth to be able to connect to a bathroom scale to tell you if the person's gained any new weight or not. You could use the same Bluetooth to be able to assess. Whether they have a reduced oxygen in their blood, you could use video and and digital to be able to look at their ankle swelling to see whether they can lie flat, to see whether they can climb the steps. And at the first sign of deterioration, the physician could be notified, change the medication, intervene in some fashion, and avoid that ER visit. The ICU and the light and the possibility of dying. That is what we're talking about. That's a very different world. That's a partnership. Generative AI can't do it, but the doctor can't do it. And we could talk about psychiatry and mental health. You see a clinician, they're worried about your depression. They say, come back in two weeks. What happens in there? Maybe something terrible happens. How do we intervene? How do we know it? This is the combination we have to move medicine from being episodic to being continuous.

David: 15:14

So you spoke earlier about how, you know costs are rising in, in healthcare and you see so many people, you know, being added, all these jobs being added on a monthly basis. Now, the converse of that, people are worried about AI taking jobs elsewhere. I saw an article, I think maybe it was in the Boston Globe, about the head saying. Hey, like the haven from, uh, from job losses in healthcare and, you know, pursuing a career as a nurse, but of course that's adding a ton of a ton of cost. I wanna ask you about, um, one topic that you have, you have talked about, which is, uh, Utah, uh, where they're piloting a program that allows the AI to renew prescriptions, not to write new prescriptions, but to renew existing prescriptions for chronic diseases without physician involvement. Is that a good idea to be doing at this time?

Dr. Pearl: 15:56

Well, I have to see the application and tested, so, so the idea is a good idea whether the current tool is adequate. I, that I can't tell you 'cause I've not actually, uh, had the opportunity to use the tool, but of course it is. You know, so much of medicine today is algorithmic, particularly when it comes to chronic disease. Well, you know, we know what the blood pressure is. We've had national societies tell us the series of medications you use where you should start the next medication to add what you should do if it's not responding. Uh, we know actually a lot of lifestyle medicine changes that can happen. The same with diabetes. You know, the CDC has looked at this question and it's come to the conclusion that if we could manage. Chronic disease as well as the best organizations do it today. And how does the best organizations do it today? Exactly what you're saying. With a lot of people. That's how we did it. In Kaiser Permanente, we became number one in the nation of a thousand programs because we had a lot of people we could invest in doing the work. Generative AI can do it to replace a lot of them. And why is replacing them? It's not eliminating the jobs. We're seeing is we have a massive population that's not getting that care. This would be a 10 or a hundred or a thousand fold expansion, and that's simply not gonna be financially viable. Particularly also the fact that, you know, rural areas that couldn't be done, or any inner cities, there's not gonna be enough resources. So I see this as being the solution. To where the American healthcare crisis needs to go, and it's why I am such a proponent of the clinician leading the way because. The insurer can put in prior authorization and restrict care, but they can't drive that collaboration of the cooperation The hospital administrator wants the beds filled, whereas as a nation, as a patient, I wanna be healthy. I wanna avoid that heart attack at the first place, and the elected officials. They can pass laws and create financial incentives, but I haven't seen that result in improvements in actual patient outcomes. I see the clinicians being the only person who can do it. I think the financial incentives have to align with it, and the degenerative AI tool is gonna be the key. Differentiator. You're absolutely right. Some of the jobs of today will be done by the technology, but that will create new jobs. We're gonna have to train people to higher levels of skill. What we're gonna start to see is primary care is gonna start to do a lot of the jobs that specialty care does today that. For two reasons. It's gonna have the expertise, it also will have the time. Imagine if 30% of what a primary care physician does today can be done technologically. We're not gonna have fewer primary care doctors. We have a massive shortage. They'll have the time to spend with the people for whom they can do the best. Good. We still have a system left over from the last century at a time when the diseases people had who were acute, not chronic. A time when the treatments we had were simpler and clinicians had more opportunity to invest in making sure that patients were able to get the right care. Now their doctors are on a treadmill running as fast as they can. Adding more and more things on top of them. I saw a statistic that said if every primary care physician did everything that every patient needed, it would take 25 hours.

David: 19:24

Yeah.

Dr. Pearl: 19:24

A day.

David: 19:25

That's a long day. It's only not possible. It's only gonna work on, I guess, on, uh, you know, daylight savings time, uh, when we switch over, so,

Dr. Pearl: 19:32

exactly. For

David: 19:33

one day. Okay. So, so, so, you know, so on your podcast and I mentioned your, your podcast host, uh, on your podcast fixing Healthcare. There's three urgent, uh, threats that you talk about and how generative AI could help. We've talked about the first two directly, which is about affordability, cliff and chronic disease crisis, but there's a third one too. And you're, I'm starting maybe to hint at it, but I wanna get at it directly. And that's about the risk of training doctors for the wrong future. So can you walk us through that and, you know, how's it going with training medical school? You know, what's happening there?

Dr. Pearl: 20:05

The problem with academia is it tends to move at a very slow pace, and that's what we're seeing today. We're still training clinicians, medical students, and residents for the approaches of the past. We're still teaching them about a doctor's office centric approach. We're not teaching them about generative ai. Now, the first thing I wanna point out is that generative AI is the first. Tool, first application that I am aware of in medicine where the students are better at it than the teachers. You know, you'll learn how to listen to the heart. The professor would told you how to use the stethoscope. You wanted to be able do the physical examination. The professor would have the expertise and would hand it off. To this next generation of clinicians, we don't have to teach this next generation how to use generative ai. I teach both of the Stanford, uh, graduate School of Business and the Stanford Medical School and at the Stanford Medical School. When I ask the medical students and residents, how many of you know how to use generative ai? All the hands go up. How many of you use it every day? Still, all the hands go up. This is not the issue. The opportunity now and again, this is a partnership, this is where culture comes in. The academic institutions have to understand their limitations today. What we wanna be doing is not teaching, uh, medical students and then residents how to do solve the problem in the doctor's office or in the hospital. We wanna be teaching them how do we solve it with the patient in the home. Chronic disease is a great example. We spend hours, days, months, saying. Teaching algorithms, the blood pressure's elevated. This is the drug to give. Here are the complications of the drug to give, and so on down the line. No, we have to say, how do we improve the blood pressure? By creating a generative AI tool that the patient can use every day. And how do we change our workflow so we don't see people on a three to four month basis? We see them when they need care. For some people it's one month when they're not getting better. For other people, it's once a year. If they're a Medicare 'cause we have to do that to meet the net, the regulatory requirements. Some other people, it might be. It happen at all if the, if hypertension is their only problem, this is a complete revision both technologically and operationally, and. I don't believe that the faculty can create the curriculum. It's gonna have to be created by a partnership. How far along are we? We've taken maybe 10% of the way to where I believe we need to go. And as part of that, we also have to teach people how to be very successful, not just in a pay for volume, healthcare financial model. Inside one that is a pay for value. Letting them understand how, by keeping people healthy, avoiding chronic disease in the first place, and more importantly, controlling it when it comes.'cause often we can't avoid it, people don't do what they need to do to accomplish that. Now we're starting to have a system that can make American healthcare achieve what it should be achieving. You know, we spend $15,000 per American per year on healthcare. On average, what you see is that Switzerland, it spends about 10,000 Germany, 9,000 everyone, less than half of us, and yet our results lag. These other countries, we have the resources. We have to just move from a, uh, from a insufficient resource view to understanding that what we're really seeing is an inefficient system. We have enough resources if we use them better and differently. So far, that process has not evolved very far.

David: 23:57

Well, we may only be about 10% of the way through the educational transformation, but we're a hundred percent done with the podcast, uh, at the moment. So I want to, uh, say that's it for another episode of the Health Biz Podcast. I'm David Williams, president of Health Business Group. My guest today has been Dr. Robert Pearl. He's former CEO of the Permanente Medical Group. He's a Stanford professor, author of chat, GPT MD podcast host and more. If you like what you heard, please subscribe on your favorite podcast platform and thank you so much. Dr. Pearl,

Dr. Pearl: 24:27

my pleasure. And if anyone wants more information, then go to my website, robert pearl md.com, where there's a lot of information on all these topics and have 'em please let me know where they disagree. I always wanna learn more.