Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again

good evening and welcome i'm ivan Oransky I'm the vice president of editorial here at Medscape and it's my great pleasure to welcome you to Medscape for a very special event the launch of deep medicine I'll do it like they do it on television latest book by dr. Eric Topol who is our editor-in-chief here at med school now in keeping with part of what I took away from this fascinating and important read I fed everything I know about our guests about book launches and about artificial intelligence into a deep neural network and I asked it to spit out a set of opening remarks first it told me to wear a blue shirt check then it said that I should be briefed check don't don't worry I'm I'm not the main attraction here but then it was what it said next it really made me nervous made me a little bit worried it said be funny and then followed by this very odd voice which some of you may recognize may find familiar it said I'm sorry Ivan I'm afraid you can't do that those of you who remember that famous line so it's that just as dr. Topol tossed out his a I informed diet advice when it suggested that I believe he cheese cake and bratwurst if you if you've missed the piece in New York Times that I'm referencing please go and find it I threw out the draft and I decided to just introduce our speakers so that we can spend the most time with them rather than with me - you're right - you're sort of far right I should say for some of you is dr. proposed surprise winning author dr. said Arthur Mukerji who's emperor of all maladies quickly became must read for anyone interested in oncology and cancer as well as a New York Times best seller he's an oncology researcher and physician at Columbia and is also the author of gene and in intimate history tonight dr. mukajee will be turning the tables a bit on dr. Tocqueville dr. Eric Topol who's sitting right here to my left who in dr. Topol interviewed dr. Mookerjee in 2015 as part of our Medscape one-on-one series of interviews with medicines leading thinkers dr. Topol is a practicing cardiologist at Scripps in La Jolla and califor and the editor-in-chief as I mentioned here at Medscape and he's really widely considered one of the world's most influential thought leaders on in digital innovation in healthcare he's also the author of two previous books just some logistics for tonight the very important one of course is that the bathrooms are over there passed that column and through that door there will be a short question and answer session following the discussion and you've may have noticed the panoply of cameras that are well not facing all of you but facing us here c-span's book TV is a filming so do be aware particularly during the QA that you may be filmed if you ask a question we hope that you do sign copies of doctor topples deep medicine will be available at the conclusion of tonight's event over by our code check where you left your your coats I know that I look forward to thought-provoking and thoughtful conversation so without any further ado please join me in welcoming dr. mukajee and dr. Topol thank you very much yes thank you Eric I thought I enjoyed your book I enjoyed your article we've been talking about some of these issues for a long time for almost a decade I would say and then finally of course this book maybe we start with some some definitions so that people are up to speed when you talk about artificial intelligence and when you talk about deep learning machine learning what do you mean and we'll come to the medicine part in a second what what is it what is it what is it if computers have been in medicine now for what twenty years maybe longer thirty years computational algorithms have been in medicine for twenty years if not thirty years what's different and and why and and explain how how deep learning and artificial intelligence is fundamentally different and define it for us well you haven't rehearsed anything so I don't know what asked me so that's a good one well really what we're talking about is the deep learning story got rooted just less than a decade ago in Toronto University of Toronto by Geoffrey Hinton and his colleagues and it was a whole new way to some type of AI really what it basically is the idea of taking data and putting it through neuron layers that no human says how many layers but the layers of these artificial neurons determine what it takes to read the features whether that speech or image and it was applied initially to imagenet which was a fantastic labeled images that facially and their colleagues put together millions of images that were very carefully annotated and what Geoffrey was able to show was that when you do this you can read images interpret them classify them as well as human beings and then over time over the last few years even better so that's really what set the potential in medicine that you could have pattern recognition with this type of their eyes specifically and that could be applied to medical scans pathology slides skin lesions and it's the nice part about it is that it's it's not really the human bias is not part of the neural network it is if you program in human bias as part of the inputs but if you don't have that it's really letting the machine do the work and I think it has potential that transcends these initial areas and then of course there are other complementary aspects of of deep learning and high AI tools that are going to be transformative so it's a very new thing the deep learning side and it really is what I think can connect the data that we are flooded with in medicine with the ability to get back to the patient care that we've lost over time so that to the patient care that we've lost I think it's a very important feature of all of this and I want to mark our time for this but I noticed that you used a very narrow definition of deep learning and of artificial intelligence you explicitly use the word pattern so Geoffrey Hinton I've been in conversation for a long time I wrote a piece about Geoffrey's work in New York [Music] and so second but let's just I'm obviously interested in the fact that you use pattern recognition you used image net and the examples you used were you know the diagnosis of skin lesions and of pathology and a radiology etc is your impression that this word will be limited in this way or will it expand outwards towards becoming wider will it ask deeper wider questions about medicine that we ask US doctors or is this in other words is this a tool which is a pattern recognition tool which of which is extraordinarily important and I'm not let's not let's not be let's not be glib or flip about that I mean the capacity the III described in in that article when a when a dermatologist a young dermatologist in training finds his or her first melanoma they go from a case study of zero to a case study of one when a neural network that has studied ingested data from five hundred and seventy eight thousand melanomas takes another one and it goes from a case study of five hundred and seventy eight thousand to five hundred seventy thousand one so [Music] we understand what the power of this data is but you have a sense of how wide this will be well I think it's a really important point because today it is relatively narrow yeah and that's partly because the data says that we have to work with in the medical sphere are relatively limited they don't have we don't have these massive annotated sets of data but it will go much more broadly but I think what we've learned one of the greatest lessons today is that we can train machines to have vision that far surpasses humans and so what was started with some of the things I mentioned is now expanded to for example in a cartogram that you could not just tell the the function of the heart but also the probability from a cartogram of a person developing this type of arrhythmia or that type of really things that humans can't see we have the greatest example that is the retina which you can tell a man versus a woman without necessarily having to look at the retina picture yes I know that but it turns out this is something that no one still has yet explained it emphasizes the black box explained ability feature but if you give retinal experts like international authorities to look at retina pictures they can tell it's 50-50 chance if we're going to get that right male or female but you can train an algorithm to be more than 97 98 percent accurate and no one knows why so when you say narrow we are only starting to imagine the things that we can train machines to do and then when you start bringing in all the different layers of a human being and then corpus of the medical literature and you know the sensor did you know Mike's a microbiome all these different things then you have a setup that's much more broad both for the individual as well as for people that are providing care for that person one of the things I will obviously touch on is privacy which is an incredibly important arena so let's just chalk out some time for that but I was obviously impressed I mean mine from my field is cancer I was impressed by the the data that's come out of the UK biobank you know in terms of breast cancer predictability and you discuss a little bit about this so just to give you a sense the audience a sense of how this world is moving and and and this is also true for cardiovascular disease but how this world is moving if you took a the the pie chart for familial breast cancer so imagine you have breast cancer in your family you know that it has crossed multiple generations in the past our capacity to predict whether you yourself a woman or a man what at risk for future breast cancer was limited to usually single highly-penetrant jeans like Braca one and braca two and people would make sometimes important decisions in their life Angelina Jolie being one of them important decisions in their lives based on that genetic diagnosis but if you looked at that pie chart of people with familial breast cancer when only talking about people with familial breast cancer only about 10 to 20 percent of that pie chart was was predictable in terms of single highly penetrant genetic genetic changes the rest of it was dark matter to some extent in other words you would come you could come to me as a physician and say look I have familial breast cancer in life in my family obviously what can you tell me what my risk is and I would say unless if you don't have bracket 1 or bracket 2 mutations I can't tell you what your risk is I can tell you whether you're at the highest quartile of risk or the lowest quartile Urbis one of the things that's happened with the UK biobank and other bio banks is that if you take genomes and then you map along that genome fate and one aspect of fate could be breast cancer you can now begin to make surprisingly deep predictions about people who are at the highest quartile of breast cancer to have future breast cancer in other words a woman who might have a nine-fold risk compared to the rest of the population of future breast cancer based on her genetic makeup this has happened in cardiovascular disease as well and these algorithms as you've pointed out are actually relatively simple they're additive algorithms how walk us through a scenario walk us through a scenario of what would happen once we create these gene fate maps right well and unleash the tools of artificial intelligence on them walk us through what what could be profound walk us through and walk us through the problems sure well I think of course you want to be careful not to put fate and genomics in the same sentence perhaps and you wrote about that eloquently in the gene book but I think the point that sit as really getting at here is that the polygenic risk score for breast cancer which is not the Braca and rare mutations about getting to score a little bit more because it's important yeah so you don't even need to do a sequence for that you can get at a I mean how many people had 23andme here know a lot of people but you know you can get out of that or ancestry if you've done that ancestry.com you can get this 1 million letters of a genome through a chip which can be run even for $20 you can find all these variants collectively hundreds of changed letters that would be the equivalent of having Braca one or two mutations so it's getting at it is so you have 88% of women who will never have breast cancer in their life and who are the 12% that are really at risk so today we had this you know remarkably wasteful way of putting all women through mammography with a 60% false positive rate now but we can already see through that between the rare mutations that are well characterized and then this collection of these variants of common variants that together we can actually predict very close to those 12% maybe 20% so you'd have a bye for all these women that wouldn't have to be screened or maybe every 10 years or something like that and the same is for all these conditions there's an actionable path it's not just for breast cancer look honey about 12 uh cardiovascular disease yeah it's heart disease is the one that's even more firmly established where you can get the top ten percent of people at risk and you know that was a very dejected for me I have no heart disease in my family and our team at Scripps made this app that's free if you want to get your results you can do it while you're here this evening takes a few minutes my gene range you can go to my gene rank and you can get your 23andme data uploaded to the app and you get your score out of 100 and I did that my score is 92 which is really high risk so I started taking a statin because it turns out statins have a much bigger impact to higher risk and there are a lot of people who are taking statins just because they have a high LDL cholesterol but it's gonna have no benefit for them what's interesting about this of course is that this is an independent risk factor from LDL cholesterol it's Independence orthogonal as a risk factor or thought complete and additive but it's actually if you were to take more risk factors better than family history or or most of the others smoking and just even plain LDL cholesterol but the point about this genomic layer of information is one layer of a lot of data that's just with the chip then when you start taking the genome sequence which already a I particularly deep learning is doing so much to unravel that humans can't do and then you start to say wow we're already doing this for a risk or for various common conditions where is this going to be in a year or two because it's zooming forward and in the book the chapter on deep discovery is about the science genomics cancer that's where the biggest advances are happening right now so you know this is someday going to get translated into far better prevention and far better parsimonious use of resources so we don't have the current scare tactics but better unwitting scared catfish but nonetheless they are well talked about talk about now the flip side of that scenario which is and there are at least two or three one of them of course is that our understanding of ourselves as human beings changes as we do this as we unleash algorithms that we've invented ourselves on ourselves you know there is a at least for me there is a significant concern that we become a kind of we imagine everyone as a locus of risk this whole audience becomes the locus of risk insurance an insurance company imagines everyone as a locus of risk and that fundamentally changes who human beings are and how we conceive of ourselves and then there's of course the proximal question of privacy you know if you're a locus of risk and you happen to leave that backpack with your with your 23andme app on the subway and someone finds out maybe that's a far-fetched scenario but they're much more in your fitted scenarios would you like to share this information with your spouse let's say you happen to have a polygenic risk score of whatever that number might be for breast cancer and you have are deciding and you happen to be in the highest quartile is this information that you'd like to share with your spouse it even if it's not you choose not to perfectly reasonable idea but is there a you know it changes the structure of of human relations if you decide not to act on knowledge that you seem to have let's tell us about what happens then well I think this privacy issue is fundamental and I think many of you seen that privacy has been declared dead in other circles not health and medicine that's not acceptable in our world of health care and there are ways to to work around this problem and the one that is I think the cyber expert who understand the hacking and breaches data being held hostage at health systems all these sorts of things the recommendation is that the data should be in the smallest units possible like one like an individual should own their data and in fact we have to get there someday like in Estonia where people each person owns their data and then it's not in a unit or mass servers that are the ideal target of cyber thieves so that's one way to preserve privacy but the ownership is really important and the reason for that is not just about privacy so nobody today has all their medical data from the time they were in the womb when that's actually really important stuff all the way through their life to this moment nobody has it because you got a differently gotta go to a lot of different people for your care in different places nobody has it but if you're gonna do AI and be able to prevent conditions or better manage them you've got to have all the inputs we know that and nobody here has them so we have a balance between maximizing privacy but also aggregating all that data and today people are generating their own data through sensors and if you get your genome sequence or even your your chip 23andme or whatever you don't want that in your medical record that's sitting in the hospital because or health system because that can be used against you with the only the only thing that's covered in this country is life insurance and your employer but disability insurance and other forms of insurance are not covered so you know excuse me health insurance and employer status but life insurance and disability are not covered so you really don't want to put that into your electronic record so we need a place right now it's a lot of these data sets are homeless we need a home and that should be the ownership of the individual we will get there someday we're way behind Estonia and now other countries like Finland and Sweden Switzerland are moving in that direction but you know we don't right now targeting obstacles in the United States I mean why why aren't we here why aren't we there was it sin as I said it was from the standpoint of cancer genetics and cancer genomics it was an embarrassment that the UK biobank was created in months years before we did ours and we've been accessing that data oh it's amazing what I mean the UK is not the only buy back the other bio banks but why why are we late to this game well we're really late I mean I just finished a almost two-year review of the NHS and not only are they the world leader in genomics they started their bio bank and now Gino's England years ago but they also have the work I did with them was to plan a digital and an AI strategy for the next 20 years and they put in you know billions of pounds of resources so in the midst of brexit they're actually planning ahead for healthcare now in this country we haven't put $1 as a national we don't have a national strategy there was an announcement by our president recently about an AI strategy without a zero dollar without any specifics and so the UK is just zooming forward you know there seems to be you know what's the what what's the problem is that you can he diagnose the problem well right now I think we're just trying to survive day to day in this country it seems like but there even with their own Stevie's also kind of survived a Labour know I'm really been impressed having done a lot of dwell time and interactions with the with the British that they they get this planning thing really seriously and they respect the power of AI so for example they have a big emergency room south keys that got rid of keyboards now you want to talk to some happy doctors and happy patient no keyboards all voice recognition and they showed it's possible and that's in an emergency room setting where you have a diverse type of patients coming in and so they're showing the world that hey we're going to get rid of keyboards first and our kind you know it used to be all the old days we talk about getting rid of paper we never got rid of paper but no one I mean keyboards are the enemy of both doctors and patients and nurses and everybody and so speech recognition which is deep learning is so advanced today and we're doing nothing well you know what we do have over 20 US companies that are on this including many tech Titans but we are as a country we're not behind the effort we're as they are and China is we're just we're slipping on this and there's a really great opportunity I don't think we'll ever have another opportunity like this I can't see in many many years even generations ahead I want to end a little bit about you know I talked a lot about patients I want to talk a little about doctors as well I'm going to read something from your book oh we this is page 306 it's a nice thing about medicine I wrote it recently a big piece on physician burnout and how keyboarding has become one of the sources of physician burnout and and why we we find ourselves automated but also dehumanized as you go through the whole day this is your writing we also need to rewire the minds of medical students so that they're human oriented rather than disease oriented hospitals and sign-out continuity rounds are all too frequently conducted by card flip whereby one training doctor reviews the disease of the patient status and relevant test results without ever going to the patient's bedside even the diagnosis of disease is disembodied that's a strange word to use about the diagnosis of a disease disembodied by looking at the scan or lab tests instead of laying hands on the person such routines are far quicker and easier than getting to know human being Ronna oddish physician in Detroit laid this out well with two groups of medical students one called pathology and other humanistic the pathology group gets extraordinary training in recognizing diseases by recognizing skin lesions listening for murmurs or knowing the clotting cascade the humanity the humanistic group gets all that training with us also trained to pursue the context to be human being letting patients talk and learning about what their lives are like what is important to them what worries them given a patient who starts crying the pathology group can diagnose the disease but can't respond the humanistic group wired for emotion even before the tears begins here is the tenth pitch of vocal cords stretched by false bravery and comforts is this is this your vision of how you know removing these burdens from from medicine will end up restoring a kind of faith that medical students are starting to fray will against affray with you know you know how it takes a while to get the book out right but it captures the whole story and that is that we've lost our way in the 40 years since I finished medical school it's been a steady erosion and we've gotten further and further away from the care their true human bond and we have an opportunity and I believe at least in my lifetime the only opportunity to turn it back to get back where we were yeah and that is the gift of time but not just that as you said in citing that it's about the the human centric aspect it is so you've just gave you know doctors and patients time together that's not enough it's got to be cultivating now how that is so incredibly important that restoration of trust the presents this price precious relationships yes there used to be intimate it used to be sacrosanct hi I gave my medical students when I have medical students and on rounds at the last day of rounds I give them a I mean I give know a week to prepare but I give them two options one option is to choose a topic pathology topic of choice triple negative breast cancer acute leukemia in the elderly whatever might be that's one choice and the other choice I give them is to say take one any of your patients that you've presented but I want them presented in the full so I want them I want to know where they were born what their real you know what their name was how many make a three-dimensional human being out of this patient and you cannot imagine the skew so 10 I think 5 to 7 years ago the skew was everyone wanted to present triple negative breast cancer now everyone wants to present the patient everyone wants to present that medical students want to present the so-called three-dimensional patient so they go and I say to them look you know what you need to do is you're gonna spend hours by their bedside and talk to them about their exam you know this is cancer force talking about their anxieties their worries their future how many you know whether they have children they don't have children you know who pays rent and then I ask them there I mean the kind of questions that you would ask you know if you if they were presenting triple negative breast cancer is saying what is the statistic that shows us the number of women african-american women who have triple negative breast cancer I say how much rent do they pay what is the cost in where they live what is the cost of a typical day's meal and if they don't know the answers to those questions they have to go back and ask those questions again anyway oh I'm really I'm not the best and I'm the most impatient teacher so this I think it's time for I mean I think you know you've we've covered a vast ground you have defined for us a a new kind of medicine a new kind of medicine that on one hand liberates patients it liberates the idea of medicine itself but also has a powerful influence on doctors and how we think of ourselves maybe I'll end with with one last question from me and then open it up to the audience who are the skeptics who doesn't believe you and why why should we why should we not believe you well there's a good reason not to believe and that is we have a history of the administrators the managers the business people to squeeze clinicians more and more so if there's more productivity and better efficiency and workflow the natural default mode is going to do see more patients read more scans read more slides so if we as a medical community don't stand up and that means both the clinical people and their patients to this force this business force then we won't see the potential here so that's the real big challenge as far as the naysayers they're most everyone because they are you know things have gotten pretty cynical and we watch the the electronic health record the single singular worst disaster to happen in medicine in recent decades it's given digital a bad name so a lot of people think that's somehow a continuum of AI when it couldn't be further from the truth I give it back to Ivan [Applause] it says it's on it is on now thank you so we're gonna take some questions now and again that was a just the beginning of what I hope will be a great conversation with all of you in the room we have some we're looking for questions from social media as well and we have someone checking into that and we do have mics so please use them be warned that the people holding the mics are going to continue to hold on to the mics because as all of us who've been to medical conferences and actually frankly any conference knows we do wish and actually require that your question begin as a question or at least for at least contain a question if not begin as a question so be warned of that but believe if everyone can raise their hands I think I saw one over here we can start there hi good evening everyone thank you so much for this opportunity I am really excited to be here just a small backdrop I graduated from residency last year I just started my newly minted attending real late last year and this just reminds me when I was studying for my boards I cooked myself up in my room and my older brother would laugh and mock at me and say by the time you're done with all your studying ai is gonna do your job so my question is the following because they're kind of scared me do you really think that a I will completely utterly take over our job as doctors or is it just a I plus MD yeah no it's a I plus MD that's why I was given Sid a rough time by the vs. but he didn't pick that title for his New Yorker piece no I think the point you're getting at though is you know today you can look stuff up on Google well tomorrow you'll have all this data processed and through neural Nets so that idea of not being able to get your arms around a person's data it's going to be simplified it's going to further reduce the time burden of trying to you know look at a person's data and also the literature will be at your disposal which it isn't today you're trying to keep up with the literature is really difficult and a lot of that is of course is most of it is unstructured text which is right at the moment hard for deep learning but as it alluded to in the outset we're moving quickly in that so that text will be interpretable so it's all going to be teed up that's why we don't need brainiacs going forward in medicine in the years ahead we need the people with the highest emotional intelligence communicative skills empathy and that's the difference I think or going forward discussing healthcare will be a heavy emphasis for his company and years ahead [Music] trove of data in terms of you know [Music] decade here's to here Eric where you think how where polls will impact health care and outcomes preventative care and years ahead and what that means for the US healthcare industry thank you oh you didn't say it no it's on the video but is know what I was gonna say on that is you know we have the detect Titans which are even beyond the ones you've mentioned are all you know in the AI they all have big AI groups and they've hired physicians and a lot of clinical people so they're on this but we've worked with the Fitbit database and the problem is it's not medical sensor grade you know it's heart rate and sleep that's rudimentary sleep metrics it's you know it isn't what you want for medical grade data and it's very incomplete and sparse yes you can get certain things out of it but the ones that you just ran through unfortunately you know including Garmin and the others they just don't have what we the kind of data need like for example we don't know what normal blood pressure is today because we don't have people that are getting their blood pressure in the wild in their real world at a high frequency and we don't even know we have organizations like the American Heart and American College of Cardiology making up guidelines with no basis and they just just last week has found that people in Canada those guidelines doubled the incidence of hypertension in this in the country of Canada and it was just done one day they decided they didn't change the guideline anyway so we are so far behind in this and so we need that we now have FDA approved smartwatches with blood pressure not through Apple well you have seen Apple with a deep learning algorithm the first consumer based algorithm for arrhythmia detection so we're moving in that direction but it's very early and I think that's the future where you have really accurate continuous or high frequency sampling of important medical metrics we don't really have that yet and steps are not really what you consider medical at all in fact that whole 10,000 steps thing is a myth there's no data to support that hi thank you guys I wrote this down because I was nervous so I think a lot of us are really excited about this medical frontier that we're all talking about but wary of issues you've talked about in terms of bias and surveillance and inclusion those sort of things do you have thoughts or recommendations for those of us who work in low resource settings or with vulnerable populations about how we can leverage some of these tools for issues of social justice and health equity the first thing to say is that this whole AI frontier could make inequities much worse if it's only used for appleman people and that's not going to help to what we have which is this big chasm but today in the new york times there's a really good piece about using the AIS for I diagnosis in India and that just is one example of how you can use these tools to make diagnoses that were not previously possible medical personnel so if we use this right we can actually you know move in the right direction so it has a lot of potential that have been available thus far involved basically white people and we don't even know how to extend those bio banks particularly the the genomic bio banks outside that and there's a good reason for that there's a long history for any in a in a very good reason for that history of mistrust of all of all the medical interventions so the first thing that has to happen if there is going to be an important concern it's the first thing that's gonna have to happen is that we have to assure ourselves and everyone else that this information will be private will be used in the right way will be used in the way that actually helps people rather than hurts people there's there's a long history of not of doing precisely the opposite so there's going to be a there's going to be a wait period and unfortunately it is during that wait period that of course you know the technology will advance rapidly so we have some we have some important catch-up to do just to give you one example prostate cancer is an arena arena where we badly badly badly cardiovascular disease badly need data from traditionally underrepresented groups including women actually to start with but also including women of color men of color etc so this is this is not gonna happen overnight but it's gonna happen it's it's going to have to involve a trust-building exercise and a powerful outreach with emphasis on powerful before we can get this done hi Eric so your book sounds fascinating I'm really excited my name is Sandeep Johar so I'm I'm really excited to to read it my question is what what do you think about the black box problem in other words when machine algorithms get so complex that we don't really know how they work how do you know what we can trust and what we can't well we have to subject them to clinical trials and if so what would be the gold standard well it's great to have you here Sundy thanks so much for joining us another noted author position often so I think the first point about that is we don't acknowledge all the black box that exists in our care of patients today so many drugs so many things we do ranging from anesthetics to electro convulsive therapy to all sorts of things we have no clue how it works but we accept that but we're holding these algorithms that are being needing to be accountable for explain ability if you will so that's you know something just to keep in mind now I'm I'm all for have no black box having everything explained and indeed there's a lot of work being done right now by deep learning algorithms to deconstruct to to basically reverse engineer what's going on in the neuronal layers to learn what is it that explains the algorithm but still today there are a lot of algorithms that are remarkably successful we have no clue like the retina example I mentioned now the question that I put in the book about I spoke to a lot of people who are they're a raw AI gurus in the field spend time with them one of them is Pedro Domingo's at a University of Washington who wrote the book the master algorithm and I said well Pedro what do you think if you if you had a algorithm that was 99% accurate for diagnosis and had no explained ability he said I'll take that over any doctor it's in the book it's a quote and so of course he's a little leaning towards you know algorithms because he understands them in general but the point being and that gets to your the essence of your question if you have a randomized prospective trial at scale that shows this algorithm does something wonderful with outcomes of patients are you going to hold it hostage to use it before the explained ability feature comes out I don't know I don't know the answer to that question we already have though in Europe the GDP our rules that say they don't want any commercial medical algorithm going forward without full explained ability which is a pretty high threshold so we'll have to see how this plays out whether Europe will reduce those standards what the US will adapt we have where to it's too early but you know the criteria of having randomized trials like we have in medicine today without requiring explained ability and adopting medical practice the question is you know what are we going to do in the algorithm a year what do you think about that and but it worked yeah there's a girl and so I have a couple of thoughts about it one is that we the medical model for the FDA as you know very well has evolved especially in the pharmaceutical universe has evolved rapidly towards explained ability and biomarker driven trials if you go to the FDA and say I have this drug I have no idea how it works and I don't have a biomarker for for patients who are likely to respond they'll say go take a hike the if the same standards were to apply to algorithms then algorithms will have to take a hike until they can explain things now that said there are several people in the AI University I've interacted with them you've interact with them people are very smart who think that this that the blackbox question is is overblown for certain kinds of simple Diagnostics for certain kinds it's not but so and there's a the community split so if you talk to Hinton he will say that there is that that he is a hundred percent convinced that the blackbox exists and will exist if you talk to others they all say that this black box problem is overblown that in fact the D convoluting algorithms are already exist and they can at least understand why the word why is important weights are being given to certain features it really will show over time but the standards for if we use the current standards for trial running that are applied to pharmaceuticals those standards will either have to change or they will have to be modified if we move forward with AI driven trials we don't have those standards for pharmaceuticals I just want to just acknowledge we had more than a hundred people watching on Facebook so not just those of you in the room and of course this will be this will be it was recorded and will be shown both on Medscape and elsewhere but I just want to take a question from one of those social media viewers Jeff will call him Jeff because that's apparently his name and sort of and dr. Topol if you could address his question which is his concern that in fact AI might and they think this was a different version of the earlier question that a I might actually dehumanize medicine that might take the human element out of medicine and he notes in particularly there was this case which was not AI but very recently I'm sure you all heard about a patient family a biller patient who was told by a doctor but through a video screen of some kind that he did not have long to live and I'm and Jeff is curious how you would respond to that yeah that's been going around the last couple of days the patient in Fremont California who was told on a video screen by a doctor remotely that he was going to die it's a new year term and of course I didn't go over too well and it's ridiculous but to call that era I is he's even more preposterous it's it's the bad judgment of the doctor to give that message through a video screen he should have you know gone over and seen the patient so this is about medical care today not about the Cure the future but artificial intelligence human lack of intelligence yeah lack of a chai so I think this is I think the crux of it could AI make it worse I mentioned it can make it worse but it's not it's in our grasp to go the other direction because the amount of efficiency workflow improvement you know the basically outsourcing of so much today to then restore a human bond that is the whole idea of enhancing the humanity the humane of what medicine is all about because we have this augmentation mechanism so it has nothing to do with that Fremont robot called robot telemedicine I'm gonna take one more question I think if I can get a mic over to Seema I don't know if I can but thanks Elizabeth so question with two things one i'm stein said God doesn't play dice so at the end of the day human AI no matter what it seems like from a patient perspective at the end of the day there is no explaining of why it happens to you it happens to me that's one but then the flip side of that is the trust so trust ability trust in the medical institution in the pharmaceutical industry in AI versus the human and so at the end of the day it seems like the medical community or health care community whatever you call it physicians sometimes are seen as blaming the industry industry's blaming the healthcare system health care system is blaming their physicians physicians are overburdened the question is how can we as a medical community healthcare community enhance trust in the patient or with the patient so the patient feels like yes at the end of the day God doesn't play dice maybe but there is enough trust in the health care system that you know there is you know I feel like I'm well taken care of well I mean the trust doesn't come when you have seven minutes with a patient and you're not even looking in the eyes of the patient okay and I start the book off with my experience of having a knee replacement and getting roughed up and that's where the orthopedist who I send all my patients to when they needed joint replacements okay so you know I have no trust in that doctor I'll never go back and see him when he recommends that I when I'm when I'm having a really rough time and can't sleep from pain and crying he says I should go see my internist to get antidepressant medicines right know the story this is an important story yeah I mean I you and I talked about it over dinner at Scripps right yes moving it's a very moving story I tell the story no I it's about the trust of when you when you're really down an hour and so what happened so you had I think the issue here is that I I went for what I thought was a routine knee operation because I had a congenital rare condition osteochondritis dissecans which my orthopedist completely forgot that I had because it wasn't in front of him in the record there was no AI tool to give him my data from when I was a chip when I was a kid a teenager so I went in and had the operation which was a grand success and then very soon thereafter I went south and you know had would look like a gangrenous leg and you know it was just horrendous the worst torture I mean I was ready to kill myself I I understood fully why people become opiate addicts and I thought it was over so when I went to see him with my wife that was what the message was I needed anti-depression medicines okay so you know that that kind of tells the story of where medicine is today because I only had like three minutes I wasn't even examined okay you know I had this horrible looking knee and leg and no one even looked at it okay so that we we doesn't matter it doesn't matter the point being is how many of you here have been roughed up in your health care yeah most people who are willing to admit it and that's where medicine is today we have a big chance to get the trust back but it means having time and a human centric approach as the machines get better which they will that you can say for sure are we going to get better that's the question thank you very much I just want to again thank our guests did you have I just wanted to say I not only wanted to thank Sid because it's wonderful to have a chance to have this conversation but to think Ivan Oransky jo-ann's drangus the people at Netscape that I've had the distinct privilege of working with for many years phenomenal thanks for hosting us thank you thank you thanks for being such a wonderful editor-in-chief in some you

Channel: Medscape

Views: 5,844

Rating: 4.9607844 out of 5

Keywords: eric topol, siddhartha mukherjee, md, pulitzer prize, author, the emperor of all maladies, deep medicine, deep medicine: how artifical intellegence can make healthcare human again, healthcare, artifical intellegence, AI, science, medical science, future of medicine, book, medical doctor, live, medscape, medscape videos, medscape live

Id: H1td1zaU-iM

Length: 54min 3sec (3243 seconds)

Published: Wed Mar 13 2019