In this episode Byron speaks with author Eric Topol regarding how Artificial Intelligence could revolutionize medicine and the health care industry.
Eric Topol is the Founder and Director of the Scripps Research Translational Institute, Professor, Molecular Medicine, and Executive Vice-President of Scripps Research, As a researcher, he has published over 1200 peer-reviewed articles, with more than 230,000 citations, elected to the National Academy of Medicine, and is one of the top 10 most cited researchers in medicine. His principal scientific focus has been on the genomic and digital tools to individualize medicine.
In 2016, Topol was awarded a $207M grant from the NIH to lead a significant part of the Precision Medicine (All of Us) Initiative, a prospective research program is enrolling 1 million participants in the US. This is in addition to his role as principal investigator for a flagship $35M NIH grant to promote innovation in medicine. Prior to coming to Scripps in 2007, he led the Cleveland Clinic to become the #1 center for heart care and was the founder of a new medical school there. He has been voted as the #1 most Influential physician leader in the United States in a national poll conducted by Modern Healthcare. Besides editing several textbooks, he has published 2 bestseller books on the future of medicine: The Creative Destruction of Medicine and The Patient Will See You Now. His new book Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again came out in 2019. Topol was commissioned by the UK 2018-2019 to lead planning for the National Health Service’s integration of AI and new technologies.
Byron Reese: This is Voices in AI, brought to you by GigaOm, and I'm Byron Reese. Today my guest is Eric Topol. He is the author of the book Deep Medicine, and he talks about how the power of artificial intelligence can make medicine better for all humans by freeing physicians from the tasks that interfere with human connection. He holds a degree with highest distinction in the study of biomedicine from the University of Virginia, and he holds an MD, with honor, in the study of medicine from the University of Rochester. Welcome to the show, Eric.
Eric Topol: Thanks very much, Byron.
Tell me a little bit about your background and how did you, obviously with the medicine background, first get into AI and see its potential for transforming the medical industry?
Well, it's been about a dozen years ago when I started this Research Translational Institute, which was predicated on understanding human beings at a deep level. That was also involving, of course, digital, wearable sensors. Very quickly we saw that there was no shortage of data being generated for each person, whether it's through different sensors or a genome or electronic health records or images, and it became clear that we needed a rescue for dealing with all this data. Clearly, AI is emerging to fulfill that very objective.
What do you mean, you set out to ‘understand humans’? Is that psychology and sociology and physiology? Is it all of that? That's a pretty tall order. You have to look at history and anthropology...
Yeah, not quite as diverse as you're mapping but rather the medical essence of a person. That would be the biological layers like DNA, proteins, the microbiome, the physiologic through sensors layer, the anatomy through scans, and then the environment you can quantify now through sensors, as well as the traditional medical information. We're not talking about anthropology or psychology as much as we're talking about what makes a person tick.
If you go to 2000, 2003 when the genome was announced, the first human genome draft, their thought was the DNA is going to have all the operating instructions. I've never thought that to be the case and in fact, we need much more information about a person. The whole concept of individualized medicine [means] being able to match up that knowledge of a person with prevention or better management of conditions, or everything we do for screening and medications and making diagnoses, everything we do in medicine, by having a deeper understanding of each person.
Where are we on that journey? If you go back [from] Hippocrates to now—because I'm always struck by how much we don't know—you can start with the brain and how a thought is encoded and what gives rise to the mind. We used to think the neurons were the story, and then it's the glial cells and then it's something else. I read recently we don't even know how the body maintains its body temperature. How does it always keep us at 98.6? Where are we in terms of our understanding of what you're trying to[do] – are we still in the era of stone knives and bearskins?
[Laughter] No, we're not. We're making tremendous headway. I think it was a remarkable study done on Scott and Mark Kelly, the astronauts, where they compared Scott – these are identical twins – who was out in space at the International Space Station, and every one of these things we just discussed, every layer, was essentially defined: the deepest phenotyping, what we call it, of human beings in history and then the analysis of what was the hit of being in space for a year on Scott, and it was quite a bit of effect on genes, chromosomes, and on his cognition, a significant impairment. We can do this now. We haven't done it at scale.
We probably now have done genome sequences of a million or so people, but it's just starting to come together. To answer your question, Byron, we can do each of these. We can do an in-depth probe of a person's gut microbiome. We can understand things that we never could before. Integrating it all for each human being is another task that is going to require AI because no human being can assimilate all this data.
Yeah I always wonder, will these systems give us more understanding of how things work? Hear me out here because I think about the antidepressant Wellbutrin, which while it was being studied, some people remarked, “You know, I don't seem to crave smoking as much.” They're like, really? They do studies and they say “Wow, this is really good for smoking cessation. Let's call it Zyban and sell it.”
It's more like we get things out of the data that we don't necessarily understand, but is it necessarily important that we understand them? We just need to know that it works. We don't know necessarily how an aspirin stops pain but it's enough to know that it does, and it doesn't seem to have terrible side effects. Do you think these sorts of systems are giving us true understanding at a systems level of what a human being is, or are they giving us just a high degree of predictive ability?
Well, there isn't one simple answer. It depends on the particular focus. In some areas, we're making significant progress across the board; in others, we're still at a pretty rudimentary state.
The one thing people are always curious about, of course, is longevity. While the number of people that make it to 100 – the percent of people that make it to 100 goes up every year, the number of people who make it to 125 is stuck at zero forever, seemingly so far. Do you think the kinds of technology you're studying are going to let us – and I'm not even talking about “curing death” but just break past 125 or 150 for a few people?
It's possible. I mean, I'm somewhat skeptical about the ability for the science of aging to have a measurable impact on extending lifespan. I don't know if there are a lot of people who are optimistic that we'll be able to change that ceiling that you refer to (of 120) and increase the number of people who are centenarians and beyond. That's really being pursued, but it's speculative. We are understanding the aging process, that science, far better than ever before and there's lots of ideas that are being pursued. So far, I don't see anything that is really making any substantive difference.
Yeah because it always seems like if you ask the people that live a really long time, “Why did you live a long time?” they always have something like, I ate a stick of butter every night, or something that's completely counter-intuitive.
Yeah we've seen that. We've had people swear that it was the Twinkies that did it. We have a big elderly program of people who are 85 average, 90 but 85 and above who've never been sick, and we've had people in that cohort that smoked two packs of cigarettes a day still at age 99. There are some genetic underpinnings that allow people, without any drugs, environmental effects, and things that we don't understand yet, that give a Teflon coating for some people, not just for lifespan but I think most people would agree it's actually ‘health span,’ the number of years you can extend where a person is perfectly healthy without any significant chronic conditions. That's the real goal, not just to be able to say you lived to some ripe old age but you had many different serious conditions including impairment of your cognition.
Absolutely. There's a sentence that goes along with your book which presumably you've – if you didn't write, you've at least seen before, and it says, “AI will create space, the real healing that takes place between a doctor who can listen and a patient who can be heard.” That's a highly personal kind of statement and yet you're saying AI creates that space. Can you elaborate on that?
Well, first let me clear up: I wrote the book, so I don't know what you're alluding to.
No, I meant it's just a line that's describing the book.
Oh, okay, alright, yeah, I mean, that line is about the story that this is a highly, right now, compromised situation where doctors are not really present with patients because they're so busy typing on keyboards, not being able to get their arms around the data, and all sorts of faddle with data clerk and administrative tasks. We need to get back to where these two human beings, patient and doctor, come together where the trust is restored, the care, empathy, communication, all the aspects that are critical for good medicine, which have been lost along the way.
We have the realistic opportunity with AI to be able to get there because we can liberate from keyboards. We can have that data assimilated, keyed up for the doctor. More patients will generate their own data and have algorithmic support, so that'll decompress a doctor's role for many non-serious conditions, so there's many different routes to get to a far better plane of human connection than we have today.
I read one thing that said the average amount of time a patient gets to explain what's wrong with them before a doctor interrupts with a question is 15 seconds, which blows my mind because that would imply that half of them are less than that. That's the kind of thing you're hoping that we can have the space to take a breath and figure out what's going on. Is that right?
Yeah, well, that's part of the problem is we don't even listen to a patient's story, which often will give the diagnosis or give the color of what's really wrong, or what's working, or what isn't working. The interruptions occur quickly and frequently, and that's part of the compromised way that medicine is practiced today. That has to change.
We have to get back to the way it used to be when a patient had the ability... there was time to tell their story and there was time to be able to really cue in to what's really going on with a person. This is what's missing today in medicine, and we have right now the greatest chance perhaps ever, but certainly for the foreseeable future, of turning back time because four decades ago or longer, there was a precious relationship and lack of this problem of inadequate time, inadequate context, presence, and trust, where there was the real chance to listen and not interrupt. We’ve got to get back to that.
We're nowhere near that now, but do you envision a time when doctors aren't even present, like I can talk to my device for 15 minutes and it can ask me all kinds of things, like do your eyes water when you eat potato chips and all these other things, and figure it out? Will we eventually get to that point?
Well, I think it's a priority that we do far better. I don't know that we want to emphasize things like what you're asking, but getting the critical features, the critical information is paramount. It's not being done well today.
I just think of all the parts of the world where they don't even have the luxury of having a doctor present. It would be great if the smartphone were good enough that it could do a pretty good job without actually having a human physician present.
Right, okay, so for that, yes. Already today in many parts around the world, including remote and places that are non-developed countries, we see the emergence of chatbots and smartphone connects and all sorts of technology that's starting to – symptom checkers, companies like Babylon Health and Ada and many others. It's starting, what you're bringing up, Byron. It's short on validation but getting increased uptake. Theoretically, though, this is a big advantage of this, which is the ability to reach a lot of people in rural areas and under-developed areas that could really bridge the divide that exists today where there's tremendous inequities; we have that opportunity. I mean, AI could make that worse by only being available to people who are relatively affluent or could actually be a great force for leveling this. Obviously I'm hoping for the latter.
Well AIs are only as good as the data we feed them, and it would seem to me that in the medical world which, as you say, by nature, generates an enormous amount of data, we’re kind of ‘in our own way’ there. I go to a doctor today and they still have paper files, and if they don't, if I need to be sent somewhere else, there's a HIPAA form, and then I have to get my records transferred over, and then there's... I don't own my records, and they're not in some format that I can just hand a flash drive to somebody and then they know everything. I suspect I know why we're not there, but I would love to know your thoughts on that. Isn't that a critical part of having AI be able to do all the things we want it to be able to do, to have the data normalized and available across large groups of people?
Oh, yeah, absolutely. I mean, I think we don't even know what normal blood pressure is now. I mean, we never capture data in people's real world. It was always in a contrived environment like a doctor's office, and it was a very one-off value rather than a continuous high-frequency captured data. That's a valuable metric that we have little insight of what is normal for a person. That's just an exemplar of the problem that you're alluding to.
Is it your hope that devices like the Apple Watch that can detect your pulse – is your hope that through these passive data collection wearables that we're going to start collecting that data on a systematic basis and use it? Is that something that’s going to happen?
No, I think that's really where we're headed. I mean, for the first time, we'll have real-world data at scale, which is going to be easy to collect, and it'll give us a whole new insight about what is normal. Right now, hypertension, since I use that example, is the number one chronic disease of man. Just in the US, we have over 75 million people with high blood pressure, the majority of whom are not getting good management of that blood pressure and all sorts of medications are being thrown at them. This will give us an opportunity to get a rebooting of that number one chronic condition, and it's just a matter of getting people to have a seamless, high-frequency collection of their measurements.
That will be the front-runner for many others. You mentioned heart rhythm. We're going to get a much better handle on abnormalities of heart rhythm in the general population. Not only are we going to learn at the individual level pulling a lot of data together to help manage a condition, conditions, or prevent them, but you're getting at the other big goal, which is to understand human beings at a much broader level. The opportunity to get from sensors as well as the ability for the analytics of that data give us whole new ways to get at things we didn't understand previously.
Yeah, I don't think it's going to be that long before you'll have a spoon that'll – every bite of food you take, it'll know what the caloric content [is] and what vitamins and nutrients are in it, a pan that detects botulism. All that kind of stuff I think – is much closer than we might guess. The third leading cause of death in the United States is – do you know? Medical error, and so do you think that AI is a path out of that? That's an embarrassing statistic, isn't it?
The error in medicine today is terribly embarrassing: 12 million serious errors a year, the fact that scans, medical scans which are used so importantly in assessing people, 30% false-negatives; that is, missed things on scans. I mean, we can go through every aspect of medical practice today and it's an error-laden problem.
The hope is that that's really the sweet spot of AI. It has the ability through deep learning, training machines to not miss things on scans, to do a far better job of processing a person's data to get to a more accurate diagnosis quickly. There's a lot of promise. I mean, a lot of this is still the promise without the proof points, but at least we have a set-up now which we didn't have before of a way to get out of this high-frequency of errors, which are far more than what people are generally aware of.
In the US, we spend something like 18% of GDP on our medical care. Most other Western nations maybe spend half that much. Paint me a picture on how AI can reduce the amount we're spending on medical care without, of course, compromising quality.
Well, the opportunities are almost limitless. In the longer term, we can get rid of hospitals, hospital rooms that is, not the – there will still be a place for emergency rooms, operating rooms, and intensive care units, but the rest of the hospital rooms could be in a patient's home. Last week, a device that captures vital signs continuously was approved by the FDA with AI diagnostics to predict a person's potential complications. That's already approved for the home today in 2019, and hospitals are one-third, almost, of the healthcare US budget a year, well over $1 trillion. If we can gut hospitals of the hospital rooms, we can get people in their home. Less staffing is required. Data plans can be given to patients for their cost of one day in a hospital. That's one far-reaching opportunity that would be enormous once we get there, and we will get there.
More near-term, the fact [is] that productivity, efficiency, workflow can be greatly ameliorated. Examples there are we've already seen that you can read scans much more quickly and accurately before the radiologist takes a look at them, or pathology slides, or for dermatologists, skin lesions, or for gastroenterologists to diagnose colon polyps that otherwise would be missed, using machine vision. So many things are being done today that have patterns that are images or image-like can be processed. That reduces the cost of professional services.
The other part, of course, is the back office operations: the people that do chart abstracts, coding, billing, and all these activities that could be reduced because a lot of this could be done through algorithms. There's lots of different ways you get at this where the common thread is by having machines trained to do things that either humans can do but don't do very well, or can't even see or do, which is what we've seen in some examples, this is a tremendous potential for economic savings.
There are limits, it seems, to what we would want AIs to do. Nobody wants to get a postcard in the mail [saying], ‘The AI has detected you have cancer. Please report at 8 a.m. Monday morning for surgery.’ Nobody wants that. It seems like that aspect of delivering news and helping people cope with it is a very human thing. How good are doctors now at that and will that become a bigger part of the training as more of the diagnostic work is maybe removed from the job?
Well, I get a lot into that in Deep Medicine, in the book, the point being that we're not really selecting doctors now the way things are changing, where still today, we focus on brainiacs with the highest grade point averages in college and the highest scores on the medical college admission tests, whereas we should be looking for those with the highest level of emotional intelligence, the best people for compassion and empathy and communicative skills, these sorts of things, because they're going to become much more emphasized when you can outsource a lot of the functions today to the support, the augmentation by machines. That's a big trend that hasn't started yet, but it will.
You're getting at the other point which is getting ‘back to the future’ where we can have doctors who today are largely burned out, [and] a very significant proportion have depression. Most of them are advising their kids, “Don't go into healthcare and medicine because it's so burdensome and you don't get to execute the mission of caring for other people.” We can get back to a point where it used to be – the profession used to be, which was the human touch, the human factor, the connection. That is achievable given the support that is coming.
The book is Deep Medicine. Who did you write it [for] – what reader did you have in mind when you were writing it? Who do you hope reads the book and what would they do?
Well, I hope everyone will read it but realistically, people who are either in the healthcare field, in the AI tech field, computer scientists. They’re going to want to work in this space because there probably can't be any area that would benefit more of all the different sectors than healthcare. Also patients, we're all going to be patients one day or another, if not have been already – that's occurred. If you know what's coming, it gives hope. It gives hope that today's fractured, chaotic, unsettling intersections with our healthcare don't necessarily have to persist over time. I lay out the blueprints in the book for how we can get to a far better state and it turns out that that was achievable in the past. We have the opportunity to get back there again.
It looks like we're coming up on time. I'll ask one last question. You say it looks like we'll have the opportunity to get back there again, and you offer other qualified statements about what's possible in the future, but in the end how optimistic are you that this is what will happen?
Well, I tend to be an optimistic person and most of the things that I've tried to project like in prior books: the digital transformation of medicine in Creative Destruction and the democratization of medicine in The Patient Will See You Now, they've taken hold. I expect this will be potential that will be actualized. Byron, you're bringing up an important point. The increase in productivity that we're going to see is inevitable. It could make things worse. Doctors could get squeezed more than [they are] already today, nurses and all clinicians. If that happens, that would be a tragedy. I'm optimistic that doctors will come together and the medical community will stand up for patients. It hasn't done so well in the past, but it does have a big chance now. I'd rather think of the positive side. It will take longer than it ought to but hopefully a very positive outcome from this new capability.
All right, let's leave it there. The book is Deep Medicine and if you're interested in the topic, it looks like a great read. Thank you for being on the show.
Thank you very much, Byron. You take care. Bye-bye.