Deep Medicine: How Artificial Intelligence Can Make Health Care Human Again

Deep Medicine: How Artificial Intelligence Can Make Health Care Human Again



so it's a great pleasure to be here with Eric Topol eric is a very prominent cardiologist very famous cardiologist author of this book deep medicine and also vice-president Scripps research and you have other roles including a project at NIH we will talk about but I guess the first thing should I should say there are certain extremely complex systems that I try really hard to learn about but I never quite get there and one of them is quantum mechanics the second is my wife who's a very complex system the third is the healthcare industry and the fourth is AI so these are these final two subjects are things I really don't understand and so I'm a little nervous about being up here but the great thing about the book is that it it talks about technology a lot they're really in the service of humanizing medicine and so I want us just to walk through a bit of what AI is what it's doing and how it actually couldn't lead to a better healthcare system but I was hoping we could start with a story you tell in the book about your knee which is symptomatic of a lot of what's wrong with our healthcare system sure well first it's great to be with you David now and I understand the complexity especially of health systems in AI and in fact it was that ladder which really led me to want to delve into this very deeply the story of the book is begins with my knee I underwent three years ago a total knee replacement because of a congenital condition rare condition called Austin dryness dissecans and I specifically lined it up to go to the orthopedist that I had sent all my patients to because I you know they had very good results and in fact the operation itself was a great success it was just afterwards that it was a disaster and I start off with a conversation my wife Susan was here with me and I went to see the orthopedist a month post-op and then I had my leg was swollen it looked like I had gangrene it was just I had been in horrible pain couldn't sleep but you know for a minutes or an hour at a time and I've been having crying spells and I really it was in desperate shape and he he looked at me and he said I think you need to get anti depression medications and that was kind of a robotic response and the other part of the layers of this are that he had he used AI he would have had all my data and would have told him this guy Topol has da stroke and rightist and he's going to have a really bad october fibrosis reaction and i've got to do special thing so he doesn't get that because that's more common than an infection which is the dreaded one which he told me about before the operation it happens twice as often and it's just scarring it's basically a kind of reaction to this artificial knee it's like a major inflammation in all the scarring so I had that and I actually now have never fully recovered but it's the problem we have today David is I think you know most everyone here has experienced there's very little time when you go to see a doctor they don't have all the data they don't have time to review it all in fact even if they could review it all it's only the data from that particular health system work that area so we have incomplete data incomplete context insufficient time and that we have the state of shallow medicine yeah now you have a sentence in the book over the course of your career you've seen the steady degradation of the human side of medicine so what's driving that and yeah there's just one case but how does it manifest itself more broadly well the I think the the the lack of time with patients averaging single digits per encounter in the u.s. is certainly a big part of it but it started back in the 1980s so I finished medical school in 79 and prior to that time when the relationship where the doctor was precious there was trust there was presence there was lots of time it was a human bond and what happened around early 80s many different factors took hold but essentially it big business and things happen like the relative value units which basically really diminished the value of interaction but increased the value of things like operations and procedures and then the latest hit in the 90s was the electronic health record which converted doctors and nurses and all clinicians to data clerks and that has basically led to this remarkable disparity a disenchantment level we've never seen with more than half of clinicians suffering burnout 20% clinically depressed the highest rate of suicides ever among doctors in history of the profession so we have seen a lot of degradation and the main problem with the time it's not just the patient's feel short-changed the doctors can't fill their mission why did they go why did anybody go into health care to look after patients if they can't look after patients and provide care so that's kind of the state that we're in right now unfortunately just maybe a little biographical detour why'd you go into cardiology or yeah yeah it's interesting I had actually set out to be going into diabetes because my father had all the complications as a insulin-dependent diabetic but when I went to UC SF my mentors there encouraged me because it was a very exciting time it was the first clot dissolving therapies the first balloon opening up arteries the first everything and so I just got captivated by it and was very glad that I could do get in a field that you could feel like you're really doing a lot it was looking I didn't DeBakey was it well DeBakey was the surgeon in Texas my mentor was kind of chatterjee at UCSF who was a legend and was a great humanist which was another big influence when he would lose a patient he would go into a crying the state and you know it took a while to get him out of it he cared so deeply for every one of his patients and then you don't see that so much today it's mainly it's not because the people don't have empathy it's just they don't have a chance to get it out to express it I've had two maybe two encounters with the medical conferences one was I saw DeBakey lecture once through a room of cardiologists and somebody passed out on the person next to him screamed is there a doctor in the house the other I gave a speech to a gigantic Conference of neurosurgeons and instead of calling the neurosurgeons just by a slip of the tongue three times I called them narcissists weird thing was they were not offended they took it as a reality anyway that's the detour let's talk about AI I sort of I wish it wasn't named that to be honest yeah it makes it seems like the rot robots are taking us over yeah you know I agree with you it's a terrible term but you know it's an old term as you are well aware it's 5060 years at least old the new term that is applicable to what we're talking about is deep learning it's very different than just general AI what it means is taking data extensive data and putting it through these neurons that are like our neurons in our brain and it basically is the data itself determines how many layers to figure out all these features so all these inputs could be an image scan it could be a skin lesion picture it could be a voice whatever goes through and then it the output is the answer an accurate answer so deep learning is the transformative subtype of AI which is just rocking it you know in the world of health care in terms of what it can do it's doing things we never thought was possible and just to give you one example if we give a picture of a retina to the leading retinal experts of the world and we say is this from a man or a woman the chance of them getting it right it's 5050 whereas if you train a machine a deep learning it's over 97% accurate now I know there are other ways to tell whether it's a man or a woman but you get the picture yeah so give us some other example say in radiology of what some of the other ways you see the the most powerful transformative effect yeah so radiology is often cited as the leading edge but and it is I mean there there are a lot of things happening in radiology so any scan a cat scan x-ray MRI any type of scan can be read more accurately and getting the accuracy of speed that would make radiologists you know feel that they have been superseded in their abilities but that doesn't mean we're don't want to have a sign off or oversight by a radiologist just means it can really catalyze the process and one thing about radiologists that most people don't know is that over 30 percent of scans have a false negative that is something is missed so that can get down to low single digits never get down to zero but maybe in the one or two percent with the use of training machines so radiologists have have had a big benefit they will derive quite a bit of augmented intelligence if you will but also we've seen a big jump in other fields especially noteworthy is in gastroenterology so having a colonoscopy is not many people's favorite thing to do and it's even worse if they miss something and it's common small polyps dominion of polyps which can be precancerous just as much as big polyps can't be missed and a randomized trial so the missing them by Machine vision a eye deep learning can be brought down to almost zero which is great and then another area that's had a lot of perhaps the most rigorous work is in eye diseases the ability to diagnose diabetes retinopathy which is frequently missed which now can be done by the receptionist doesn't even require a doctor at and also things like many other conditions like macular degeneration and so I disease are going through revolution using deep learning ok let me take advantage of my ignorance here so I have a pain in organ X I go into you you could put me through the whole AI experience you or whatever data you can get what is the out what is the machine what is the machines outcome is it there is a diagnosis here's what you should do or is it just here so diagnose yeah great point so today it's really mostly diagnosis it's not really coming up with recommendations for therapy diagnosis this is something again that most people aren't aware of we have a rate of 12 million serious misdiagnosis a year it's hard to imagine that we live with those and many of them could be you know life-threatening so we want to get the diagnosis story straight that's where AI deep learning has its biggest impact treatment you know later but and the other thing about this is is isn't just on the doctors side we're starting to see AI deep learning take hold on the consumer patient side so it's but that's all in the in the realm of Diagnostics yeah one of the most exciting parts for me was about mental health yeah and about being able to hear the tones of a person's voice or even go through Instagram and look at facial expression and predict suicidality and things like that to tell us about what's what's good and bad it's it seems extremely potentially extremely important and completely horrifying all the same time right well I think you've summed that up well every day so mental health is a problem with depression being the number one cause of disability and a grossly insufficient number of professionals to help people with depression besides that our treatments for depression which often rely on medications often they often don't work and it takes a long time to figure out that they're not doing anything and they have lots of side effects and cost so it's kind of a mess of a field and it's critical that we get this straight now and their deep learning has potential because it turns out that the things that we do naturally like talk our speech is rich about our state of mind the tone intonation all the aspects of our speech relative to our baseline you can tell a person if they're depressed better than they know themselves subjectively then you add on things like your breathing pattern if you sigh a lot this is saying that you're depressed and you may not even be in touch with it you're sighing then you have the keyboard of your smartphone when you're texting or doing an email or on your lap your laptop the way you touch the keys and then it goes on your you know it could be your heart rate and your blood or your vital signs and you know facial recognition so there are so many ways that we can objectively determine and track continuously state of mind so it's really exciting because for the first time we have objective means which we've never had before the only problem is they haven't been applied yet and we have we don't know how many of these things you really need that is when does it get saturated and is it different for certain people like for certain people just knowing that they are reclusive that they don't go anywhere they have no physical activity that is what the ticket is that you've made the diagnosis and the you can quantify it for other people it could be their speech so there's a lot of work to be done but it's it's rich so is that so just for example is the tone of voice flattered and a person right definitely and then you can you know there are many algorithms now because speech is just as developed as images with deep learning so there are many ways to quantify that you know – like the decimal point we're not talking about you know a zero or one we're talking about really shades of gray so it's impressive what we're that's headed the other thing that David about the mental health thing that I think most people would never have predicted I certainly didn't people are more comfortable to share their innermost secrets with an avatar instead of a human being so if we can exploit that and in fact some of these entities that are working with avatars are starting to develop ways so that we can get some of that help some of that bolstering through a non human to deal with the lack of professionals that we have today and the preference of people I actually found that horrifying the fact that that you know people are more comfortable talking to a robot than a human being just seems well a it seems frankly let a narcissistic seduction because you're talking to something and you will never have the demands of a relationship with that thing right it can't be good long term actually well and it gets back to the creepy side do you really want to have your mind being tracked but maybe if you're got significant depression yeah and maybe also we want to get away from medications treating depression and there's lots of other approaches so you know perhaps for a short term maybe people would want to go through that and you had mentioned that Facebook will occasionally privately intervene depending on the posts of the people on Facebook and and you had mention made a good point they're gonna be Facebook people here that they don't actually release how they the algorithms they don't release any of their internal workings of how they figure that out that's right no it's a real problem particularly since this data is not protected the privacy issues are really undermining the the success of the things we're talking about and so that has to get straightened out there's one thing about privacy for non health and medical matters but this this is something that's really critical yeah I want to take you back to the 90s when the internet and all that was just getting going and there was this great burst of optimism that we'll be able to really communicate with each other world peace will happen we'll all get to know each other and it hasn't really worked out that way and Neil Ferguson this historian went back and said we should understood that because when the printing press was first invented people had the same reaction now we can all share books we can we'll all communicate with each other and we'll have world peace instead we had 200 years of religious wars because getting to know each other maybe wasn't such a good thing I don't know but so what's the what's the are we too optimistic and what's the thing that worries you about all this well you know I think idea that technology has negative consequences is is real but you know we're kind of in a desperate situation right now in health care we have gotten it down to a situation which is not only untenable but really how can this be sustained and potentially get worse so my biggest fear you know I have concerns about the privacy and security obviously things like the bias that's embedded it's mainly human bias into algorithms things like worsening inequities which we already have serious problem lots of those problems that are front and center but to me the biggest issue is that if we don't get this right that is a gift of time all these efficiencies accuracies productivity ability to process information for both patients and doctors at a speed and completeness that never was conceived before what will happen is if we don't stand up for patients now the administrators the managers who have nothing to do with patient care they're just the bean counters they will make things worse they will make the squeeze continue and we will see even further attrition so this is the time where we have to say this opportunity is not likely to represent itself for generations if ever so I do agree that overall there is a hit of technology but here we have do dehumanize health care we have gutted the care of healthcare this is our only shot to get it back that I know yeah why should we think as long as the financial center of its fee for service and all that are in place why should we think anything technological will shift that basic logic that there's an incentive just to speed through people and that'll really get to know yeah well I think it comes down to the point that our intelligence maybe is sobering but it's not going to change human intelligence is kind of fixed right and machines are just going to get a lot smarter in terms of the tasks that they can and so we as humans need to become more you mean and to take advantage of the gradient that exists now and it's going to get ideally if we emphasize what are our qualities what are our unique things that we bring and there's nothing more important than our health so you know I think that that's why we have to look at technology not with a blind eye by any means but to take the things that it can do to give us this gift of time which is the beginning of the restoration the beginning of reestablishing trust and presence and a relationship because the patient-doctor relationship is almost non-existent some of you may have that but most people are roughed up when they go to a doctor today because they feel like hey I didn't even get examined or I had just a few minutes I never even saw the doctors face they were pecking on a keyboard that's kind of the problem we have today you have a good anecdote where you're practicing with another surgeon I guess and he does the the touching and the diagnose the touching of the patient you don't and the patient's upset with you yeah no I had a patient from back in the days in Cleveland come to visit me in San Diego on a kind of urgent basis and then I was in the room but my colleague was doing the exam because they wanted to whisk him away to the cath lab and I didn't want to hold things up and so when I went to visit him in the hospital hours later for the first time ever in her in a relationship that we had for you know Oh more than a decade I look he looked cross I said well what's the matter he said you didn't examine me oh my gosh no I was watching and being examined and I didn't want to hold up the works anyway we you know I apologized and I understand people do want in fact that's a ritual when you go to the doctor and they don't even listen to your heart or they listen to it through your shirt or blouse you know you know that you're not getting the real deal and in fact you are submitting this ritual which is a you know a time-honored thing when you go to the doctor you actually want to have a real exam and that doesn't happen that much anymore it's amazing yeah so you make this parallel to driverless cars yeah that there there probably won't replace humans but there's it's the interaction that matters yeah so you I think you know you follow this whole world of tech well and there was this amazing hype about how there's going to be this driverless cars that pick you up like uber and lyft and and there any conditions they have no human backup it's called level five by the Society of Automotive Engineers total autonomy well now we have the realization that will never occur because there will be weather conditions there will be road conditions and all sorts of things that will never get rid of human backup and the best we can ever achieve is so-called level for conditional human backup now with medicine there are certain things like patients can now get if you're in the UK or other countries you can get your urinary tract infection diagnosed with deep learning very inexpensively going just to the drugstore to get the kit you can get your child's ear infection diagnosed a skin rash Banga so the list is increasing quickly dr. list but these are non serious not not the kinds of things that you have to rely on a doctor in fact you can get them done more accurately without a doctor quicker and less expensive so we'll see these things develop for consumers patients but all these serious matters are going to require a doctor so that's basically going to get to level three which is this intermediate form we're not going to go doctor list only in special non serious circumstances yeah I'm gonna open the Florida questions shortly but how how our doctors doing in empathy is something that's trained it's not a innate thing how are doctors being trained in empathy and how should they be trained well it's interesting because you can cultivate empathy it isn't something you're just born with genetics and medical schools don't generally the 150 of them in this country they don't generally put much into this but empathy is vital basically the idea of being able to be compassionate being able to put yourself in the patient position and perspective is so remarkably critical but I think the problem we have today is there's plenty empathy among doctors and nurses and clinicians it's there it just isn't able to come out but also we don't spend enough time to cultivate it we do have much more of this idea that looking at the scan instead of the patient or reviewing the lab data instead of the patient we don't listen to the patient's story so the way it works today in this limited time is within seconds something like 18 to 22 seconds the patient is interrupted so how can you get empathy if you don't even let the patient tell their story and by the way that story is never gonna get digitized it's not an AI think it's something that you really that's a human thing for someone to tell their story about what their symptoms and what their concerns are so empathy is listening step number one and you know you're really cueing in in that presence presence is it's more absence today yeah finally talk about what you're doing with NIH and the scarce resource here is actually information yeah so we were just talking before we got started there's a I don't know how many of you have heard of the all of us program a few people ok quite a few it's a program of a million Americans and probably will wind up being well more than a million there's a couple hundred thousand that have been enrolled all of you are welcome to enroll it's a study where the people come in of more than half are of non-european ancestry underrepresented minorities they will all have eventually at least if they choose their genome sequence their gut microbiome sequence all sites that with types of sensors all the data will go back to them and this is a the biggest program ever to be launched in medical research in the United States and it's going to take about a few more years to get the whole million people in but we'll be communicating on a frequent basis about the results of that person back as well as people in the whole cohort and it's very exciting so we have a big role that at Scripps research and it's now about a year and a half into it okay I started wearing a Fitbit and it was telling me I was getting really good sleep between 8:00 and 11:00 in the morning and I realized I wasn't asleep I was writing in those days so I don't know if my heart rate goes down or something like that and when I'm doing what I should be doing I don't know it's what you're do bring up a good point is the sensors that a lot of people have experienced today you know are not really medicalized and they're not all that accurate and counting steps or you know crude measures of sleep or not to kind of you know the sensors that we would want to rely on that's depresses me okay let's open the floor right here they'll start in the front we have some microphones coming around thank you I'm Daryl gray with the Ohio State University can you comment a bit on the potential for AI in reduction of health disparities yes so this is a big issue that inequities are as bad as you can imagine in this country I just was looking at the Washington Post today and it was about people in Tennessee waiting outside in the cold to be getting free care whereas that's available you know most any other country without having to wait and for days outside in the cold so we have serious inequities in this country and the question is can an AI improve that or will it worsen it as Darrell's pointing out so I think it's fair to say that you can markedly reduce inequities we have examples of this so one of the my favorite apps is smartphone ultrasound where you just plug your smartphone a probe into the base of your smartphone and then you can take pictures of any part of the body except the brain so in Africa this is being done in places where there wouldn't have any technology and getting algorithms to read that ultrasound so they're getting medical grade scans in places that would never be able to afford that type of technology so you can reduce inequities there's lots of other examples because we're talking about relatively cheap chips apps software but you can also make it worse if you only make it available to people who are affluent so this could go either way but we need to really push on it to use this and exploit the opportunity to reduce the inequities because it's such a big problem and over there yep my name is Judith Barnard this is anecdotal but I hope it will lead to something else as we get older we collect a whole bunch of doctors in different fields all of our sessions with our doctors are leisurely there's a lot of phone calls beep back and forth between us and we have never had the feeling that we are anything less than a human being and what I wanted want you to address is how much of this and I'm sure a lot of it is as economic status geographical and how you can put emphasis upon those places that don't have the privileges we have instead of just spreading it across the whole spectrum well I guess so you're saying that you've never had a negative experience yeah you're fortunate how many people here have had a negative experience I think you're in the minority all right so the first thing in what you could do which is already being done in the UK and select centers and in many centers throughout China is to liberate from keyboards keyboards are the common enemy of patients and doctors and now with natural language processing which is a subtype of AI a different than deep learning but it's a lot of the same principles we're already at a point where that can be done it's just a matter of rolling it out and there are select places it's being piloted here in this country if we can get rid of data clerk functions for doctors that frees up a tremendous amount of time also it restores the face-to-face contact so we have to think of the ways that we can do this that our keyboard liberation is about voice processing and you get much better notes since synthesized notes from that conversation then you get from the cut and pasted notes that are in your electronic record which are 80 percent of the notes are cut and pasted and most of them have significant errors that are propagated from one note to the next so these are things that cut across all health encounters not just people who are happy with their care but for everyone right hi I wanted to you you talked about being liberated from the keyboard but I wanted to get your feedback on and I think an organization that's using keyboards quite effectively I don't know if you're familiar with crisis text line what is the name crisis text oh ok I've heard of it yes so yeah this is a a mental health crisis line that uses text interaction as opposed to getting on the phone with a counselor though you are live with a crisis counselor but they are a pioneer in using deep learning to better connect people with services on their platform for example like they program their platform with certain words that they thought would be indicative of a suicide attempt like die suicide etc and what they actually found was that the word ibuprofen was 14 times more likely to predict a suicide attempt so these are places where we can use machine learning to actually better inform people who who may be very empathetic but would probably miss something like that because it seems sort of counterintuitive but on a disintermediated sort of interaction platform I think they're just an example of somebody who's really doing this combination of machine learning and human intervention in a really good way and I'd love to get your yeah actually I think I referenced it in the chapter on mental health in the book which is what you're doing the point is a great one and that is within text not just speech there are cues that we wouldn't know just like you're pointing out with the ibuprofen that actually are indicators of suicidal ID ideation so this is really important is that and it really summarizes the whole opportunity here that machines can be trained to see things learn things that we can never do and who would have ever thought that was going to be the signal right and this is I think really important because it predicting suicide psychologists and psychiatrists are not very good at this at all you know I review that data in the chapter but we can actually get help from deep learning to find that the the cues that we should be zooming in the Sun I'm Bruce McGovern I'm a cancer survivor and as a cancer survivor there's I've been sequenced I've been x-rayed I've been cat skiing I've got all the information that could possibly have been has been gathered from me and yet every time I go to the doctor I spend the first half hour filling out a from two to four to six pages of information why isn't that traveling with me and why isn't it travelling with everyone in this room yeah after a while you've got it all the information they don't need any more information right and and you know why what is what is no matter regardless of where you go even in this hospital where I got my surgery they want to get another x-ray I've got to have all that fill out all these pages of information yeah so so I'm glad you mention this and it's so true it's so real that you have to just fill out the same forms all the time and this is kind of that shallow medicine that we have today right so I've made the case that everyone should own all their data it's your data you paid for it and you have a vested interest and it could even save your life one of the things that people don't know which is remarkable is that 10% of scans in this country we're talking about billions and billions of dollars are repeated unnecessarily because the patient can't get a hold of their scan so just have 200 we'll have to just do it again and those are really expensive by the way so if you had all your data and you wouldn't be filling out all these forms because you would just send that data here here's my baseline boom and you would just give it in Estonia of all places that's the way it works right and other countries like Finland Sweden Switzerland that's how it works but in this country we still have these forms with the clipboard and you say wait a minute this is 2019 what's the matter here so we can we can do better than this that does ium require artificial intelligence that just requires human endeavor to make it better maybe in the red in the middle there in the back lady in the blue shirt in US sweatshirt right there yeah one of the problems with moving the use of artificial intelligence in medicine is federal reimbursement for the costs associated with it and the doctors willingness to say that it can contribute to the diagnosis process do you have any comments on that was the first part the first part is you need the federal reimbursement for all the artificial information data that would be made available to the doctor to telemedicine aspect of some of it all of these parts are not always reimbursed either through the federal government or the insurance system well at the moment there isn't much of this being used in the US so I don't think the reimbursement thing is come into play the only area where it's starting to take hold is in radiology there's about 20 FDA approved algorithms for various types of scans largely otherwise there's been clearance for things like the Apple watch that detects your atrial fibrillation rhythm which obviously isn't a reimbursement from the from the government so it's all done behind the scenes that is the health systems that adopt this just to improve accuracy and speed for the radiologist they don't they don't get reimbursement for it they've just decided that it's going to become this the standard of care so we haven't really confronted that that issue at this juncture over here maybe even tables written in French thank you – two questions one is do you think deep learning is actually going to enable people on in rural health centers where a kind of pair of physician personnel can now become more engaged in bringing the care to healthcare yes so this is another correlation Carlat that is – the inequities so we have a big rural population that by many metrics has substandard care today telemedicine is pretty primitive it's basically just a video chat which you know you can only do so much during a video chat but where that can take us into the next phase is to exchange data so yeah there are lots of sensors and even doing some lab tests that people can do on their own that are coming alive quickly and so when that is a data exchange that's a way to reach people anywhere and in fact going back to the all of us program a lot of our people are coming from rural areas so the other thing that I think is really vital to this story about the gift of time and improving care is that all people have more autonomy relative autonomy so whether wherever you live if you are able to generate your own data ahead of algorithmic support that's going to not only help you but it's also frees up doctors because they obviously will have to devote less time to that so there is a lot we want to do to empower patients irrespective of their geography it's my second question by the way I'm Robbie Diaz Brenton from University of Arizona is do you think that deep learning could be applied to clinical trials I work in the Alzheimer's field those trials you know will last typically 18 months maybe longer that's a long time to try to detect a signal that there's been target engagement or movement on a biomarker do you think deep learning can actually be applied to data very early on in a clinical trial to determine responders versus non responders yeah in fact there's been a few studies to show that exactly the problem we have is not so much deep learning to predict in ways that we couldn't today but we don't have anything to do about it so yes the problem we have with with this technology is you want it to be highly actionable so you want to apply it to things where you know you can make a difference but you're right it may help in the future clinical trials to get at an earlier point in time okay let's go like a woman in a blue sweater like I think it's blue looks blue to me it's great actually weatr how prepared is our research enterprise to deal with these things especially when they fall outside of our our current paradigms of research how will the research enterprise deal with it it's actually it's gonna explode a lot of the current ways we're thinking about yeah it's interesting you bring that up but it's actually we have to rely on the research to make all this stuff real because if we don't have prospective studies most of the studies so far were deep learning our retrospective in these data sets big labeled pristine data sets but that's very different in the real world clinical testing ideally with a randomized trial or compelling data so we really need the research enterprise to do this right and you know go through the peer review and and also study it and implementation so one thing that you're also touching on is that just you have this great algorithm and it's now you know it's really accurate and it's validated and patience and it's done prospectively that's great but then how does it work in the real world that's the implementation phase because we've already learned for example that an algorithm is very site-specific so it's only done in one Health System and may not work in another or ancestry specific so you want to see how it works in roller and you never want to let your guard down because remember in a doctor/patient thing it's a one-to-one thing and if we make a mistake somebody gets hurt it's low numbers of people but here an algorithm could hurt a lot of people really quickly so this is we need research that's kind of a continuous research to detect that let's go over this Headroom hi thank you very much David Rosenthal from Yale School of Medicine quick question about sort of the time horizon of when this was gonna happen and we're gonna see some of these realization gains there was just a recent nature paper a couple of months ago you saw about speech generation from neural nets where they put some electrodes on people who are having epilepsy surgery and the computers were able to make human understandable speech and the question is these amazing things that are happening with AI neural nets how soon will we see real application of them for stroke victims is it years is it decades yes well David bring up a great one and with your interest in the mental health side it's really I don't know if you saw this but so the the study that was striking it was done at UCSF and there were five people were they were working on epilepsy surgery to deal with the site of the seizure but what they did was they they recorded the electrical activity of the brain to see if they could go from brain activity to speech directly and he did in fact they had the recordings and it's just amazing so without the person talking just from the brain signals they were talking now why is that it so important well as David's pointing out if you have if you're paralyzed and you and or you've had a stroke and but your your it's only affecting your ability to talk not your in your brain center this is amazing technology so this whole idea of the the brain unlocking things and who would ever thought you could just go from brain signals to actually talking so this is exciting and so the unti the follow-up that was just in five people but the the recordings were were mind-blowing literally so there that's gonna happen quickly I would say in the next couple years certain patients this will be it's not that difficult to do and it would really change the lives of those people okay one more and then I'm gonna ask a question a Shireen Gabriel rush University a lot of the illnesses we deal with today are related to the health behaviors so can you comment on how a I or deep learning can help us improve these health behaviors whether it's around diet or exercise or smoking yeah that's a big one so the toughest nut to crack is to get people to practice healthier lifestyle we are seeing some things that are heading in that direction so for example in Finland they did a very large study where they gave people their heart genetic risk scores which we now can do very inexpensively and accurately and a very large proportion of the people who did change their behavior they stopped smoking they lost weight they increase their physical activity so having data for some people is enough to to make a difference but one of the things that we Dave and I were talking about this before he gets started it's about diet it's a fundamental thing and we all would like to have food as medicine and we never had a way to do that and we don't have it yet today but we're inching our way to that so the whole idea is that you can now with these multi-layered data of your gut microbiome your physical activity what exactly you're eating and drinking everything of that your sleep your stress level all these things can be analyzed through machine learning and they already have shown and with sensors like the glucose sensor to give what is the right diet for you to glucose spikes or now last week triglyceride spikes and insulin levels that isn't exactly or preventing cancer or eventing Alzheimer's or or heart disease but it's for you a customized bespoke diet recommendation and it come I I did this experiment and I had amazing the surprises as part of it a lot of it was sobering my favorite foods got rated at D and things that I thought were lethal got a a-plus but it is a wake-up call and weird there is no commercial entity that's that's out there that's doing it right and it isn't quite ready yet but just to give you the sense that that might be instead of saying everybody should do this you know this activity this diet to have it done on an individualized basis maybe that will help yeah I'm hoping bourbon and chocolate will be my A's less I actually want to ask a metaphysical question on that exact point as we learn a lot of we get this information my question is are we more disel ike than we thought or are we more alike so and diets we've learned we're pretty we're much less alike yeah so I think the startling thing is the more we probe this with this so-called deep phenotyping the more we realize that we're incredibly unique and you're going back to that study that was just reported on the diet identical twins had markedly different responses to foods as far as their glucose triglyceride and insulin and every way we look at it we are so remarkably unique and the problem we have is the way medicine evolved is it treated everybody the same everybody should have this screening test everybody should eat this food everybody should take this medicine if you have this particular diagnosis it just turns out that's just wrong and it's a lot of waste a lot of mistakes we can do better than this okay the book is deep medicine Eric Topol [Applause]

Related Posts

Leave a Reply

Your email address will not be published. Required fields are marked *