a16z Podcast - a16z Podcast: Health Data -- A Feedback Loop for Humanity

Episode Date: December 5, 2016

"We live in a world where we use millions of variables to predict which ad you're going to click on. Whether or not you deserve to get a loan. What movie you might watch next. But when it comes to our bodies and even serious diseases, we want to reduce things down to just one or two variables." It's insane that we actually collect so little data about our bodies. The modern day physical is downright spartan in what it captures, not to mention that we're using 200-year old tools to capture that very limited data. Which is why we need to borrow from other domains of science and data and apply that to our bodies, in more ways than one, argues Q founder and CEO Jeffrey Kaditz with a16z bio fund general partner Vijay Pande (in conversation with Sonal Chokshi) on this episode of the a16z Podcast. But how do we get there? What would data "rights" look like -- and could we possibly donate data much like we currently donate organs? And for catching diseases like prostate or breast cancer early, how can we use data captured over multiple points in time -- something not really done right now in medicine -- to be more predictive, sensitive, and specific beyond so-called "representative" population samples? What IS a 'diagnostic', really, anyway? The views expressed here are those of the individual AH Capital Management, L.L.C. (“a16z”) personnel quoted and are not the views of a16z or its affiliates. Certain information contained in here has been obtained from third-party sources, including from portfolio companies of funds managed by a16z. While taken from sources believed to be reliable, a16z has not independently verified such information and makes no representations about the enduring accuracy of the information or its appropriateness for a given situation. This content is provided for informational purposes only, and should not be relied upon as legal, business, investment, or tax advice. You should consult your own advisers as to those matters. References to any securities or digital assets are for illustrative purposes only, and do not constitute an investment recommendation or offer to provide investment advisory services. Furthermore, this content is not directed at nor intended for use by any investors or prospective investors, and may not under any circumstances be relied upon when making a decision to invest in any fund managed by a16z. (An offering to invest in an a16z fund will be made only by the private placement memorandum, subscription agreement, and other relevant documentation of any such fund and should be read in their entirety.) Any investments or portfolio companies mentioned, referred to, or described are not representative of all investments in vehicles managed by a16z, and there can be no assurance that the investments will be profitable or that other investments made in the future will have similar characteristics or results. A list of investments made by funds managed by Andreessen Horowitz (excluding investments and certain publicly traded cryptocurrencies/ digital assets for which the issuer has not provided permission for a16z to disclose publicly) is available at https://a16z.com/investments/. Charts and graphs provided within are for informational purposes solely and should not be relied upon when making any investment decision. Past performance is not indicative of future results. The content speaks only as of the date indicated. Any projections, estimates, forecasts, targets, prospects, and/or opinions expressed in these materials are subject to change without notice and may differ or be contrary to opinions expressed by others. Please see https://a16z.com/disclosures for additional important information.

The content here is for informational purposes only, should not be taken as legal business tax or investment advice or be used to evaluate any investment or security and is not directed at any investors or potential investors in any A16Z fund. For more details, please see A16Z.com slash disclosures. Hi, everyone. Welcome to the A6 and Z podcast. I'm Sonal and I'm here today with our general partner for our bio friend, Vijay, and we have a special guest joining us. We're just really talking a lot about machines. learning and bio and all the trends in computer science meets bio right now, which is fun. I'm excited. And joining us today, we have Jeff Kaititz, who is the co-founder of a company called Q. The website is Q.Bio. They're building tech that measures, digitizes, and simulates human physiology. So first of all, how the heck do you guys do that?

And why does that even matter? Well, I think we live in an era where the amount of information we actually collect about our physiology is, like, exceedingly small. if you think about what the annual physical consists of. And actually, if you ask most physicians and most research, there is actually no real correlation between outcomes and consistently doing annual physicals. But at the same time, look at something like dental care, which whether you're sick or not, your mouth is sick or not,

twice a year you go and they measure a set of semi-quantitative standardized metrics. And as a result, we have these longitudinal profiles on hundreds of millions of Americans. Nobody goes to dentists twice a year for the record. Is that really true? I don't think so. Do you go Dennis Wraising? I try to. Well, you're lying.

I don't believe you. I do try to. It might be 1.7. Okay. But even if it was one, you know, I think if you look at the trends by most metrics, even adjusted for afflation, the cost of dental care is staying flat or going down. But at the same time, the quality of our dental health is going up, which means there's an example of a system that is very personalized.

It's very preventative. And it's really just based on the simple. idea of tracking longitudinal changes in some standardized set of metrics over time in order to make forecasts about the general trajectory of your oral health. You know, that's a great point because you think about it, you go to a dentist, you get x-rays and you get them, you know, once every few years and you have this record over time. And so you can see things develop where maybe it starts off small and you don't know what it is, but you can see it change.

And that added dimension of time, I think could be such a huge advantage. I think we want to reinvent health care to be preventative. We believe that healthcare can be reinvented so that it gets better and cheaper over time. And I think our vision for the world is really one where no one dies of a treatable disease. In the scope of what is treatable is actually increasing, in the scope of what is diagnosable is increasing. So we think that if we have the technology to treat something, no one should die from it. In medicine, you know, we talk about diagnostics having a specific sense. sensitivity and specificity.

And by diagnostics, you mean just test that measure? Well, that's actually, I think, an interesting question because we have specific opinions about what a diagnostic is. But in general, a diagnostic is a predictive model. Okay. Right? And when we talk about machine learning and computer science, like effectively, you do a clinical study, you come up with a variable and you say, this variable has some predictive power.

And the way the medical community communicates this is in terms of sensitivity and specificity and that tells you something about its true positive, false positive, false negative rates. But typically in practice, the way that we apply that those clinical studies is to take a point measurement, typically if you're symptomatic and then compare it to some clinical study that was done potentially a long time ago on a quote-unquote representative population. And if you were above or below some threshold, some decision is made. And that's what the sensitivity and specificity is based on. The reality is that we have a lot of diagnostics that are recommended not to be used regularly because they don't have great sensitivity

specificity as a point measurement. PSA is a perfect example of this, which might have in the... What's PSA? Prostate antigen. It is the gold standard test for determining if a man has prostate cancer, but it has a very high false positive rate. Now, you know, I think if you look at this as a time series, right? And by time series, you mean like more than one point in time.

Right. Then you're effectively understanding. you're really looking at changes in time then. And for us, you know, looking at changes in time is what personalized medicine is about. There's a huge focus on genomics. And we think genomics is important. But the idea of a representative population, I think is fundamentally flawed, especially it's kind of ironic in the age where you're embracing the idea that everybody has this unique genetic code. But the reality is that everybody's evolution of their physiology is unique to them, even if you're a twin.

Even if two people have the same genome, they actually won't necessarily express the same phenotypes, which to us means that there's more information encoded in the evolution of your physiological state, and the trajectories in your physiological state, then your genetic code alone. So that effectively, from our perspective, that means that human physiology is a long-tail distribution. So your genetic code cannot be the kind of silver bullet that is magically going to bring us to the age of personalized medicine. For us, personalized medicine is much more about understanding. the history of your physiology and how it changes over time and how doctors can use that to interpret that based on, you know, you as an individual. And then your genome can add extra information about how to interpret, you know, whether you're not, you have a high level of some biomarker based on a genetic variant you might have. Do you think doctors, to some

extent, are doing this already? We talk to more and more doctors who, you know, who, and honestly patients who say, you know, we would love to do this. There's a few problems. And the biggest one is the fundamental thing that governs our healthcare system right now is based on retroactively looking actuarial models. And as healthcare costs increase, that means that next year, in order to cover the population, premiums have to increase. Rather than thinking about how much does it cost to take care of a specific population, because last year, over the last 10 years, they've gotten sick at this rate, and it costs this much for these drugs and these drugs are going up. And so look at it more like, well, we know we have therapeutics for these certain diseases.

that kill some percentage of people, what would we have to measure and how sensitive would those measurements have to be and how predictive would the models have to be in order for us to save money in the long run rather than cut costs in the short run? This is where the healthcare reimbursement model is kind of really the key thing. When you reimburse for services, you only think about doing that when you have a real reason. And especially it puts the providers in this situation where they have to all. argue why this service is important to get the payers to do it. And for something where you have a healthy patient, it may give the impression that you're doing services that are not necessary.

And the system is, in the old model, is really kind of stacked against this. In the new model where you're trying to develop and give the best value, that's where actually doing this could make a ton of sense. What's the name of the new model? So, you know, what's come up with the ACA and other modifications to healthcare laws where being value-based versus just fee-for-service. based. It actually will create the opportunity where this actually makes a ton of sense, because the total value to the patient, the value to the payer is in making sure people don't get these diseases, which are very expensive. And I would add, interestingly enough, almost an irony to the idea of value-based care is if you're going to have a value-based care model, if you're not

monitoring the population continuously and regularly, you really don't have a way to know if somebody's getting healthier or not. So unless we have a way to quantify what it means to get healthier, we can't even develop a metric for what value-based care is. And so I think that not only does this make sense in terms of preventing illness and catching it at early stages, if we want to shift to a value-based model, I think it's required in order for us to develop metrics that would allow us to do that. I'm hearing you guys say, in reality right now, we don't really have a way of really capturing a person's longitudinal trajectory. Not with clinical data.

Like, there's obviously an explosion in wearables, but, you know, that obviously isn't necessarily always clinical quality data. Or there's some patchiness. I mean, you get a test here and there, but it's just very anecdotal. And what's in the record? Is that because there's different doctors taking the measurements?

Or just you don't get every test on it all the time. Right. I mean, you just get it when you're symptomatic typically. And that's sort of when it's already potentially getting to be late. Yeah, I mean, most of the most lethal. diseases, by the time you're symptomatic, you're in an advanced stage. So if we wanted to prevent them. That's a theme that comes up over and over again. I can't say that I know anybody who doesn't have some story of some relative or some friend

going in for some unrelated test and finding out they had some terrible thing wrong with them. And if they're lucky, it was caught early. But if they weren't, it was caught late. And so I think one of the questions to ask is if we have the technology to detect these things early and we just sometimes get lucky for unrelated reasons and catch it, why wouldn't we figure out a way to use that more frequently? And I think the answer actually is costs, right? And false positive problems. Two aspects of cost, right? So cost is one thing and false positives or another. Just even if it didn't have the false positives, the cost could be quite, it seems like it could be high. Let's talk about that. So let's first of all define what a false positive is and why it matters.

I remember them from my days of psychology studies. Yeah. Well, yeah, so the issue for a lot of cases is that you get a false positive, you know, where the doctor thinks you have the disease, but you really don't. I would even further segment it to two levels where there's the risk once you have a false positive of having an adverse outcome because of the false positive. Right, because of that. Because a lot of times you can have a false positive and then do another non-invasive test, which eliminates the need to do something invasive. The real problem is a subset of the false positives, which is, you know, depending on, you know, the pathology. and the diagnostic and what the standard is for care, it's different. And so when we talk about the cost of a false positive, you know, there's the cost of any

incremental test. And that's one thing. I think that part is actually less concerning than the part that where a person's health is negatively affected because of the false positive. A false positive in terms of prostate cancer can mean a biopsy that is unnecessary, which could damage, you know. your prostate. Same with breast cancer and yeah. You know, this is where time is actually really interesting because if you're looking at PSA one shot, you may or might not do well compared to

the population. But if you have the time series data, you know, that looks like it could be useful for improving the accuracy, reducing the pulse positive, right? Yeah. And, you know, the reality is that we don't think it's novel. Actually, we would argue that if you look at almost any discovery in any scientific discipline, the approach almost built, defined into the scientific method is you have a system, you don't understand, you have the ability to measure some parameters of the system, you regularly measure it, you build a model, you see if that model predicts the next part of the time series. If it doesn't, you measure more, you find a better model, whatever it is, and you iteratively get better and better and better, and that's how you

eliminate false positives. Well, a classic example playing out right now, and this is actually a very heated debate among a lot of all kinds of scientists, doctors, and various associations, the recommendations for women in mammograms? I think one of the issues with the tests on mammograms is the data suggests that it has too many false pauses and that you think about both the cost to the system and just the cost to these women, that they think they have breast cancer, but they don't. That's been a huge motivation to deprecate the test for only women that are older. Right.

But this is the opportunity where having an improvement on the test's accuracy would be the key thing. Because if the test isn't accurate enough, then we get into this problem where the medical system will naturally, and I think correctly, suggest that it doesn't make sense to get the data. And this is maybe the irony. The irony is that if you think of this test from a one-off point of view, it may look less useful. But if you think of it from a longitudinal point of view or longitudinal in the context of many other tests, suddenly now this is actually potentially useful information. But that's perhaps the case that has to be made and so on. And so I I think that's the challenge that lays ahead.

One of the issues I would just point out is mammograms aren't completely non-invasive. Yeah, exactly. There is some radiation. And so that's an example of something that I think would potentially be dangerous to recommend doing using like a time series to increase the sensitivity and specificity. You know, I don't think that's necessarily true for all of the imaging modalities. But I think that is important. Going back to thinking about it from a data science perspective and, you know, built a number of businesses based on data science. where you do these analysis of the predictive power or information value, you might say, of a single variable.

And I can tell you, like, when you talk about, and I don't have the exact number with me, but if mammograms have a predictive power or information value in terms of defining or predicting whether or not somebody has breast cancer with a 70% accuracy, no data science in the world would throw out a variable that gave you that amount of information. They would just say, well, we need to find the other ones. to combine with this that have some orthogonal information value that will bring us to a much higher level. And we don't have a way of doing that right now. Well, I would argue we do. We just don't do it. That's a really interesting point. I think we have to, you know, it's a combination of this idea of

how can you increase the sensitivity and specificity of diagnostics? And when I say, and I want to make one other point that's, I think, really critical. And one other issue we have with even the notion of a diagnostic in medicine is that, you know, when we talk about measuring a system and modeling it, that requires deconflating two ideas that currently kind of bound together in medicine, which is a just the simple idea of a diagnostic is actually two separate ideas that are jammed into one thing. And that is a measurement and then an analysis. Now, one of those things is immutable in some sense. It can be done once. It doesn't even matter if you do it wrong. You can't change it. It was just done. An analysis,

can be rerun. Like, right now in medicine, a lot of times we store the result of a diagnostic, but we don't keep the measurement, which actually prohibits us from going back and building models that, you know, in reusing the measurement. Right. To prevent. The store of data. Exactly. I come from the world of developmental psychology. And our bread and butter is longitudinal studies where you track a human being's progress over time. That's the only way you can get that information. People would actually do these things to try to control for genomics by using twin studies. And the interesting thing is that there are fields that are actually biological that do this too. So sometimes I hear, oh, that works in physics and chemistry where there's, you know,

there's not the biological or life element. But ecology for predicting population growth or deforestation. We use longitudinal data to predict what the climate is going to be in 100 years. We use it to predict what the weather is going to be like next week. For some reason, this has completely been, you know, ignored for the purpose of healthcare. And I think that's why, and when we look at success of something like dental care, we believe that it's possible if you step back and think about and kind of redesign the way healthcare is delivered, especially at the primary care level and the relationship between your primary care doctor and you, that healthcare can get better and cheaper over time. It's really about creating a positive information feedback loop, you know,

not to go, you know, harp on this dental care, because. I've got our teeth more than twice a year, apparently. There's dentists even are trained now and know that they are effectively the front line of health care. There are papers and research studies that show that there's a lot of diseases that somebody can have no symptoms for, like cardiac disease, that dentists get the first look at because it's correlated to gum disease. So the next question to ask is, well, how do we know that gum disease can be an early

indicator of cardiovascular disease? You know, we would argue that it's not because your gums are the the best biomarker for determining whether your heart is healthy or not, but we have the most longitudinal data on our mouths, which makes it easy to correlate to an outcome. So our question is, if that's not the first order biomarker, what are they? And why aren't we tracking those, right? And how many people could we save if we figure out what those first order biomarkers are? Because they're likely inside your body, not on your gums.

I mean, part of the other challenge, even beyond value versus fee for service, is really demonstrating that this will affect value. And this possibly is a chicken and egg problem, right? Because you need the data to prove the value and you need the sense of value to get the data. So how does one deal with that? I think that's more than anything an ethical issue. And, you know, that's something we spend a lot of time thinking about. We fundamentally believe that there are a growing number of people and there's, I think,

a lot of data that supports us. and not just patients, but actual doctors who believe that a patient has a lot more rights to information about their body. Right now, there is a, to some degree, there is a paternal position taken by, you know, organizations like the AMA that says there's certain information that is dangerous for you as a consumer to have about your own body. We find that a little bit ironic considering we also live in a country where you're allowed to drink and smoke and do things that definitely damage your health. but we're at the same time told that there's certain information we're afraid to give you because you might misuse it. So. Yeah, had a person's cigarette but not information. So when you talk about the chicken egg problem, we think a big part of that is changing what

a patient's rights are. People should own and control information about their own bodies and have a right to whatever information they want about it, especially if it can be non-invasively gathered and, you know, they're willing to pay for it. That also suggests the obvious opportunity. The ability for patients to better use the data and have something to do with the data, I think might be part of the solution. I'm still not clear, though, on where this data is coming from. How is this data going to get into the system, this longitudinal tracking? And how does it affect, say, a concrete case like prostate cancer?

Hopefully this isn't too long way to an answer, but just kind of take it one step back. This year is the 200 year anniversary of this stuff. Really? The 200, 200, 200 years old instrument that we are still using in almost the same form. of technology that is used in the annual physical. What? It's true. Let's pause on that for a minute.

So the 200-year-old stethical, first of all, the thing is 200 years old, that's shocking. And it's pretty much in the same form for the most part than as it was 200 years ago. There's been some modifications where you can now hook it to your iPhone. But otherwise, it's pretty much the same. The form factor, the style. And it is the most advanced piece of technology in a regular physical. I don't know if it's the most advanced. When the doctor puts the light, the LED light in your eye, that's probably slightly more advanced.

The frequency of light coming out of the LED is not the point, though. But in terms of an actual measurable thing. For the most part, like, those are kind of the cutting edge tools that we use to assess our general physical health. If you think about when that was, that we did not have the technology of the tools to non-invasively measure anything about our physiology. We didn't have the ability to look inside of our bodies at high resolution without. ripping you open. I have a feeling that the first dentist kind of said, well, how are we going to assess somebody's health? And they're saying, well, we can just open their mouth and look. You know, doctors didn't have that. When the physical was invented, doctors couldn't just say,

well, we're just going to open your chest up and look. Now we have an unbelievable set of technologies that allows us to look at our physiology at everything from an atomic scale to at varying degrees of scales. It's certainly this explosion inomics. Genomics is just the tip of the iceberg. It's the tip of the iceberg, actually, just in terms of information content, your genome only has about 10 to the 9th bits in it. We think your physiological state at a point time has probably 10 to the 18th more information than it, which is a million. And what is physiological state? There's conjecture that the way you define the complexity of a system is in terms of the number of bits it takes to represent its state. The complexity of the genome

is on the over of 10 to the 9th. You know, the best estimates we have for a complete representation of your physiological state at a point in time is that it's a million trillion times more information. It's like every voxel that describes you. Although you could argue it's really the diffs, right? So the genome, you know, maybe

there's a million base pairs on a billion in the, or actually millions between us and chimps, so even less between you and I. But then, and then the voxel difference might be less, but still it's huge. The voxel, you know, is like a 3D pixel. The 3D pixel, right, yeah. If you imagine you could break us down in our bodies into a whole bunch

3D pixels, what the information content of that is. And that's, you know, that's about how your arteries are getting clogged. That's about how, you know, you're getting tumors. That's about, you know, all these things. We now have genomics, transcriptomics, proteomics, you know, proteomics, metabolomics, microbiomics. You know, so you, you can imagine dividing your body up into, you know, as small volumes as you want and then say, well, I want to do a proteomics on just this point in space. And if you continue to squeeze the size down and down and down, you can start to imagine the amount of information that's contained in your body. And don't forget the fact that your genome, while it is relatively static, you know, and you do

have mutations in it, the environment that your genome is in are things like methylation, which actually change the way your genetic code is interpreted, but doesn't necessarily change your genetic code. All those things are part of your physiological state that accumulate over time. And we would argue that it really is part of the explanation so for why twins diverge. you are constantly interacting with your environment and accumulating complexity. We keep on thinking about DNA as being fixed, but our DNA is changing over our lifetime and that a little, little tags get put on that affects how the DNA behaves.

And this is where twins can start off with the same base DNA, but the methylation can change very much over time. Your point is that even with that methylation, there's just a physiological pattern that we express in our lives, that is more indicative of health in some ways. So, like, when we talk about 10 to the 9th bits in your genome, that doesn't include the complexity that's added by the number of ways methyl groups can attach to your DNA and how that is different in every cell in your body. And you have 5 trillion cells and then another 5 trillion that are microbiome. That's only the genomics. So you can start to see, I think, so if you actually think about the percentage of the complexity of our physiological state we measure on an annual basis, I don't think it's any surprise. that we have very poor ability to predict when we get...

But here's the key part, which connects it all together, which is that it seems like a lot of bits, but the fraction that's changing over time is relatively small. And so what you go from is what may look like a lot of data, but it might be a lot of noise. By having the change over time, suddenly this huge amount of bits goes down to be relatively small. I have a background in particle physics,

and that's, you know, what it was all about is it's everything's relative. There is no absolute. Doing a clinical study and coming up with an absolute threshold is inherently we know from a lot of other science is less sensitive than tracking deltas in the same system. And the reality is we're all different systems and we should probably be treated that way. I think there is also information about it in the population analysis, right? But I think it's the kind of combination of being able to statistically analyze a person's changes with respect to the entire population, but also within the context of. their past healthy self. And I think those are the kinds of things that we can do today to improve the

sensitive and specificity of these tests we have, and which hopefully would reduce, you know, if we had perfect tests, obviously, people wouldn't be worried about doing them as much as we wanted. And I think that that is, you know, a, you know, a, you know, a noble goal. But I think we definitely know how to set up systems that have this positive information feedback loop and get better and better. I mean, Facebook and Google, almost every modern business that's made, like large modern business in tech is based on the idea of creating a positive feedback loop, right, where the product gets better, the more people use it. And there's really

no reason health care should be the same way. So just concretely, it's a very simplified example. My hemoglobin is low. But there are other people who may have the same exact absolute level of hemoglobin. For them, it's, yeah, it's broadly speaking, low, but it's not as dramatically low as mine is because mine is significantly lower than its baseline that they've tracked over time because I've taken a ton of blood test over time. And that's sort of like a simplified way of thinking about it. Once you have a baseline down, you can get a sense that you are getting into some danger zone without the level needing to get high in some absolute level. I think that brings up the second part of this is so you can add time as another dimension to any specific biomarker

and increase the sensitivity and specificity. The other thing you can do is have multivariate models for pathology. It's kind of ironic that we live in a world where we use millions of variables to predict which ad you're going to click on, whether or not you deserve to get alone, right? What movie you might watch next? But we want to reduce things like cardiovascular disease to one or two variables. God, that's so true. It's a little insane that the most mundane and yes, still important in your day-to-day routine things in your lives have so much data going into them. And to your point multiple variables over time. So combining all these things, but for our physicals and our exams, it's like two variables and rudimentary tools. Your hemoglobe and being at a certain

level, it could be low for you. And there's an enormous number of reasons why that could be. I know the reasons. It might be measurement error. But the point is, is that we need to move beyond and take what has been tremendously successful in almost every other domain of science in terms of modeling, you know, measuring and iteratively modeling systems, we need to apply that to our bodies. I think that there's a little bit of reticence to do that because people think, oh, our bodies are so complex, you know, we'll never understand it. And I think we have to decouple our thinking part of our body from the physiological state, which is governed by the same laws of physics. In all of science. And also because it's complex is why we need all the data.

I think the separation of measurement and analysis is absolutely critical. And I think it's one of the reasons that I think we have issues. And I know there's a lot of company spending a lot of money throwing a lot of computational power at all of the medical data that we have. Or technically all the analyses we have. Exactly. But I think part of the problem is that we don't actually have the measurements, right? Even when you get your blood glucose measured, right, the output that that, little device gives you is an analysis. It doesn't store the actual sensor readings that

it used to compute that number. We would be better served as a society if we recorded the actual sensor reading. And some people might say, oh, it's expensive to destroy much information. And if there's one thing that's like I'm confident of is it's that storage is getting freer every day. Yeah, this is a key thing. But our analysis gets more sophisticated. By the way, just a really interesting example of this. I read this amazing article on Lanset and how for the last like 30 years they've been you know obviously collecting all this GPS data all the satellite data but the sensors and they have all the information encoded from these sensors but they didn't have the technology to be able to read that that data yet and so now they're going

back and trying to figure out how to back analyze 30 year old data and because they stored the information in its original form and the sensors they're able to now go back into these weird things like bake the data like literally bake it to like figure out Like, read it out. At the end of the day, you know, if we can do this and transform the kind of way we capture information in the health care system, what we really need at the, you know, a lot of people are organ donors. What we need is data donors, right? And we need ways that I can put on my license. I am anonymously donating my health outcome from whatever I die from, as well as the longitudinal profiling of myself.

Because if I donate an organ, I can save one person's life. If I donate my data, I can help save the lives of every person who ever is born ever after me. And that's a network effect. That's how you as a person can improve the health of every generation after you. Yeah, that's amazing. And how do you then control for the fact that the sample, because you're not getting to a population yet, is so small. Like you might have a few random people who might take on some project like donating their data. How many people do you need for it to get interesting?

Exactly. That's a really good question. There will be a lot of debate on what that is. I mean, just to give orders magnitude, my gut feeling is a thousand would still be a huge data set. I mean, you compare that to the people who donate organs. It's a relatively small fraction. You know, I mean, a million would be a dream, but like a thousand, 10,000.

Or think about the people who donate their bodies to be cadavers. What if those same people, I would much rather donate my data. I feel a little bit weird about doing my body. I would love to see the government sponsor a public data repository where people could donate data and they could put like you go to the DMV and you'd say are you a data donor and your health care data if you die is released heavily anonymized. We're talking about a feedback loop for humanity. It's amazing. Hopefully. Thank you for joining the A6 and Z podcast.

Yeah, thank you. Thanks.