In this fifth installment of the Fortress and Frontier series on Discourse Magazine Podcast, Robert Graboyes, a senior research fellow at the Mercatus Center, speaks with Dr. Eric Topol about the slow progress of medicine, how machine learning will improve healthcare, the importance of the doctor-patient relationship, the triumph of the mRNA vaccines and much more. Topol is a cardiologist, scientist and author of several books, including “The Patient Will See You Now” and “Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again.” He is the founder and director of the Scripps Research Translational Institute, a professor of molecular medicine at the Scripps Research Institute and a senior consultant at the Division of Cardiovascular Diseases at the Scripps Clinic.
Previous installments of the Fortress and Frontier series include two conversations between Robert Graboyes and Temple Grandin. The first can be found here, and the second can be found here. The third installment is a conversation with Pradheep Shanker on COVID-19 data. The fourth is a discussion with Devi Shetty on India’s Narayana hospital system and innovations in healthcare.
ROBERT GRABOYES: Welcome to all our listeners, and a special welcome to our guest, Dr. Eric Topol, world-renowned cardiologist, scholar, educator and perhaps America’s preeminent healthcare visionary.
I’ve subtitled today’s conversation “Healthcare’s Reluctant Revolution,” because at least by my perception, one of the central themes of Eric’s writing is that world-changing improvements in healthcare are right in front of us, ripe for the picking, but our society, laypeople and healthcare professionals are moseying toward those innovations at a painfully slow pace. Before I tell you about our guest’s background, let me just say, Eric, it’s wonderful to have you with us today.
ERIC TOPOL: Bob, thanks so much. I really appreciate the chance to have this conversation with you.
GRABOYES: Delighted. I can’t do justice to our guest’s accomplishments because there simply isn’t enough time to do so. So here are just a few items plucked from his biography. Eric received his medical degree from the University of Rochester in 1979. He’s been a tenured professor in internal medicine at the University of Michigan; chief academic officer and chairman of the Department of Cardiovascular Medicine at the Cleveland Clinic; founder and provost for the Cleveland Clinic Lerner College of Medicine; professor of genetics at Case Western Reserve University; and professor of genomics, molecular and experimental medicine at Scripps Research. The list goes on and on and on.
Doctors, researchers, journalists, public officials and friends often ask me to recommend books that explore the problems of healthcare and which peer into the remarkable future barreling our way as we head toward the middle of this century. As of today, my top five books include three written by Eric Topol. In our discussion today, I’ll ask him about all three books.
And in passing, I’ll note that later this year I’ll be interviewing the author of at least one of the other two books, and I hope two of them. [Editor’s note: Jason Hwang, co-author of “The Innovator’s Prescription: A Disruptive Solution for Health Care,” and David Goldhill, author of “Catastrophic Care: Why Everything We Think We Know about Health Care Is Wrong,” are now scheduled to record podcasts in this series.]
All three of Eric’s books provide a symphony of astonishing possibilities with a somber countermelody of disappointment with the slow pace of progress and stubbornness on the part of healthcare professionals.
Lastly, before we start, I’m proud to mention that in 2017, I was privileged to co-author an essay with Eric entitled “Anatomy and Atrophy of Medical Paternalism,” which is available on the Mercatus Center’s website. Again, Eric, thanks for joining us today.
TOPOL: Yes, I’m looking forward to your questions. I appreciate your comments on the books. That means a lot to me, really.
Dr. Topol’s Proudest Accomplishments
GRABOYES: Oh, they’re fantastic. I recommended them when I was interviewed on a podcast yesterday, and I rattled them all off.
So, I’ve given some routine biographical information on your education and affiliation. I went through your CV and bios online, and I simply couldn’t pick out what was most interesting. There’s an awful lot of things. So, why don’t you share an item or two from your background that will help the audience understand what a fascinating career you’ve had—accomplishments that make you the proudest?
TOPOL: Well, that’s interesting. I think starting a new medical school, the first one in 26 years in the United States, was certainly something I’m especially proud of. The look to the future and the disappointment of how long it takes traces back, for me, to college. I actually was at the University of Virginia, and I wrote—my thesis was called “Prospects for Genetic Therapy in Man” in 1975, and it took 40 years for that to be actualized. I’ve often picked out things that might occur, and I’ve been way off when they would occur. That seems to be thematic, and for reasons that you’ve already mentioned.
GRABOYES: And I’ll just mention that to my regret, I didn’t know you at the University of Virginia, but we were there at precisely the same time.
TOPOL: Oh, wow. I didn’t realize that. Did you finish in ’75?
GRABOYES: I finished in ’76.
TOPOL: Oh, so that makes me . . . [chuckles]
GRABOYES: I’m sure we bumped into each other on the Lawn at some point.
TOPOL: Yes, exactly. I was privileged to live there for a year, and I think I peaked out back in early age.
GRABOYES: I actually ended up teaching there in the nursing school for, I think, 12 years in the early part of the century. So, it’s still connected.
TOPOL: It’s such a phenomenal institution, by the way, yeah.
What Does ‘Creative Destruction’ Mean for Healthcare?
GRABOYES: That it is, including the medical portion.
Next, I’m going to ask three questions, one on each of your three futurist books. After that, we’ll drill deeper into some of the topics of those three books. So first, the physicist Max Planck famously declared that science progresses one funeral at a time. [chuckles] That’s a good one. That is, scientists don’t change their minds in light of new evidence. They die and are replaced by younger scientists who were bathed in the newer ideas during their educations.
The economist Joseph Schumpeter popularized a similar notion in economics, that progress in economics occurs when old businesses fail and new ones take their place. In 2010, Eric borrowed this concept for his book “The Creative Destruction of Medicine: How the Digital Revolution Will Create Better Health Care.” Eric, please explain why you use the title and what it tells us about healthcare.
TOPOL: You really nailed it, Bob. The creative destruction, although many people in the medical community were upset with that term—obviously, it’s derived from Schumpeter, and that’s what we need still in medicine. Basically, it’s an analog entity. It hasn’t moved with the world as it should. It had a very disappointing start with digital with these electronic health records, which have been an abject failure based on billing and have essentially made doctors and clinicians, nurses, into data clerks and made for profound disenchantment.
There are so many other parts of the digital revolution that medicine is very reluctant to be part of, no less to adapt to the threats like we’re seeing with this ransomware and health systems being taken hostage throughout the United States. I think what we have not done is make—still, even though that’s several years ago that book was written, we have not made the adjustments that we should to capitalize on the digital revolution.
Doctors don’t have any respect for sensors, for patients to be able to generate their own data. There’s so much of what we could be doing that it’s stuck in the kind of paternalistic, ritualistic and this ossified, sclerotic type of medicine that we practice day to day.
GRABOYES: I’m going to give you a little small aside, utterly irrelevant but kind of interesting to me. I recently discovered that the term “creative destruction” straddled medicine and economics long before your book. Schumpeter popularized it, but he didn’t invent it. He borrowed the term from another economist, Werner Sombart, whose daughter married Hans Gerhard Creutzfeldt, who discovered Creutzfeldt-Jakob disease. Just a little interesting aside there.
TOPOL: Was it actually applied to medicine by that person?
GRABOYES: I don’t think so. I just know he was an economist with a son-in-law who discovered the disease.
TOPOL: That’s fascinating. Thank you.
Centering Healthcare Around Patients
GRABOYES: Eric’s next book was “The Patient Will See You Now: The Future of Medicine Is in Your Hands,” a play on the phrase “The doctor will see you now,” which all of us have heard many times after sitting for hours and hours and hours in a waiting room. The book tells how patients can and will reverse who controls their encounters with physicians. Eric, please elaborate, and for reasons I’ll make obvious afterward, I’ll encourage you to mention the AliveCor Kardia device.
TOPOL: I will. You know, I think this is where we had the great opportunity to converge our thoughts in the essay that’s on the Mercatus Center website that you mentioned. It’s sad, but there is a tremendous amount of paternalism in medicine. That is, it goes back more than two millennia, whereby the doctor was in control and knew everything. The patients really had little regard for what they could learn or know, and we’re still stuck in that mode today, largely.
So for example, two-thirds of physicians in the United States are unwilling to give copies of their office notes to patients, even though it’s the work product that patients pay for and it’s their body and it’s something that should be an entitlement. We don’t give patients their data in general; they have to beg and grovel for it. We’ve even had healthcare regulations from the U.S. government to try to prevent this information blocking, and still it is occurring, and you have to pay to get data via fax machines. What’s that? [chuckles]
We have this idea that doctors are in control, that they are in a higher tier of knowledge. Sure, they have training and knowledge, but there’s not enough respect for the fact that the patients are the ones with the symptoms, the conditions, the health issues. It’s their bodies, and they deserve to own their data.
They deserve to be on a level playing field because this is where—again, you insightfully mentioned how the younger physicians get this. They understand that the world is different, and we’re no longer in the era of Hippocrates or his predecessors. This goes to, in many respects, as the practice of medicine today, that we have to change gears and get rid of, eradicate, this paternalistic problem.
How Technology Is Improving Healthcare
GRABOYES: Well, I’ll mention, in your book, you mentioned—right at the start of the book, if I remember correctly—an amazing device called the AliveCor Kardia. This is, for listeners, there’s a condition called atrial fibrillation; it is an arrhythmia of the heart, and it can lead to strokes. It can be kind of a silent killer and throw off clots that go to the brain. It can incapacitate you; the strokes can lead to death; they can lead to permanent disability. And sometimes you’re not actually having a case of AFib, and you run to the emergency room and get checked out, which immediately adds a couple of thousand dollars to the American healthcare system.
So, shortly after I read Eric’s book, I was just enthralled by this idea of the AliveCor Kardia. I was speaking in Arizona to a meeting of 250 health insurance executives at the highest levels, and I threw up a slide of the Kardia, which is this $99 cell-phone-based device that allows anybody to do their own electrocardiogram to find out if they’re experiencing this arrhythmia.
And I threw it up on the screen and asked 250 executives, “Does anyone know what this is?” I don’t believe there was a single person in the room who had any idea what it was. I said, “This is a little device that costs $99, and you could give it to every patient on your rolls with AFib. And every time one of them says, ‘Nope, I’m not having it’ and doesn’t go to the ER, your insurance company has saved, I don’t know, $5,000. Once in a while, one of your enrollees is going to say, ‘I am having it,’ and you might avoid having a $500,000 case that you are paying for over the next few years. So I would ask all of you executives, ‘Why haven’t you heard of this thing?’”
It was interesting because the person who had invited me, she almost got tears because one of her very young friends, a young mother of about 35 years old I think, did have AFib, didn’t realize she was experiencing it, stroked out and died. Anyway, I gave my talk, flew home from Arizona to Virginia, and the next morning I woke up feeling something I had never felt in my life, a thumping in my chest.
My wife rushed me to the emergency room, and I was lying on the gurney, and the doctor came up and looked and said, “Sir, you are experiencing atrial fibrillation. Do you know what that is?” I said, “Yes, I gave a lecture on it yesterday.”
GRABOYES: So we skipped some of the early things. The fact is, I can tell you of at least three episodes where—I’ve never had another episode; it’s the only one, thankfully. It’s five years out. But I’ve had—in the early days, especially, I was scared to death of this. So I kept thinking, “Maybe I’m having it again.” And I would pull out my—I immediately bought the device the same day.
And a couple of days later, I was out in a remote area on the way to give another talk. I thought I was having it, pulled the device out, found out I wasn’t. It relaxed me. I drove home six hours, and I have had at least two other episodes where the same thing happened. I cannot say what peace of mind this thing brought to me. Thanks to your book, I knew about it. Actually, I told a friend about it, and it, in a roundabout way, saved his life.
For a couple of years, I would give talks on this, and I would show it to cardiologists, and for a while, not one of them had ever heard of it or seen one. Some of them had heard of it, but they had never seen it. Now, of course, it’s on everybody’s Apple Watch—the same sort of device. For me, it’s certainly the most vivid example out there. There are loads of others that diagnose diseases, things that only a doctor used to be able to do.
Let’s move on to the third book.
TOPOL: Well, maybe I just say a word on that.
GRABOYES: Sure, please.
TOPOL: Because you had asked me to comment, and I got into the bigger picture, but I think it is a great example, Bob. I had not heard the story about the lecture of the insurance executives; that’s fascinating. But this is basically emblematic of what the opportunities are, giving people the ability to make a diagnosis. That is just a fingertip sensor that you—not only that has changed, where you can actually get six leads of the cardiogram.
So I use it in every patient I see as a cardiologist because I find all sorts of things I would miss just by taking the pulse. Now it’s the fingertips and putting it on the left leg, and you get all these leads. In fact, we’re working on a project to see what we can get rid of 12 leads with all the wires, and where they wheel in and do these fancy tests, where we probably can get almost all, if not all, that information through fingertips and having this on the left leg.
As you point out, this is now—it was adapted to the smartwatch so you don’t have to have the fingertip sensor. In so many respects it uses AI to interpret the data so anybody could get a preliminary sense of whether they have the arrhythmia or not. And you’re bringing out the most important part, which is the reassurance it provides for people who have atrial fibrillation.
Most people think, “Oh, it’s going to diagnose this condition and help prevent a stroke.” Yes, in the right people with risks, that’s true. But the thing that I have found that you’ve so duly emphasized is how many people will have fluttering and concerns, their heart is racing, and then this will tell you, “You’re in normal rhythm.” So that’s so great to preempt emergency room visits. Who wants to go to the emergency room as a patient, even putting aside the cost? So, thanks for bringing that up.
GRABOYES: Been there, done that. Actually, after this happened, one of my former students, who is herself a distinguished professor of health administration, came to my office just to pay a visit. I showed her this device, and her eyes just went like saucers; she was astounded.
Her husband is a doctor, and he was driving—the two of them were driving home from his med school reunion, and he was driving and said, “I’m having atrial fibrillation right now. I’m a doctor, I know what it is, and I’ve got it.” He went to straight to his doctor when he got home, and of course, the EKG was perfectly normal. He said, “Well, no hint of it here.” He had it again, ran to the hospital, got checked, “Nope, you’re all fine. It’s not.” He said, “Look, I’m a doctor, I know what this feels like. I need to get treated. I probably need an ablation.”
They said, “Well, you’re going to have to show us that you’ve got it before we’re going to start messing around in your heart.” She saw this device and she was staggered. She bought it for him that day. He got it that week, and the next time he had an episode, bang, there it was on the thing. If people haven’t seen this device, it also allows you to turn the EKG into a PDF.
He shot it off to his cardiologist and said, “There,” and he said, “Okay, you’ve got it. Now we can work on it.” The very odd twist to the story was the doctor decided he needed an ablation, and during the workup, during the blood work, they said, “You’ve got bladder cancer.” And it was just at the point where if it had gone any farther, it probably would have taken him out, but they were able to treat him and save him. I actually called up Dave Albert, who invented the AliveCor, and said, “Well, I just want you to know that your device also saves people from bladder cancer.”
TOPOL: Wow, that’s quite a story.
GRABOYES: Yes, it really was.
TOPOL: Hey, that emphasizes the point that people don’t have just one condition. Their holistic story is very important, and that’s a vivid example.
Artificial Intelligence in Healthcare
GRABOYES: Yes. Feel free during this talk—your books are so chock full of fascinating examples. Maybe in the next one, we get to—about your own example. Eric’s next book was the third in what is now a trilogy, and I hope will not stop at trilogy, is “Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again.”
I just read this recently, and I would say the progression of the books is fascinating. In some sense, I found this book almost mystical at times. It’s about the developing symbiosis between human intelligence and machine intelligence. I think I searched to see if you use the word “singularity” there; I don’t think it was in there.
TOPOL: No. [chuckles]
GRABOYES: But this idea of the machine brain and the human brain almost melding to the point that you can’t tell one from another, the so-called Turing test. I think you may have mentioned Turing in there. But this book is a roller-coaster ride into topics that have been fascinating to me for a while: neural networks, autonomous cars, natural language processing and mortal fears that artificial intelligence will dehumanize medicine, diminish doctors and take away their jobs.
Now, I will tell you that the message of the book is very different from those fears. So I’d love to go on that, let you riff for a while on the promise and the perils of artificial intelligence and machine learning, the promises they pose and also the risks they pose.
TOPOL: You’ve done a nice summary already, and that’s a great foundation. I think the first thing to note is, when I was starting to delve into where we’re headed with artificial intelligence, I came across the sense that this could and likely will be the biggest thing ever to happen in medicine. Because we are accumulating more data, generating more data than ever before, which exceeds the human capacity—even though some doctors think that they are superhuman—that’s not going to be capable of keeping up with the data that is aggregated for any given individual.
This big-data world where the data weren’t analyzable because we didn’t have the methods—now we have the lean-on-machines era that’s just beginning in many respects. The main sweet spot here is that images like a cardiogram, but more typically an X-ray, a CAT scan, an MRI, these sorts of images, or a PATH slide or a skin lesion—things that are a pattern can be read by machines when they are trained by hundreds of thousands, if not millions, of images with extraordinary accuracy, more accuracy than human eyes—and, interestingly, complementary. That is, the additive of some of the parts here is striking.
It’s exciting because there’s lots of medical errors, like 12 million serious medical errors a year in the United States. And if we can reduce those errors, that would be great, wouldn’t it? I think we can. A lot of this is still early, but to me that was interesting, and I think it’s still a work in progress. But what was the most exciting thing which led me to, really, in the culmination of the book, the last chapter, “What Is Deep Medicine?” is the return of the human connection, the trust, the presence, the precious patient-doctor relationship.
If we go back to the ’70s, when you and I were back at the University of Virginia, and then we go to when I finished med school at the end of that decade, medicine was very different then. That was when the patient-doctor relationship, it was precious. There was a lot of time to visit with patients; there wasn’t a sense of being rushed. There weren’t things like electronic health records or relative value units or big business of medicine. It was still preserved as central in our lives, and our relationship with our doctors was extraordinary, for the most part.
Now, what happened over these decades, unfortunately, has been a profound erosion of that relationship. And the biggest factor—there are many, but the biggest factor is there’s no time to be with patients. The average appointment is less than seven minutes, and the typical situation is there’s not even face-to-face communication because doctors pecking on a keyboard, looking at a screen with their back to the patient, that doesn’t work. So there’s little examination, there’s little presence, there’s little trust, and what we have is just a terrible degradation.
AI could actually bring that back, that is, this gift of time. It’s very exciting. We have to go after it; it’s not going to happen by accident. In fact, since the administrators are the overlords of medicine in the United States and many other countries, they could say, “Oh, well, now you’ve got the machines. We want you to see more patients, read more scans, read more slides.”
So this is going to take an active effort to achieve this humanistic benefit that we could derive. It isn’t just about improving accuracy, but it’s also about restoring a back to the future of the patient-doctor relationship. And I think if we go after this, it’s attainable, and that’s what gets me excited.
Importance of the Doctor-Patient Relationship
TOPOL: You alluded to it: I opened the book with my experience of having a knee replacement, and this was five years ago almost to the day. I had a horrible experience, which still in many ways plagues me today. The worst part about it was the fact that my orthopedist, who I used to refer many patients to for knee replacement, but when I was a patient, I learned that there wasn’t any care. When I would go into clinic for follow-up, I’d be in a lineup of 20 people all being scheduled like in an hour, and there was no examination.
And the time when my wife came with me because my knee was purple and severely swollen, and I was in unmitigated pain, unable to sleep. And I told him my story, and he said, “Well, everything really was good in the operating room,” like it was my fault. And there was no empathy, zero. What he said was, “Eric, I think you need to see your internist, and you should get some prescriptions for antidepressants.” That’s what he told me.
And my wife and I looked at each other saying, “Oh my God, this is incredible.” If there were ever underscoring of how bad our patient-doctor relationship is, my own experience reinforced that.
GRABOYES: The lesson I took away from that is, if you’re going to be a doctor, don’t irritate a patient who’s a best-selling medical author. [chuckles] I’ll have to dive into one more of my stories since we have this mutual connection, and I think you will know the person involved and appreciate it.
So when I was at UVA, some years before that I had had a positive TB test. At that time, the standard was, go get a chest X-ray every year just to make sure it’s nothing. And this was probably 10 years after the test, and I was still getting the X-rays. I went back to where I was living, and the phone rang, and it was the receptionist over at Student Health. She said, “Dr. Camp would like to see you.” I’m guessing you remember Dr. Camp, who I think the world of.
I said, “What’s up,” and she kind of nervously giggled and said, “Oh, it’s probably nothing, he just wants to see you.” I said, “Fine.” I mosey on over to see Dr. Camp, and I said, “What’s up?” And he said, “It’s probably nothing.” I said, “Okay.” He sent me down the hall and I got—literally there was like five people who gave me the precise same nervous giggle and said it’s probably nothing.
Finally, a very young doctor—I’m guessing he was a resident or an intern—was looking at my X-ray, and I said, “Would you please tell me what this is? I’m tired of that response.” He said, “You really want to know?” I said, “Yes, I want to know.” He puts the film up, shows me a discoloration on a couple of vertebrae in my neck. He said, “I’m guessing this is just sometimes at the edge of film it doesn’t develop correctly, but it could be something else.”
And I looked at it, and I had no medical background, but I knew enough to say, “It’s discolored. I’m guessing that’s a different bone density, and you’re checking to see if I have bone cancer.” He said, “That’s exactly what we’re looking at.” I said, “Good, I feel better now. Now I’m worried about one thing instead of 600 things.” I just wished they had told me that at the beginning, and of course, it was just the film.
I have carried that lesson and been adamant with doctors for decades since saying, “Just tell me what the problem is. It’s my body; you don’t have to protect me.”
Excessive Testing and False Positives
TOPOL: That’s a great story. Also, in the AI era, the chance of a false positive would be markedly reduced because basically the neural networks are trained to not give a positive output. We talk a lot about this term, and I know you’ve heard it a lot, “precision medicine.” Precision medicine means if you make the same mistake reliably, consistently, that’s precision medicine. [chuckles]
That’s not what we want; we want accuracy of medicine. [chuckles] Precision isn’t going to get us that far. We don’t hear that term, accuracy of medicine, and there you had an inaccurate reading of your X-ray. So we’ve got to get rid of those because look how many women are plagued with their mammograms that they get, and the false positive rate is over 60%. Look at men who have false positive PSAs and wind up getting all sorts of procedures for their prostate unnecessarily. We can do so much better. This could affect not just false positives and false negatives, but also the cost of healthcare because so much is overcooked and overdone.
GRABOYES: Yes. You talked to the—I think it was in “Deep Medicine,” about the corporations that buy these expensive batteries of tests for their top executives and turn them into worried basket cases. You get filled with false positives and checking everything else out there and probably spending a fortune on unnecessary items.
TOPOL: Exactly. I mean, they have these executive physicals where they do every possible test from head to toe, often over $10,000, $20,000. The CEOs of a lot of companies go and get these. They’re basically a recipe for an incidentaloma to find things and go down rabbit holes. I can’t tell you how many people I know as patients and as friends who have been injured from these tests, this battery of unnecessary tests.
As you well know, it’s this whole Bayes’ theorem that you shouldn’t do a test unless there’s a high pretest probability. For example, if you use the AliveCor device and you’re 25 years old and you have no symptoms, you’re going to get a false result just by—statistically because you have no risk. Whereas the CEOs and affluent people that get these executive physicals, they’re just looking for trouble unwittingly. They think they’re doing things to prevent, but no, they’re actually doing things to potentially induce trouble.
GRABOYES: I taught at various medical centers for about 19 years. I’ll probably do it again soon. I used to walk the students, who were doctors and nurses mid-career usually, and I would walk them through the Bayes’ theorem stuff.
And there was a case—unfortunately I can’t find it online, and I regret it because I’d like to use it in some writing, but it was in the very early days of the AIDS epidemic. Some poor fellow went in, got an AIDS test, and the doctor said, “It’s positive.” He said, “Okay, how accurate is the test?” The doctor said, “99% accurate.” The patient apparently went out and killed himself. In the autopsy, they did subsequent tests and found, no, he didn’t have it. Given the penetration of the disease in the population at that time, a 99% accurate test showing positive meant he actually had a 9% chance of having the illness.
That’s a very subtle piece of statistical mathematics, and it used to confound my students to pieces. I will tell you, it scared the dickens out of some of the doctors in my class when I taught that. I said, “The problem is that humans weren’t designed to speak in percentages. We don’t do it intuitively, and you can draw some terrible, terrible conclusions if you don’t understand it.”
Mysterious Mental Processes
GRABOYES: I think one of the fascinating things in your book, and I’d love you to talk about it a bit, is—and I think this gets into when I said it was almost a mystical experience reading it—that when you talk about how neural networks work, how machine learning works, the key to it is we don’t know, you can’t know, any more than I can know how my mind works, how my wife’s mind works, how we actually process information.
I’m, on the side, a musician, and I play piano, and my fingers will just fall on the right places on the piano. People ask, “How do you do that?” I say, “I have not the faintest idea how I do it; it just does it.” Similarly, the way you were describing the way machine learning works is quite similar. The reason machine learning is so much more interesting than carefully programmed algorithms is this mystery of we don’t know what the machine is doing or how it’s doing it. Could you talk about that a bit?
TOPOL: Sure. Mystical is a good term here, Bob, that I think applies. One of the best examples I can think of for this phenomenon is the retinal photo. When you take a picture of the retina when you have an eye exam, and if you take that picture and you show it to the international leading retina experts of the world, and you say, “Is this picture from a man or a woman?” the chance of them getting that right is 50%—50%. But if you put hundreds of thousands of these retinal photos through a deep neural network, the accuracy of the gender is 97%.
Now, we have a better way to determine gender than the retinal photo, [chuckles] but the accuracy is stunning. We still don’t know—and it’s been replicated in different studies—we still don’t know why we can get to 97% with a machine, whereas the human eyes, expert human eyes, have no clue about what is the difference between a male retina and a female retina.
Now, what’s fascinating is this effort that’s starting to happen of demystification: that is, using the neural network and going backwards to find out what features in these hidden neuronal layers, these artificial neurons, in these various layers going backwards, what were the features that it picked up that human eyes don’t see. And so there are attempts now to basically deconstruct the networks.
It’s early. We don’t have great examples to say, “Oh, we cracked the case,” because still today, we don’t know how the neural networks are so great at interpreting retinal photos. But I think that is a work in progress as well, which is a lot of people want explainability, which is a pretty high order because a lot of things—as you mentioned, how you play the piano—a lot of things in medicine and the treatments we give, we have no clue how they work. We know they work.
To say machines have to do better, we have to know exactly how it’s making that diagnosis or that interpretation. That’s a pretty high threshold. This explainability feature is an issue in medicine. Are we going to demand that? Are we going to settle for that it’s validated extensively? There’s a lot of debate about this concept, but I think over time it will be less controversial because we’ll demystify the issue.
GRABOYES: One of those fascinating cases that I’ve ever heard of, I wrote a couple of little pieces on it some years ago. The Cerner Corporation came out with an AI software system called St. John Sepsis Agent, and it is a sniffer. You wire it up to a patient in a hospital, and it will identify the presence of sepsis, a MRSA infection or something like that—if I remember correctly, something like five or six hours before any human being will ever detect that something is awry. Of course, there’s very little in a hospital that’s more dangerous than a superbug getting loose in the place. Somehow this software system would look at the patient’s vitals and I guess some atmospheric elements and pick up the fact that, yes, this person is in the early stages of sepsis.
Well, one of the odd things they discovered in the course of it was, for reasons—as far as I knew, they had no idea why it was able to do this—the same software system whose purpose was defined a biological organism floating around in the air, also was an accurate predictor of the onset of PTSD and suicidal tendencies. No idea why something that’s out there to sniff out a bug is going to be able to look into the deep recesses of the brain and understand your behavior. I love that story just because as far as I know, there just is no answer to why it could work.
Imagination and Machine Learning
TOPOL: Well, you’re bringing up what I think is a particularly enthralling aspect of AI in medicine, and that is that we need to be imaginative because we can train machines to do things we didn’t know we’re capable. Going back to the retina example: Recently, it was published that not only can the retina predict things like kidney disease, Alzheimer’s disease, glucose control for diabetes, blood pressure control—just this picture of the retina.
Also, interestingly, you’ve heard of the calcium score of the heart that you can get for risk of heart disease, which is a test that a lot of people get which I don’t recommend because it can be misleading. Nonetheless, you have to go through a CAT scan and get your calcium score. What’s interesting is you can get that from the retina too. You can get the calcium score of your heart through the eye, so basically, we have to be thinking much broader than ever before.
The Mayo Clinic recently tried to see if you could diagnose COVID through the electrocardiogram. That was a reach. I think the data are very questionable. But, nonetheless, I give them credit for at least asking the question. Machines, because they are looking at pixels and processing data that human eyes can’t—we just can’t do it—also, with multi-layered data, they can take the electronic health record and sensors and genome and microbiome and put all this data together. Humans just can’t do this stuff, right? So we’re going to learn a lot about these things, like you just mentioned, that there’s this unexpected relationships or unexpected yields that we previously wouldn’t even have conceived.
GRABOYES: At one point I took an EKG using my AliveCor, and after owning it for a while I know what a normal rhythm looks like. I don’t even have to wait for the thing to tell me what it is. However, it gave me a very normal-looking wave pattern, but it said, “Can’t determine whether you’re having AFib or not.” It’s kind of like, “Reply hazy, ask again later.”
Anyway, I just got on the phone and I called Dave Albert. I said, “Dave, I’m getting this reading, and I figured you might want to know this because I know that that’s a normal wave pattern, but it’s saying that it’s ambiguous.” He said, “Shoot me a PDF of it.” I shot it to him, and maybe two days later, he wrote back and said, “What you had was—” it was a premature something, something, something, and I forget what specifically.
He said, “The machine learning is good, but it’s not that good. So sometimes we have to tweak it because someone like you calls and says it’s not quite functioning right.” He said, “Thanks for your call. We’ve now added that in, and if it happens to you or anyone else again, it’s going to know what it was.”
I love that idea, again, of the symbiosis between human intelligence and machine intelligence. It is not going to make humans obsolete. I tend to think that AI machine learning is going to make us more important in—which I read as the central message of, certainly, your third book.
TOPOL: Yes. This is, I think, fundamental, and that is our performance as humans is kind of fixed. We’re not going to get a whole lot more intelligent over the years, over the next generations. But machines through these neural networks are autodidactic, so there really is, like the example you just gave—well, just think if you had all the cardiograms that the AliveCor were seeing, and you just kept feeding it into the network to get better outputs, and you get better, better, better so that accuracy was just 99.999 for everything. That’s where we’re headed.
The tricky part here is that a lot of these things go through a regulatory review at FDA, and what happens there is unfortunate because what they do is they freeze the algorithm. They say, “Well, we can’t let it be autodidactic. We just got to stop right here.” This is a problem, right? Because we are anticipating something so powerful that we can’t deal with it.
I understand both sides to this. The FDA doesn’t want to necessarily just give a blank check for an algorithm. On the other hand, we’re eliminating the improvements that can be made through these, so it’s fascinating. We’re embarking in an era the likes of which we have not seen before.
The COVID-19 Vaccine Triumph
GRABOYES: Let’s look at some really happy news. A year ago, I was absolutely convinced we would see no COVID vaccine for several years, and that when it finally came out, it would only be moderately effective, and you’d have a pretty high percentage of people who wouldn’t be helped by it.
I have never been so thrilled and ecstatic to have been proven wrong. I’d like to defy Max Planck’s dictum there. I like to learn from new data. For me, the rapid development, approval and deployment of mRNA vaccines is almost an unprecedented achievement, both at scientific level but also at the administrative level, the way the FDA reacted to it and then the way the supply chain reacted.
I’ve seen some of your tweets on the subject, and I gather that you’re of a similar mind, but I’d love to hear what you think of that. And how does this whole experience with the vaccines over the last year fit with the message of your books?
TOPOL: I’m glad you brought this up because I had been thinking about it, but I don’t know if I ever really made a comment on this point. That is, when we were hearing that there was going to be a vaccine within 18 months when the SARS-CoV-2 virus was sequenced and identified, “18 months,” I said, “Oh, come on. That’s impossible.” The average vaccine that’s been successful takes eight years from the time the pathogen is found to when you have an effective vaccine proven to be worthy. And as you know, Bob, many diseases, we have no vaccine, like malaria and tuberculosis and a long list, HIV.
TOPOL: Yes, exactly. When I heard this, I said, “This is impossible. They’re just trying to set us up here so that you get in the car for a long journey. Are we there yet? We’re not going to get there. We’re not going to get there.”
Then we did, and it is something that stands out to me as the biggest medical triumph because so much of humanity was facing—is facing—an existential crisis, and we’ve lost millions of people because of COVID. But think if we didn’t have these vaccines, how many millions more would be dying and suffering? Not just hospitalizations, but this long COVID.
Now, the silver lining of the pandemic is two dimensions here. One is that we can do this. We can do this. That is, we went from the sequence of the vaccine to a template Moderna and then, later, Pfizer vaccine within days, days. Then we were basically going into trials in humans within weeks and completing the largest trials ever conducted in history of vaccines, and a rapid review at FDA, which, fortunately, was completed. It wasn’t cut short because of political issues, so it was a success story that was just in many ways a miracle, fantastic outlier like nothing we’ve seen.
But it sets a new precedent in two ways: one, that we can do things quickly in the midst of a crisis in life science and medicine. But two, we’ve got a new platform mRNA in a nanoparticle. Just the other day on Saturday, this past Saturday, the 26th—I remember that day because it is my birthday. On the 26th of June of 2000 was the first human sequence of DNA, our human sequence. On June 26th, 2021, was another major event. It was the first time we gave an intravenous infusion of mRNA nanoparticle genome editing to fix the problem of hereditary amyloidosis, and just think where that’s headed.
We’re going to see mRNA nanoparticles, which is the platform for the vaccines, not just for other vaccines, but it’s being treated now with genome editing for diseases that were incurable. It has been used to knock out cholesterol in primates for life, one shot; for cancer, for neurodegenerative diseases.
This is really exciting stuff. When you think about it, there’s two of the most powerful tools we’ve ever seen in the history of medicine. We’ve talked a lot about AI, but the other is this CRISPR genome editing, and there’s actually a convergence of the two that you can make the editing better with AI. But this is fantastic stuff. It’s so exciting. It’ll take longer than the vaccines because the vaccines were just so urgent and some of these matters are not as urgent, but I hope we move faster, and I hope we learn from this pandemic.
GRABOYES: I think we’re learning a lot from the pandemic. I’ve been a health economist. I switched out of international things, monetary, into health about 25 years ago. I wrote, I think, sometime around a year ago that in the first three months of the pandemic, I had seen more innovation in healthcare than I had seen in the previous 25 years of my career in health economics.
First of all, obviously, the behavior of the FDA with these wonderful vaccines was already evident that things were moving quicker than in the past, but also things like the massive rapid acceptance of telemedicine, of remote monitoring. In Singapore, of chatbots explaining what to do to COVID patients who couldn’t be brought into public facilities. Or the release of nurse practitioners: “Go practice autonomously. We don’t have the luxury of stopping you right now.” Dropping of a variety of regulations, allowing doctors to practice telemedicine and in-person medicine, even across state lines. I would say it’s the most massive unplanned experiment in the history of healthcare.
GRABOYES: I think it worked brilliantly. I’ve had a few experiences with telehealth myself in this period, and what a godsend it was. I don’t know. Your thoughts on these?
The Future of Telehealth
TOPOL: Well, it was interesting, going back to the “Creative Destruction” book, I said we’re ready for telehealth, telemedicine. It’s important. It’s complementary to the classic way of medicine with a face-to-face physical visit, and it basically was going nowhere. Then, boom, we had physical distancing, and there was by necessity having to adopt telemedicine as part of the way to see patients and try to help them. Out of necessity, out of desperation, we turned to something that was available years prior, but we had no interest.
Unfortunately, Bob, here the reluctance to use telemedicine was many things that were controllable that—for example, the reimbursement issues, the fact that the United States had no telemedicine law that would allow people to practice across state lines even though we all have a national license; that’s the same exam that we take.
We got over the barriers quickly. I hope that—your point about remote monitoring—we can avoid hospitalizations to a very large degree. As you well know, as an economist, the number-one part of our near $4 trillion line item is hospitals. We could reduce their need substantially if we started to develop algorithms for remote monitoring of patients in their own home, in their own environment, which is so much more preferable. There’s much more we can do, but we have seen the light of some of the good things that medicine can change when there’s a sense of a real need and desperation.
GRABOYES: Just prior to the pandemic, I gave a couple of talks to really top-level med students around the country. And I talked about my views on telehealth. In particular, I don’t have time for it now, but there’s a story you can—anyone can Google up in my writing about how my mother’s life was saved because my nephew, who’s a doctor, was just talking to her on FaceTime and picked up on the fact that she was in early stages of MRSA and got her to the hospital very quickly and saved her.
The students I was talking to were astounded by that and some of the other stories I have about telehealth. And one of them—I think he was a fourth-year med student at a top-flight school—said, “You’re making me wonder, why is it that telehealth has never been mentioned the whole time I’ve been in med school? It just doesn’t come up in the conversation. None of these things that you’re talking about really come up. How does one do an examination through a laptop?” Any comments? You founded a med school. Are our med schools behind the curve on catching up with these technologies?
TOPOL: Yes. Well, one thing that’s interesting is, you know, we talk about genomics and about AI, and there’s about 150 med schools now. There’s been a lot of them that started in recent years, and they don’t have any AI or genomics essentially in their curriculum. But they have put telemedicine in because even prior to the pandemic, they recognized that that would be a complementary way to efficiently see patients for nonserious matters.
That is, a lot of things you can do with what’s a video chat today but what’s increasingly going to become integrated with data. For example, you could show your AliveCor and discuss it during a visit, or you have an algorithm that’s tracking your lung function and you have asthma, or your blood pressure single screen that has your last 50 readings on it over the last few weeks, and on and on.
For many things, we’re going to see the video chat of today, telemedicine 1.0, transformed to 2.0 where we use neural networks, algorithms, all sorts of ways to transmit, process data for oversight of physicians. And then, do we need to have a regular visit? This is so much analogous to people that want to work from home now because they experience, “Wow, isn’t this nice? I can work in my pajamas, and I don’t have to deal with [what] the commute is, and it’s so nice. I can just go to my refrigerator anytime and whatever.”
This is what medicine is figuring out, that “Hey, you know what? We can do this.” We can do this. A lot of stuff we relied on a face-to-face visit, putting patients in terrible inconvenience, having to go to not only get an appointment, which average takes three weeks for a primary care doctor. Then you have to go to the waiting room and wait an average of an hour behind schedule, and on and on.
Look at this. You have a telemedicine, and you’re seen at the time that the appointment’s set up, and you’re not dealing with all these middle people, and you’re actually seeing the doctor. You don’t have a mask on. All these things. It’s really interesting. It’s kind of the work-from-home attitude that’s taking shape in medicine.
Looking to the Future
GRABOYES: The biggest problem that you named was that the refrigerator is always 20 feet away from me. I just wrote a piece last week based on a colleague’s conjecture. He had said, “I’m wondering. Is this the first pandemic in human history that ended up with an increase in the human biomass rather than a decrease?” I think it’s a wonderful conjecture.
I know you need to get on. We’re just about at that break point, but just any last thoughts? I don’t know, any visions of what you hope to see 50 years from now? We probably won’t be there, but any thoughts on where this is all going? A good message to leave the audience with.
TOPOL: Well, Bob, first, let me just say I so enjoyed this discussion with you. You’re remarkably articulate and insightful, so it’s a real pleasure to have this. The hope I have, which will probably take longer than it should, is that we get back this patient-doctor relationship. That’s what I really would love to see. We have the tools to do that now. That’s so exciting. We have the gift of time that can take shape. If I had one major wish that will be the outgrowth of medicine’s next phase, it will be that.
I know some people think I’m ridiculously optimistic. I don’t think so. I think that we have to, as a medical profession, seize this opportunity for our patients. I think you know that there’s a global crisis now of burnout, not just among doctors, but also among nurses and all walks of people in health professions. In order to remedy that, in order to turn that around, we really have to have that connection, that human connection.
And so, as I close the book of “Deep Medicine,” this is a time for us to get deeper into our human connections because we aren’t going to outdo machines. We’ll work symbiotically with them, but machines will never get truly humanoid. We need to get more like that.
GRABOYES: That’s great. Sometime offline, I’ll send you a couple of emails. I have been thinking about that burnout problem and that loss of spirit among doctors and the pall of depression that has settled down on the profession. I’ve got a few odds-and-ends ideas I’d love to pass by you.
TOPOL: It’d be great.
GRABOYES: We’ll have more conversations. But again, I just want to thank you. This was a fantastic conversation.
TOPOL: Terrific. Great to be with you, Bob.
GRABOYES: Take care. Thanks again.
TOPOL: Thank you.