The Peter Attia Drive - #91 – Eric Topol, M.D.： Can AI empower physicians and revolutionize patient care？

00:00:00.000 Hey, everyone. Welcome to the drive podcast. I'm your host, Peter Atiyah. This podcast,

00:00:15.480 my website, and my weekly newsletter all focus on the goal of translating the science of longevity

00:00:19.800 into something accessible for everyone. Our goal is to provide the best content in health and

00:00:24.760 wellness, full stop. And we've assembled a great team of analysts to make this happen.

00:00:28.880 If you enjoy this podcast, we've created a membership program that brings you far more

00:00:33.280 in-depth content. If you want to take your knowledge of the space to the next level at

00:00:37.320 the end of this episode, I'll explain what those benefits are. Or if you want to learn more now,

00:00:41.720 head over to peteratiyahmd.com forward slash subscribe. Now, without further delay,

00:00:47.740 here's today's episode. I guess this week is Dr. Eric Topol. Eric is a very famous cardiologist,

00:00:55.080 geneticist, and digital medicine researcher slash pioneer. He's the founder and director of the

00:01:01.120 Scripps Research Translational Institute, TSRI. And prior to coming to Scripps, he served as the

00:01:06.640 chairman of cardiovascular medicine at the Cleveland Clinic, a post he held for about 15 years.

00:01:12.120 We actually start the interview by talking about the story that led to him leaving the Cleveland

00:01:17.880 Clinic. And actually, we spent quite a bit of time on this story, which is something that I certainly

00:01:23.020 remember following during its unfolding in the early part of the 2000s. He's also the editor-in-chief of

00:01:29.940 Medscape. And in 2012, he published a book called The Creative Destruction of Medicine. What we talk

00:01:35.400 about today, though, is his, I believe his third book, but it could be his fourth, called Deep Medicine.

00:01:40.900 And this is something I've been wanting to talk with Eric about for some time, because it really goes

00:01:44.840 into, in a non-sci-fi way, the application of artificial intelligence, deep learning, machine learning,

00:01:52.780 in medicine, which is, as you probably realize, a field that is, at times, upsettingly slow to adopt

00:01:59.440 to technical change. We talk about a lot of things in this episode. And in a surprisingly brief period

00:02:04.680 of time for one of my podcasts, actually, we talk a lot about the gut biome. And I actually have a

00:02:09.160 great and spirited discussion about it, because as some of you may know, I'm kind of a skeptic of this

00:02:13.780 whole gut biome is going to be the answer to all of our woes. But I think in the end, Eric and I

00:02:18.560 really kind of end up being much more closely aligned in our views of the utility of this tool

00:02:23.920 as a way to provide predictive insights. In fact, there are a number of things that I came out of

00:02:27.880 this episode with some follow-up notes for myself, as far as people I want to connect with, researchers

00:02:33.580 that I want to connect with, to better understand how I can utilize that information for some of my

00:02:38.940 own clinical interests. I think what comes across in the end of this discussion is that doctors aren't

00:02:43.760 going anywhere. And in fact, Eric has a slightly contrarian view of what the impact of AI in

00:02:50.300 medicine will be. He argues that it's not at all the doctors are going to go away. It's just the

00:02:54.560 doctors are going to change their focus and frankly focus on the one thing that doctors and humans in

00:02:59.280 general can do far better than machines. So with that said, I hope you enjoy my conversation with

00:03:04.860 Dr. Eric Topol. Well, Eric, thanks so much for coming over on a Friday afternoon.

00:03:14.540 Great to be with you, Peter.

00:03:15.640 This is kind of funny because we've both lived in San Diego for over 10 years and your name comes

00:03:21.160 up all the time. Everybody says to me, you must know Eric Topol. And I say, well, of course I know

00:03:25.200 of him, but no, I don't know him. And where it really comes up is every time I'm at Dexcom.

00:03:29.740 And obviously you're on the board there and I know Kevin Sayre very well. And I met several folks on

00:03:35.060 the team. So it's hard to believe this is the first time we're meeting.

00:03:38.740 It is. I've heard a lot about you over the years, Peter. Now you mentioned Dexcom. I've been on their

00:03:44.100 board for nine years and I've watched this early medical wireless world of sensors be transformed.

00:03:52.380 So that's been really a privilege. Whereas most people think about steps as a wearable sensor and not

00:03:58.700 glucose. So, you know, it's been a great company that started nine years ago, at least when I

00:04:04.280 started with them, they were really having a rough time to get people with type one diabetes to use

00:04:10.500 continuous glucose. And that's really changed a lot.

00:04:13.440 Yeah. Kevin and I met about four years ago on an airplane. I still refer to it as certainly one of

00:04:18.940 the top two luckiest seating assignments I ever had to be sitting next to Kevin and immediately clicked

00:04:24.240 and I've never taken the sensor off since. You can see I'm wearing my G6 right now.

00:04:28.700 Wow. And I agree. I mean, it's sort of comical when people think about the number of steps one's

00:04:34.340 taking as a quote unquote wearable or a valuable insight when you think about what could be measured

00:04:39.760 in the interstitial fluid. And glucose, of course, is just the thin end of the wedge on that.

00:04:44.160 Exactly. I think a lot of people haven't realized how where this is headed. The more concern,

00:04:49.600 of course, is using in the right people. That's like, for example, the Apple watch for heart rhythm,

00:04:55.160 where so many people are using it and it's a recipe for false positives. But if you use these

00:05:01.620 sort of things, these more advanced sensors properly, they can be really a big difference.

00:05:06.040 You're one of the earliest adopters of mobile telemetry and mobile devices or outside of

00:05:12.500 hospital devices, ambulatory devices to be able to measure heart rhythm. I wasn't even planning to ask

00:05:17.560 you about that, but I can't resist at this moment. Give people a little bit of background about you.

00:05:21.320 Obviously, you're a cardiologist. We're going to talk about what you've done at the Cleveland

00:05:24.680 Clinic and what you've done here at Scripps. But what interested you in cardiology in the first

00:05:28.560 place? Well, it wasn't really cardiology that got me into medicine. It was the interest in

00:05:33.380 endocrinology. My father had been a type one diabetic and had every complication you can imagine.

00:05:40.640 Was blind by age 49. So I decided that since that seemed to be such a primitive area as far as

00:05:48.500 no prevention and lack of treatments outside of insulin, that maybe that would be the way to go.

00:05:54.000 And so when I went to UC San Francisco for my residency, it was really to get geared up to be

00:06:00.120 a diabetologist. And what happened there was I was completely transfixed by what was going on

00:06:06.560 in cardiology, particularly kind of Chatterjee, who's been a medical hero of mine, had a big influence

00:06:12.920 in really changing the path. It was also a very remarkable time in that it was the first balloon

00:06:19.480 angioplasty to the coronary arteries, the first clot dissolving therapy for heart attack, and so many

00:06:24.940 other things that it was captivating. So I've never regretted that change, but it wasn't what I had

00:06:30.380 initially in mind.

00:06:31.360 The field of cardiology today is so specialized. I mean, to say that one would do a residency in

00:06:38.360 medicine followed by a fellowship in cardiology would be as broad as doing a residency in general

00:06:43.620 surgery today, where the field has maybe stated in another way, an interventional cardiologist

00:06:50.880 versus a lipidologist would have virtually nothing in common outside of their foundational

00:06:56.480 training in cardiology. I mean, I don't know that many people actually appreciate that outside of

00:07:00.740 medicine. No, you're really right, Peter. There's these subspecialties. It could be a heart failure

00:07:07.640 or prevention or the plumbers, interventional or the electricians, electrophysiology, and on and on.

00:07:13.280 So it's a very broad discipline. The field has matured so much in the last decade or two.

00:07:19.520 There are obvious benefits to that. What do you think are the limitations of that stratification?

00:07:24.960 Well, the major limitation is that you get this ice pick view of the patient. You know,

00:07:29.980 when you see a patient as an interventional cardiologist, you're thinking about what arteries

00:07:33.960 can I fix? And the electricians are thinking about the person as having an arrhythmia. So the general

00:07:41.200 cardiologist, which is the group of people that would be the advocates to prevent unnecessary

00:07:47.000 procedures, they don't get enough respect. It's just like, you know, primary care internal medicine

00:07:52.040 doctors. And so we really want to boost them up because they are the ones that are really caring for

00:07:58.520 patients and looking out for their overall cardiovascular health.

00:08:03.280 How long ago did you sort of get the sense that we didn't have to be in the hospital with a 12 lead

00:08:10.940 EKG on a patient to appreciate what was happening in the conduction system of their heart? And in fact,

00:08:16.440 I feel like I even remember seeing you on TV 10 years ago on CNN or Good Morning America for all I

00:08:23.200 know, I honestly can't even remember, but you were in a clinic and you were sort of saying, look,

00:08:28.020 there's going to be a day when a patient at home is going to wear a device and it's going to send me

00:08:34.320 a warning sign that something is going on. Right. Well, it's interesting you bring that up because

00:08:39.360 it was kind of a serendipitous connection to San Diego. So it was, I think 1999, I had gotten to know

00:08:47.460 the folks at Kleiner Perkins pretty well and this guy, Brooke Byers contacted me and he said, you

00:08:53.760 know, we're looking at this company called CardioNet. We don't really understand this thing about being

00:08:59.360 able to do electrocardiogram monitoring over the internet. Could you look at this thing? I'm going

00:09:05.980 to send you the slide deck. So it was, you know, 1999, 20 years ago, I looked at this thing. I said,

00:09:12.920 whoa, this is an eye-opener. And it was a San Diego-based company. And you were still in Cleveland

00:09:17.600 at the time. Yes. Yes. And the whole idea was things were starting to really converge of the

00:09:23.180 idea, at least, of medicine and the internet. But the sense that you could monitor people remotely,

00:09:30.600 continuously for multiple leads of their cardiogram was really exciting because up until that time,

00:09:36.740 the only way we could do that was to put on one of these bulky Holter monitors. This is Norman

00:09:41.920 Holter from the 1950 version. We still talk about his name. They're still using it. And, you know,

00:09:48.480 that is, you can't exercise really. You can't take a shower. You know, it's, you can't wear this too

00:09:54.240 long and it isn't real time. You then send it in and, you know, you get, or you go back to the clinic

00:10:00.700 and take it off. And so it's so antiquated when you think about it. So the idea that we could

00:10:06.400 transcend that era with this mobile continuous monitoring. And by the way, there are people

00:10:11.520 who have been on a Holter who died suddenly. And the Holter, of course, serves no purpose other

00:10:16.200 than to tell you what arrhythmia killed them. Exactly. So now it's a whole different world.

00:10:20.860 And then you start to say, wait a minute, why don't we just do everything? Not just a cardiogram.

00:10:26.340 We could do all the vital signs. We, we get rid of hospitals or at least those hospital patients

00:10:32.400 who are not in an intensive care unit. And I think that's where we're headed. That is 20 years ago

00:10:39.020 was kind of the entry point with this first dedicated wireless company, CardioNet. And it's

00:10:45.400 just where we're going to build on to ultimately eradicate the need for most hospital rooms, which

00:10:51.600 is a pretty big deal. And it's probably the most transformative aspect of where we're headed

00:10:56.920 because that's the number one item for healthcare costs. It's not just the facilities, but the

00:11:03.340 personnel. And so they account for a third of our $3.6 trillion annual healthcare budget.

00:11:10.780 Medicine seems to always take longer to adopt things. I mean, I think in general,

00:11:14.780 as people, we tend to have optimism that exceeds the pace at which technology moves forward. I mean,

00:11:20.920 in some ways, engineering examples get overused. We talk about, uh, the Manhattan project,

00:11:27.720 which is kind of remarkable when you really stop to think about it, that in such a short period of

00:11:31.480 time, they could go from a proof of concept in the early forties to a finished product, you know,

00:11:37.480 in 44, even the space race is kind of remarkable in terms of the accuracy with which they were able

00:11:43.420 to sort of project and map out the steps. It doesn't seem that healthcare follows that curve.

00:11:48.820 It doesn't just follow a straight Moore's law.

00:11:51.660 Well, it's actually defines Moore's law. If you plot that out, you look at, wow,

00:11:58.100 the cost of chips has gotten so incredibly low over the course of now 50 plus years and healthcare

00:12:05.600 costs are going the opposite direction. So their lack of embracement of the digital era and also the

00:12:13.480 lack of it having the impact of lowering costs is notable. It's palpable. It's a general resistance.

00:12:21.420 You know, I liken it to a sclerotic or ossified nature of the medical community, very resistant

00:12:26.960 to change. The only time you see lack of resistance when it is tied to markedly improved reimbursement,

00:12:34.560 for example, you know, the adoption of robots like Da Vinci or, you know, something like that.

00:12:39.720 Otherwise there's just no real incentive to change. And of course we want to be careful

00:12:45.720 because we don't want to adopt a significant change when it isn't validated or proven. But

00:12:50.300 when we, when we see things that are unquestionably advances and still unwillingness to move in that

00:12:57.120 direction, that's disconcerting. Yeah, it is. And there are a few problems I've contemplated

00:13:02.000 where no matter how much time and energy I put into it, I really can't even see the direction of the

00:13:08.060 solution. I think there are some problems, even the political system, when you think about how

00:13:11.480 broken our political system is, I don't think you, you don't have to be a student of political

00:13:15.200 science to appreciate that. But I think most people who spend a lot of time thinking about it,

00:13:19.540 if given a magic wand would know how to move in the right direction, right? If you stopped

00:13:24.240 gerrymandering, if you maybe discarded the electoral college, like there were like five things that you

00:13:29.460 could do structurally that would bring politics back into a sort of more civilized era.

00:13:34.820 But if you said to me, wave a magic wand, what, how do you fix medicine? The only idea I've ever

00:13:42.180 had is one that involves changing behaviors directly, which of course becomes a bit of a

00:13:47.560 tautology. But it seems to be that the disconnect between the driver of demand and the one who pays

00:13:53.080 the bill is the biggest problem. Does that kind of resonate with you? In other words, in this system,

00:13:57.740 which is not a budget driven healthcare system, like it is in the UK, it's a demand driven system.

00:14:02.620 So the system will rise to the cost of the demand. The demand is mostly driven by the patient and

00:14:08.920 the physician, the patient requiring the care, the physician ordering the treatment, but they bear

00:14:14.400 less than, I mean, they bear maybe, I think the last thing I saw said about 11% of the total cost

00:14:19.240 is born by those driving demand, which is sort of like you walking into a car dealership and knowing

00:14:25.480 you only have to pay 11% of the car price. It's going to completely uncouple any reality.

00:14:31.880 To me, that seems like the elephant in the room. And that's why I think the problem is,

00:14:38.080 as you stated, which is why are hospitals so expensive? Well, if people actually saw the bill

00:14:44.960 of, you know, a hospital stay and realized what the, you know, the gauze and the pillowcase were being

00:14:50.200 charged for, I mean, they'd, they'd scream, but of course they don't have to pay it directly.

00:14:54.760 They're only paying it indirectly.

00:14:56.960 Right. It's a really messed up system in so many respects. You touched on one big one, but

00:15:04.040 the basis as this absurd healthcare charges, it's just unfathomable. And all the things that have

00:15:13.800 been done to date like Obamacare and the debate now about Medicare for all or whatever doesn't

00:15:20.640 get to the root of the problem, which is the cost. That's right. It's politically in vogue to deal

00:15:25.320 with access, which is an important problem. Absolutely. But if you expand access without

00:15:31.500 reducing costs, you trade one bad problem for another bad problem. That was so educational for

00:15:36.160 me because over the past year and a half, I worked with the NHS to review their health system in

00:15:44.780 particular, the impact of technology, AI, digital medicine, genomics. And as you already mentioned,

00:15:52.400 Peter, they have a system which can be changed like a light switch. They don't have a single payer

00:15:58.580 and they have far better outcomes than we do at about a third of the cost per person.

00:16:07.240 And what's interesting is they have the will to make these changes. They're adopting things

00:16:12.340 at a rate that is no comparison to the US. Like for example, they already have places in the UK where

00:16:18.800 they've gotten rid of keyboards instead of doctors typing and being data clerks. But they're also

00:16:24.980 interested in making things more efficient than they already are more efficient than we are.

00:16:30.560 But the difference is the incentives, just as you outlined. This is not employer-based healthcare.

00:16:36.280 This is not co-pay related. This is healthcare for everyone. And we're going to make it as good as we

00:16:42.800 can and as least expensive as we can. So that country is in many respects, a different model.

00:16:50.040 Canada is like it. And many countries in Europe are similar. And we're so remotely disparate. It's

00:16:56.820 just unfortunate. You know, I grew up in Canada. So I have sort of mixed feelings about the discussion

00:17:02.400 of a single payer system because I've seen the advantages of it and you've outlined them. I don't

00:17:07.540 know as much about the NHS, which may be better than in Canada. It's not national. It's done by each

00:17:13.900 province. So, but it's still universal coverage within each province. But that said, my whole family's

00:17:18.680 still in Canada. And I will say the following, when they get care, it seems to be pretty good.

00:17:26.620 But boy, is it hard for them to get care sometimes. It depends what you need, right?

00:17:30.560 Right.

00:17:30.760 If you need a coronary artery bypass and you have critical stenosis, you'll get excellent care in

00:17:36.900 Toronto. You know, it'll be no worse than it would be in the best hospital in New York or San

00:17:41.640 Francisco. But if you have a torn ACL, it might take you six months to get that MRI. Now we could

00:17:50.100 debate whether or not in the long run that matters as much, but you know, I've always found it

00:17:56.180 interesting that at least a country like Canada, and again, I don't, you can speak to the UK for me.

00:18:01.420 There's great resistance to have a second layer of private insurance on top of the public

00:18:05.820 that would allow that, which to me seems like the best hybrid solution, which is you have to have

00:18:12.760 a safety net that provides universal coverage for everyone. But if an individual decides, look,

00:18:18.560 I'm willing to pay an extra $10,000 a year, which by the way, is still a fraction of what I pay to

00:18:22.840 insure my family. You now have a separate queue that you can go into, you know, so it's, I mean,

00:18:28.140 it's like Disneyland, right? It's like in Disneyland, you can do the special pass where you don't have

00:18:31.740 to wait in line, you pay an extra, whatever it is. Why do you think that places like Canada for

00:18:36.820 sure, but maybe even the UK have resistance to secondary insurance? Well, they have secondary

00:18:43.080 insurance. People have that. In many respects, it works like you just described. I see. So the NHS

00:18:49.480 has a second tier that one can buy privately. Yes. The main difference is there's a philosophy that

00:18:55.460 if you're a citizen, as you said, healthcare is a right. It's a right, not a privilege. Yeah.

00:19:00.620 Yeah. And then there are people that have this added insurance. It does separate a small fraction

00:19:07.660 of people into this other class of getting access to more rapidly, to perhaps a different queue as

00:19:15.720 you outlined it. But for the most part, it's a small proportion of the people in the UK. And I think

00:19:21.740 it's somewhat similar in Canada. But I think the difference really is that there's two big

00:19:28.440 poles of problems. If you look at it in the US, there's the indigent who either don't have access

00:19:34.920 or if they do, they're just not getting the kind of care that you would like to see. And then there's

00:19:40.300 the affluent who get too much. They get overcooked. They get executive physicals and they get all this

00:19:46.840 stuff that shouldn't be done. And the outgrowth of that is bad outcomes. They get incidental illness.

00:19:52.440 You don't see that in the UK and in a lot of other countries. So we have this problem at both ends. And

00:19:59.240 most of the recognition has been on the end of the underrepresented and indigent, not on the people who

00:20:06.420 are getting overcooked. And that's a problem. I interviewed Marty Macri a little while ago, and he's

00:20:12.940 written very eloquently about this problem, specifically with respect to pharmacotherapy.

00:20:17.780 You know, you mentioned executive physical. Well, your alma mater, of course, is one of the places

00:20:22.340 that, you know, is certainly has to be regarded as one of the finest hospital centers in the United

00:20:27.140 States. And then by extension, the world. And I feel like I've had half a dozen patients come back

00:20:32.500 from their executive physicals there, or more so ask me if they should go and get it.

00:20:37.260 So when did you, when did you get to Cleveland? Early nineties?

00:20:40.180 Yeah, I got there in 91.

00:20:41.520 And what was the Cleveland Clinic in 91?

00:20:43.760 It wasn't as well known as it is now. It was particularly well known for bypass surgery.

00:20:52.140 It was the place where Rene Favalaro essentially invented bypass surgery. And also Floyd Loop had

00:20:59.820 really brought in the internal mammary artery, which was a big advance. It also had some other

00:21:05.060 traditions. I mean, Mason Sones had been there, discovered coronary angiography.

00:21:09.640 Was it considered at that time ahead of Minnesota, which was also arguably one of the earliest

00:21:15.300 pioneers? You know, Stanford and Minnesota were really these huge pioneers in earlier cardiovascular

00:21:19.600 medicine.

00:21:20.540 Well, I think in cardiovascular, that was its signature contribution. I mean, obviously there

00:21:25.080 were others, but it was...

00:21:26.800 Lillehi, of course, was, you know, Shumway and Lillehi were kind of these gods.

00:21:30.800 Exactly. I think the main, you know, for coronary disease, because there had been so many,

00:21:35.980 a cluster of remarkable innovations. That's where it had its biggest footprint. And when I went there,

00:21:43.400 there was by far more bypass surgery done at Cleveland Clinic than anywhere in the world.

00:21:48.620 And you had gone, after UCSF training, you went to University of Michigan?

00:21:53.100 Well, there was one stop in between. That was at Johns Hopkins where I did cardiology.

00:21:57.300 Oh, I don't think I knew that.

00:21:58.340 Yeah.

00:21:58.620 So we have this one overlap in our backgrounds.

00:22:00.940 Yeah. And then I went to, I was seven years at University of Michigan. My first job, my second

00:22:05.840 job at Cleveland Clinic, it was almost 14 years. But when I went there in 91, it wasn't a really

00:22:12.500 strong academic center. It was, in fact, I'll never forget my chief of medicine at Michigan. When I

00:22:20.220 told him I'm going to Cleveland Clinic, he said, well, that's the end of your academic career.

00:22:24.320 It was viewed as almost going into private practice?

00:22:26.560 It was viewed as, well, high volume, factory medicine, you know, high quality, but, you know.

00:22:32.920 Did you have a strong residency fellowship system underneath you? Were you going to be

00:22:36.180 heavily involved in training?

00:22:37.580 There was a medicine fellowship system, but I don't know that I would qualify it as strong

00:22:42.400 academically. You know, large, lots of them, but they weren't doing cutting edge research and

00:22:48.400 it wasn't that kind of scholarly environment. So my mission when I went there to refute the

00:22:55.520 Yamada's view that it was the end of career was actually do just the opposite and enliven

00:23:03.080 it and wake up the curiosity and the innovations. So that's what we did. And, you know, it was

00:23:08.680 a big transformation because it involved a whole new team, you know, bringing, it was like an

00:23:13.760 exchange transfusion because they hadn't written a paper in the cardiology division in a couple

00:23:18.960 of years. Really?

00:23:20.640 Yeah. It was, it was pretty, it was very much dominated by cardiac surgery and it was just

00:23:25.120 limited high productivity in academic side.

00:23:29.240 So it really comes down to incentives again. I mean, today we see the opposite problem,

00:23:32.760 of course, where we have the proliferation of total nonsense journals and absolute horrible

00:23:38.000 things that don't pass for science being written constantly because of course the pendulum on the

00:23:43.720 incentive is you have to publish. Right. And so presumably at Cleveland, that was simply not

00:23:48.540 that the, the pendulum was the exact opposite where you're, you were probably compensated based

00:23:52.300 on clinical productivity and nothing more. Yeah. I mean, I think the cardiologist, when I talked

00:23:57.320 to them, when I was interviewing and then when I got there, beached in, they said, you know,

00:24:02.060 we're the handmaidens of the surgeons. And they were so busy taking care of the patients because

00:24:06.580 the surgeons didn't really see the patients outside of the operating room. And they needed

00:24:10.540 obviously at this high volume of patients need a lot of care. And there weren't that many

00:24:14.800 cardiologists. So the cardiologist would run the critical care and the step-down units and

00:24:18.500 everything. Everything. Wow. So they really were giving great care, but they were consumed by that.

00:24:24.520 So they didn't really have the time and nor did a lot of them since it was highly inbred then

00:24:28.840 really have the knack of asking questions and chasing them down and whatnot. So we brought in a whole

00:24:35.020 group. I mean, I started, there were 30 cardiologists. When I left, there were over 90.

00:24:39.560 Were you brought in as the chairman of cardiology?

00:24:42.320 Right. Right. No, I was age 35. Actually, it was really funny, Peter, when you think about it,

00:24:47.820 Bill Belichick and I started the same day. Bill Belichick was the youngest football coach in the

00:24:53.440 NFL history head coach. And I was the youngest chairman in the history of Cleveland Clinic. So

00:24:58.020 we got to know each other a little bit. It was a very different era for Bill Belichick.

00:25:02.400 Who's my favorite coach, by the way.

00:25:03.800 Is that right?

00:25:04.220 Oh, I mean, I'm obsessed with Bill Belichick. I met Tony Gonzalez recently and he told the

00:25:12.780 absolute funniest story about his experience with Bill Belichick at the Pro Bowl one year,

00:25:18.260 which I won't restate now, but in the show notes, we'll link to a video of Tony telling

00:25:22.920 that story along with an article that was written up about it at some point. But I'm fundamentally

00:25:29.260 just obsessed with Belichick. Well, he's a really interesting guy.

00:25:32.560 It's a bucket list for me to meet him at some point.

00:25:35.260 Wow. No, he's a kind of fascinating figure for many reasons. Actually, one of the most memorable

00:25:40.660 things that happened regarding Bill Belichick was he benched Bernie Kosar. That didn't go over well.

00:25:48.040 And Art Modell, who was the chairman of the board of Cleveland Clinic, we were good friends and he was

00:25:52.920 over for dinner at our house. This had all happened and it was an uproar, you know, with the dog pound

00:25:58.040 and everything. And so Art and Pat were over and they said, you know, well, we had to put a sign

00:26:02.380 in front of our house. Bill Belichick doesn't live here. But, you know, there was as much fear of what

00:26:10.140 was happening then as when, you know, Art Modell moved the Browns to Baltimore. Yeah. Yeah. So

00:26:15.060 during those almost 14 years, it was great to see, you know, this kind of renaissance of-

00:26:21.020 And you were supported.

00:26:22.060 Yeah.

00:26:22.380 Because, I mean, presumably you would not have left Michigan without an explicit understanding

00:26:27.640 that you were not coming to implement the status quo. You were coming to rattle it.

00:26:32.180 Exactly. And, you know, it was really because of Floyd, as we knew him, Fred Loop. He was a very

00:26:38.060 progressive CEO of Cleveland Clinic. He, even though he'd been a cardiac surgeon throughout his career,

00:26:43.580 and in fact, in the early years was still operating, he wanted to see cardiology thrive.

00:26:48.540 He's not alive anymore, is he?

00:26:49.900 No, no. Unfortunately, he died of a rare cancer a few years back, but at a young age,

00:26:55.560 because it was a surprise. He had such longevity in his family. He said, Eric, you know, I want you

00:27:01.680 to come in and, you know, just completely get this place, supercharged, make cardiology the greatest

00:27:07.460 anywhere, and I'll back you 100%. And not only that, but in 2000, in year 2000, when I was thinking

00:27:14.300 about leaving, actually to go to Stanford, he said, why do you want to go to Stanford in medical school?

00:27:20.140 Let's just start one here. And so that gave me the green light to work with Case Western to get

00:27:24.840 a new medical school, and there hadn't been one in Cleveland or in the country for 26 years.

00:27:30.440 So that was the show of Loop to be a great leader. I mean, he wasn't threatened by cardiology. He wasn't

00:27:37.440 threatened by making it a far more academic environment. He actually saw those odds pluses.

00:27:42.540 He was, you know, just an extraordinary leader.

00:27:46.200 We would have had that second overlap if you'd come to Stanford, because I graduated from Stanford

00:27:49.880 Med School in 01.

00:27:50.960 Oh, wow. When I was looking there to be the dean, it was just after the divorce. It was like a low

00:27:56.600 time morale was broken.

00:27:58.280 It was. So what was his name? I'm blanking. There was a dermatologist who was the dean.

00:28:03.600 Eugene Bauer.

00:28:04.560 Exactly.

00:28:04.880 Is that his name?

00:28:05.380 He was the one.

00:28:06.120 He was the one that I think basically in the failure of that merger, it made sense that there

00:28:10.760 was it going to be a regime change? I don't know if Phyllis Gardner was ever in the running for it,

00:28:15.280 but I always liked her. She was top drawer. I was super impressed by her, but a pediatrician.

00:28:21.140 Yes. From Boston. And he was there for a number of years. And I know who you're talking about.

00:28:26.580 Yeah. Yeah. Last name begins with a P. I don't recall.

00:28:28.540 Yeah. Pizzo.

00:28:29.480 Pizzo. Yeah.

00:28:29.980 Yeah. Yeah. He had been at Boston, infectious disease, pediatrics, and he was kind of opposite

00:28:36.180 of the Stanford way. He was anti-entrepreneurial, anti, in many respects, innovation. So it was

00:28:43.000 interesting to see how that worked out. But before I had decided I didn't want to go there, mainly-

00:28:49.160 So the job you were potentially going to take at Stanford was to be the dean, not to be the

00:28:52.340 division chief of medicine or cardiology. Got it.

00:28:54.320 Yeah. No. And I was, at the time, I thought it was a dream job. I thought Stanford, even

00:28:59.640 though it was coming at a tough time in the wake of this UCSF-Stanford breakup, I thought,

00:29:05.860 you know, hey, it's unilateral. It can only get better.

00:29:08.480 Absolutely. But of course, knowing what I know, the little bit that I know about what it means

00:29:12.260 to be the dean of a hospital, it seems like that would have not allowed you to thrive in

00:29:16.340 the way that you ended up ultimately finding a second home.

00:29:19.920 That's an astute point. You know, you have to know what you're good at, and that might not

00:29:24.040 have been a good fit in retrospect. But I was restless. I was looking for a change. And

00:29:28.920 in fact, working on getting the new medical school at Cleveland, which we basically got

00:29:34.180 in 2002, and the first class came in 2004, that kept me busy. I always need a kind of

00:29:40.160 big project to know something that's a reach to keep me going. And so that was important

00:29:46.260 that it actually was a four-year run on top of the other things I was doing, is to get

00:29:51.060 that new med school off the ground.

00:29:53.100 And there's something else that happened in the twilight of your career at Cleveland

00:29:56.700 Clinic that I want to talk about, because it's, on a personal level, it was very near

00:30:00.380 and dear to me, which was your involvement in the uncovering or elucidation of the challenges

00:30:06.100 with a medication called Vioxx. This is such an interesting example of, it's a case study

00:30:13.560 in so many things, right? Because I'll state for you at the outset my bias in this entire

00:30:18.720 story, and then I want to go into the story in detail. If I could go back in time and

00:30:23.660 be czar for a month or a day or a year, I would have put a black box warning on Vioxx.

00:30:29.480 I would have left it on the market for most people who could have tolerated it and made

00:30:34.040 sure that it was very transparent that this is going to increase the risk of a subset of

00:30:38.180 the population and everybody's happy. Unfortunately, that's not what happened.

00:30:42.580 Merck, I think in the hubris of wanting to deny that there was any potential patient subset

00:30:50.820 that could be harmed by this drug, ended up spending, by my calculation, at least four

00:30:56.280 years probably concealing data. You'll tell us the story and it may be longer. And in the

00:31:01.340 end, a lot of people lost what I still consider to be probably the best COX-2 inhibitor that

00:31:06.640 was ever out there. So that's my bias. I could be wrong.

00:31:09.280 Actually, I agree with everything you said. Totally. And we never discussed it. We never

00:31:13.700 met before.

00:31:14.440 No, no. So now let's talk about the story. So tell people what a COX-2 inhibitor was and

00:31:18.740 why was it such a big deal when these drugs came out in the late 90s?

00:31:22.200 Well, this was, at that time, viewed as the most important blockbuster in medicine, Vioxx

00:31:27.200 and Celebrex. They were competing with each other. They, I think, introduced right around

00:31:31.940 99. And it was a race because there's multi-billions of dollars for each drug. The promise

00:31:38.240 was instead of the Advil and Aleve and other non-steroidals that they would replace, they

00:31:44.200 would spare the stomach. They would be more potent to relieve pain and better anti-inflammatories.

00:31:50.340 That was how they were built.

00:31:52.100 And the reason, because they were selective. So these cyclooxygenase enzymes that the Aleves

00:31:56.800 of the world indiscriminately block them. And one of the problems is, yes, you get the anti-inflammation

00:32:02.380 that relieves your pain, but you also rip apart sort of the gastric lining and a whole bunch

00:32:08.200 of other things in the wake. And of course, as you said, Celebrex and Voxx came along and

00:32:12.040 said, we're going to selectively target just cyclooxygenase 2, which almost seemed too good

00:32:17.800 to be true, by the way. In medicine, it doesn't often work that that happens, that you can selectively

00:32:23.560 hit one of these two enzymes. But nevertheless, that was-

00:32:26.280 Yeah. I mean, they did have some selectivity, but not as much as advertised. But nonetheless, I wasn't

00:32:32.200 really paying attention to this because-

00:32:34.180 Right. You're not a rheumatologist or an orthopedist. This was out of your wheelhouse.

00:32:37.400 No. And I'm not even into drug safety. That was not the kind of thing I was into. And in fact,

00:32:42.220 it was only because this remarkable fellow of ours, Deb Mukherjee, who now is the chief of

00:32:48.320 cardiology in Texas. But at that time, he came to me and said, Dr. Topol, I'm looking at this data

00:32:53.540 from the FDA. And what they're saying is that Voxx is really not at all causing any heart problems.

00:33:02.440 It's actually that the comparator, the naproxen, was the one that is decreasing-

00:33:07.440 Providing a benefit.

00:33:08.460 Yeah.

00:33:08.620 That was the argument.

00:33:09.720 And I said, well, Deb, this FDA, they approved this drug. It was a year plus after they approved

00:33:15.420 it. And I said, how did you get this data? Because back then, to get into the bowels of

00:33:19.680 the FDA website wasn't so easy, but he did it on his own. So I give him credit. And I looked

00:33:25.500 at it. At first, I didn't believe it, but then we spent quite a bit of time-

00:33:28.500 Now, do you remember the numbers? Because I remember that it was naproxen, but do you

00:33:31.960 remember what the absolute risk difference was between naproxen and Vioxx in that first

00:33:37.000 cohort?

00:33:37.320 Yeah, I don't. There was this trial called Vigor. I don't remember the exact numbers, but

00:33:42.220 it was something like an excess of heart attacks in the Vioxx arm, rofococciv, that was not trivial

00:33:51.580 if you look at it per hundred people.

00:33:54.040 My recollection, I could be wrong, and we will link to all of this in great detail. So it'll

00:33:57.880 be, for those of you listening, this will be completely accurate in the show notes. I

00:34:02.000 want to say it was like 15 deaths per 10,000, but I don't remember what the baseline, I don't

00:34:08.140 remember what the naproxen number was.

00:34:09.400 Yeah, I think it was 15 per 1,000.

00:34:10.940 15 per 1,000, okay.

00:34:12.260 Or up to 20 per 1,000, depending on how you interpret the data. But there was a definite

00:34:15.980 gap. And so first I was questioning, Deb, and then we raked over the data.

00:34:21.420 And I said, you know what? You're onto something here.

00:34:23.400 But at that time, Eric, did you think that naproxen provided cardioprotection?

00:34:28.000 No, actually, I said, where did that come from?

00:34:30.380 Okay, so in other words, you sort of questioned the premise.

00:34:32.580 Very peculiar, because there was no data to really support that. And it seemed like a very

00:34:36.860 odd explanation for this excess of heart attacks. We went over the data, and I said, you know what?

00:34:43.680 We've got to publish this, because this is really important. And so we put together a paper.

00:34:48.940 It went to JAMA. It was published-

00:34:51.840 This was 01, or 02, right?

00:34:53.620 No, it was published in 01. In fact, it was, I think, August 30th, 01.

00:34:58.000 You know what? I remember it so well. I remember where I was reading it.

00:35:02.740 Yeah, it was the summer of 01.

00:35:04.500 It was on the front page of the Wall Street Journal. It was on a lot of other front pages. But there,

00:35:09.100 it quoted me as saying, we could be facing a public health disaster.

00:35:14.240 Now, did I ever know that that would be the case? Did I ever know that it would be three years later

00:35:20.600 to the date, September 1, that Merck had this abrupt withdrawal? And in the process, by the way,

00:35:27.860 Peter, before we published in JAMA, Merck came out to try to intimidate us to withdraw the paper.

00:35:34.580 Once they heard-

00:35:34.980 How did they know? Did the reviewers give it to them for comment?

00:35:37.620 The reviewers apparently communicated to them that there was this hatchet job on Vioxx coming,

00:35:44.280 you know? And so, they came to us and tried to intimidate us. And they also then,

00:35:48.060 what I learned from the editor, then the editor, they tried to intimidate her,

00:35:52.620 that they would sue JAMA. And it was unfounded. And as soon as we published the paper, and that wasn't-

00:35:59.060 What did they actually say to you guys?

00:36:01.100 Well, they said that, you know, we did data dredging. They had all the lines. They basically

00:36:06.020 said that we were hacks and didn't know how to hack.

00:36:08.200 P-hacking, data dredging, et cetera.

00:36:09.240 Yeah. All we did was basically review the data that was filed on the FDA. And by the way,

00:36:15.260 some of the things that didn't get out in the public, there were other small studies

00:36:19.360 that never really got in the spotlight that also showed the excess of heart attack.

00:36:24.340 So, the signal-

00:36:24.920 So, did your JAMA study include a meta-analysis of those smaller ones as well as the original FDA-

00:36:29.760 Yes.

00:36:29.960 As the Vigor trial?

00:36:30.900 Exactly. And we saw this consistent signal.

00:36:32.900 And you saw this pattern.

00:36:33.720 It wasn't a question. And we also saw a lesser signal for Celebrex. It was in the paper. But the

00:36:39.920 one that was, you know, just so consistent, and you couldn't deny it, was with Vioxx. And that was

00:36:46.640 not just compared to naproxen. It was compared to other things, including placebo.

00:36:51.440 And so, your feeling at that point was the naproxen comparison is a red herring. And whether you're

00:36:58.000 doing this against a placebo or Advil... And by the way, was there a belief at the time that just

00:37:03.040 general ibuprofen had slight prevention or was neutral?

00:37:07.020 Neutral.

00:37:07.580 Neutral.

00:37:07.880 Neutral at best. There wasn't any hint that naproxen afforded benefit or protection. So,

00:37:13.280 that whole premise was off base.

00:37:15.420 And so, we were talking about a difference of one in a hundred and absolute risk?

00:37:19.140 Two in a hundred.

00:37:19.780 Two in a hundred. So, one in 50 additional. And at that point in time, because I think later on,

00:37:25.400 we knew more. But in 01, did you have a sense of which patients were the ones that were at risk?

00:37:31.320 No. I think that we still don't know that, who was at risk. We do know that 80 million people

00:37:36.080 took Vioxx, which is a lot of people.

00:37:37.960 Yeah. But it wasn't necessarily those with hypertension or those with dyslipidemia. I mean,

00:37:43.000 were we able to sort of stratify it at all?

00:37:45.140 No. In fact, that's the hardest thing is that when there were all these lawsuits of people that had

00:37:49.260 heart attacks, you know, Merck defended it saying, well, it could have been their hyperlipidemia and

00:37:53.920 their high blood pressure. And it's very hard in an individual person to ascribe the hit to Vioxx.

00:38:00.120 That's difficult because most people that have severe osteoarthritis are also having comorbidities

00:38:06.840 that would put them at risk for heart attack. The signal kept showing up though. Like when Kaiser

00:38:12.600 looked at their patient base, database, they saw it everywhere. It looked, it was a heart attack

00:38:17.500 problem and stroke problem, by the way, but heart attacks especially.

00:38:21.120 And the strokes were hypercoagulable strokes?

00:38:22.940 As best we can tell. Yeah. In fact, when I did the 60 minutes segment brings me to that idea about,

00:38:30.460 we talked about, it was right after Vioxx withdrawal. And I was upset because Merck was

00:38:35.140 claiming they did everything right. And I knew much better that it wasn't true. In fact, we had

00:38:40.660 called this three years before and they still never took it seriously. And as you said, they could have

00:38:45.340 just admitted there was a problem. It was in all their emails. It was clearly they knew about it.

00:38:50.020 Wait, there's evidence they knew before your paper?

00:38:52.800 Oh, absolutely.

00:38:53.620 Oh, I don't think I realized. I thought it was, your paper came out in 01. That was the shot across

00:38:59.840 the bow. Then they just completely denied it, concealed data until it became undeniable by 04. I didn't

00:39:07.980 realize prior to 01 internally, they had seen the same signal.

00:39:11.580 Absolutely. No, they had emails. They recovered from the, all the way back to, from 99 when the FDA

00:39:18.520 approved the drug, 2000, well before our paper.

00:39:21.680 Because they did make the argument in 99 that naproxen was risk lowering and that's why there

00:39:26.940 was no signal.

00:39:28.260 Yeah. In fact, the term signal was used as the head scientific officer and all the people involved in

00:39:34.960 the Vioxx development said, well, let's turn it on. Let's flip it.

00:39:38.320 The communications experts showed up and yeah.

00:39:41.140 Yeah. No, the whole thing was just so incredibly contrived and it was all clear that they were in

00:39:47.040 this race with Pfizer, with Celebrex. They didn't want to lose it. $5 billion was on the line and

00:39:52.220 whatnot. But when I went to 60 Minutes to discuss this right after the turbulence of the withdrawal,

00:39:59.460 the interviewer, he had just had a stroke on Vioxx. He never revealed it on the show. I said,

00:40:05.460 well, why don't you, you know, in the, just like we're talking before we actually went on the air,

00:40:09.660 I said, well, why didn't you tell people that? He says, well, I'm not part of the story. I said,

00:40:14.980 well, you had a stroke. I mean, that's kind of a big deal. Ed Bradley, you know, I think there was

00:40:20.460 a lot of hits out there. It's a shame because, you know, up until that time, Merck had been.

00:40:25.820 And what finally, what finally sunk the ship in 04?

00:40:30.180 Well, when they withdrew the drug, there was another new trial. And this one, again,

00:40:36.340 the same exact signal.

00:40:37.540 This was a phase four. This was a, yeah.

00:40:39.600 Actually, I think it was phase three for an expanded indication. Whereas the early one was in,

00:40:45.200 you know, one condition. This was in another, it was a large trial. The heart attack thing was

00:40:49.000 right there again. And they just couldn't deny it anymore, especially on top of everything they've

00:40:54.940 been trying to suppress for years. So they just pulled the plug on it. But-

00:40:58.700 Was there any ramification?

00:41:00.180 No. That's actually, when you mention it, you know why I never should have been involved with this.

00:41:05.820 I regret it because-

00:41:07.680 You do regret it?

00:41:08.460 Oh, absolutely. Because nothing ever happened. I mean, no one at Merck ever-

00:41:13.080 In other words, you believe that had you not written the paper in 01, they still would have

00:41:17.540 withdrawn it in 04?

00:41:19.000 They might have because after we wrote the paper and published it, others started to come alive,

00:41:24.360 like Kaiser and others about this signal. So it was getting more and more undeniable.

00:41:29.520 So I don't know that our paper, even though it was the first one and it was in a high profile journal,

00:41:34.100 I think they still would have had a hard time keeping that drug. Well, they might've done what

00:41:38.400 you suggested, Peter, which is put on a warning and keep marketing, which is what they should have

00:41:43.040 done. It was a good drug. But the problem was that the doses that they were recommending,

00:41:47.220 certain people were getting exposed. And you say, well, two out of a hundred is not a lot,

00:41:52.260 but when you have tens of millions of people, it's a lot.

00:41:55.380 Yeah. One out of 50 absolute increase in risk for a hard outcome like mortality is a huge deal,

00:42:00.960 especially if you can't know who that patient is. So this is where, again, because I never,

00:42:05.500 you know, I was in the middle of my residency when this was going on and I was a surgeon. So it's not

00:42:09.020 like this was top of mind. I just had a personal interest because I remember using Vioxx and finding

00:42:14.180 it so efficacious and finding it to be personally much better than Celebrex and much better than,

00:42:22.040 you know, single day dosing. I think he took 50 milligrams once a day. I mean, it was like,

00:42:26.500 you know, and I had just had a horrible back injury in 2001, which is actually another story where

00:42:32.160 they'd operated on the wrong side. And I had multiple trips to the OR. So I was really debilitated.

00:42:36.920 And in the midst of a surgical residency, Vioxx was the saving grace for me. But my recollection was,

00:42:44.620 oh, but there's a subset of patients in whom you could sort of carve out to not take it. And that

00:42:50.160 would have been the interesting question. That would have been the clinical question, which is like,

00:42:53.320 for example, like if you, if you look at drugs that cause birth defects, something like Avodart or

00:42:58.440 Dutasteride or something like that, like don't take it. If there's a pregnant woman nearby kind

00:43:03.020 of thing becomes a very clear and obvious way. Not that that causes birth defects, but that

00:43:06.960 interferes with androgens. I don't know. So it's interesting to hear you say that, that basically,

00:43:10.940 I don't want to put words in your mouth, but it almost sounds like you said, if you go back in

00:43:13.860 time, you wouldn't have done it. No, because it wound up being a horrible phase in my career,

00:43:19.240 the true nadir, not only during that time after the withdrawal, whether threats from whether it was

00:43:27.220 Merck or friends of Merck, you know, calling up saying, if you don't stop talking about this,

00:43:32.460 you know, bad things are going to happen to you. I remember being out of town one night

00:43:36.440 and my wife got a call like that. You better stop. You better tell your husband to stop saying

00:43:41.820 things about Merck or you're going to regret it. It's hard for people to believe what you're

00:43:45.280 saying, right? It sounds like the sort of thing you'd see in a mob movie. Yeah. No, it was the worst

00:43:50.340 experience. And then I even had my own institution. Unbeknownst to me, the chairman of the board,

00:43:57.220 of Cleveland Clinic, a fellow named Malachi, who was the CEO of InvaCare, but he and Gil Martin,

00:44:03.720 the CEO of Merck, were best friends from Harvard Business School. And so he and the CEO of Cleveland

00:44:09.140 Clinic were basically getting up to suppress me and gag me and also to turn on me. So I had my own

00:44:16.240 institution. I had Merck against me. It was a nightmare. I mean, a veritable nightmare.

00:44:21.360 But without redemption, when Merck finally pulled the drug, you would think that

00:44:26.440 one, it would sort of give people pause to realize that this was, as you said, probably inevitable.

00:44:34.120 And two, it was the right thing to do. Unfortunately, it was too big a hammer for,

00:44:38.920 you know, like I said, I was naive.

00:44:41.000 But they were backed into a corner.

00:44:42.460 No, they were in a corner, but, you know, to be able to-

00:44:45.220 But nobody came around.

00:44:47.140 You know, to me, it's kind of nowadays, everybody talks about truth and fake and whatnot. But to me,

00:44:54.480 then was the beginning of seeing that syndrome because here was truth and it was just being

00:45:00.340 basically turned into fake news by Merck. And they had gone years of marketing a drug,

00:45:07.800 mass marketing a drug. You couldn't turn on a TV set without seeing ads for Vioxx. And they never

00:45:14.700 fessed up. And they just, every single patient case that went to court, they basically prevailed

00:45:20.640 eventually, whether it was the original case or the appeals, by this whole inability to proof,

00:45:27.620 for proof in an individual patient. So they didn't pay anything.

00:45:31.720 There was no restitution whatsoever?

00:45:32.940 Nothing that I know that's significant. And most importantly, the executives who oversaw this,

00:45:38.220 who knew exactly what they were doing, they didn't go to jail. They were never indicted. There

00:45:43.080 was never any charge.

00:45:44.820 So no civil suits at all?

00:45:46.340 Nothing, nothing.

00:45:47.260 Which is interesting. It tells you something about how difficult it is when the complication

00:45:52.760 is a ubiquitous disease. You see, it's different when you're dealing with, well, and of course,

00:45:58.800 we think of the examples that turned out to be wrong, right? Like the use of silicone breast

00:46:03.320 implants and lupus. Well, it turned out to be incorrect, but you at least had a signal to talk

00:46:07.380 about because lupus was so rare or what other connective tissue disorders they were talking

00:46:11.160 about. But as you said, like, how can you possibly look at any individual and make that case

00:46:16.780 probabilistically? You would need a very large trial to determine that.

00:46:20.940 Oh, yeah. No, you can't single it out. It's almost impossible. If you had assays to show

00:46:26.220 that the selectivity of the COX-2 inhibitor was prothrombotic, making a clot in a person,

00:46:32.320 and that person then had a heart attack or stroke, but who had that? I mean, these were sudden events,

00:46:37.820 and no one had a proof in that person that their clotting status changed from the drug.

00:46:42.440 It seems like that's got to be the most likely mechanism.

00:46:44.960 Oh, yeah. The mechanism of how these people went down is not elusive. But what's sad about this too,

00:46:52.440 Peter, is I had known Roy Vagelos to some extent. I had the highest regard for this company. We were

00:46:58.600 doing trials with this company when it happened. And so just to see a company that was viewed as

00:47:05.000 one of the most ethical-minded companies, not just in pharma, but across all companies,

00:47:11.860 to see it take these tactics of marketing, it's really sad. But of course, that's many years ago.

00:47:19.000 You know, that's 2004. We're 15 years later now. Yeah, but I can hear it in your voice, Eric. It's

00:47:24.140 still quite traumatic. Yeah, it was almost the end of my career. You know, that's what precipitated

00:47:28.640 leaving Cleveland and fortunately, you know, coming to San Diego, which was the greatest thing ever.

00:47:34.600 But who would have known for a year or two, it was a question of whether there would be a new

00:47:39.400 position and whether it would be suited to things that I would want to do. So how did San Diego,

00:47:45.500 I mean, were you a pariah at the time? Yeah. You know, one of my people who I regarded

00:47:50.460 in extraordinary, the Pope of cardiology, Gene Brownwall, he was trying to help me. He says,

00:47:56.460 you know what, Eric, you're radioactive right now. And I was. And I even had people at Cleveland Clinic,

00:48:02.220 by then there was a new CEO and others who were charged to try to nuke me. That is any place I

00:48:09.400 interviewed for a position. They were calling them and actively trying to take me down. So

00:48:15.740 ultimately, over that course of a year, when I was looking to move, I started realizing I have to do

00:48:22.020 this in stealth mode because, you know, I've got people who are trying to get me. And fortunately,

00:48:26.780 a very close friend of mine here in San Diego, who I'd known for decades, Paul Tierstein,

00:48:32.900 who is at Scripps. And I had collaborated some with the people at Scripps Research. And so

00:48:37.120 we started talking, they were excited about what I had a vision for. And then I was ultimately

00:48:43.320 recruited in fall of 2006. And what was the role they brought you into at that time? Because at the

00:48:51.520 time wasn't Scripps, I mean, it wasn't really a clinical powerhouse. It was a research powerhouse.

00:48:56.780 Well, kind of both in some respects. The research-

00:49:00.020 Was the affiliation with UCSD on the clinical?

00:49:02.420 No, no. Purely Scripps Health.

00:49:04.680 Purely Scripps Health, yeah.

00:49:05.640 So Scripps Health was on the move. They had, Chris Angorder as the CEO had basically put together,

00:49:13.040 stitched together many different Scripps entities into one called Scripps Health. They were and are

00:49:19.440 completely different entity than Scripps Research.

00:49:22.060 So TRSI, the Translational Research Institute, that was totally separate.

00:49:25.840 So totally separate. Although prior to 2000, year 2000, they were one entity, but it was Scripps

00:49:31.680 Clinic then, not this big health system with multiple hospitals and 30 clinics and whatnot.

00:49:37.140 So what I did was to come in to be a cardiologist at Scripps Clinic, but also to develop a new institute

00:49:45.020 that was dedicated to translational research, particularly genomics. That's why I came here.

00:49:50.440 And it was only, you know, within weeks I realized, wait a minute, what about wireless? What about digital?

00:49:55.840 Because you don't want to just rely on a genome, even though back in 06, there was tremendous-

00:50:01.400 Now what was Craig Venter doing at that time?

00:50:04.220 So Craig, he had the Venter Institute in Maryland. He was also working in synthetic biology,

00:50:10.780 had a synthetic biology company here. And I think he was aiming to develop another Venter Institute

00:50:17.480 in San Diego. But he, as a pioneer of pushing the whole sequencing project, of course, in year 2000,

00:50:25.760 announcing it with Francis Collins and Bill Clinton at the White House. But he had moved

00:50:30.640 not just from sequencing, but also to writing the genome with synthetic biology. That was his

00:50:36.000 interest at the time.

00:50:36.940 I see. Got it.

00:50:37.680 So I came here to try to make human genomics and genetics center stage for the two Scripps

00:50:43.760 institutions.

00:50:45.320 Right. And you wanted to translate this as quickly as possible to basically patient care.

00:50:49.760 Yeah. To change practice, which is, we're still working on that. But that was the goal. And we

00:50:55.140 basically very quickly, fortunately, were able to get a big grant called the CTSA grant, one of the

00:51:01.680 57 now hubs of that in the country. And we're the only one that's not a university or without a medical

00:51:07.800 school. But basically, Scripps Research is a storied institution with some of the best life

00:51:13.540 science in the world, ranked number one in nature for innovation and influence above some of the very

00:51:20.240 top known centers. So it has had a phenomenal track record. And to work with them, this great brain

00:51:26.800 trust of scientists, and to try to bridge that with this big clinical entity, Scripps Health, which is

00:51:32.500 a dominant player in the San Diego region, a big region. For me, it was perfect. And basically,

00:51:39.220 the big grant we were able to get led to an innovation space, you know, just to do whatever

00:51:44.380 you think would be appropriate to make medicine better. And there's no shortage of ways we could do

00:51:50.680 that.

00:51:51.800 And you're also the editor-in-chief of Medscape. Is that right?

00:51:54.080 Right, right.

00:51:54.760 How did you get involved? And I think anybody listening to this who's ever gone onto Google

00:51:59.660 and searched for something will notice Medscape is usually coming up with information. So what is

00:52:05.140 Medscape?

00:52:05.760 Yeah, well, Medscape is the professional side for healthcare professionals of WebMD. The way I got

00:52:12.220 into it was in the mid-90s, when the internet was kind of warming up, I started with a couple of

00:52:19.640 friends, the heart.org, which was the first cardiovascular website for cardiologists and

00:52:26.520 anyone working in this space. So we started that and, you know, it's all about getting great content,

00:52:32.820 getting journalists. And it was, for many years, a big magnet for not just the information, but also

00:52:41.200 a forum for education and for debates and whatnot. So ultimately Medscape started to cover every

00:52:49.660 specialty and they acquired the heart.org. And in that acquisition, being the editor-in-chief of that,

00:52:56.360 they ultimately asked me, would I be the editor-in-chief of Medscape? And so I've done that now for several

00:53:01.720 years. It's been great.

00:53:02.700 How much time does that take? I mean, that's, you must have an editorial staff under you because

00:53:07.360 it's such a voluminous, I mean, it's like an encyclopedia.

00:53:11.280 Yeah, no, they have an amazing crew of medical journalists and they cover everything that moves

00:53:16.220 in medicine. I don't do so much day to day. I set general direction. We have a monthly call to go over

00:53:23.460 features that I usually try to introduce ideas for that. I do a lot of interviews. I try to find,

00:53:29.220 like, you know, this week was the big Wall Street Journal issue with a Penn Medicine former dean

00:53:37.140 taking on medical education today, saying that it was completely off base to nurture students

00:53:43.440 on climate change or gun control or any social injustice. And of course there was a revolt that,

00:53:51.380 you know, and we're going to have a lot of that in Medscape. So I tried to bring up my, you know,

00:53:56.500 when I first got involved, the website was much more pharma oriented. And what I've tried to do is

00:54:02.880 round that out with not just devices and medical education, but also, you know, the whole genomics

00:54:09.580 and digital medicine and AI and all those sorts of topics. How big is the staff?

00:54:14.700 Oh gosh, there's probably over 30, 35 journalists. Ivan Oransky recently joined as the VP for editorial.

00:54:22.340 He runs retraction watch, which is really formidable, but no, it's, it's a big staff.

00:54:28.120 It's a for-profit or not for-profit?

00:54:29.800 Well, it's part of WebMD.

00:54:31.020 It's part of WebMD.

00:54:32.500 WebMD used to be a publicly traded company, but they were acquired about a year ago by a

00:54:37.620 company called internet brand. So they're now a private, but for-profit company.

00:54:41.880 Got it. I didn't realize that I should have known that I suppose, but I didn't realize

00:54:45.120 Medscape was under that umbrella.

00:54:46.600 Yeah. And I've always tried to weave in the WebMD side because WebMD has a big reach to

00:54:54.000 consumers, as you pointed out, to kind of go to search for lots of common things in medicine.

00:55:00.200 And we don't do that enough. I'm hoping that over time we'll, we'll see better crosstalk

00:55:05.980 because we may have some really interesting things on the Medscape side or the opposite

00:55:10.720 on the WebMD. We don't get enough trying to get that mixed audience.

00:55:14.160 Yeah. It's funny that it's taken us this long to get to your, your most recent book,

00:55:18.520 but, um, I think it was a worthwhile route to get here because I think that the story

00:55:23.840 of Vioxx alone, I think is, um, well, I learned a lot because I, again, I think I knew parts

00:55:29.160 of it, but I don't think I appreciated the severity with which you've paid a price.

00:55:34.160 Fortunately, you passed tense and didn't hold out.

00:55:36.520 Yeah. And it sounds like it's worked out for the best, but that's phenomenal.

00:55:40.520 That seems to be one of those experiences that falls in the category of you're probably better

00:55:44.900 for it, but would be, you'd never want to redo it.

00:55:47.920 Exactly. You know, you get much stronger, you learn that, learn who your friends are and aren't. And

00:55:52.600 basically I, when I got here, it was like being in the witness protection program

00:55:56.520 and, uh, you're starting all over. I remember had this big lab where we're going to do all the

00:56:01.580 sequencing and, you know, and I, I'm sitting in this big lab and I'm the only one in the lab.

00:56:07.140 And I got a lot of recruitment to do. Now we have just in our group, you know, well over a hundred

00:56:11.500 people. We have one of the largest NIH grants in history to do all of us, the big million person

00:56:18.640 participants, a diverse group that we're doing so much with over the years ahead. So, you know,

00:56:24.400 things are really humming. So it's been great.

00:56:26.900 So your book is called Deep Medicine and the picture on the front really points to AI,

00:56:33.460 but the book is about more than that. But I want to start with that. Now let's assume for a moment

00:56:38.120 that someone listening to this has heard the term AI, but, and sort of knows from science fiction

00:56:43.740 movies, what it kind of means, but that's the limit of the knowledge, right? So they don't maybe

00:56:49.360 necessarily know the difference between machine learning and artificial intelligence or those

00:56:53.000 terms synonymous, let alone, how would that even factor into medicine? And how do you separate out

00:56:58.560 the sci-fi from what's how already happening, right? And to, you know, what you think a path looks like.

00:57:02.680 So take that in any order you like.

00:57:04.940 Well, I mean, I think the problem with AI is it's been around the concept since the fifties,

00:57:09.740 1950s, and it's diffuse. Yes, there's lots of sci-fi and movies and misunderstandings, but

00:57:16.200 what we're talking about now is a specific subtype of AI, which got its birth just over 10 years ago

00:57:23.880 called deep learning, neural networks that allow for inputs. And they could be millions,

00:57:30.580 billions of data points, could be images, could be speech, could be text. And then it goes through

00:57:36.080 these layers of artificial neurons, which are not very much like neurons, but nonetheless,

00:57:41.160 they can distinguish features progressively as they go through this, this network, and then you get

00:57:47.640 outputs. And what's remarkable about this era and why it recently won the Turing prize for

00:57:53.460 Jeffrey Hinton and his colleagues from University of Toronto. But the thing that's so-

00:57:59.500 The Turing prize being basically the Nobel prize for computer science.

00:58:02.540 Yes, I should have mentioned that. Exactly. The reason why this is such a big advance in medicine,

00:58:07.960 the biggest advance I've ever seen as a student of medicine for many decades now, but it's so big

00:58:13.820 because you can take particularly now images and you can get accurate definition of the image

00:58:22.200 better than experts, doctors. So whether it's radiologists or dermatologists, pathologists,

00:58:29.300 cardiologists, I mean, you go down the list, ophthalmologists, and you will see studies now to show

00:58:35.380 superiority of accuracy, or at least as good through a machine.

00:58:40.340 Now, to be clear, this is in initial recognition, not comparison. I mean, I think this is an area I

00:58:45.720 don't know very much about, Eric, but the last time I thought about this and did some reading about

00:58:50.140 this, I came away with the impression that if you took an MRI of a person and you showed, so this first

00:58:57.240 time this person's getting an MRI, you get the best radiologist to look at it. You get the best

00:59:01.520 computer to look at it. The computer still struggled for macro context. It still didn't even realize that

00:59:07.780 was the liver per se, but it could certainly with greater fidelity and resolution once told that

00:59:14.180 was the liver identify and maybe be more clear about, well, what's a cyst versus what's a hemangioma

00:59:21.800 versus what's a hepatoma. So it had superiority there. It also had superiority when it came to serial

00:59:27.620 studies. So, you know, Mrs. Smith had a chest X-ray a year ago. She has a cough now. She has another

00:59:33.220 chest X-ray. Is there a difference? But am I right in my recollection? No, no. Actually, I'm really glad

00:59:38.700 you put some anchoring on that because what we have, deep learning is in many respects extraordinary,

00:59:45.060 but it's very narrow. So if I say, find me pulmonary nodules in a chest X-ray, that's where I say it can be

00:59:53.260 superior. And clearly the best is the combination, the synergy, the symbiosis of what the machine can,

01:00:00.140 quote, see versus what the doctor could see. So yes, it's a very narrow thing. But what we're

01:00:07.340 talking about here is there's so many mistakes in medicine because things are missed or are

01:00:14.300 inaccurate. And, you know, this extends through pathology and every different specialty.

01:00:18.620 Yeah. Your thesis is not, I mean, many people have said to me when they talk about this sort of loosely

01:00:24.420 that the radiologist is the first doctor on the chopping block. That's not really your thesis.

01:00:29.500 No, I actually think that's completely wrong. Jeffrey Hinton said that once, and I think

01:00:34.540 he will ultimately regret it. The point being is that it basically tees it up. That is, you get a

01:00:41.520 different complementary read of something, and that helps for speed and accuracy, and it could

01:00:49.280 ultimately lower cost, and it could ultimately improve medicine. The thesis of deep medicine is

01:00:53.640 if we lean on machines more. In many respects, we can get into that. But if we do that more,

01:00:59.200 we can free up to have time with patients, and we could get the doctor-patient relationship back to

01:01:05.320 where it ought to be, where it was, you know, some 40 years ago. That's the main premise that is unique

01:01:11.760 about the book in which, you know, I really build up to deep empathy with the last chapter. But the real

01:01:18.220 thing that's different now is that we have lots of promise, lots of potential for AI. We haven't

01:01:26.340 actualized that. We haven't proven it for the most part. You know, one of the only randomized trials

01:01:32.040 to date is in colonoscopy, because a lot of polyps, particularly if they're flat or sessile or small,

01:01:38.000 are missed. And it's very much operator dependent, how much time they take to do a thorough colonoscopy.

01:01:44.660 And so now there's a Chinese randomized trial that shows, hey, you know, if you use deep learning

01:01:49.880 machine vision, you can pick up polyps that are routinely missed. And so then people say, okay,

01:01:56.300 so what? Maybe the ones that are missed are not important. Well, you know, that's where we are

01:02:01.080 today. That's the study, is you look at the denominator of the missed versus the not missed,

01:02:05.040 the machine caught versus not, and what's the prevalence? Because if the prevalence of pathology

01:02:09.580 in them is at least the same, you could argue, they shouldn't be missed. And it might be higher

01:02:14.820 if it's sessile. Yeah. So, you know, 20% of polyps were picked up by machine vision. And then

01:02:21.680 we still don't know how much of that were true disease likely. I've always felt the field, if

01:02:28.600 radiology is the first pit stop on this journey, I've always felt like the ICU needed to be a very

01:02:34.440 close second. How much is really being done there? Because A, it's the, obviously the most data rich

01:02:41.300 environment after radiology, radiology also, of course, informing the ICU, but in terms of just

01:02:47.380 raw numbers coming out about a patient, you know, if you think about a patient on a ventilator with,

01:02:52.520 you know, CVVHD and like you pick, pick every device strapped up to a patient. It's not the same

01:02:58.960 as a formula one car, but you're, you're in the ballpark of that much data. Yeah. And you're

01:03:04.020 touching on one of the big deficiencies of AI and deep learning today, which is multimodal data.

01:03:11.000 So when you have all these inputs of varied types, you know, not just their vital signs,

01:03:17.800 but, you know, could be machine vision of their facial recognition. It could be, you know,

01:03:24.020 so many different parts about that person, no less their prior electronic record. And we don't do well

01:03:31.300 with that because deep learning today is, as I say, narrow task. It's like, you know, what's in this

01:03:36.940 eye ground for ophthalmologists? Is this a diabetic retinopathy? Is this something else? So the ability

01:03:44.260 to take many layers of data, which would be the ICU story is in the early stages, even more so than

01:03:50.300 the image recognition. Yeah. What realistically, where do you think that is in terms of, again,

01:03:56.600 caveating it with the, it's always going to take longer than we think it is. Is this something

01:04:02.140 where, I don't know, in 10 years or in 20 years going to an ICU will afford a patient, the luxury of

01:04:11.900 a true supercomputer. That's basically assimilating the CVVHD data with the ventilator data,

01:04:21.060 with the swan GANS data, like stuff that, as you point out, like it's, it's even the most

01:04:28.700 analytical physician can't really recognize the patterns that are deep within all of those data.

01:04:36.120 Well, you just touched on with that statement, the essence of why we need AI support, not just in an

01:04:43.240 ICU patient, but in every patient, there's more data than we can handle. Especially when you say people

01:04:49.580 are wearing sensors, they're going to be wearing more, people are going to have their genome sequence

01:04:53.340 or they already have a genome chip or array, a microbiome of the gut. I mean, no less all their

01:04:59.880 records, not just the one healthcare place that they happen to be visiting that moment.

01:05:04.400 So this flood of data per person, no less the intensive data collection in an ICU setting,

01:05:11.880 this is overridden human capability. We need machines. We have to fess up that we can't do this,

01:05:17.780 but we also have to acknowledge that we're not there yet now. So when are we going to get there?

01:05:22.040 Well, you know, Fei-Fei Li and the group at Stanford has done ICU work, machine vision to see

01:05:28.380 whether- Is it single machine or is it integrated?

01:05:30.760 Their studies have been mainly single whereby, for example, they're looking to see risk of

01:05:36.620 extubation so that you don't have to have a nurse in the room all the time that what's going on with

01:05:41.900 that patient that they're going to self-extubate or others have looked at likelihood of sepsis or

01:05:47.980 different pieces of the story, but no one has integrated it all yet today. And I think that's

01:05:54.180 where it's headed. We're seeing these hybrid models of bringing the data together, but a lot of the

01:06:00.100 problem with this field has been way out of bound type of where it can go. And when I did the research

01:06:07.400 in the book, which involved a few years of work cumulatively, I spoke to a lot of the real gurus

01:06:12.400 in the field and they made it clear that, you know, we are going to get there eventually, but we're not

01:06:18.140 there yet. That is the challenge because when you think about other big breakthroughs that we look

01:06:23.080 back on, we don't realize that they were more discoveries than creations sometimes. So for example,

01:06:31.440 look at germ theory, right? This is, again, it's something we take for granted today. And it's hard to believe

01:06:38.020 there was a day when you wouldn't wash your hands before operating on a patient or you wouldn't wear

01:06:42.420 sterile gloves. So we acknowledge how that transformed medicine in a step function manner, but two things

01:06:49.340 are a little bit missed when we contrast it. One is we didn't have to build it. We accepted it and

01:06:58.040 discovered it. And two, it didn't happen overnight. Like there's still a generation that it takes to

01:07:04.140 implement these things. And so that's the best case scenario, right? It doesn't get any better than

01:07:10.300 that. This is something that has to be built. This is almost, this is, I can't think of an example, maybe

01:07:15.960 I'm wrong. And if anyone's going to think of it, it's you, but is there an example before when we had the

01:07:20.960 idea and the promise, and then we set out to engineer the solution building into it? So for example, I'll give

01:07:27.480 you, I'll give you a failed example, which is the EMR, right? Like literally the worst thing on the

01:07:33.960 face of the earth. I heard you once talk about this and I think, I think it was you actually,

01:07:38.000 but maybe it wasn't, but I'm going to give you credit for it. But I think you best summarize it

01:07:42.900 by saying, look, the EMR was created as a billing solution, not a clinical solution. And I couldn't

01:07:50.420 agree with you more, but there's an example of, okay, we have a problem. Medical records are so

01:07:55.500 cumbersome, so voluminous, although really they're just a two-dimensional problem. It's really not,

01:07:59.480 it's three dimensions if you include time, I would say. And we now have computers, quote unquote. So

01:08:04.920 computers will solve the problem. Let's build it. Well, one, it took a lot longer than people expected.

01:08:10.400 It took much longer to implement it and it sucks much more than people could have ever imagined it

01:08:16.020 could suck. When I think of those examples, I keep saying, is there a positive story? Is there a great

01:08:22.500 case study in medicine where the engineering solution lived up to its expectations?

01:08:28.040 I think you nailed it. I don't believe there is one. This is a unique story that's being written

01:08:34.920 as we speak. It's so different than we have a technology and we just want to implement it. This

01:08:42.240 is one that there's a lot of construction that's still required. We know what the house is likely going

01:08:48.800 to look like when it's built, but we're, you know, still on the foundation stage.

01:08:53.960 Do you think that these are problems that are going to be solved by the giants? Is IBM, is Google,

01:08:59.500 I mean, are these the entities that figure this out or is this going to be solved in more of a pharma

01:09:07.040 model where the early discovery and the early stage, you know, even the stage one, the safety trials are

01:09:14.820 done by small companies that ultimately get acquired by rolled up into larger companies.

01:09:19.440 You know what I mean? Today, like the, the Merck's and the Pfizer's of the world aren't really doing

01:09:22.960 drug discovery anymore. They've decided we're going to outsource that to more nimble companies.

01:09:27.220 And basically the private markets now subsidize that while the public markets subsidize late stage

01:09:32.160 drug development. Do you think that's the way this is going to be? Or do you think this is going to

01:09:35.800 have to start and finish within the behemoth companies that have their enormously deep pockets?

01:09:40.080 I think this is a story of innovation from the outside. I think it's very different than what

01:09:45.180 you're seeing now with, you know, the consolidation and pharma and outsourcing here. You see the big

01:09:50.620 titans like Google and Microsoft, Amazon, and the rest of them, they all recognize this is the

01:09:56.400 greatest opportunity for growth and also some, a noble mission of improving health. So you have

01:10:03.740 that group, you have startups that are, there's no shortage of those. And you also have some

01:10:10.400 governments like in China, in the UK and other places that are nurturing this, that are investing big

01:10:16.080 in this area. So, you know, I think this combined force of multiple entities is where we're going to

01:10:23.160 see this, you know, really take off. It's starting to happen much more in China out of need. That is, the

01:10:29.960 implementation is way ahead of what's going on in the U.S. because they have so few.

01:10:34.340 What are some examples?

01:10:35.460 Well, the radiologists, whereas here, we're just starting to get a bunch of FDA approved algorithms

01:10:40.360 for reading various types of scans. They already have that widespread throughout China. They already

01:10:46.840 are doing, you know, many things on the, well, you know, we hear the only FDA consumer approved or

01:10:54.000 cleared is the Apple watch for heart arrhythmia, a deep learning algorithm for atrial fibrillation.

01:11:00.420 Whereas.

01:11:00.800 So explain how that works. Let's use that example because that's near and dear to everybody's wrist.

01:11:04.840 And I see you're wearing your Apple watch there as well. So let's just say you went into the

01:11:09.400 Apple store today for the very first time and you bought an Apple watch.

01:11:13.500 Okay. So first of all, it's on the back surface of the wrist, the volar surface of the wrist.

01:11:18.240 And what is it shining through? And I assume it's shining it onto the veins in the back of your arm.

01:11:23.300 Yeah. No, it's, it's picking up optically each heart rate and you can see the light that it's

01:11:28.620 used. And for the deep learning algorithm, which actually was first cleared by a startup,

01:11:35.180 a live core. And then a year later, Apple, which they didn't even acknowledge that they had been

01:11:39.360 a year after the first, but nonetheless on their watch, they get heart rate. So at rest,

01:11:45.660 and then when you are active and then they basically for you, it has your data whereby when

01:11:53.240 you have heart rate at rest, that's off track for you, it says, Hmm, get a cardiogram and you get a

01:12:01.140 one lead cardiogram when you press the crown on the EC and you get a good quality cardiogram.

01:12:06.080 And then if it has atrial fibrillation, which lead does it most closely approximate on the 12 lead?

01:12:11.400 It's a lead one. You get a cardiogram read for atrial fibrillation, which one thing it's pretty

01:12:19.400 good for that. I was about to say, not to minimize that, but AFib seems like about the easiest thing

01:12:24.160 to pick up because of the irregularity of it, right? Yeah. Although there are, there is some false

01:12:28.980 positives and negatives because sometimes the P ways are that you're, you're looking to be

01:12:34.940 absent, you know, sometimes you can get faked out. And so it's reasonably good. And, you know,

01:12:39.720 it's in the 90 plus percent accuracy level, but it's all about the base theorem of, for people,

01:12:46.220 more information. Well, for people who are not risk, a lot of people have an Apple watch who are,

01:12:51.040 you know, young and have zero risk of atrial fibrillation and they get a cardiogram and gets

01:12:55.100 anxious and they may even get workups by a cardiologist. So this is a problem where we have

01:13:01.000 marketing of an algorithm, the first deep learning algorithm.

01:13:05.440 How long does it take by the way, um, to learn a person well enough that it would be willing to

01:13:10.880 make a recommendation like that? Oh, just a matter of hours. Wow. Or certainly by a couple of days,

01:13:16.380 it's got it down. Okay. But yeah, I mean, you know, your heart resting heart rate by the

01:13:21.340 accelerometers, it knows that you're not moving and why did your resting heart rate used to be 60?

01:13:26.420 Why is it a hundred something? And then it'll tell you to get a cardiogram, but...

01:13:30.400 And it can't make any other diagnosis. It can't diagnose any ventricular rhythm.

01:13:34.140 Not now.

01:13:34.680 Or atrial tachycardia or anything else.

01:13:36.880 Ultimately it should be able to, but those algorithms haven't really been validated yet.

01:13:41.940 But ultimately we know now I use a six lead cardiogram. It isn't on the watch,

01:13:47.140 but you can just do that with sensors and put it on the leg.

01:13:50.300 Wait, wait, how do you do that? That's interesting.

01:13:51.820 Yeah. You know, it's basically half the size of a credit card.

01:13:55.360 Where do you get this? This is, it's an aftermarket product or...

01:13:58.880 No, it's actually marketed now by a live core. The one that came with this ECG on the watch first,

01:14:04.740 they actually put it on the Apple watch, but it was their algorithm. They came up with a six lead,

01:14:09.680 which you then put that on your leg, your left leg, and then you get six, all limb lead.

01:14:14.880 And you do this with your patients as well?

01:14:16.500 Yeah. Every patient, when I see them, instead of just taking their pulse,

01:14:20.260 I also do a six lead cardiogram. It's been remarkably insightful because it's free.

01:14:25.300 It takes a second. And then I can really be much more certain about if they have an arrhythmia,

01:14:30.720 but also diagnose conduction system abnormalities.

01:14:33.880 So it's accurate enough that you can measure your intervals perfectly.

01:14:37.600 Oh my gosh. It's, it's the quality is amazing. Yeah. I mean, the six lead...

01:14:42.420 Now, can you send them home with the same kit and then can they get a six lead on themselves at home

01:14:47.500 and let you see the data?

01:14:49.240 They could. I haven't done that yet, but that's probably where this is headed.

01:14:53.240 The reason why this is actually funny, you mentioned it, Peter, you can even do your own

01:14:58.160 stress test with this.

01:14:59.700 Yeah, of course. In fact, you could do a real stress test, which is in the actual environment

01:15:03.420 under which you need to be stressed.

01:15:05.040 Yeah. I did that the other day. I did a rest ECG and then I got on a bicycle, stationary bicycle,

01:15:10.980 and went really hard. And then I, just after I got off and did a six lead again, I, so I said,

01:15:15.520 wow, you can do a stress electrocardiogram, high quality, six lead, and never go near a medical

01:15:21.460 kind.

01:15:21.640 Where's the output? Where are you seeing the output?

01:15:23.080 Oh, on your phone.

01:15:24.260 Okay. And you can make it a PDF and send it off to your doctor.

01:15:27.860 It makes it automatically. Yes.

01:15:29.740 Huh.

01:15:30.460 Yeah. It's pretty cool.

01:15:33.560 So, I mean, that strikes me as proof of concept now.

01:15:38.600 Yeah. Well, and that's where we're going to get, like you alluded to, when are we going to get all

01:15:42.960 the heart rhythm abnormalities diagnosed and the heart conduction, you know, which is a precursor

01:15:48.100 to arrhythmia. So that's where we're headed because one lead, it's hard to do that. But when you have

01:15:52.660 all the limits, in fact, with AI, you could impute all 12 leads. You don't even need to get the other

01:15:57.880 six leads. So pretty soon we're going to see that six lead become really valuable entry for what's

01:16:05.180 going on in a person's heart. In fact, Mayo Clinic just published a series of papers on 12 lead

01:16:10.220 cardiograms that you could get heart function. You could predict from a cardiogram whether they're

01:16:15.020 going to have atrial fibrillation. You can get the potassium level of the blood through that.

01:16:19.680 Wow.

01:16:19.860 I mean, the amount of data that's sitting in this pattern, which we can't see, is amazing.

01:16:26.780 Well, think about the number of times I see this once a month, and my practice is really small. So if

01:16:32.280 I'm seeing this once a month, let's extrapolate to how many times this happens in the United States.

01:16:36.960 The blood hemolyzes slightly on a blood draw and the potassium comes back at 5.5.

01:16:41.820 Oh yeah. Or higher.

01:16:42.820 Yeah. And you don't know what to do.

01:16:44.460 Right.

01:16:45.040 Well, imagine you had that EKG, you wouldn't have to panic because every time that happens,

01:16:49.920 you have to call that patient, send them into the ER, get a blood draw, confirm what you know is

01:16:55.640 likely true, which is the stupid sample hemolyzed. Their potassium is really 4.7.

01:17:00.120 And, but imagine you didn't have to do that. You could just say, push this button on your watch.

01:17:06.600 That exemplifies-

01:17:07.960 That's a great example.

01:17:08.620 It exemplifies what we can't see, but having a machine trained by a million cardiogram,

01:17:15.980 what it can see. And in the book, in deep medicine, I have a chapter and it starts out

01:17:21.320 with that story. How did they discover the potassium story? Something we can't, we can tell

01:17:26.820 the potassium is really high with the tall T waves, you know, but we can't get to the decimal point.

01:17:32.500 Right. We can't distinguish between 4.9, which is do nothing and 5.6, which is you better be

01:17:38.040 careful.

01:17:38.320 Right. And that's what machines are good for. And we're going to be seeing a lot more of that

01:17:42.900 kind of stuff. That is the eyeopening thing to me is to learn about all the things that we humans

01:17:48.700 can't do, that machines can be trained to do. And they're just going to get better over time.

01:17:54.440 So if you, do you wear a Dexcom sensor?

01:17:57.220 I have. Not regularly. I'm not a diabetic, but I have, and I've learned a lot from it. I've,

01:18:02.000 you know, I've tried Dexcom and the Libra. I mean, I've, I've really found this glucose thing

01:18:07.640 because of how it interacts with what you eat, with your sleep, with your physical activity. It's

01:18:13.780 amazing. Yeah. It really is. You know, people ask me why I still wear it because I'm not diabetic

01:18:19.600 and even my patients, so about a third of my patients to a half my patients wear this,

01:18:25.780 none of whom have diabetes.

01:18:27.320 Wow.

01:18:27.840 And I always ask them for 90 days. If every one of my patients would wear it for 90 days,

01:18:31.740 at least I'd be happy. And then we could decide, but what's in, what invariably happens

01:18:35.780 is people realize the, they go through the following cycle, which is Peter, you've been wearing this

01:18:41.340 thing for four or five years. Haven't you already figured out what to eat and what not to eat?

01:18:45.540 And I say, well, yes and no, but it's more complicated because like, for example, let me

01:18:50.700 show you this. I have not eaten anything since 4 PM yesterday.

01:18:56.440 Wow.

01:18:57.100 It's 1130. So I'm coming up on 20 hours of no food.

01:19:00.260 Yeah.

01:19:00.540 Look at the variability in my glucose for the last 12 hours. It's been as high.

01:19:06.340 It peaked at 118, which was, it peaked at 118, which was right after a workout

01:19:14.440 this morning. And by the way, it was just weight training. It wasn't like high intensity interval

01:19:19.160 training. If it's high intensity interval training, it's going to go much higher.

01:19:21.960 Now it's sitting at 94 and you'd think, well, if knowing that it's 94, like if I ate a bagel right

01:19:30.700 now, could I predict what it would go to just knowing it's 94? The answer is not a chance.

01:19:35.480 You see, just knowing it's 94 isn't enough to tell me my response to the bagel.

01:19:40.380 I have to know how much glycogen I have. I have to know how much cortisol I have. I have

01:19:45.040 to know how much insulin I have. Like there's so many variables and that's why four or five

01:19:50.800 years later, there's nothing about this that is boring to me because I'm constantly learning

01:19:55.980 a new physiologic experiment. I mean, if there's anything that's ripe for AI, it would

01:20:01.340 also be CGM coupled with other data. So in other words, I don't think the CGM data as

01:20:07.720 a, as the input feed would be sufficient. You would have to constantly be pairing it with

01:20:11.760 your activity and other sensors because if we had like, if you had the cortisol sensor

01:20:17.300 and the lactate sensor, I mean, that starts to become remarkable predictive power. And when

01:20:23.060 you could get to the point where, cause this is my dream, I want to know, can I eat that

01:20:27.260 right now or not for my parameters? So this is my pipe dream is I want to be able to say,

01:20:33.540 go into the algorithm and tell it your desired average glucose, your desired variability. So I

01:20:39.560 want to average glucose that's below a hundred milligrams per deciliter or below 110 milligrams

01:20:44.860 per deciliter. I want a standard deviation that doesn't exceed 15 milligrams per deciliter.

01:20:49.880 And now you tell me what I can eat. Spend the first month watching me eat, learning how my body

01:20:57.160 responds to every different food and go from there. I mean, directionally speaking, how long

01:21:03.220 would it take to get us there? We're getting there. I mean, we're chipping away at that. So

01:21:07.180 the gut microbiome is a big part of the story too. And I know you're such a proponent of this and I

01:21:13.160 am, I call myself a gut skeptic because, well, why would I say that? I certainly don't

01:21:21.400 disregard the importance of that. I think I'm waiting to see a great example of how I can use

01:21:28.760 it outside of like the really clear clinical ones. Like certainly knowing how to change the gut

01:21:35.140 microbiome in the context of C. diff colitis is profound. It seems very likely that something about

01:21:42.000 the gut changes in patients with diabetes who undergo gastric bypass, that seems to really

01:21:47.900 suggest, but it could be as high as the duodenum. And the most compelling evidence I've seen is that

01:21:51.900 it's actually the change. It's the duodenal bypass that specifically gives them this incredible

01:21:58.000 remission out of the gate more so than the lower GI tract. But I think most of my skepticism comes from

01:22:05.220 the fact that it's not clear to me what to do with all those data, which may be exactly your point

01:22:10.120 that when I see patients constantly show up to me with their gut sequence and they say, well,

01:22:16.380 look at this pathology state here. And I say, well, first of all, I don't know that that's a pathology

01:22:20.000 state. And if it is a pathology state, is taking a probiotic the answer? I don't have any evidence

01:22:26.060 that that's the case either. So is it more a readout state or is it a form, is it a malleable

01:22:32.980 state that we directly interfere with?

01:22:35.280 Right. So those are all important questions. I think the real insight here is that up until

01:22:41.880 when Aaron Siegel and his group at Wiseman Institute in Israel, up until they did what

01:22:47.440 now has to be seen as a classic study.

01:22:49.740 This was the cell metabolism paper from about a year ago?

01:22:52.680 Well, there was a paper in Cell 2015, which was really the seminal work. And now there's been

01:22:58.280 several more and it's been replicated by many others. They took now thousands of people healthy

01:23:04.320 like yourself and they went ahead and got gut microbiome, but they also got exact same amount

01:23:10.840 of food, exact same time. They also got all their labs and every piece of data they could get on these

01:23:17.280 people. And they found that you could predict if they had a bagel, which ones are going to have and

01:23:23.320 what level of glucose spikes they're going to see. And they, they found that so many spikes, you know,

01:23:29.180 even very significant spikes, 160, 180, 200 in healthy people with no sign of diabetes.

01:23:35.680 And how, how durable do you think the knowledge is from the sequence? So for example, like if you

01:23:40.640 sequenced that patient Monday morning at 9am, how much do we know that Friday at 5pm, the data are

01:23:50.360 still relevant, assuming you could even, cause you can't get the data in real time.

01:23:53.860 Right. So they didn't do any DNA sequencing. And of course that wouldn't change. So we don't know

01:23:58.780 the genomic side of this, but we do know the microbiome, unless you do something significant

01:24:03.740 like that. Oh no, sorry. I didn't mean their DNA. I meant the DNA of the bacteria.

01:24:07.680 Okay, good. Yeah. Yeah. The DNA of the microbiome is pretty darn stable everywhere you look at it,

01:24:13.900 unless you change radical change in your diet, like change fiber content, or you take antibiotics,

01:24:19.300 but it's very stable from day to day. I see. So you would say that, look, Peter,

01:24:24.400 only if the patient does a course of antibiotics, do we need to recheck them or make a radical dietary

01:24:30.840 change. But if a person's in quasi steady state, you could sequence them every quarter and basically

01:24:37.980 update your pretest probabilities of what the distribution is. That's right. And another tier

01:24:43.420 of complexity, because it's good that you're a skeptic on this, but early on these various

01:24:49.460 companies that would do microbiome assessment, they just said how much you have of this bacteria or

01:24:54.760 that bacteria. It was like a density of bacteria. It turns out that you touched on it. If you sequence

01:25:00.220 the bacteria, the changes in that bacteria species sequence is just as important as the density of

01:25:07.080 the type of bacteria. So this is not easy. It's expensive to do it right.

01:25:11.160 And we've already seen a big fraud on this front quite recently, right?

01:25:14.520 Yeah. Ubiome.

01:25:15.880 Ubiome.

01:25:16.460 Is the Theranos of this space, it seems like.

01:25:18.700 Yeah. Well, they were billing people illegitimately, and they were only reporting on

01:25:22.740 density of bacteria. I don't know that they're reporting-

01:25:24.660 So they weren't object fraud. It was just bad practices?

01:25:27.420 Yeah. I don't think they were doing things wrong with respect to the microbiome density. It's very

01:25:33.240 rudimentary. They didn't do sequencing. They did basically a bacterial density of-

01:25:38.380 I mean, I found them to be the most useless company in the history of civilization because

01:25:42.100 back in 2012, 2011, I was, I mean, at least acting like I was on the forefront of this,

01:25:50.420 trying to understand it, and ordering these sequences on myself and all my patients. And

01:25:55.580 I don't understand how this company stayed in business. I mean, they didn't, but they couldn't

01:26:00.040 run a sequence to save their lives.

01:26:01.420 Well, yeah. I think the biggest thing is they were fraudulently billing people, double billing,

01:26:07.200 triple billing, you know, all that sort of thing. That's what got them into, you know,

01:26:10.800 basically collapse mode. I don't know enough about their sequencing.

01:26:14.380 I mean, I always found Larry Smarr's stuff to be the most interesting because Larry's doing it at

01:26:18.520 a level that you couldn't do commercially.

01:26:20.740 Yeah. So shotgun sequencing, where you do true metagenomics, you know, there are only

01:26:25.240 certain labs, like the one I mentioned in Wiseman. Here in San Diego, the Knight Lab does

01:26:30.340 metagenomics. That is-

01:26:32.620 K-N?

01:26:33.060 Yeah. K-N-I-G-H-T, Robert Knight. So these are the centers that are doing it right,

01:26:38.160 that are sequencing each species of every organism that's found. And we now know that

01:26:44.400 that sequence is equally as important as the type of bacteria. So that's the sort of data. Now,

01:26:51.120 the other thing you're bringing up that's really important is we have no idea how to manipulate

01:26:56.000 the gut microbiome. The only thing we know is a fecal transplant in certain people with,

01:27:01.620 you know, pseudomembranous colitis, C. difficile. Outside of that, we don't have,

01:27:06.300 we have crapsules that are being made that are being tested.

01:27:10.660 So I think you and I definitely see probably more closely on this than I would have guessed

01:27:15.360 initially, because we agree that at this point, it's an output of data, not an input to manipulate

01:27:20.760 necessarily.

01:27:21.280 So I probably need to go back and look at the Weissman paper again, because I don't think I've

01:27:25.620 looked at it in over a year. And my view was, which is probably incorrect by the way, that CGM

01:27:32.700 and dietary logging would have been sufficient. So what I really want to do is go back and look

01:27:39.160 at that paper and see what the gut biome added above those things, which I'm guessing there is

01:27:44.540 something there.

01:27:45.180 Yeah. No, there is. And we need more. I mean, basically right now is you could predict if you

01:27:50.620 had all the data and the right algorithms, you could predict which foods you'll spike from.

01:27:56.340 And then this was taken to another level by the group in London, King's College, led by Tim

01:28:01.040 Spector. He brought in all these twins from all over the country, identical twins. So they had their

01:28:06.800 gut microbiome, and they also put in a line to a vein to get blood samples for triglycerides.

01:28:12.720 Yep.

01:28:13.220 And they saw the same thing.

01:28:14.720 Which by the way, you could get out of a sensor.

01:28:16.860 You could, you could.

01:28:17.600 Ultimately.

01:28:18.200 Yeah.

01:28:18.420 We don't have one yet.

01:28:19.640 No. And you know why I mentioned lactate earlier? If you have real-time lactate, you

01:28:24.980 are estimating with really great precision, mitochondrial oxidation. Now you understand

01:28:30.460 fuel partitioning. You see, to me, if you asked me a year ago, how would you want to

01:28:35.880 best estimate fuel partitioning? I'd say, oh, it's tough because you got to have somebody

01:28:39.160 basically walk around with a respirator or something that can measure oxygen consumption

01:28:44.800 and CO2 production. But I think lactate's telling you that. I think if you really know how to

01:28:50.060 calibrate lactate, you can estimate fat oxidation versus glycolysis and, or glycolysis through

01:28:57.880 to lactate. And so all of a sudden you now get into this. So the reason that right now

01:29:02.020 my glucose is 94, but if I ate a bagel, it would go to like 104 is because I'm so glycogen

01:29:08.360 depleted because I haven't eaten in 20 hours and I've worked out very hard, or at least for a long

01:29:13.980 enough duration. Conversely, it's not uncommon after dinner. Let's say you have dinner, you have

01:29:19.380 a glucose spike up to 120, and then it comes back down to 90, 94. You eat that same bagel, you'll go

01:29:26.780 higher. Well, a very important input into predicting that is knowing glycogen stores and insulin

01:29:33.620 sensitivity of the muscle and all these other things. So what I need to better get is how many

01:29:39.760 different phenotypes, macro phenotypes of gut biome are there that really matter?

01:29:45.900 Right. The bigger picture though, I agree with all your point and it'd be nice to see a lactate

01:29:50.740 sensor that's tested at scale and is accurate. It, you know, took a while to get that for glucose and

01:29:56.040 we're earliest stages on the lactate, but there's still a lot of naysayers here and I understand their

01:30:01.760 perspective and that is, so what? So what if your glucose goes to 180 or your lactate goes to this

01:30:07.320 or your triglycerides go to that? The point is, do we know that changing that, that keeping everything

01:30:13.700 nice, level, keel? You don't think the diabetes literature has made it clear enough that normalizing

01:30:19.680 glucose and insulin is beneficial? Not enough, no. The only way you can get at this, inferentially,

01:30:27.340 yes. But you know what we've... Oh, you're saying I can tell the story that a glucose of 120 is better

01:30:33.620 than 180 because I have clinical trial data to demonstrate that all day long. And I can even

01:30:38.800 tell you that how you achieve that matters. But you're saying I don't have the data to tell you

01:30:46.640 that 100 is better than 110. No, no. Another way to put it is I don't have the data to show

01:30:53.320 that if you wear a sensor for X number of time, 90 days in your case or forever or a week,

01:30:59.680 that you're learning about avoidance of glucose spikes changes your prognosis. We don't know that.

01:31:06.580 And the same thing for triglycerides, which by the way, they don't correlate. And we're learning that

01:31:11.200 each person's individual response throughout their day is so incredibly unique. And we're learning some

01:31:19.160 of the factors. I don't even know. We know all the factors that influence that. You've mentioned,

01:31:23.020 of course, glycogen stores and physical activity and the microbiome.

01:31:27.140 And cortisol, in my experience, plays a staggering role.

01:31:30.120 Stress. No question. Stress. I mean, you know, just if you get an intermittent,

01:31:34.060 intervening cold, no less stress, you know, in your family, your life, whatever experience.

01:31:40.280 So yeah, this is a really interesting area. We're learning about ourselves. It's like, you know,

01:31:44.680 know thyself sort of thing. We're in the early stages. And I see the skeptics. I understand their

01:31:50.220 perspective. I think that we have to prove it. I lean where you are, which is why not have this

01:31:56.560 information? I've learned a lot about myself, no less the feel from it. But I think we have to admit

01:32:03.020 that we got a ways to go. What do you think's a significant blind spot in medicine today at the

01:32:09.060 macro level or at any level? No, I think the biggest blind spot is how poor we are in diagnosis,

01:32:16.660 no less in treatment. I mean, I think that when you really look hard at the data,

01:32:20.960 what's amazing, Peter, is you see all these clinical trials that declare, you know, a triumph.

01:32:26.780 And they're helping three out of a hundred people. I mean, a great example of statins that,

01:32:31.920 you know, primary prevention of statins, if not the number one, close to number one class of drugs

01:32:37.820 that are used today. And you see that out of a thousand people for primary prevention,

01:32:43.920 988 derive no benefit for five years of taking a statin and 12 out of a thousand get benefit.

01:32:51.460 So whether you look at the diagnosis where, you know, if a doctor...

01:32:54.820 That's another topic I know that we may disagree on. My view on that has always been that

01:32:59.660 because the time course of atherosclerosis is so long, you know, it's a disease that begins in

01:33:05.040 infancy. We certainly know from, you know, the starry stuff of the, you know, the seventies that,

01:33:11.340 you know, basically by the time you're 18 years old, you've, most people have a stage three lesion

01:33:16.880 at that point that the challenge with studying primary prevention is you could never study it

01:33:22.060 long enough to really see where those curves start to diverge.

01:33:25.720 Well, no, they diverge. But the question is, are they going to keep diverging? And, you know,

01:33:29.820 most of the benefits starts to kick in right around 18 months. And yes, they're still slightly

01:33:34.940 diverging, you know, after five years, but we don't have any data beyond that. So I should restate

01:33:39.840 that is that we don't have any proof that more, you know, you can suggest that instead of 18 out of

01:33:46.960 a thousand people benefit, that it goes to 36. Exactly. But what about the 970 that don't derive?

01:33:53.260 No, no, no, no. That's fair. I think the point is, do you believe the Mendelian randomizations

01:33:58.940 or do you think that the Mendelian randomizations have artifacts in them, which any Mendelian

01:34:05.280 randomization will have an artifact if that which changes the variable of interest also changes

01:34:10.320 something else that you don't know? That's always, that's, that's the blind spot of the

01:34:13.400 Mendelian randomization. No, they're, they're a neat way to get a readout, but they're not perfect.

01:34:17.120 You know, I think that whether you look at diagnosis where, you know, if you take people

01:34:22.260 who have died under a doctor's care and you say, why did that person die? What was the cause of

01:34:27.860 death? 40% of doctors say, I absolutely know the answer are wrong. 40%. That's how many it autopsy are

01:34:39.040 show a different reason for the death as what was pre-mortem. Wow. That's, that's a, that's a,

01:34:44.980 that's a big gap. Yeah. If you ask doctors to make a diagnosis, if they don't think about it in

01:34:50.600 the first five minutes, five minutes is 95% accurate. But if they don't, it doesn't come

01:34:56.000 to mind in the first five minutes, it drops down to 24% accuracy. Basically what we have is type one

01:35:02.740 system, type one thinking system, one thing I should say at Kahneman's work. And we just are

01:35:08.420 reflexive. We don't reflectively go over anything. We don't have time. We don't simulate all the data.

01:35:14.260 We can't because of so much. And so we have to basically the whole, the big whole is acknowledge

01:35:21.600 that we can do far better and it can't all be through human support. We need help.

01:35:28.340 Do you think there's any area where, I think you've made a compelling case that machine plus

01:35:33.380 human should be better than human. Eventually it will be. I mean, I think even in radiology today,

01:35:39.960 it should be. Oh yeah. Yeah. Hopefully critical care would be an amazing place where machine plus

01:35:45.140 human should be better than human. Do you think that there are extremes on either way? Do you think

01:35:49.620 there are places where humans will always be better than machines plus humans?

01:35:53.880 Yes. And that's in being human, which is the bond, you know, just like our conversation,

01:36:00.520 this deep conversation, it's really, uh, illustrates the human connection. We don't have

01:36:06.660 those kinds of conversations in seven minutes or 10 minutes where the patient can't, and, uh,

01:36:11.760 you can't get to know a person's life history. That's never going to get really digitized story,

01:36:16.600 their story. We interrupt patients within 18 seconds from, we don't listen. So the point being is that

01:36:23.900 machines, that's the, what we don't want machines. They don't have context. They're not going to be able

01:36:29.100 to truly understand all the nonverbal communication and the real issues of a person that are deep.

01:36:37.740 And that's where the humanity, we need to bring it back. I mean, the essence of medicine is that,

01:36:44.000 and it's been lost. It's become this big business and it's,

01:36:48.360 Do you worry about the, um, I don't know, well, it's not really a brain drain, but do you worry that

01:36:52.840 the war for talent has shifted? And back when you went into medicine, it's probably safe to assume

01:36:59.760 you were one of the best students in your high school, the best student in college. Like it

01:37:03.680 would have been skimming at the very, very top of the pyramid of students. Is that still the case

01:37:09.360 today? Or has, I mean, you've already alluded to a number of things that I've seen. Luckily I don't

01:37:15.080 experience them just on the nature of my practice because it's private. But I mean, these stories you

01:37:20.120 tell, I know so well, the doctor that gets seven minutes to see a patient. And in that seven

01:37:24.920 minutes, six minutes is typing into an EMR, the moral distress and the absolute erosion of a belief

01:37:34.780 in what you're doing is a huge cause of burnout amongst physicians. And I don't understand like

01:37:41.880 if you're the top student in college and you're interested in life sciences, why you'd pick medicine

01:37:46.240 today? Unless you had a profound confidence that you could carve a path distinct from what most are

01:37:52.220 probably going to do. I mean, you have to be an optimist, I think, to pursue medicine today.

01:37:57.120 Do you worry that it's going to be hard to recruit the most talented kids out of college into medical

01:38:03.820 school? Well, I'm hoping it won't be, but you know, I do hear constantly friends who are doctors who

01:38:10.580 tell their kids, don't go into medicine. Yes. I hear that. I mean, that's really bad because here

01:38:17.660 is the ultimate profession for sense of really helping people. And then you have the people who

01:38:24.120 are in it saying it's horrible. And as you know, Peter, the physician burnout, no less all clinician

01:38:30.300 burnout is at peak. And why is that? It's because it become data clerks and they're squeezed for the

01:38:37.600 time. They can't care for people when you don't have time. So this is, of course, going back to

01:38:44.640 that main thesis of gift of time. We can get that. But I think we have to restore medicine

01:38:50.900 the way it was in order to attract the talent that you're referring to. And I think it's doable.

01:38:58.360 It's not going to be easy. It's going to require a lot of activism, which we haven't had that much of

01:39:03.040 in medicine. Yeah. This is something doctors are quite ill-equipped to do. It seems.

01:39:09.160 We're seeing the light on activism. The gun control, NRA really brought it out when they said,

01:39:15.460 stay in your lane. This was when the AMA said a physician should be asking a patient if they own

01:39:20.600 a gun. It was American College of Physicians. They published it in the Annals Internal Medicine last

01:39:26.140 fall. And then NRA said, you know, these doctors should stay in their lane. And then you had all

01:39:32.100 the doctors came out. One of them, Judy Mellinick, saying, this is my fucking lane. And, you know,

01:39:37.220 it went everywhere. It went viral. The idea being, if doctors are going to be killed potentially by

01:39:41.920 patients, it's not an unreasonable question to say. Well, that and caring for all the gunshot

01:39:46.420 victims. Yeah. You know, the emergency room. And the suicides. And the suicides. Which is probably the

01:39:50.940 greatest cause of gun death. Yeah. You've got both suicides and homicides and mass killings and,

01:39:56.420 you know, AR-15s and all this stuff. So, you know, the fact that this was taboo, that doctors

01:40:00.980 didn't, weren't allowed to talk to patients and they weren't allowed to do research. I mean,

01:40:05.500 there was no research in this area. So this was a void that now has brought out a lot of the activists

01:40:11.980 and social media. And it's this new era of young physicians, a lot of them women. And we're seeing

01:40:20.300 activism like never before. I wrote about it in the New Yorker recently about this and, you know,

01:40:24.920 should doctors organize. And this, what we're seeing, it is happening. And the hope is that

01:40:30.300 we're going to see an organization take hold where all doctors can join as well as ultimately

01:40:35.780 patient advocates and others. And that we will turn this around because this is the biggest concern I

01:40:43.300 have, Peter, is that we're going to see AI kick in more and more over the years. But what will it do

01:40:49.080 if the administrators who are the overlords, who are overrepresented as compared to the people

01:40:57.000 taking care of patients, if they keep the squeeze on, we're just going to see things get worse.

01:41:02.600 So we have to override that. And the only way we're going to do that and turn inward and get the

01:41:07.640 humanity back in medicine is to have doctors organize. And the gravitas of a million doctors

01:41:14.620 in America, all being part of one entity, it could be enormous.

01:41:19.140 You know, and this, this gets to another problem within medicine that I alluded to earlier, which

01:41:23.360 is the patients aren't footing the bill based on the system directly and therefore demand in a

01:41:30.480 demand-based system. So if it's not a demand-based system, it doesn't matter. Like in other words,

01:41:34.340 in the NHS or the Canadian system, the patient is not footing the bill, but they're also not driving

01:41:39.760 the demand. The demand is budget set. But in the US where you have this paradox, it also means the

01:41:45.800 patient's voice doesn't matter. And that's the irony of it, right? So it's the opposite of the DC

01:41:52.120 license plate, you know, which is taxation without representation. It's like no taxation and no

01:41:58.320 representation. And that's why I don't, I worry that patients can't be the ones to drive this change,

01:42:05.240 which they should be able to, like the patients should demand this change, but because they're

01:42:11.440 not the ones writing the checks to feed that $3.8 trillion machine directly, they're only doing

01:42:17.260 it indirectly. In other words, they don't get to control it, right? They're paying their taxes

01:42:21.840 and their employers are withholding it, but they, it's not the same as saying, here's my dollar,

01:42:28.020 go and do this thing with it.

01:42:29.880 Yeah, no, that's why if you get the doctors to come together to start this, and the sole purpose

01:42:35.240 is the patient-doctor relationship. It's not about better reimbursement or all the other

01:42:41.760 trade guild activities, but rather it's about, we want to fix this relationship and bring the

01:42:46.980 humanity back in medicine. Then we start to see, you know, that the ability for that patient interest

01:42:53.080 to be recognized. Because now all you have is you got a lot of patient advocacy groups,

01:42:57.720 but they're, you know, they're just like the doctor organizations, they're all balkanized.

01:43:01.760 We need one entity to stand strong. And, and I, I hope we'll get there.

01:43:07.120 Well, that's a good point, right? It should be made up of physicians and patients, actually.

01:43:10.760 It really shouldn't be, they shouldn't be separate.

01:43:12.480 No, no, I think, you know, it started with the physicians because, you know, there's just a million

01:43:16.920 of them that we can identify and get them together as many as possible. Then you start adding on

01:43:22.620 Are there really a million physicians in the United States?

01:43:25.120 Yeah. You know, the fact checking of the New Yorker is amazing when they, I've never experienced

01:43:29.480 that before, but they tracked down everything and they got to the numbers that I never could get

01:43:34.620 to. So not all of them are practicing. There's about 900, almost 900,000 are actually practicing

01:43:40.420 in some respect, but there are a million docs in this country.

01:43:43.520 Have you done the math about how many doctors we would need under this new regime of empathic,

01:43:52.420 intelligent, artificial, this symbiotic relationship? In other words, because, because

01:43:57.380 there's a bunch of moving pieces, right? It's radiologists still exist, but now they're there

01:44:02.740 to talk more with patients and interpret the diagnosis as opposed to make the diagnosis, right?

01:44:08.840 And you're still going to need the internist, but now they have more time with the patient and

01:44:14.140 they don't have to worry about the diagnosis as much as they have to worry about the treatment

01:44:18.780 directionally. Eric, do we need the same number of physicians? Are we going to need more physicians

01:44:24.800 or are we going to need less physicians in 30 years on a per patient basis?

01:44:29.400 You know, when I did the UK review, we got into that and we had economists and all sorts of brilliant

01:44:35.980 people modeling on that. And I think what we'll see is even though everything now would suggest we

01:44:42.440 need a lot more doctors because of the aging population and all the comorbidities and the

01:44:46.600 complexity. I mean, if you just look at like, for example, how do you care for a patient with cancer

01:44:51.100 today? It's gotten very complex. So you would project we're going to need, you know, a steep growth

01:44:57.420 curve. But what we're going to see, I think, is a big blunting of that because we are going to be,

01:45:02.780 the machine story is not just about doctors relying more on machines getting support. It's also about

01:45:09.240 consumers, patients. And so when you get the outsourcing and the offloading, you start to

01:45:14.960 see a pretty big, and then when you get rid of the hospital story and just have surveillance centers,

01:45:20.260 remote monitoring, you start to see a very less need for expansion. So I don't see what we're going to,

01:45:26.440 you know, be a decline, but just a difference in the curves as they go forward over the next few

01:45:31.980 decades. And there will also be this, I don't know, for lack of a better word, kind of a growing pain

01:45:37.560 as people transition. I mean, most of the radiologists I know today, you know, for example,

01:45:43.480 have very technical backgrounds. I mean, any of the MRI folks I know, they usually have a great

01:45:48.860 background in physics and things like that. So all of a sudden there's going to be a different

01:45:52.660 selection criteria. For example, you may want to choose radiology independent of how technical

01:45:58.120 your background is in physics or mathematics. And so it's like, it takes a generation to make these

01:46:04.020 switches. Do you see any other types of changes in how people will select into different specialties?

01:46:10.960 Well, that's hard to know. I think we have theorized, Saurabh Jha and I from Penn Radiology,

01:46:18.080 that there might be a new specialty that would just be radiology and pathology combined,

01:46:23.060 at least the pathologists who work with slides, because it's very similar interaction with the

01:46:29.040 computer. And they're often very much, as you know, integrated. So that might be a whole new specialty

01:46:33.960 over time. But, you know, overall, one thing we don't want to forget here is that the empowerment

01:46:40.060 of patients to do doctorless diagnoses of most common conditions, whether that's an ear infection

01:46:46.660 of a child or a urinary tract infection, a skin rash or skin lesion, and on and on. The routine things

01:46:54.520 are not going to have doctor in the loop, only if, you know, treatment is needed perhaps. And that's

01:46:59.660 only in the U.S., not in a lot of other countries. So that is going to change also specialties. Because

01:47:05.300 today, you look at pediatrics, you know, it's a wonderful specialty, but a lot of that could be

01:47:10.280 decompressed if you give parents more autonomy for their kids. So we're going to see lots of changes

01:47:17.000 based on the patient side or the parent side of things, which I think has not been adequately

01:47:23.240 appreciated. If you could conduct any experiment and there were no limits on the resources,

01:47:29.420 you had. So you could, and it could be a, an experiment within the real recesses of basic

01:47:36.700 science. It could be a real translational experiment that takes something to the, from the cutting edge

01:47:43.320 of the bench to the bedside, or it could be the largest clinical trial ever done to test a question

01:47:50.320 that vexes you without any new introduction of a new technology, but just simply asking a question

01:47:56.220 like the Vioxx one, for example, you know, take as much time as you want, but I'd love to know

01:48:01.220 what would be a dream experiment for you if this is the one shot on goal where you've got billions of

01:48:08.540 dollars and no holds barred. Yeah, no, it's easy. Oh, it is. Yeah. For me, I mean, I'm, I, it's a dream.

01:48:15.080 You know, I wrote about this with Kai-Fu Lee and Nature Biotech last month, and it was called,

01:48:20.520 it takes a planet. And basically it would be to, the experiment would be to develop a

01:48:25.300 planetary digital infrastructure with all the data of each individual and continually being assessed

01:48:33.380 and assimilated process inputs, federated AI. So the data never leaves the country, whether it's China

01:48:39.540 or the U.S. or wherever. So it's not a privacy security. But the point being is when a new person

01:48:46.000 comes in and we want to prevent a condition or better treat it, and we have billions of other

01:48:51.260 people to draw from, and we have these digital twins, if you will, because today we learn from

01:48:58.400 clinical trials and it's really farcical in many respects because those clinical trials are contrived

01:49:03.620 and the benefit is three per hundred or something like that. What about the other people?

01:49:07.540 Well, not only that, it's often three per hundred because of the heterogeneity of the population.

01:49:12.100 Exactly.

01:49:12.520 It might be 30 per hundred if you knew who to apply it to.

01:49:15.540 Yeah. So, so my experiment would be just for that. You have pinpoint precision because I know

01:49:21.060 Peter's twins, all of them around the world and how, what treatment they got and what outcomes they

01:49:27.580 got and what, how I could prevent their issue that they otherwise had. And so it would be to develop

01:49:33.080 the ultimate learning health system.

01:49:34.800 And the twin you're defining is obviously not just genetic, but it's every layer, every layer.

01:49:40.820 So it's my gut twin, my epigenetic twin, or my approximate genetic twin, my phenotype twin,

01:49:47.820 my metabolic twin.

01:49:49.580 Right. And that's, I think where we can go in the future. And it involves many different types of AI.

01:49:56.780 And I think we'll get there someday. And it's an experiment.

01:49:59.280 Is that a 50 year? I mean, what realistically is.

01:50:02.720 I think it could be done in 20 if we were, if we really were going after it, because it's doable.

01:50:08.340 You know, the question is.

01:50:09.140 That strikes me as bigger than any one country though, right?

01:50:11.860 No, no, but if you get, if you get U.S. and China to start it because of the diverse population

01:50:16.920 and the largeness, the U.S. being third in population in the world, and you get that going,

01:50:22.220 then the rest of them join on, you know, you pretty quickly, you will have twins.

01:50:25.980 Do you have a sense of how much that could cost per person?

01:50:29.440 It just depends on the layers of data and the analytics. I mean, it's not going to be

01:50:33.940 trivial, but less than you would think. This is mostly, you know, in silico work. It's not,

01:50:40.180 you know, it's happening anyway. This data all sits in places that's all fragmented.

01:50:46.120 What's the, what's the price of a full exome sequence today? Both, a full genome sequence

01:50:51.840 today. So not a 23andMe, but where they're doing the full sequence.

01:50:54.440 Yeah. Well, exome 400, a full genome, eight, 900. So, you know, they're-

01:50:59.340 So they're sub a thousand.

01:51:00.420 They're going to keep coming down. And at scale, perhaps, you know, that's going to happen even

01:51:05.780 faster. But you start having all these things done at scale and, you know, right now we don't

01:51:12.560 have enough. The reason why a genome isn't valuable is because we don't have a billion people with a

01:51:18.460 sequence and a phenotype. Once we start to get into those big numbers, then we start to decode it.

01:51:24.020 So you don't think it's that the genome is not deterministic enough? You think it's

01:51:28.260 too small an N so far?

01:51:29.960 That's a bigger problem?

01:51:30.920 Big part of it. Yes. I mean, a genome will never be fully deterministic. I mean, that's

01:51:34.960 not possible. Probabilistic, yes. But the probabilistic side of it is hampered because

01:51:40.580 of inadequate numbers.

01:51:42.180 Yeah. Because I got to tell you right now, I find every time my patient sends me their genome,

01:51:47.360 I just roll my eyes and say like-

01:51:49.640 No, there's a limited amount of value.

01:51:51.380 Yeah. Well, you know what I usually say to them? I say, look, anything in here that matters,

01:51:55.820 we already know. You know, if you're a 40-year-old, unless you were adopted, you know,

01:51:59.840 sometimes you figure out, you know, this person has Lynch syndrome or something that you had to know.

01:52:04.660 So there are some numbers on that. If you look at the Danville, Pennsylvania,

01:52:09.540 Geisinger Health System, where they've done over 100,000 people, they have excellent,

01:52:14.920 not full genomes, but the coding elements. And they have 5% that they find something quite

01:52:21.100 important, so-called pathogenic. So like Lynch syndrome or BRCA or sudden death arrhythmia.

01:52:27.820 Which again, my point is, if you're doing your job as a doctor, you should have figured that out

01:52:32.500 in the family history and gone looking for it, but not with, I mean, in other words, you should

01:52:36.680 have gone to the genome to be confirming what you suspected, hopefully.

01:52:39.880 Yeah, but you know, a lot of that's missed. Like BRCA is a perfect example. You know,

01:52:43.760 what about BRCA men carriers? You know, you just don't know.

01:52:47.780 You're taking a good family history, don't you? I mean, I guess it depends on how much the patients

01:52:51.880 know about their family too, but you're right. No, those are good examples. But when I look at the,

01:52:57.100 like how many times do I look at Prometheus and it spits out, oh, you're at a higher risk for

01:53:01.060 atherosclerosis and a higher risk for diabetes. And I'm like, this is such nonsense, right? Like if you

01:53:06.140 actually understand how to evaluate lipids and you're wearing a CGM, you certainly don't need

01:53:10.240 this thing to tell somebody. No, Prometheus, I mean, I think there's gross deficiencies of outputs

01:53:15.900 because again, going back to, we don't have one central repository of data. Oh gosh. How small is

01:53:24.640 it? In terms of how many millions of people have had sequence, it's small still, you know, it's in

01:53:30.720 the 10 to, well, it's in the less than 10 million for sure of whole genome sequence.

01:53:37.360 And then of course, how many do we have accurate phenotyping on? Because if the phenotype is

01:53:41.960 not that accurate, then it dilutes the quality of what you're trying to do, right?

01:53:46.140 That's essential because what phenotype we have is fixed at the moment the genome was assessed.

01:53:51.920 Exactly. We don't know. The phenotypes change. So all the studies that we have-

01:53:55.520 So this has to be living, breathing, and longitudinal.

01:53:57.720 Exactly. And that's why I'm trying to see our way through, like, you know, the all of us study

01:54:03.520 of the million people that we're onto right now is the beginning of something like that,

01:54:07.280 where all the layers of data for long-term follow-up, it's still tiny, a hundred thousand.

01:54:11.980 I mean, a million people is tiny, but it's a start. But if we could get the leading countries in the

01:54:17.520 world to get behind this, you know, this is something that should override concerns about

01:54:23.720 competition in countries, this is about, you know, for mankind, humankind, then we might be able to

01:54:30.160 really develop something that would promote health of all human beings. That would be far-reaching.

01:54:36.860 And you know what? I actually think this is going to happen someday. I know it sounds far-fetched.

01:54:40.740 I know you think, looking at me like I'm a little cuckoo.

01:54:43.120 No, I mean, look, a lot of things seem far-fetched in the moment. I think it's,

01:54:47.640 truthfully, I think it's technologically more capable, it's more possible to me technologically

01:54:53.700 than it is politically.

01:54:55.360 Okay, well, that's good to hear. I like that.

01:54:57.000 Because I don't, I think the biggest challenge, I'm doing the back of the envelope math,

01:55:02.080 just, I think it's a couple trillion dollars to do this in the United States alone.

01:55:06.940 I don't know. Depends on how long, I mean, this is over forever.

01:55:11.900 Exactly. Which means it's a moonshot. And I don't feel like our political environment

01:55:19.200 is capable of moonshots anymore.

01:55:21.300 We're just trying to live day to day here now.

01:55:23.020 Yeah. So long gone are those days when you could make a bold, we're going to spend the

01:55:29.200 equivalent of a couple trillion dollars over 20 years, long after I'm gone, meaning me,

01:55:35.160 meaning the politician who's going to be the torchbearer of this to make this a reality.

01:55:39.820 I don't know. Maybe I'm just a jaded, skeptical guy when it comes to our political system.

01:55:43.840 Well, it's actually healthy to be looking at it that way. And also, I want to also frame it,

01:55:49.640 Peter, as it is an experiment, because you still have to, once you develop it at some scale,

01:55:54.540 you still have to prove that it's helping people.

01:55:56.660 Right. So you'll need to use a biased and an unbiased subset of these.

01:56:00.220 Yeah. So, you know, it's theory, it's intriguing, it's doable, and it's going to get progressively

01:56:05.820 better, you know, what we can put in as inputs, but it's still a question mark. Will it improve?

01:56:11.940 I just think that our complete reliance on clinical trials is misled.

01:56:17.560 I completely agree. I think the heterogeneity problem and the exposure, the time under the

01:56:22.140 curve, the exposure problem, make clinical trials very difficult to extrapolate from. I mean,

01:56:27.760 here's one of my favorite pet peeve examples is people are so quick to dismiss Zetia as a useful

01:56:34.120 drug. But in reality, it's never once, to my knowledge, been targeted towards patients that

01:56:39.340 are hyperabsorbers of sterols. Right.

01:56:41.580 And yet, you know, so Zetia gets sort of diluted in clinical trials because you're giving it to

01:56:46.260 patients that have normal and abnormal absorption of sterols. And so on balance, it doesn't look like

01:56:54.000 a very interesting drug. It seems to work okay with a statin, but I'm convinced there are patients

01:56:58.820 out there taking statins who should be taking Zetia because if you, you can phenotype this,

01:57:03.160 you can really see people who don't make that much cholesterol, but they absorb it like crazy.

01:57:08.580 It's amazing.

01:57:09.440 But who's going to do that clinical trial, right? No one's going to do that clinical trial.

01:57:12.640 The lack of interest of the people that manufacture the drug because of dilution of the,

01:57:17.720 of the market. I mean, it's really unfortunate, but you're, you're absolutely right about that.

01:57:21.340 Well, Eric, this has been, this has been a really interesting discussion and I'm glad,

01:57:25.260 uh, it's crazy that it took a decade for us to, I mean, I, I'm amazed you knew my name. I'm flattered,

01:57:31.040 but I, I certainly known about you for, from the day I got to San Diego and I'm just glad that the

01:57:36.700 podcast and your book really became a good excuse to sit down. So thank you. Thank you for your work,

01:57:41.060 most importantly, but also thank you for making the time today. I know you've talked about this a lot

01:57:44.920 and I'm sure you didn't necessarily feel like talking about the book. Oh,

01:57:48.020 and some of these stories over and over again, but I know that people listening to this are

01:57:51.240 going to appreciate it. Well, thanks, Peter. I, I, we talked about things I actually haven't really

01:57:56.000 gotten into in the past, but I also, you know, really enjoyed great intellectual thinking with

01:58:03.140 you. It's fun. And I hope we'll have a chance to get together a lot more in the years ahead.

01:58:07.100 Oh, we certainly will. We're, we're almost neighbors. So it has to happen. Thanks, Eric.

01:58:10.940 Thank you.

01:58:12.680 Thank you for listening to this week's episode of The Drive. If you're interested in diving deeper

01:58:16.820 into any topics we discuss, we've created a membership program that allows us to bring you

01:58:21.220 more in-depth, exclusive content without relying on paid ads. It's our goal to ensure members get

01:58:26.720 back much more than the price of the subscription. Now to that end, membership benefits include a

01:58:32.060 bunch of things. One, totally kick-ass comprehensive podcast show notes that detail every topic, paper,

01:58:38.060 person, thing we discuss on each episode. The word on the street is nobody's show notes rival these.

01:58:43.360 Monthly AMA episodes or ask me anything episodes, hearing these episodes completely access to our

01:58:50.000 private podcast feed that allows you to hear everything without having to listen to spiels

01:58:54.660 like this. The qualities, which are a super short podcast, typically less than five minutes that we

01:59:00.260 release every Tuesday through Friday, highlighting the best questions, topics, and tactics discussed

01:59:04.860 on previous episodes of The Drive. This is a great way to catch up on previous episodes without

01:59:10.180 having to go back and necessarily listen to everyone. Steep discounts on products that I

01:59:15.500 believe in, but for which I'm not getting paid to endorse, and a whole bunch of other benefits that

01:59:20.080 we continue to trickle in as time goes on. If you want to learn more and access these member-only

01:59:24.680 benefits, you can head over to peteratiamd.com forward slash subscribe. You can find me on Twitter,

01:59:31.260 Instagram, and Facebook, all with the ID, peteratiamd. You can also leave us a review on

01:59:37.200 Apple Podcasts or whatever podcast player you listen on. This podcast is for general informational

01:59:43.140 purposes only and does not constitute the practice of medicine, nursing, or other professional

01:59:47.580 healthcare services, including the giving of medical advice. No doctor-patient relationship is

01:59:53.720 formed. The use of this information and the materials linked to this podcast is at the user's own risk.

01:59:59.640 The content on this podcast is not intended to be a substitute for professional medical advice,

02:00:05.120 diagnosis, or treatment. Users should not disregard or delay in obtaining medical advice

02:00:11.460 from any medical condition they have, and they should seek the assistance of their healthcare

02:00:16.060 professionals for any such conditions. Finally, I take conflicts of interest very seriously. For all of my

02:00:22.940 disclosures and the companies I invest in or advise, please visit peteratiamd.com forward

02:00:29.620 slash about where I keep an up-to-date and active list of such companies.

02:00:59.620 Thank you.

02:01:01.160 Thank you.

The Peter Attia Drive - February 03, 2020

#91 – Eric Topol, M.D.： Can AI empower physicians and revolutionize patient care？

Episode Stats

Length

Words per Minute

Word Count

Sentence Count

Misogynist Sentences

Hate Speech Sentences

Summary

Transcript