Vinod Khosla, renowned Silicon Valley entrepreneur and investor, talks about how technology will reinvent healthcare as we know it.
Chrissy: Vinod, so we’ve met a couple times over the years at different healthcare conferences. It seems like healthcare is one of your top interests these days. Can you talk about kind of why you’ve been so fascinated by the sector and what do you think the role of technology will be in the next few years?
Vinod: I think we are starting a new era in health care. So far, health care has been mostly experiential, incremental improvements. Clearly healthcare today is much better than it has ever been. But we’re going to see, because of new technologies, a quantum jump on healthcare improvement. Data science, data, machine learning, all the things we hear about, the hype will actually, in my view, come to bear on real solutions for real patients. Not only for sick care but also actually make healthcare more about health. I think all that will be possible and cost effective.
Chrissy: Got it. How do you separate so much of this hype that you mentioned, this sort of machine learning, the data science from reality when you’re meeting with entrepreneurs and they’re telling you they’ve got the next great AI tool for healthcare?
Vinod: It is important that we take an experimental approach to learning as long as it does no harm. Healthcare is very old fashioned and slow to move and everybody knows that. In fact, the fundamentals of traditional healthcare institutions’ approach to medicine are mathematically wrong. So I rail against the Hippocratic Oath that every student takes. It says if you save 100,000 people and hurt 1,000, you shouldn’t introduced that new therapy. To me, that works when you have dumb patients. When you have smart patients, you give them the choice.
The fundamental change we will see is the consumer will be much better informed and the patient will become the CEO of their own health, making the tradeoff. So which risks they want to take and which risks don’t. Some may opt to pass that decision making on to a doctor. But the really smart people will want to control their own destiny, make their own choices, and make the decisions on whether to take a procedure, a therapy, a drug, based on understanding what the tradeoffs are.
Having said that, to your original question of what’s hype, what’s not, it’s very hard to tell. Quantum leaps are possible but not every proposal for a new approach to health will work. In fact, I would suggest that most won’t work. And if you have 1,000 attempts and 1,000 claims, 20 or 30 will work very, very well and most will fail. But the 20 or 30 that work will completely revolutionize health care.
Chrissy: Interesting. So it sounds like you’re talking about a different kind of doctor-patient relationship that we’re gonna see in the future. Which reminds me of one of the statements that you made several years ago that got a lot of attention in the media, around how technology will replace doctors. I think you sort of clarified that later and said that it’s more about augmenting the role of the doctor. Can you talk a little bit about how this doctor-patient relationship will change with all this new technology coming in that brings more information, more data to the patient, than ever before?
Vinod: Let me explain it the following way. A few years, I was talking to the Dean of the Harvard Medical School. And I suggested they change their admissions criteria. To get into Harvard Medical School or Stanford Medical School, you need IQ off the charts. IQ off the charts, sort of the doctor Houses of the world, the skilled diagnosticians are also the guys with the least EQ. I suggested they change their criteria from the highest IQ to the highest EQ and make their admissions criteria look much more like the admissions criteria for the USC Film School. Because the human element of care is what doctors will need to do. We know the placebo effect works. Statistically, I can prove that the human element of care, the belief in whether something will work or not, is an important part of medicine working. That is more likely to be the role of doctors then saying, “Here’s the diagnosis,” or “Here’s the prescription,” or “Here’s what’s right for you,” even if I only spend five minutes with you in a patient encounter and don’t know your history, your preferences, whether you like risk or not, whether you’re optimizing for the next 10 years or the next 50. All those things a patient knows about themselves and they will use that to make the right decisions.
Chrissy: Just to recap a lot of what you just said, which is really fascinating advice around how the next generation of doctors should be selected and trained, what you’re saying is we need to sort of change how doctors think about care from just being the doctor houses of the world who are making these smart diagnoses to also having the bedside manner. Does that mean the patient is more involved with the shared decision-making rather than the more paternalistic approach that we’ve seen in healthcare so far?
Vinod: You know, the paternalistic approach is the wrong thing but it’s also plain wrong. The Journal of the American Medical Association, JAMA, recently took a couple of cases. They classified cases both hard cases and easy cases. And they asked a few hundred physicians to do the diagnosis. The accuracy of diagnosticians, clinicians, even when they could ask follow-on questions, asked for extra data, extra test results, on the easy cases the accuracy was 55%. On the hard cases, the accuracy of diagnosis was 6%. They’re killing a lot of patients with this wrong diagnosis and nobody seems to acknowledge it.
But that wasn’t the bad news. The news actually got worse because they asked these clinicians how confident they were in their diagnosis. Even in the hard cases where accuracy was 6%, the confidence in the diagnosis was extremely high. It’s stunningly bad where we are today and it can be improved with much better tools and systems. That’s what I hope all those entrepreneurs out there trying to change medicine will do. I jokingly told a class of medical students at Stanford a few years ago, “If you want to be a great clinician in 15 years, then you should go to the Stanford Math Department, not the Med School.”
Chrissy: Tell me a little bit more about these tools that are getting you excited. We’re talking about the sort of Fitbit activity trackers? Are we talking about much more sophisticated mobile medical devices that can track things like your EKG, your heart health?
Vinod: Well, I’m talking about something much more than that. There’s some wellness things and there’s some data. If you wear an Activity Tracker, you’ll exercise a little more or take a few more steps. But that’s not what I’m talking about. Should a patient need a stent, should they do bypass surgery? Do they need an ACL repair? Do they need a medication or not? All those are real clinical decisions with life or death consequences. You know, just drug interactions kill more people than car accidents in the United States every year. So that’s the kind of thing I’m talking about. If a patient is being administered the wrong drug which will interact negatively with something else they are already taking, a system should make that decision, not a physician who forgot to look at every other drug. Or even if they looked at it, didn’t remember that there’s this interaction between two esoteric drugs.
We can’t expect a normal human being, which a physician is, to remember the last 5,000 drug interaction articles or the last 1,000 papers which were published in the last year on the cancer therapy that they’re about to administer. You can’t expect that. It’s not human. What humans are good at is getting out of that white coat, being really friendly and being advisors, mentors. That’s what the doctor should be doing. Not trying to do things that computers do so much better than humans once we get to that system.
Chrissy: What you’re saying is you’re not advocating for doctors to go away entirely and to be replaced simply with machines where I sort of type in my symptoms and I get a set of drug treatments evaluated for me, based on what I’m already taking. You’re saying that the doctor is supposed to be sort of a friendly, human face but totally using these tools to guide their own decision-making with the patient?
Vinod: Well, a doctor should have very high EQ and be really helping a patient think about things and explicitly guide them, not make decisions. A doctor’s role should be bring out a patient’s preferences in making a choice. If I have my third ACL surgery on my knee, the decision I might need to make is a very personal one, “Do I want to keep skiing and injure it again or not? And if not, maybe I don’t need that surgery. And if I want to then, yes.” But that’s a personal decision. It’s a lifestyle decision. It’s not a decision a physician should be make…or a surgeon should be making. That’s the kind of missteps we have because we have this paternalistic vision of medicine, as opposed to saying, “Let’s enable the patient to make their decisions fully informed and fully informed in normal language.”
Chrissy: What would you say is holding us back from this sort of future scenario in which algorithms and AI are presenting this data to the patient and the doctor? Do we have this data or is it sort of buried in silos and do we still have a task ahead of us to even get it out and bring it into one place?
Vinod: Well, there’s no one single answer. There’s maybe 15,000 procedures or drugs or therapies one can follow. There’s probably somewhere between 10,000 and 15,000 diseases and diagnosis that are possible. It’s a complex map in medicine. And in fact, that’s part of the reason a human being can’t keep that in their head. In some areas, we definitely have sufficient data and people haven’t worked on the right problem or worked on it the right way. So take reading an MRI chart or a X-ray or a CT scan. There’s no reason a human should be doing it. We have the data. We just haven’t developed all the algorithms. We hear stories about point algorithms. The role of the radiologist should be obsolete in the next five years.
In other areas, we just don’t have enough data and we need to collect data. You look at a diagnostic blood test. You know, most physicians need how to look at 30 different blood markers. An immunologist may know how to look at 30 or 35 more immune-related markers. A cardiologist might look at 10 or 20 additional cardiology-related markers. But the fact is there’s probably 1,000 to 3,000 biomarkers that are relevant. We haven’t collected that data because physicians have not known how to use them. Until we collect it, we won’t even know if it’s useful. So that’s where the experimental approach needs to come in. We need to collect 3,000 biomarkers on every blood sample, if we can do it cost effectively. And if we do, we will discover which of those are useful for what diagnosis.
Almost certainly, your symptoms won’t determine what disease you have, it’s your biomarkers and your physiologic markers like heart rate variability or others that determine what disease you have. I’ll tell you a story. A friend of mine, Larry Small, who has IBD, he tracked weekly his biomarkers for 10 years. And he discovered at some point his biomarkers were going out of kilter. And he said, “Something’s wrong with me.” He went to his doctor. Doctor said, “How do you feel?” He said, “I feel great.” Said, “Go home.” Two or three years later I think, they discovered that he had IBD. And he talks about it publicly, he gives lots or talks on this. If the disease had been caught years earlier from when the biomarkers started to change, we’d have much better treatments than catching diseases in the late stage. So that’s an example of where medicine needs to change.
Now, we don’t have that data to answer your question. But if we started collecting data on a million people, we would be able to characterize IBD and cardiac disease and diabetes long before it becomes visible today. And we’d have much better treatments or lifestyle changes that would help one. That’s what your annual exam should be about, not the current annual exam which has repeatedly been shown to be worthless in improving health. So anybody who does a physical exam is wasting time and money today according to most of the published literature.
Chrissy: Absolutely. And there are certainly some that agree that we should get rid of our annual physicals, which is kind of a crazy thing to even imagine.
Vinod: Crazy for a doctor to imagine, not for somebody who cares about what the facts really show.
Chrissy: Yes. Exactly. Presumes that we’re rational animals there. So I did want to ask you…back to this idea of algorithms and you mentioned radiologists being obsolete in five years. I’ve seen quite a few case studies in my own reporting of, take a case of something like a patient with diabetes and some of these patients end up getting a retinopathy and go blind. You can see it in the image scans, which tended to pile up on the radiologist’s desk. There have been algorithms that have already been developed that have done a decent job of pinpointing these early signs of retinopathy and making these early interventions. But the problem has been the regulation. How do we get these algorithms into routine clinical care? What do you think the next steps are to change the system, so that we can even start introducing this technology and bring it to use?
Vinod: We take a very U.S. centric point of view. In fact, the U.S. FDA covers 5% of the world’s population, not the other 95%. There’s probably 500 to 700 million people in India who’ve never seen a doctor. They just don’t have access to one easily, I would guess. That’s where this medicine’s more likely to start. I suspect, and I’ve said this many times, I will get better cardiac care and advice through a mobile phone in a village in India than at the Stanford cardiology department because Stanford Cardiology in 20 years will still have experts and the mobile phone will have much better knowledge. I think that’s likely to happen. Today, if you’re in Tanzania and you’re a woman and you need a C-section, almost certainly, you’re going to die because there just aren’t enough people to do a C-section.
So the world is a place that needs much more medical expertise than we can conceivably train even to get to the level of the median physician. I think there’s plenty of places these kinds of technologies can start. They won’t start day one at Stanford Hospital. In fact, they’re likely to be the last to adopt such new technologies. Need will drive adoption, I think. In the U.S. context where medicine’s pretty good, we’ll have to prove that the new medicine is much better before we adopt it. The process will be slower and will take much longer.
Chrissy: Interesting. Yes. I completely agree that we do tend to take a U.S.-centric approach. Do you see many in the venture capital community thinking now about healthcare in other countries? Places like the village in India or even in my native Europe?
Vinod: I think it’s starting to happen. I saw a number of startups at YC and YCombinator that were addressing health, many from India or Africa or other parts of the world. I think it is the beginning of a revolution and the more success examples we get, the more startups we’ll get because start-ups like role models of success. I think it’s an engine that’s barely starting. We’ll see slow progress, I suspect, for the next five years. Then we’ll see acceleration as more role models, as examples show up that can be predicates to newer approaches.
Chrissy: So no one really wants to hear that their job is gonna become obsolete or be replaced and doctors are certainly a powerful bunch. How have they been responding to some of the ideas that you put forward around the role of algorithms and AI and how that will change care?
Vinod: Well, doctors are humans. You’ll see a wide range of responses. There will be doctors who only care about their patients’ welfare and they will take whatever best tools that are available and apply to their patient. And if they’re digital, so be it. There’s others who want the same thing but are more conservative and wait longer for proof that new things actually work. Then there’s others who want to do more surgeries and will do more surgeries to make more money. So there’s the whole range. One has to realize that the world of doctors is the world of humans and they’re represented of the really well-meaning ones, there’s conservative ones, there’s forward thinking ones, there’s mercenary ones. And it’s like any other profession and we’ll see a range of responses.
Chrissy: Interesting. What would you say…radiology is certainly one but what are some of the other medical specialties that you’ll see becoming less and less relevant as we find that algorithms can take on a lot of these jobs?
Vinod: One of the things about innovation in technology is you can’t just predict on a timeline. It very much depends on a entrepreneur deciding they’re gonna make that change happen. So electric cars may not have been where they are today had Elon Musk not started Tesla. In fact in 2012, the Department of Energy made a forecast for the number of electric cars in the U.S. that was smaller in 2030 than Tesla shipped last year. Why? Because Elon made that happen. We need those entrepreneurs among your listeners to say, “This world is possible. And despite the obstacles, despite the problems, despite the uncertainty, despite the difficulty, I’m gonna make it happen.” As those entrepreneurs try to make it happen and hopefully some of them make it happen, we’ll see this new world. It won’t happen by itself. It won’t happen through the medical establishment. In fact, Walmart didn’t change retailing, Amazon did. NBC didn’t change media. YouTube and Twitter and Facebook did. Some entrepreneur or entrepreneurs need to decide to make this vision happen. Those are the ones I keep looking for.
Chrissy: Are you finding that there’s an increasing number of folks in the sort of engineering, computer science, data analytics world who do want to get into healthcare? Or are you sort of desperate for a whole new infusion of talent here?
Vinod: I think there’s a lot of people in these new areas, in traditional, what I call “Information technology areas,” the world of Google and Facebook or Twitter who want to do this. If you take what’s called a quant on Wall Street doing hedge fund, trading algorithms, or somebody at Google or Facebook doing algorithms for selling more stuff to more people, better ad targeting but they’re not motivated by that job. I see many of them saying, “Why do I want to do more effective ads when I can work on saving somebody’s life? Why do I want a hedge fund to make more money and develop an algorithm for that when I can work on something more meaningful?” I’m seeing more and more of that and I’m encouraging more and more of those really smart, machine learning, AI people to work in this area because it’s much more rewarding, it’s much more meaningful.
Chrissy: When you see a group of entrepreneurs walk into the room that have a technology background and they want to build algorithms in health care, would you be put off if there was no one on the team that had any healthcare experience or would you see that as an advantage because of the fresh thinking they might bring?
Vinod: I actually would be discouraged if they had a CEO who was a healthcare person. I find that most teams are more innovative if they don’t know healthcare. So I encourage people not to do the traditional thing and hire a senior person from healthcare who then fits them back into the old shoe. A healthcare person doing digital health will be like General Motors doing the EV Van instead of the Tesla. General Motors worked longer, more years, and more dollars spent than Tesla and had zero impact because they fitted into the mold of cars as the automotive industries understood them. We can’t have that happen in healthcare. We need the Teslas that reinvent the notion of an electric car.
Chrissy: Although, you know, sort of my counter to that would be Tesla’s an expensive car. Most of us can’t afford it and if you’ve got Silicon Valley kids who’ve got plenty of money and opportunity ahead of them, in healthcare, how can we ensure they’re actually thinking about Medicare and Medicaid populations, and those in healthcare that are getting left out in system versus just building yet another technology for this rich and worried world?
Vinod: Let me just suggest politely that your view is very naive. Let me not be shy about saying yes, it was an expensive car. But they have a much more affordable car coming. And that car would not have been possible, the 30-something thousand dollar car would not have been possible had they not started at the top end. In digital health, since many of these technologies will be software, we won’t have that problem. For expensive blood tests, we will have that problem. For a diagnostics software program, we won’t have that problem. So let’s not shoehorn all notions and not do things for the reasons you just cited. Let’s try them all, see which ones work, which once start at the bottom and scale up because the person in Tanzania had no other alternative and which one start at the top, at the very expensive end, and then scale down.
Chrissy: Do you have any examples sort of in your own portfolio of a case like that where something was able to scale up or scale down?
Vinod: Well, Color is a great example. A BRCA1 breast cancer case was thousands and thousands of dollars not very long ago. That technology as Color provides it is so cost effective, it’s almost trivial. It’s less than the cost of your physician visit effectively.
Chrissy: Great. Well, I think we are just at time. I super appreciate your time and this is a fascinating conversation. Lots to think about going forward and I’d certainly love to hear from some of our entrepreneurial listeners, physician listeners and see what they think.
Vinod: Thank you.
For Vinod’s full white-paper, “20-percent doctor included” & Dr. Algorithm, go here.