Back

MIS2019: Fireside Chat with Vinod Khosla & Dr. Toby Cosgrove

Video

MIS2019: Fireside Chat with Vinod Khosla & Dr. Toby Cosgrove

Toby Cosgrove:

Vinod, I am so pleased that you’re here. I have to tell you that about three or four years ago a mutual friend John Doerr gave me your tome, if you will, of about a 100 pages saying that we probably didn’t need doctors much anymore. They were going to be taken over by the various computer capabilities. That was a tough swallow for me. Happily, I hope they don’t learn to do cardiac surgery right away, so there may be some people employed. But tell us about the theory and tell us about why you wrote it. Let’s start with why you wrote it.

Vinod Khosla:

About eight years ago I was on a ski vacation, tore my ACL, took an MRI, took it to three surgeons that recommended very different procedures for recovery. So I said, the treatment I get is a function of the doctor I get, the surgeon I get, not the condition I have, that’s just patently wrong on its surface. At the same time I was looking at the other end of the spectrum, which is how do you scale healthcare in India? And with an unlimited budget, with hundreds of billions of dollars, I couldn’t come up with a way within 25-30 years of getting the same patient doctor ratio, just a simple metric, in India as it exists in the U S. So it forced me to think about nonlinear solutions. How do we broadly provide highest quality care?

As I did more work on it, and it was eight years ago, I first wrote this blog called Do We Need Doctors? I realized that AI, which was still well buried within the surface is talked about a lot now was going to be the solution that would scale skills, deep skills, whether it’s physicians or technicians or cardiac technicians. The way to scale that skill to make it broadly accessible to all 7 billion people on the planet was through technology and AI. So, four years ago I wrote this piece called 20% Doctor Included, which is a 100 pages or so. I drew heavily from the Institute of Medicine studies, in that it’s heavily referenced, many other sources, and I came up with this idea that we could dramatically increase the quality of care while dramatically reducing the cost of care, only through technology. It was the only way to do it and we had to get past intermediate steps in the way.

The role for the humans, and I actually define it in this paper, is the human element of care, not complex disease diagnostics. If Craig Mundie gets 5,000 proteins, you can’t ask a physician to look at 5,000 proteins, and you can’t say that a disease is caused by one biomarker when there’s 3,000 connected metabolic pathways in the human body. It’s not complicated, which is how physicians explain it, the human body’s complicated. It’s mathematically defined as a complex system. Complex systems. Complicated is hand waving, complex systems is a science. So, I like to reduce it and I said this in my paper, we’ve had very good practice of medicine and every year, every decade it’s been improving for the last 100 years. It’s better than it has ever been, but it is far below what it can be. If you take a complex systems view of both healthcare and sick-care, sick-care is what we do today.

Toby Cosgrove:

So what are the factors that are going to limit our ability to do this?

Vinod Khosla:

Probably more than anything else, humans.

Toby Cosgrove:

Humans in terms of the complexity of their system?

Vinod Khosla:

Well, their resistance to change.

Toby Cosgrove:

Ah, Really? I never noticed that.

Vinod Khosla:

It’s often stated as why this is not going to be important. People cite to me, you develop something new, like a 23andMe test, to say, “Are you at risk of breast cancer for a woman or a carrier for a male?” People will say, “Take 25 years for it to become standard practice.” That is the biggest barrier. The other thing to keep in mind is, we worry much more about acts of commission than acts of formation. I like to say, for all physicians here, if you took that Hippocratic Oath, please renounce it because it is mathematically doing more damage to society than it’s saving lives. So if you first do no harm, then you’re saying, I’d rather by through an act of omission, not save a 1,000 lives as long as I avoid a 100 deaths. That’s the Hippocratic Oath. It’s fundamentally wrong and we as humans still love it. We need to experiment a lot with human safety or patient safety in mind, and that’s the balancing act while allowing for innovation to progress.

Toby Cosgrove:

Okay. Now let’s talk a little bit about innovation, because I think that’s the key about really why we’re all here. Why do you think it takes so long and why is healthcare so resistant to innovation?

Vinod Khosla:

Well, there’s a complex set of reasons. First, we as human beings are comfortable with what we’ve done in the past.

Toby Cosgrove:

Absolutely.

Vinod Khosla:

Extrapolate the past in a new patient case. If it’s a radically different approach, then we tend not to like it, but as organizations, imagine if you could make healthcare twice as cost effective tomorrow. That’s half the revenue at the Cleveland Clinic tomorrow. I suspect you will have strong resistance. The American Medical Association today in most States, except for nine States, doesn’t allow RN, a registered nurse to prescribe a Z-Pak.

Toby Cosgrove:

Right.

Vinod Khosla:

Right? Because it takes away a primary care visit.

Toby Cosgrove:

Right.

Vinod Khosla:

The American Dental Association doesn’t allow a dental hygienist to provide preventive care to patients except under the supervision of a dentist in most States. So, at every level making healthcare more cost effective is a revenue reduction, it’s that revenue for a PCP if they can’t do, if you eliminate a face to face wizard, it’s institutions like the AMA, they all have resistance to change and they’re not looking at aggregate benefits to society and you can always story tell why their benefit … you can find an example why it would hurt one patient, and in that story you lose the fact that might save 10,000 lives this year.

Toby Cosgrove:

Do you think the selection and the training of physicians has anything to do with their reluctance to change, allergy to change?

Vinod Khosla:

Yes. There’s many, many characteristics and we’ve talked about this. You get through organic chemistry as you like to say by memorizing things, and then you get through med school by following rules and then get through early training by doing what the chief of staff tells you, to use your words. You’re taught to behave a certain way not explore and experiment, which I can see the reactions of people saying, “Well, that’s bad for patients.” Except every conceivable experiment you can think of is being tried by some physician somewhere.

Toby Cosgrove:

Okay. So, let’s realize that we are reluctant to change and innovation is not a gene that is dominant in most of us. What do we do about that? How do we begin to take what now is the 13 years from proven to standard of care and shorten it?

Vinod Khosla:

Yeah. Now this problem is not endemic to just healthcare. I looked across the last 40 years that I’ve dealt with technology and innovation and that’s all I’ve done in my life. I couldn’t think of, and this is surprising to people that challenge me. Any major example of large innovation coming from within the system in that area in 40 years and doesn’t matter if you look at biotechnology, no pharma company innovated, in fact, Bob Swanson was an associate at my previous firm, Kleiner Perkins, left to start Genentech and biotechnology sort of originated there. Space innovation came from SpaceX and Rocket Lab not from Airbus or Boeing or Lockheed. Media didn’t come from CBS or NBC or Fox, it came from Twitter and Facebook and YouTube. Retailing didn’t come from Walmart or Target, it came from Amazon.

I couldn’t think of one major innovation in any area, electric cars didn’t come from Volkswagen or General Motors, they came from Tesla and self driving cars came from Google and Waymo. So if no example exists in the last 40 years of a major innovation coming either from people who were within that industry or even had backgrounds, all these examples I cited of innovation, major innovation happen from people. Nobody from the taxi industry could have created Uber.

Toby Cosgrove:

What about Skunk Works?

Vinod Khosla:

Have they worked? 

Toby Cosgrove:

I don’t know. Have they? 

Vinod Khosla:

I don’t know of a single example of major innovation. Now, minor innovation, like going from one semiconductor process to the next generation, Intel can do pretty well. And the financial industry is interesting because they’re so profit motivated and transaction oriented, they’ve done things like collateral debt obligations, which are financial innovations. But I think you need somebody outside the system to rethink how something should be done.

Toby Cosgrove:

Okay. So let me ask the hard question. Should we forget our discussions with Microsoft, Google and Amazon and turn to the startups? Do they have a place in the healthcare universe?

Vinod Khosla:

So I should say innovation-

Toby Cosgrove:

You have to understand he’s a venture capitalist, so you know where the answer’s coming from.

Vinod Khosla:

Well, I’m just looking at the historical fact.

Toby Cosgrove:

Okay.

Vinod Khosla:

Right? Because I have spent a lot of time on this question of expert opinion, and for those of you who are interested, there’s a great book called Expert Political Judgment, makes the argument that no area, whether it’s political science or economics or radiology have experts generally done very well with their opinions. They can talk very well on stage and the better they are on stage, the worst they are in practice. He makes the argument, this is a guy called Philip Tetlock, who’s done 20 years of research. He’s a social scientist, first at UC Berkeley and I think now he’s at Yale or something, on expert opinion. That generally is very poor at predicting the future when there’s any change involved So yes, you have to look at startups, you have to look, all founder-driven companies. I mean, Waymo is from a large company, but it’s the vision of a single person. I can tell you Larry Page drove that one thing.

Toby Cosgrove:

And Sebastian Thrun.

Vinod Khosla:

And Sebastian, but Larry got Sebastian to drive it. So, founder-driven companies do much more innovation, whether they’re a startup or a larger company. If I were to say, how do you make Google Health much more effective? And I’ll get in trouble for saying this, you don’t hire Dave Feinberg from Partner’s Health to run it, right? It is likely to make it much more compliant with healthcare norms than to radically reinvent healthcare.

Toby Cosgrove:

Okay. I’m trying to avoid the hits on Google at the moment but-

Vinod Khosla:

I love Google. One of my favorite companies, very good friends with everybody there.

Toby Cosgrove:

So, let’s change the topic here a second, because there’s now in healthcare a tremendous discussion that is important around the social determinants of health. We know that your health is essentially 20% the medical care you get, 20% your genes and the rest of it is the social determinants. What is the potential for machine learning to begin to help with dealing with those social determinants?

Vinod Khosla:

Yeah. So what you’re really asking is how can you live five miles apart and have 10 year difference in life expectancy?

Toby Cosgrove:

Actually, here in Cleveland, in the city of Cleveland, there is a 23 year difference in life expectancy.

Vinod Khosla:

Okay. So, the first thing you have to say is, this is a very high dimensionality problem.

Toby Cosgrove:

Absolutely.

Vinod Khosla:

Is it the street you live on, and everybody’s attitude? Are there trees on that street or is it bars on windows because people are afraid, and if there’s bars on windows, is it the bars or the psychology of the person who has to feel they have to protect themselves in this high levels of chronic stress. It’s a very high dimensionality problem, and only machine learning with the right data collection can really say what are the social determinants of health. How do you force this complex problem and say, “What is it? Is it income levels? Is it your unwillingness to spend that extra $20 for a vaccination, flu vaccination, or is it something else? Are you avoiding visits because you’re afraid of copay and it’s an income problem? Is it the loving care of a family that Dean Ornish would argue for? Is it changing gene expression because your environment is more or less meditative as Deepak Chopra?”

Vinod Khosla:

This is all signs we’ve hand waved our way into saying, “Hey, food is the best medicine, but we don’t know what food does, parsing it into details. So, the FDA database on food has 9,000 foods characterized, if you look at FDA labels, about 150, 147 components to be precise. There’s a guy called Barabási doing network science at Northeastern. He’s found about 28,000 components of just the first 1,000 foods he’s analyzed. So we have a project with him called the Foodome to do a full network map of food, and interestingly people will love this. We can hand wave and say, processed food is bad, he can actually prove it. The ratio of zinc and potassium and other components of any food follows a curve. When you process it, certain components become outliers in this map.

So you can tell the degree of processing or ultra processing or not, of food, by doing a mass spec analysis of all 28,000 compounds. There’s 100s of versions of flavonoids or polyphenols, and they make a difference. Then you can turn this hand waving about nutrition of food into a science of food, and we have a project to do that and hopefully in two or three years we’ll have better results to show. And Greg talked about doing the same thing with your proteome.

Toby Cosgrove:

Yes.

Vinod Khosla:

And we’re also doing it with your metabolome and your lipidome and your methylome and your … I can go on.

Toby Cosgrove:

A lot of omes.

Vinod Khosla:

Lot of omes.

Toby Cosgrove:

So let me … I want to change the topic again and I’ve got two things more that I really want to ask you. First of all, how do you decide as a venture capitalist who you’re going to invest with? Is it the idea? Is it the people, is it the timing, what is it? Because I’m sure a lot of people here have gotten an idea that they’d like to get some capital from.

Vinod Khosla:

Well, first it’s highly uncertain and speculative. So, if I’m right 20% of the time I do really well. I’m very comfortable with the ambiguity of being wrong 80% of the time. Large institutions can’t do that, which is also a reason large institutions don’t innovate as much. There’s a 100 reasons something can fail, something can fail because the science failed, something can fail because its clinical implementation failed, something can fail because its organizational impact failed. 

Many, many years ago we did a procedure called heartport that we’ve talked about. Instead of spending weeks in the hospital for open heart surgery patient like a valve replacement, it was a less invasive approach that let the patient walk out in two days. It failed because it dramatically reduced the revenue for cardiac surgeons, that really was the reason, if you looked at why a Heartport adoption rate was so slow. That then becomes … I always look at, is it a large enough innovation to impact patient care? Personally I am a little bit different, I don’t start with CPT codes, I don’t look at CPT codes. I start with the level of change and impact on patient health.

Toby Cosgrove:

Okay, now wait a minute, level of change. So are you more after investing something that has a very high level of change or something that is going to be easier to change?

Vinod Khosla:

I much prefer very high levels of change. So, this is probably a good point to say, there are enough innovative healthcare organizations. So if it is incremental change, the organizational infrastructure of large organizations actually does better at incremental change than a startup does.

Toby Cosgrove:

Yes.

Vinod Khosla:

A startup has too many disadvantages, and so if it can be adapted as an assistant for radiologists to better read and reduce the error rate on radiology graphs, a perfectly reasonable thing for a larger organization to leverage some startup that’s doing the algorithms, incremental change like that is pretty easy. If you sort of say, and I’ll give you an example, we’re reducing a one hour cardiac MRI, which is not done in most MRI centers because it’s too complex, a 100 or 200 button presses on a windows like interface on an MRI machine. We’ve reduced it to one click and 15 minutes for much more data content than a standard cardiac MRI, and we don’t need cardiac MRI technicians. That kind of change is best done outside the system.

Vinod Khosla:

So, it depends if it’s large change, you’re better off starting and controlling the process and not requiring others to adapt it. And especially the AI driven changes where the AI is superior to human judgment, you can’t have human doctors say, “Oh, I trust this more than I trust myself, or that I’m willing to change my workflow to accommodate it.” So what do you do? You build it outside the system. And no question we can provide far better mental health care at $2 per patient or per member per month than Magma can at a much higher price point. When we do that, our peak load doesn’t start till after 8:00 PM and continues till 3:00 AM, mental health care when you need it, not when you can schedule an appointment with a psychiatrist in three weeks.

Vinod Khosla:

Same thing with cardiac care on a live call, you can do it and our average patient is taking more than 10 average ECGs a month. You can’t do that within a healthcare system. Not only that, we have their voice imprint so we know what their voice is saying, if they have a comorbidity for depression at that time or a manic depressive episode at the time, they are feeling something in their heart. Did they just finish exercising? Did they just eat a heavy meal? We have all that data, not when you can schedule an ECG in three weeks when your condition may or may not exist. So that kind of broad scale data collection is much more important to this new model, and I think it’s best done outside the system.

Toby Cosgrove:

Okay. So I want to give a little fodder to the group here. Tell us your favorite of your investments in healthcare, your favorite five.

Vinod Khosla:

My favorite five, like asking me to pick my children?

Toby Cosgrove:

Yes, I am.

Vinod Khosla:

I’ll think of the top of my head. The annual physical has been repeatedly, both in studies and meta studies proven to be totally worthless, expense with no benefit, maybe negative benefit. You can reinvent it, I’d recommend trying an annual physical called Q.Bio, full body MRI, which will soon be less than 10 minutes in the next year or two for a full body MRI with a lot more information. And they’re using old machines and new physics, which seems odd, but that’s what they’re doing. They can use a GE magnet, but do new physics with it and not be just restrained to T1 T2 pulses.

Toby Cosgrove:

Good. 

Vinod Khosla:

And computations-

Toby Cosgrove:

Do you want to tell us the name of the company?

Vinod Khosla:

It’s called Q.Bio.

Toby Cosgrove:

Q.Bio?

Vinod Khosla:

Yeah. DiscernDx can track your health care along the lines Craig was talking about 10,000 biomarkers for under a 100 bucks per dried blood spot, including shipping of kits, blood samples, everything. Right?

Toby Cosgrove:

Okay.

Vinod Khosla:

You can do that.

Toby Cosgrove:

What’s the name?

Vinod Khosla:

DiscernDX

Toby Cosgrove:

Okay, that’s two.

Vinod Khosla:

Okay. Microbots, about the width of your hair, 300 or 400 microns untethered that can travel anywhere in your body, magnetically guided from the outside. You can watch them move, so they could go up your spinal cord, deliver a drug at the nerve that’s causing you pain instead of washing your whole body with opioids or travel into your brain, pick a brain cancer sample, bring it back out and have no tether and very little damage because of that.

Toby Cosgrove:

And the name is?

Vinod Khosla:

The company … I don’t know if they have announced their name yet.

Toby Cosgrove:

Okay.

Vinod Khosla:

They’re still only in animals, not in humans yet.

Toby Cosgrove:

That’s very exciting.

Vinod Khosla:

All right?

Toby Cosgrove:

Okay, four.

Vinod Khosla:

Fourth. Both in ultrasound and I will cluster this as a category because I think this is important, we have replaced the technician for either cardiac ultrasound or other ultrasound or MRI machines. So, what do you do? You change the cost of an ultrasound by 80-90%. If you do that in a primary care visit, you should do an ultrasound, you shouldn’t be doing a physical exam, because many papers to establish much more data can be done in an ultrasound and, you can Google this, then a physical exam that a primary care physician can do with their hands.

Toby Cosgrove:

I’ve been telling cardiologists for years that the heart is a pump, not a music box, and get rid of that stethoscope.

Vinod Khosla:

Yeah. The last example I’ll give you…

Toby Cosgrove:

You didn’t give him the name.

Vinod Khosla:

Oh, HeartVista does cardiac MRI, self-guided, so autonomous cars instead of driving a car, you drive a machine, you don’t stop in between. You can cut the time from more than an hour for-

Toby Cosgrove:

So, what’s the name of the company?

Vinod Khosla:

It’s called HeartVista.

Toby Cosgrove:

Okay, great.

Vinod Khosla:

And the ultrasound machine company is called Caption Health. Last one is my personal favorite. I wrote about it eight years ago when I said Do We Need Doctors?, four years ago when I detailed the 100 page thesis on Transforming Medicine, and I got my son two years ago to start doing this, which is build AI-driven, free primary care for the world. The only way to do free primary care for all 7 billion people on this planet is through AI. So my son’s working on that, they’ve just announced in California. Unfortunately, when somebody needs a prescription, they have to have a California physician do the final prescription, but the cost per session will approach $3 or $4, and even in $3 or $4, the biggest component of cost by the way, is medical malpractice insurance per session. If you do it at $4, $2 goes to medical malpractice and $2 to everything else. So, I’ve given you five, I could probably give you a dozen more.

Toby Cosgrove:

All right, all right. So, you gave me the five, you didn’t tell us the name of your son’s company.

Vinod Khosla:

It’s called Curai, C-U-R-A-I.

Toby Cosgrove:

Good.

Vinod Khosla:

But look, primary care should be accessible 24/7, anywhere in the world at any time without needing to go to urgent care or an emergency department.

Toby Cosgrove:

At home.

Vinod Khosla:

At home, over text messaging because when it’s text messaging, telemedicine is really bad for AI learning. When it’s text messaging, you can structure the messaging to maximize the rate of learning for the AI. And what we found, and this is really interesting, literally there’s an exponential curve of the percentage of the text messaging dialogue that you can automate. So the way they do this, and this is the way to do it, the patient only talks to the AI, the AI only talks to the physician, and they have 50 physicians in India, and at the right time they can bounce to a position in California to do the prescription, but the physician only talks to the patient.

So every text message that AI generates, sort of your suggested reply in Google mail, the physician either forwards it to the patient, changes it and forwards it or discards it, whichever of those three it is, it’s a learning opportunity and the percentage that the physician forwards without change is measured every month to make sure they’re making progress. This is the way to build AI systems with humans in the loop training the AI and eventually it’ll get to 95% AI does and 5% humans do.

Toby Cosgrove:

Vinod, I know that you and I can carry on this conversation for the next two hours and I’m sure that the audience would probably be champing for more. I can’t thank you enough. You have stimulated all of us enormously and I enjoyed very much sharing your thoughts with us.

Vinod Khosla:

Well, it’s great to be here and talking to an audience.

Toby Cosgrove:

Thank you very much. Thank you sir.

Vinod Khosla:

Thanks.