First Look - Vinod Khosla Stanford Medicine X Closing Keynote

Ladies and gentlemen, with great honor, we welcome legendary entrepreneur, investor, and mentor Vinod Khosla with the closing keynote. Good afternoon. Hope everybody is still awake. I'm an odd person to speak here because I know very little about the healthcare system, and I know very little about medicine. But I contend that the system won't change from the inside. And what we will see will be radically different if it's in— invented by entrepreneurs versus invented by people within the system. That's not to say there won't be collaboration, there won't be people from both sides. The starting point of view matters a lot. Let me caveat everything I will tell you with It is speculation and musings of somebody who's a technologist or technology optimist, I should say. I probably don't know much about technology either. So let me go through this relatively quickly. We know this problem. Lots of misdiagnosis. It's like saying every day you go you go to drive every day and once a week you have a car accident. In fact, this misdiagnosis is as big a killer as breast cancer. That's surprising to many people. This one's my favorite. People argue with me. So I had to find a source for it. I'm really talking not about the 10-20% of really good doctors. I want to start with the 10-20% of doctors who are really bad or missing. So the bottom half of the doctor pool is what we really are talking about, especially in a place like India. See, human doctors, they have a few problems: cognitive limitations, lots of biases. The more often a term is mentioned in the press, the more often it gets diagnosed that year. Plenty of data to support that more ADHD gets mentioned in the press, the number of diagnoses go up. We can spend plenty of time talking about why that's a normal human behavior. So, and everything I say, there's in small writing, and you're welcome to have a copy of the deck. There's a source. So, Feel free to read the articles associated with these. In fact, I, I reviewed them today to make sure. Lots of sources for errors, but cognitive factors are a big part. And system factors. And my hope is we can eliminate both those. One of the most surprising POTS was premature closure. You have a set of symptoms, you walk in, in the first 5 or 10 seconds the doctor decides what you have, what your problem is, what the prescription is. They might ask a few other questions, but they fail to really keep an open mind, and that's often a cause of misdiagnosis. Here's the other surprising part. Now, to be feared, this is the value of second opinions for people who are seeking a second opinion on a Cleveland Clinic service. But it's shocking how often getting a second opinion changes the diagnosis. It shouldn't happen. If this is really medical science, The answer should be the answer. A second opinion or a third opinion or a fourth opinion leading to different answers says it's not medical science. I'll get back to that. This is data from the American College of Cardiology. They have 7,000+ recommendations. Leading to 53 practice guidelines on 22 topics. 40— almost half of them have no real clinical evidence to support these recommendations. Only 11% have real science behind them. Data-supported science. Of the Class I recommendations, only 19% have real data to support. We are talking about opinions, which Hippocrates called ignorance. This is another one of my favorites. Surgeons were given detailed diagnosis and asked if patient should have surgery. In this case, it was cardiac surgery. Half said yes, and the other half said no. So whether you got surgery or not was a function of who you asked. You think that's bad? It gets worse. They were given the same data 2 years later, and 40% changed their mind. This is random coin flipping. This is the state of state-of-the-art medicine today. Now, in this case, sample size was really small. They were asked to diagnose stenosis. They disagreed with each other 60% of the time. Guess what? 8 to 37% of the time, they disagreed with themselves when given the same data a second time. By the way, the 8 to 37% depended on which vessel they were judging stenosis in. This was a major procedure that was actually banned. All the experts were asked to judge the probability of a negative consequence Major negative consequence. That was the probability distribution among the experts who decide whether it should or shouldn't be retained. This is medicine today. Same thing with colon cancer. There are estimates of the impact of colon cancer screening on cancer deaths. I hope you get my point. This is the tradition of medicine, not the science of medicine. Even the most basic things, like should your child get an antipyretic, should fever be reduced when they're ill, hasn't really been tested. I have some extensive notes in the notes to this slide deck, so if you get it, It's worth looking at. But something as simple as this is a practice that hasn't really been tested. One study that was done a few years ago, they did it in a critical care ICU for children with the use of antibiotics in a sample of ADP kids, deaths were 7 times higher among people whose fever was— kids whose fever was reduced than when it wasn't. Yet that is still the practice in all of medicine, almost all of medicine. So when my son called me the other day and said, "I got this high fever. What should I do?" My answer was very simple. I said, "If you want to feel better, If you want to get better faster, don't. That is, in fact, probably the real choice. And I say probably because I'm not a doctor, thankfully. There's additional problems. Americans generally understand medical instruction at the 5th-grade level. Very few health professionals communicate with patients at the 5th grade level. This basically says that when you get an opinion, you can't count on it being right. This is Dr. Ioannidis. This is a PLOS article. Dr. Ioannidis is right here at Stanford Medical School, somebody I've followed for many years. From before he came to Stanford. Think about this. If medical studies is the basis of medical practice, and if a study is more likely to be wrong than right, or at least a research claim, that's pretty damning. And then there's the conflicts of Interests that we haven't yet talked about. Johnson Johnson just introduced a new system called Sidaceus. Anesthesiologists are objecting to it on the grounds that they should be present in every colonoscopy procedure, needed or not. The AMA is fighting nurse practitioners being able to do simple prescriptions. I can go on and on. So how do we change this system? It's through naive questions. To try things. To try them from the outside. So my view is the system only changes if we empower the one person who cares about their health the most. It's the patient. I do believe definitely over the next decade the patient will become the CEO of his own health. 80% of what MDs do will be replaced with a better standard of care than the average MD. I'm not, again, talking about the best top 10, 20%. Now, the question whenever I say something like that is the human element of care is important. But then I always answer, shouldn't the human element of care be provided by the most humane humans? Why only MDs? That's not to say MDs can't be humane. I'm saying let's select for the most mirror neurons in the person providing the human element. Let's not just select what college degree they have. Then I get this thing about more integrative medicine. Absolutely false. Human beings can only keep so many variables in mind when making decisions. Ask any psychologist, any neuroscientist. You can't integrate across the last 10 visits you had with the patient. If you, if you happen to get the same doctor, you'll almost always get the same system. If you know how often a person texted or when they called or where they eat and if they change their pattern or change the amount of activity, You're going to do more integrative care. There is no possibility that the MD will, one, be able to get all that data, or two, get it accurately. My point is machines will do the diagnosis, the prescription, even alternative recommendations more accurately. Maybe tell you much more accurately the prob— the probability of an outcome and the probability of adverse outcomes from whichever course of treatment and let you, the patient, make those choices. So where are we on these systems? Let's start with one very simple one that was done in Oregon. They found very high accuracy in diagnosis in a relatively trivially designed system. I didn't think of this as a sophisticated system. And when it came to doing triage at an acute care clinic, it was very, very successful. There's plenty of examples. But the thing to keep in mind is these are all what I call V0, version 0 systems. These are toddler MDs. They're more like digital first aid kits. They're more like the cell phone version 0. Don't think of this version 0 cell phone as today's iPhone. Think of it as this floor-mounted thing that where the handset cord was bigger than the iPhone of today. That was my first cell phone. That's where we are in these medical systems, and people tend to believe that because they have some faults, and they do have plenty of faults, that somehow those won't be fixed. That's a failure of imagination. Lots of companies trying lots of approaches. Let me talk about a few of them. The simple ones. If your kid has an ear infection, a little attachment to your iPhone can solve that problem. You don't have to go call a doctor, get an appointment, wait a few days while your kid's ear is hurting, then take— go to school, pick them up, get away from work, go to school, pick them up. I suspect many of you have done this routine. As somebody with 4 kids, I've done it often enough. And then waste half a day of my time and my kids out of school just to get a 5-minute appointment so the ENT can peek inside the kid's ear. You could actually do a new exam every 5 to 6 hours— every 6 hours and know actually if a prescription is working. It would dramatically cut down on the use of antibiotics. 'Cause you could decide very quickly if the course of treatment needed to change to an antibiotic instead of assuming you need it. So this is one of my favorite devices. This little iPhone— I don't need Siri. I'll give you a quick demo. So I presume most people have seen this device, right? It's one of our investments, sort of becomes very quickly, if you can hold it steady, an EKG machine. Now this is cheap enough where you could give 100 of these away for the cost of one ECG. Why shouldn't every cardiology patient have one? And yeah, it's probably not as good as an EKG with 12 leads, but there you can do that once every 6 months. This you can do 6 times a day. And I guarantee you there's a lot more information when you take this 6 times a day and you know how your heart EKG does when you wake up, after you exercise, when you're not— after you eat, when you're not feeling well, when you're stressed in the office. Think about it. Otherwise, you get that EKG, 12-lead EKG, when you happen to get an appointment, and your heart may or may not be doing anything interesting then. So— This is a little device. You put your iPhone in there and you can self-prescribe glasses. Now remember, all these are version zero systems, like that floor-mounted cell phone. This particular one, which is not announced yet, I prototyped it last week. I wore it for 2 days and they collected 20 20 million data points off me. 20 million data points. There's no internist who can either know how to get that much data or know what to do with it. We are talking about drug adherence and things like that. You know everything about a person's physiology from something that costs less than the cost of registering at a hospital. This little device can analyze your breath. Now, what can you tell? You can tell if a diabetic is having ketosis. You can tell if you're about to have an asthma attack. If you have COPD, your nitrous oxide will go up. You can tell whether you have lung cancer or breast cancer or what type of lung cancer you have, which an MRI couldn't tell you. Of course, my favorite is the new DSM, another attempt to just make everybody a mental health patient. The standard of acceptance of a diagnosis is a kappa of 0.2. I don't have enough time to explain what a kappa of 0.2 is, But Scientific American, I think, called it pathetic. That was the term that was used. How about reliable, consistent diagnosis? I have a Ginger.io app on this. It knows what I do, when I do it, who I text, who I call, when I call them, when I eat, where I eat, who I eat with. It knows if I called somebody on Thursday to make plans for Saturday night or Friday night. It knows when I called my mom or dad or brother. It knows— it collects thousands of behavioral data points a day. There is no possibility, even without a kappa of 0.2, that a psychiatrist can have as much information. It even knows if you walk from your bedroom to your kitchen. Way more insight into your mental health than any human being could possibly have. It's the rough equivalent of what the NSA is doing, hopefully for good. But it's true. It's the same stuff. All of medical literature captured, much like IBM Watson, in a graph of medicine, but available to consumers. We talk about evidence-based medicine, but if most of the medical research claims are wrong, or more likely to be wrong than right, Then how about first doing practice-based evidence? We got millions of patient records. So this particular company is mining the data to say what evidence should— is reliable. Not just what the drug company got approved through a fairly limited study. You get the idea. Health app, most of you probably know. And these are all companies we've invested in. Instant answers. But my point is, we've just started the innovation curve. These will look like that floor-mounted cell phone 10 years from now. These are trivial compared to what we will see. These are point innovations. They'll help. I mentioned cardiology, AliveCor, dermatology or ENT with CellScope, optometry with EyeNetra, psychiatry with Ginger.io, internal medicine with Quantu— every single medical vertical will go through reinvention. These point innovations, harmless little things that can be critiqued in many, many ways, and I suspect some of you are sitting in the audience reacting fairly negatively and critiquing this or that. One, they'll be fixed. They'll turn into a wave. And explode into a tsunami. What is version v0 will start to— oops, sorry— look like the new iPhone equivalent. This is what you have to imagine. These little toys, these point innovations, these seemingly vulnerable to criticism devices and technologies and apps will become. These clumsy point innovations incited by machine learning, where data will in fact be reduced to actionable insights, will have us discover things. If we have enough time, I may be able to answer a few questions. Here's the great part. The best MDs will train the system. The top 5 or 10% will be the individual trainers of these systems, so they will capture all the expertise of those. And for many years, they will become amplifiers of those bottom 50% of doctors. They will assist them make better decisions. Maybe even make good decisions despite the doctor. It'll let a nurse practitioner or a less trained medical professional do way more. In fact, this amplification is probably the only way the developing world can get the medical resources it needs. In the interest of time, I'll skip past that. It's hard to imagine what data can do. Let me suggest the following. In fact, let me not suggest it, let me claim it. And I challenge anybody to bet any amount with me on this. In the next 10 years, the data sciences will do more to improve healthcare than all of the biological sciences combined. I'll give you one example that comes from Kaiser Permanente. A data science just looked at the incoming stroke victims' data. By just changing protocol based on just data analysis, Kaiser reduced death rates by 40%. Now think about it. Death is a very measurable thing. You either go to the morgue or you go home. There's only two outcomes. It's very quantitative. It's not fuzzy numbers. But just by changing how they administer the protocol around statins, a data scientist created a larger decrease in death rates than any drug for stroke, period. Which other drug for stroke has reduced deaths by 40%? The practice of medicine will become the science of medicine. That's my quest, my hope, my speculation. I'll be wrong on many of the specifics. But I will be directionally right. I'm not claiming to sit here and predict exactly what path we will follow. Half of the things I talked about will not work. But the ones who don't try will fail for sure. I like to say, you can fail to try or try and fail. If you fail to try, you have 100% probability of failure. If you try and fail, you have some probability of success. And so the critics, the pundits who say why this won't happen, I think are on the wrong side of this battle. And I will fail often, but because of my willingness to try, I will have opportunity to succeed. In fact, Martin Luther King said, "All human progress depends on the socially maladjusted." That's a direct quote. In fact, George Vinod Khosla said, "Human progress depends on the unreasonable man." He should have said man or woman, but I quote him with fidelity. It's happened in other areas. No plane is flown by a human being anymore. Way too complex for a human being to do. Almost as complex as your human body. 15 years ago, only 15 years ago, if you said stock trading could be done with machines, you'd be laughed out of the room. There isn't any respectable firm that doesn't use automated systems. And the best ones don't hire anybody from Wall Street. They don't use stock traders. They don't use knowledge. They just use machines and analysis. And of course, Google is doing the driverless car. Having ridden on it on 101, it's a much more complex problem than medical diagnosis. And of course, I've mentioned this before. I'm running out of time, so I'll try and stop. It will be— these kinds of systems will be a solution for one of the largest problems in medicine, the global under-resourcing of this. The doctor-patient ratio in India is 10 times worse. That may actually let them get much better care much faster with fewer people to get in the way. So let me stop there. I don't know if you have time for any questions or I should get off the stage. Somebody please tell me. Well, since nobody's kicking me off, are there any questions? Raise your hand. We'll just— Ask questions. No questions. Yes, there's a question there. Speak up loud and I'll repeat the question.

My name is Jolene Chambers.

I'm from Dallas, Texas. And the mission that I'm working on as an ePatient is— thank you. I'm working on a mission as an ePatient to have safer and more effective implanted medical devices. And I noticed that was one of the categories that you were looking at. And I was hoping that maybe you could elaborate on that and perhaps talk about how people could partner with you to do that. We— I've been partnering with Consumers Union that just recently wrote letters to the top 6 producers of hip and knee implants and asked for a warranty. We know that they have the analytics.

So let's keep the question short since this wasn't scheduled.

Okay.

Let me try and answer that. I did lie a little bit that I know nothing about medicine. I don't, but I am trained— my master's degree was in biomedical engineering. I was an electrical engineer by training and then did biomedical engineering because I must have had ADD. I could never do the same thing for very long. But let me give you a simple illustrative example that applies to a range of questions of that ilk without going into implants, because then we'd get into which implant. Almost certainly, immediately after an implant, you can detect how well it's doing if you measure physiology. If you're collecting with your watch 10 million data points a day, I bet you know a lot more about how your implant is doing than the doctor does today. More importantly, I talked about practice-based evidence. If we analyze the history of these implants, of any particular implant, across millions of patients, hundreds of millions, whatever the number, per patient population, We would know more about the effectiveness of those implants, when they work, when they don't, who they work in, than any quote-unquote medical trial the FDA approves, because those are all rigged to just be approved by the company that's trying to get approval. That's another case where we will do something I call studies after the study. That means after introduction to practice of a new technology, a new implant, a new drug, we will start to do the real study of how well they effectively work in patient populations. Nigam Shah did an analysis of just the Stanford data. And it was clear years before the FDA banned Vioxx that It should have been banned. Just from a small data set like Stanford. Imagine if he had access to all US data. Any other questions? One last question. One last question.

One last question.

Oh, okay.

So you have a room full of engaged patients who have done a lot of really cool stuff to really get revolutionary things going in healthcare. So as an investor in a lot of health tech and as a BME, what would you tell them as, you know, what their next step should be? And how do they get to scale and get your attention? What do you think they should do?

Well, the first thing I would say is just because we know things will change doesn't mean we know the path they will take in changing. Really, important to realize that I find it impossible to predict where things will go. I find it relatively simpler, not simple, but simpler to predict that they will change. So I'm convinced, I'd say with very high probability, that things will be radically different. If I wanted my kid to be a doctor, I'd send him to the math school, not to the medical school today. That's pretty radical. I would ask them to try lots of things. That means if we can't figure out the exact path, it means we can't predict what technology or what iteration of a technology will be successful. And we will discover these through zigs and zags, and some things we'll try will seem foolish in retrospect, and the critics will say, "I told you so," or critique you for this or that, that it doesn't have, and the FDA won't like it because it doesn't fit into a convention, but that does not mean you have to give up. Keep trying, keep experimenting. Unique things will evolve. And don't listen to the experts. There's way too many experts. So let me finish by giving you some stunning data about experts. Um, Professor Tetlock at UC Berkeley did a study on experts. He took 250 of the best experts in a number of different fields. Uh, he followed them and he, uh, the names are anonymous because they didn't want to cooperate if they were going to be measured. He followed them for 20 years across 28,000 forecasts. These are the expert panels I talked about, the expert opinions you get. Of those 28,000 forecasts, which is a statistically relevant number, he found the average accuracy of experts was the same as dart-throwing monkeys. So don't let them— by the way, if you want to read the study, if you're statistically inclined, read his book called Expert Political Judgment. If you want to just have a fun read, there's a great book called Future Babel. Okay, Future Babel is a very fun read that captures the essence of his research for normal consumers. But it says, please don't listen to these experts. If you think you can break the norm, please attempt it. Thank you all very much.