The value of biosensing for preventive medicine

SiPhox Health founder Michael Dubrovsky discusses personalized health data, the value of continuous metrics, and the future of healthcare.


Article highlights

  • Traditional blood tests were designed for an annual doctor's visit, but more frequent at-home testing of biomarkers like hormones can provide personalized data to optimize health.
  • New technologies like silicon photonic chips can miniaturize blood analyzers to provide rapid at-home testing that was previously only available in central labs.
  • Direct-to-consumer health products that provide value to individuals can create a feedback loop to improve preventative care through viral distribution.
  • The healthcare industry lacks competition and a financial incentive structure to drive rapid innovation compared to markets like semiconductors and aerospace.
  • Continuous biomarker monitoring could eventually allow personalized interventions based on data pooled from millions of people rather than trial and error experiments.

Michael Dubrovsky is a co-founder of SiPhox Health, a health-tech company that went through Y Combinator in the summer of 2020. 

SiPhox offers an at-home product that measures 17 different biomarkers. We know glucose is an important molecule when it comes to measuring and monitoring metabolic health. But it is only one molecule. 

Michael and Levels’ co-founder Josh Clemente discussed why we want to monitor all these different biomarkers. What’s the point of biosensing? Which molecules should we be monitoring? What are the implications of diet, sleep, exercise, stress, environmental factors—all these things that can affect our biomarkers?

Measure What Matters

Josh Clemente: What is so important about biosensing, what should we be measuring, and why is monitoring multiple molecules so interesting?

Michael Dubrovsky: That’s a big question. It’s worth going back and asking, “Why did people start doing this?” It’s probably pretty recently that people seriously decided it’s useful to start measuring things in the blood. I’m actually not sure what the first thing was—maybe blood typing for blood transfusions, things like that. 

They used to look at the erythrocyte sedimentation rate. If you take the blood out of the patient, how long does it take for it to settle out? That length of time is proportionate to your inflammation a little bit. They used to do these types of things.

Ideally, you wouldn’t have to pay attention to your body at all. There are some people doing it for entertainment, and I’m one of those people. But most people probably don’t want to pay attention to it at all.

You’re taking this ape that’s supposed to sleep in a cave and probably wake up at sunrise and walk around all day and then fall asleep exhausted or something like that, and sitting it down in a chair indoors, and having it do digital marketing, say, for 12 hours a day. 

Because we’re operating completely out of the range our bodies are designed for, but also because we’re trying to get much more out of them—maybe we’re all trying to get the ideal outcome, not the average outcome—it starts to make sense to make measurements.

Josh Clemente: Why does it make sense to make measurements? What is it about measuring specifically what’s inside the body that gives us any additional information that might be useful?

Michael Dubrovsky: That’s a good question. The typical approach is more symptomatic. If nothing feels wrong, then everything’s fine. But over the last hundred years, as we’ve started collecting much more information about biomarkers in general, including blood biomarkers, we’ve found you can see things in the data long before they manifest as symptoms. Things are labile, and you can see this. You can see incredible things happening.

It might not be a good idea, but you can take hormones and become enormous, right? People are putting on 50 pounds of muscle by taking a tiny catalyst that completely changes their body. It’s not just bulk things like weight gain; it’s also these things that are present in very small quantities that can catalyze enormous changes in your body. Being able to monitor that can, first of all, allow you to know what’s going on and how your body’s responding. It can also allow you to make changes you want.

Josh Clemente: We’re playing with these levers that are manipulating these molecules circulating in our bodies every day, hundreds of times. We’re making all these decisions about what to eat, how to sleep. We’re exercising, we’re eating, we’re working, and the goal is to achieve some outcome. Meanwhile, we’re kind of ignoring the complexity of the human body. We’re flipping all these switches, but we aren’t actually measuring the state of our bodies, which, as far as we can tell, is the most complex system in the universe. 

By simply assuming we can control our health over long time periods, but not measuring our current state or how it’s changing, we’re really missing the forest for the trees around what it means to be healthy and what optimal might look like.

Your point about how we’re outside of our comfort zone as a species is absolutely right. Our environment has transformed 100x over the last hundred years, and yet many of our health-monitoring strategies have not. Most people’s health strategy is, wake up in the morning and look in the mirror and figure out if something’s totally wrong. We’re using eyes and ears to see if we’re really going off the rails. 

We can improve. We can take the same complex system sensing we have in our cars and in our airplanes and in our diagnostic machinery and apply them to the body. That’s maybe where we’re heading.

Michael Dubrovsky: That’s one side of the argument. If you take the devil’s-advocate side, you don’t really need to measure things that work. Take a building, for example. Parts of the building that never fail don’t really get tested. You put sensors where you expect to have problems. With wind turbines, you’re putting a sensor on the motor because that fails; it catches fire. There are all these epic videos of wind turbines spinning on fire. 

People used to get nutrient deficiencies that are now completely solved. In many ways, the food system is worse, but in some ways, it’s better. People aren’t getting goiters due to a deficiency of iodine. That doesn’t happen. You don’t need to monitor that. 

On the other hand, for more subtle things that haven’t been solved, that’s where you have to monitor and tweak and do all these things. It’s the gray area where we have solutions, but they’re not perfect. That’s where you have benefits for monitoring. If there’s no solution, knowing that something’s going wrong doesn’t really help you. But if there is a solution, but it’s subtle—there’s no one-size-fits-all solution, for example—that’s where monitoring is really beneficial.

But over the last hundred years, as we’ve started collecting much more information about biomarkers in general, including blood biomarkers, we’ve found you can see things in the data long before they manifest as symptoms.

That’s actually most of medicine. It’s mostly not that effective. That’s the problem. There are certain things that are super effective: if you break your leg, they can put it back together. But there are many—especially long-term things—that, if you look at the studies, work for only about 50% of people. That’s where monitoring can really help you see and tweak and actually get the 50% result that’s good, rather than being in the 50% that got no value out of something.

Josh Clemente: Even in cases where there isn’t a solution necessarily, we have to reframe our opinion around whether it’s worth knowing that, because there’s a situation where somebody doesn’t necessarily have a cure for a condition. For example, there are a number of hormonal deficiencies where the body just does not produce certain hormones. There isn’t necessarily a cure, but there are solutions that can manage those conditions. Or you can see the onset of them and make better preparations. You can set up a better quality of life if you understand what’s going on. 

We’re talking in generalities here. But what it comes down to is the reason we would want to measure what’s going on in the body is so that we can understand the health state and how it’s changing over time. It’s setting a baseline and understanding it in relatively high resolution. And then, like you said, in the areas likely to fail,  looking at markers and watching how those progress is a really powerful tool to know maintenance intervals and/or potential failure modes. 

For some of us, the cardiovascular system might be the first to go, and we can start to see the indicators of that relatively early. This is one area that the current medical system actually looks at, whereas for cognitive decline, we don’t really have markers of how that system starts to degrade. We use things like cognitive function as opposed to the biochemical signature of what underlies that to know what’s happening. 

It’s generally an argument for increasing the amount of information, but not just overall volume. Data’s not data, in this case. It’s actually looking in the areas where we know failure is possible and already playing with the levers with our decisions.

When—and How Often—You Should Test Depends on the  Biomarker

Josh Clemente: Measuring makes sense, but what about the frequency? How do you choose what to measure? How do you choose the frequency? 

In general, in the US, most people get blood work maybe once a year or once every other year. They get a panel of results and you get a PDF with some numbers printed on it. That is a frequency you can measure. Why should one measure more frequently than that, and what additional information does it unlock to do so?

Michael Dubrovsky: It’s useful to give some context. Before starting this company, I hadn’t had a blood test in five years—maybe six years. At the beginning, we weren’t sure we’d be building a blood-testing device. After we realized we’d be building it and we got past the first technical hurdle—we knew it could work, and could be built—I started thinking about what could be the biggest impact of something like this. 

I wanted to make it a little bit easier to do the same blood test that’s normally done. Let’s do that once a year. If you ask a hundred doctors, “What do you want from a blood-testing tool?”, they’ll say, “I want to do the yearly blood test while the person’s in my office so that I can just get the data and discuss it with them while they’re there, because people just do that blood test, and they disappear.” 

That’s considered the burning issue in healthcare, in terms of blood testing. Because many markers fluctuate frequently, if you’re testing them once a year, you don’t know if you’re looking at the noise from the fluctuation—depending on what happened to you that week or what time of day it was, that type of thing—or a real change.

Unless you have somebody who’s a functional-medicine doctor or you’re paying a lot fpr longevity care or something, doctors are typically just looking to see if you are out of range. Have you gone completely off the normal human range and do you need to start doing a bunch of tests to figure out what’s going on? Or do you need to completely change your life because this is really high risk? 

One of the reasons the frequency is so low is that, if they were responding to small changes, they might just be responding to the normal variation. You’re going to have variations based on season. You’re going to have variations based on the time of day. The peak for testosterone or cortisol levels is one hour after you wake up. If you show up, a half-hour could make a 10% difference. Doctors can’t respond to that because they can’t even time your blood test perfectly.

For all of these reasons, they’re taking a very blunt approach: “You’re within this general range, which is the 95% two-standard-deviation interval for the population. Maybe the population’s very unhealthy, but at least you’re within two standard deviations. Come back, we’ll check you again, and we’ll see what happens.” 

I’m not the first person to say this. This is a common complaint— hey just wait for you to become very unhealthy before doing anything. But which tests does this matter for? For many of the tests they do yearly, you don’t really expect to see changes. It doesn’t make sense to test some mariners, like blood counts, more than once a year.

The complete metabolic panel happens every year. It started with what they could test. You can take the person’s blood, put it on a microscope slide, and count the cells. People used to do this visually. There were technicians that would look into a microscope and count your cells and write the number down on a piece of paper. That’s the type of blood test they could do. They could measure very high-concentration stuff. 

A good example is CRP. CRP is C-reactive protein, an inflammation marker. It goes up when you have a bacterial infection, but normally it’s supposed to be very low. Until recently, all they had was the CRP test, which could only measure it at very high levels.

Then about 10, 20 years ago, a higher-sensitivity test came out. It’s called hsCRP. It’s really the same marker, but at a lower range. That’s when they started seeing that, actually, people are chronically elevated on this marker for no reason. It’s not as high as it would be if someone were super sick, but it’s much higher than it’s supposed to be in a healthy person. 

Even to this day, they don’t really address that part of the range. But it varies from marker to marker for what makes sense to measure frequently or even continuously. A lot of the ones in the yearly panel are relatively stable, and that might be why they’ve evolutionarily ended up there.

Josh Clemente: We should contrast the annual stable markers. You raised good points about most of the molecules that are circulating in our bodies. They’re a kind of combination of a half-life, meaning the molecule’s changing over time—degrading, essentially—and being replaced. 

You’ve got circadian rhythms, your day-and-night cycle. Let’s say people work second shifts. They might have two circadian rhythms that are overlapped. There can be this dysfunctional double circadian rhythm and variations on it. 

Then you’ve got the impact of decisions. My cortisol spike may be naturally circadian driven, with the peak at one hour after I wake up. But if I happen to do CrossFit 45 minutes after I wake up, maybe my unnatural cortisol peak ends up being an hour and 30 minutes after I wake up, meaning I have stacked two peaks on top of each other with this unnatural elevation.

If I want to understand how cortisol levels impact my quality of life—my mental health, my sleep—I know that sort of marker fluctuates on a 24-hour basis, and we are able to measure this. We measure it on our annual panel. 

But what do we unlock by measuring those more labile markers on a more continuous basis, which we don’t get otherwise? The body is a dynamic machine. We’re moving around, we’re making all these decisions, we’re putting it under stress. As such, the annual panel is designed to be taken in the lowest load condition: you’ve fasted, you’ve slept well. Right after you wake up, go in—no coffee, nothing—get your blood work. In many cases, that’s by far the best scenario. Why do we test that way and what would a continuous—or maybe not continuous, but higher-frequency—detection unlock for us?

Michael Dubrovsky: That’s an interesting question. If we’re going to make progress on the 120-year lifespan in general, or whatever the health span is now—80 years or something like that at best—we have to start figuring out what’s really going on. 

For example, if a drug or a procedure is going to get us to a 150-year lifespan, that’s not going to happen without measuring the effects of that drug because, as with any longevity drug, you won’t see any symptomatic effects for decades. It has to be in the biomarkers. There’s no way around it. 

But for most people, it’s probably the end use. They have an end goal—they want to sleep better or they want to lose weight or they want to prevent diseases X, Y, Z. Six out of 10 Americans have a chronic disease at this point. If you don’t have one now, there’s a good chance you’ll have one in the next 20 years, let’s say. 

There are 3,000 proteins in blood, and then there are also tens of thousands of metabolites. If you start counting metabolites, it goes really high. The good news is that it’s also over-determined. A lot of them are parts of chains. 

CRP is the inflammation protein, but it’s actually induced by what are called cytokines, which are signaling molecules that tell your body it needs to have higher inflammation. If you get poison ivy, what’s really happening is the oil from poison ivy is binding to proteins in your skin, which your body starts to recognize as foreign proteins, because they change shape from the binding of the oil. Then your T cells, which are immune-system cells, release cytokines into your blood or into your interstitial fluid locally and it just explodes. A bunch of inflammation is triggered this way.

If you’re measuring CRP, you’re actually measuring the cytokine levels from one day ago. You’re a day delayed. In a way, you don’t need to measure all 3,000 to get the data. In many cases, the systems are over-determined, and you can get a lot just from looking at time-series data. You can look at one marker over time, and it tells you something about what’s going on in the more complicated systems. 

There’s a zoo of things in your body, but they’re connected. We’ve learned a lot about them. Taking all that information, you can zero in on a couple of markers and get a lot out of that use case.

Josh Clemente: That’s a great callback to why one would measure more than one thing. You can start to take a few properly selected molecules and use them to interpret and act as proxies for a whole network of things. You don’t have to measure everything in the body to know what’s going on. You can find the high-leverage elements like CRP, which can then be used to interpret T cell activity and cytokine activity. 

There are these specific molecules—hormones are a really good example—which are really information-dense. Hormones are basically the messenger molecules to the distributed network of our cells. If you can interpret that signal—what a specific hormone’s concentrations are—you can kind of approximate what is going on in a complex system. Then you can even remeasure a byproduct of the cells to see if they followed the instruction set. That can tell you whether you have a central nervous system issue, where you’re not sending the right signals to produce the right hormones, or you have a cellular issue, where the cells are just not responding to the hormone signal. That’s a really powerful, high-quality amount of data that is interpreted from a few results. 

People Are Heterogeneous, and So Is Technology

Josh Clemente: How do we decide what to measure today? If we can measure this stuff now, why are we not measuring it in higher quantities already? What’s holding back a huge industry here?

Michael Dubrovsky: The big progress in blood testing happened in the late ’90s, and maybe early 2000s. If you talk to somebody who was working in blood testing 30 years ago, they’d tell you it’s not a precise science. 

One of the issues is that people are very heterogeneous. If you take a thousand people and test them using the same test, it’s actually a massive battle to have that test give the same result, or a correct result. Again, there are 11,000 things in the blood. All kinds of things could be going on.

The industry has gotten a lot better at dealing with that. The FDA decrees we have to be dealing with something now, and that then improves all the tests. But even to this day, the best instruments don’t perfectly agree with each other. It’s nothing like electrical engineering, where somebody sells a product that says five volts, and when you measure it with a voltmeter, it says five volts. That doesn’t exist in medicine. But within some error, that’s starting to be the case, especially for the more well-established targets.

Josh Clemente: When somebody goes to get a lab panel and measure their total cholesterol, what should they expect the realistic accuracy of that test is?

Michael Dubrovsky: We’ve done a ton of testing on this ourselves. I don’t have perfect proof of this, but the longer a test has been on the market—total cholesterol, for example, has been on the market for decades—the lower the coefficient of variability. If you send in your sample 10 times, you’re going to have a 5% difference.

But if you take a vitamin D test or a testosterone test, those tests have really not been offered for that long and at that high a volume. You have to take those with more of a grain of salt. If you have a change of 5%, you shouldn’t see that as a real change, especially if you’re making one or two measurements and you just see that they’re essentially equivalent. That’s my personal belief. This is not based on a big study or anything like that.

Josh Clemente: People might find that counterintuitive. They might think newer tests are going to be made with better technology, and be of much higher quality. What do you attribute that error to?

Michael Dubrovsky: People have done an amazing job of automating and scaling up relatively complex assays in what are called clinical analyzers, these giant robots that take in tubes of blood, and pull out fixed volumes. All the competitors have almost the same thing. They’re all running the same types of assays. 

There are three types: chemistry, immunoassays, and blood counts. With chemistry assays, you just mix a bunch of things up and shine a light through it and look at the change, typically in the absorbance or scattering. You’re monitoring a chemical reaction going on inside. 

There’s a zoo of things in your body, but they’re connected. We’ve learned a lot about them. Taking all that information, you can zero in on a couple of markers and get a lot out of that use case.

Immunoassays require proteins that recognize the target. Roche, which sells an analyzer, validated their assay for vitamin D against their previous model. They basically sent the FDA a document that said how their Roche Analyzer 1000 works exactly like the Roche Analyzer 900, but they never validated against the Abbott. 

If you send your sample to two labs and one is using the Abbott, and the other one’s using the Roche, I’ve been told that sophisticated doctors will always send your sample to the same lab. If they’re serious, they want to know the differences. They want it to always be tested on the same instrument.

It’s these types of things—these subtleties—that don’t completely harmonize. It takes years to find the differences because, again, people are heterogeneous. There might be a small difference between these two machines for only 1% of the population, but these are the types of things that get ferreted out over the course of a decade.

Josh Clemente: Process improvement and technology improvement—the actual underlying equipment matters. There’s this whole industry and machinery behind the simple process of going into the doctor’s office and getting blood pulled. That stuff vanishes somewhere. We don’t really know where it goes. Then a week or so later, you get a phone call and you have a PDF in your hand. 

Meager Progress, Theranos, and Looking Ahead

Josh Clemente: You had mentioned that doctors want to be able to provide results in real time, or at least in the same visit. Explain why there is currently such a long lag. Why is a blood tube being sent across the country to get a result and what’s actually happening to it?

Michael Dubrovsky: I’m by no means an industry veteran. It’s very fun to talk to the blood-testing industry veterans. It’s kind of like they’re sailors or something. It’s actually a pretty down-to-earth industry, and they’re definitely doing their best. In biotech, therapeutics are generally what make money. You can sell the same drug to a billion people. It’s like software: it’s incredibly profitable. Diagnostics and instruments in general are considered pretty low status because the margins are super thin and the equipment is difficult. 

The history of the industry is that people realized they wanted to have the whole test done in the doctor’s office. Everybody knew that. It’s kind of obvious. It’s like saying I want my iPhone to run ChatGPT locally. It was obvious to everybody, and there was a vacuum for somebody to say they could do it. That’s what Theranos did. 

To say you’re doing it was the minimum viable product. Of course, nobody in the industry eventually believed them because they couldn’t show anything. But that was enough to do years of fundraising and everything. That killed a lot of other companies that were working on real solutions because investors got very scared of the space. But some of these companies survived.

I got into this only in early 2020, so I wasn’t around for the whole fallout of that. That’s one of the reasons we don’t have these things. The other reason is that if doctors need a single result that you can’t produce on your device, they’re going to send a tube of blood to the central lab because at the central lab, they have three different tools: the blood count, the chemistry, and the immunoassay, that do all the blood tests they want. 

Basically, they’re just sending the tubes there. They don’t care; they just want the results. They don’t care what machine it’s run on. People have struggled to build a machine that will do everything they want so that they don’t have to send a tube, because as soon as they want one result that is not possible, that’s it. But there are a few companies like Truvian, Genalyte, and Vital Bio that are trying to package up the complete blood panel and put it into the doctor’s office. 

This might be a hot take, but it’s taking so long to deliver this that you might as well just go to the next thing. By the time they deliver this, it’s going to make money. Whoever cracks this is going to make a lot of money. But it’s not going to have as big of an impact as people were hoping because medicine is changing. 

This whole doctor’s visit thing—the yearly blood test—is starting to look stale. But that’s the situation. It’s delayed partially because of Theranos, and partially because it’s a poorly funded, low-status industry in general.

Josh Clemente: Just for background context, Theranos is the company that claimed they could do a litany of tests with a single drop of blood, right there in a microwave-size box. What they were trying to replicate—and what this whole industry you just described is trying to replicate—is the central-laboratory industry. The goal is to replicate, in the office, those three giant analyzers in a room that take blood tubes and can break them down and get the whole menu of tests a doctor’s interested in.

You’re saying you might as well go to the next thing because it’s taking so long to replicate that entire central-laboratory process in a small box that can be put in a doctor’s office. Things have evolved. 

It’s pretty crazy that a single example like Theranos can have as much of an effect as you indicated. What went wrong there and what lessons are we going to have to take from the Theranos story when we’re talking about, for example, what you’re working on next?

Michael Dubrovsky: I don’t really know what went wrong. What’s interesting is that once you spend some time in the industry, you realize lots of people worked at Theranos. You start meeting people who worked there. I’ve interviewed people who worked there at relatively senior positions. It seems they were doing some kind of work in a silo, which was going fine. But there was some kind of integration problem.

This is something my co-founder says, and it’s a really good point: Every hard-tech industry has a Theranos, but Theranos got very famous because they gave incorrect results. If they had never delivered anything, they probably wouldn’t be very famous. But they were actually giving critical results that were wrong to people, and they did all these things. But if they had just raised a billion dollars, spent it, and disappeared—there are lidar companies that did that. You can give a bunch of examples.

Josh Clemente: Just another Juicero.

Michael Dubrovsky: The Juicero founder is not being covered in the New York Times every year, right? With any deep-tech product, it’s so difficult for investors. But what they did wrong fundamentally was just trying to chase the obvious, impossible problem people want, versus being realistic and saying, “Today it takes three enormous instruments to do this. What can be done with the technology we have or can develop in five years while people are still paying attention?” If you collect a good team, and give them five years of runway where they’re really focused, what can be done in that time? That’s really when people deliver.

A good example of a company that does a one-drop, at-home test is Athelas. They also went through YC and they’re FDA cleared now on the market for blood counts. For cancer patients and some other use cases, you really need to keep track of blood counts. Athelas has an FDA-cleared device that does that at home. This has been unbundled. Basically, instead of doing the whole thing having no idea how to do it, people are focused like; they’re dedicated to solving this piece of it for a particular niche where it actually works.

Josh Clemente: What is that next thing you were hinting at that the industry should start to shift its focus toward?

Michael Dubrovsky: This is going to happen in different parts. My family’s from the Soviet Union, where healthcare was socialized. Both my grandparents were doctors. My grandmother ran a pediatric hospital. I’m pretty familiar with the system. It’s interesting that you can spend nearly 20% of the American GDP on healthcare and get what you get. What we get is really bad for the amount of money we’re spending, considering what the Soviet Union was able to achieve—and what many other countries today are able to achieve—with fewer resources.

If you look at it from that perspective, the chances of doing something super fundamental to fix the existing system are very low, but there are all these other things happening that are trying to circumvent it, or just improve little parts of it. That’s working out great. Telemedicine and all kinds of gray-area services, which represent wellness bleeding into healthcare, seem to be working. People are actually getting results from them. They’re really low cost because they’re self paid. 

If you collect a good team, and give them five years of runway where they’re really focused, what can be done in that time? That’s really when people deliver.

There’s a lot of progress there, which can eventually be used by the healthcare industry. And it eventually does—they adopt things at work. It benefits the main healthcare industry over time. It’s just hard for it to start there.

But what do I think is the next thing? We don’t really know how to cure chronic diseases. If we knew how to cure them, it would be a moot point. Prevention is probably the most interesting thing because of the difficulty of curing and managing chronic disease. Management is also interesting, because we can do much better. But prevention is where a lot of the value is, and that’s also where there’s a lot of room for innovation, because it just hasn’t been touched.

Josh Clemente: What’s the next thing in detection?

Michael Dubrovsky: Focusing on doing the complete metabolic panel is a strategic error in my opinion. That panel is designed for that yearly visit, and the yearly visit in itself is not that valuable. 

I went to the doctor recently just to try it out, because I usually don’t do the yearly visit. I didn’t even see a doctor. I saw a nurse. She never talked to me. She gave me a form that asked me if I was depressed. Then she said, “Alright. I’ll see you next year.” Maybe there’s a lot of innovation that can be done around that yearly visit, but then that innovation would also change what the blood test looks like or what it’s doing.

They should probably be reviewing a year’s worth of blood data that’s taken weekly or monthly rather than looking at that single-point measurement. It all needs to be reimagined. You may be building something that takes five or 10 years to build to a spec that’s not working anyway, and financially, you could make money from it, but it’s not going to change people’s lives that much. 

In terms of what’s next, it’s just looking at actually solving problems. People have problems that are directly labile. Let’s actually look at what can be done about that. Rather than just asking how we can slightly improve the cost of doing something that’s already being done, the question is: What can be enabled? 

It broadly falls into the categories: issues with hormones, and issues with metabolic health like weight loss. Weight gain is the symptom of whatever’s going on, like metabolic dysfunction. There are all the inflammatory diseases and then cardiovascular diseases, which sit at the nexus of inflammatory and metabolic problems. These buckets are where most of the value is in trying to help people manage those things, improve them, prevent disease, and so on. 

In terms of technology, these areas benefit enormously from just giving the power to the person to do the test. They can test themselves and have the context around that data and have some kind of support, which doesn’t even necessarily need to come from a person, in some cases. But just having the ground-truth data already makes such a difference in figuring out where the person’s at, how they can make an improvement, and if whatever they’re trying to do is actually working for them. None of that is answered at a yearly visit with a complete metabolic panel.

The Power of Personalized Data at Scale

Josh Clemente: There are all these circumstances people find themselves in. They’re trying to manage, but today they do not have the abilities. It’s sort of, wait until the yearly checkup to get a test or a litany of tests, some of which may be adjacent to the problem you’re trying to solve today. 

If I want to have a baby, let’s say, and I go to the doctor and get my cholesterol panel and CBC and all these other things and I’m told, “You’re healthy,” but I’m still having trouble conceiving, the overlap is not what I need it to be. You’re saying we need to expand what we’re detecting, and the rationale for it.

The traditional blood tests were developed for a specific reason and have some information value, but there are all these other real-world examples of how understanding the health state and the underlying molecules driving it can be really powerful. But what you’re describing is, people could be given the technology to measure this themselves and more frequently so they can see changes—specifically changes due to an intervention. 

In the fertility case, maybe the intervention is managing PCOS through a different diet. PCOS is polycystic ovarian syndrome, and it’s very closely related to insulin resistance. Perhaps the intervention is trying different dietary approaches to see if, for example, insulin relative to glucose starts to adjust. 

Can you give some more examples of these sorts of real-world levers that you could build through testing to give people this kind of Healthcare 3.0 world, where you’re not just focusing on symptoms and you’re not just doing an annual blood test?

Michael Dubrovsky: For many of the things people care about, you can typically find a menu of five or 10 things to try. If your sleep is very dysregulated, there are all these tools. You can go as far as going to a concierge service that will completely analyze your sleep, or keep it as simple as taking magnesium before bed, which is $5 a week or so. 

You can just try these things, and you can even look at sleep scores. But the reason you would do a measurement is to really see if there’s a hormonal or metabolic reason behind your bad sleep. That kind of ground-truth information can be a lot more valuable than symptomatic things. Even the amount of hours you sleep is a symptom of something, of an underlying process.

I’ve been doing a ton of blood tests because we sell blood tests and test blood tests. In my own data I can clearly see when I screw up my circadian rhythm or other things of that flavor. You can see the connection between circadian rhythm and inflammation, circadian rhythm and testosterone, and things like that. That’s not true for everybody. 

Rather than just asking how we can slightly improve the cost of doing something that’s already being done, the question is: What can be enabled? 

We do correlations across a lot of blood tests. Let’s say inflammation is connected to sleep for 50% of people. For the other 50%, it’s not connected. You can improve your sleep, but your inflammation will not drop. The cause is something else. 

Every engineer that’s built anything has had to do a hundred rounds of debugging. If you’re not able to measure anything, you’re not able to debug what’s going on. It’s very unlikely you’re going to try one of these 10 things and it’s going to fix it. When that happens, people become evangelists of that solution. All you have to do is fix your sleep. But these things are so complicated and interconnected that that’s not actually true.

Josh Clemente: That’s anecdotal, but the data trends you’re pointing to are really interesting and open up a bunch of questions about how we can start to tackle a problem like that. You have interpersonal variability, for example, when trying to develop a new technology or test that has to show good results over a very wide swath of the general population. Some of that is reflected in the intervention, meaning sleep doesn’t necessarily change inflammation for everyone.

But it points to the challenge of getting a new technology to market when there isn’t necessarily an intervention associated with it. We’re not tracking a symptom. You’re actually measuring for the measurement’s sake. You’re measuring to describe the person’s health state better. The system really isn’t set up for that. The regulatory bodies don’t think about measuring all the molecular milieu inside the body just because you want to understand it and see how sleep affects inflammation. 

How do you make the case that this should be measured? How do you produce compelling evidence that it’s worth measuring, for example, inflammation or the underlying sleep molecules or cortisol when you aren’t actually doing so to prove the effectiveness of an intervention?

Michael Dubrovsky: Historically, the way people have gotten things approved for these home-use cases, or even wearables and things like that, is they never really achieve the same performance as the central-lab instrument. They make the argument that people need pregnancy tests at home. People are not going to come in for this test. They just need it at home. For some things, that argument is very easy and, over time, gets approved. 

But for many markers, that’s not the case. The best strategy, at least the strategy we’re taking, is to develop things that are the same quality as the central-lab instrument. If you can hit the same quality as the central lab instrument, regulation is actually not as hard.

The difficulty with regulations has really been that home testing, because it’s typically paper-script based, has not been able to hit the same level of quality as central labs. Because of that, it would essentially have to get exceptions from the FDA. The tools aren’t very accurate, but maybe the FDA is going to put it out because there’s a great medical need or a public-health need. 

But how do you make that argument for sleep? They might not care about sleep at all. In their framework, like you’re saying, that’s something that’s way upstream of anything they care about.

Josh Clemente: I’d love to have an entire conversation about the regulatory framework and proof of necessity—proof of value—for the next generation you describe, where the person is in the driver’s seat. 

We’ve shifted the locus of control about what we’re measuring and how we’re measuring and when. The individual whose health is at stake is measuring things that relate to their quality of life and their health span and the avoidance of conditions they don’t want. That is a value statement to that individual. 

It’s not a public-health-at-scale, general population sort of health mission, which is what the healthcare system is currently thinking about. They’re thinking about how we can acutely lower the rates of stroke. That may be connected, and if you look at the Daylight Savings examples, every single year, when the clocks shift backwards and you lose an hour of sleep, heart attacks and strokes go up like clockwork. The opposite happens when we shift back in the other direction and you gain an hour of sleep.

We’re trying to extricate ourselves from a situation where we are only thinking at the scale of everyone, and shift to a condition where I can only think about myself and you can only think about yourself and we are each looking at many data points about ourselves and adjusting our lifestyle levers along the way. 

There’s a lot there. The whole system is not structurally set up for the kind of proof of value that I as a consumer want when it comes to healthcare or the diagnostics we’re talking about.

Michael Dubrovsky: We generally are in agreement about this, but to play devil’s advocate, today you don’t really have a choice. If you want to optimize your sleep, there’s no metaobject that’s optimizing your sleep. Nobody on a national level is trying to figure out how to get people to sleep better, or how we figure out what’s happening. That’s not going on. 

It is in some small way. People are tracking it, but it’s not aggressively being managed. If you want to fix your sleep, you have to do it yourself, make the measurements, get the trial, the hacks—whatever. But there might be an optimistic world where, at some point, there’s enough data on enough people that you don’t need to do all of these experiments on yourself. At some point, you should be able to pool data from millions of people and start getting real insights that are forward compatible. 

You bring in the millionth person, you make a measurement on them, and you tell them, “Look, this is going to work for you.” And that’s already true in some cases. They’re trying to do this for cancer therapy. Where it really matters, the medical system is pretty good. They’ll sequence your cancer, tell you this immunotherapy is not going to work on you—that type of stuff. 

Over time, there is this transition from “you’re on your own, you have to fix this” to it being much more solved and something where you can leverage the experience of millions of other people. But for that to happen, this all has to be built up. But there’s an optimistic world where that happens.

Josh Clemente: I would argue that’s not even a devil’s-advocate condition for this situation. It’s a both-and, where I can both focus on my own selfish concerns and, simultaneously, my data could be aggregated and pooled into our general-population understanding of how inflammation and cortisol are connected, or how fertility is affected by insulin resistance. Then we can start to understand and potentially phenotype people based on a few very selective tests and quickly understand what intervention is most likely to work.

Today, you just don’t use data in that way. There are obviously exceptions, but at least not in a wellness context where I’m trying to optimize my diet for weight loss, for example. We generally do not understand what levers to pull for me versus for you. And that can certainly change.

The Future of Healthcare

Josh Clemente:  This isn’t just a conversation in the abstract. At SiPhox, you and your co-founder, Diedrik Vermeulen, are working on a technology. I’ll let you introduce the tech and what you guys are most focused on right now and the major challenges.

Michael Dubrovsky: There are the three pillars of blood testing I talked about: blood counts (lipids), chemistry tests (ions), and then immunoassays, which are proteins and hormones. We’re really focused on proteins and hormones. 

We’re taking all the optics that go into one of these instruments—with a lot of lasers and lenses and things like that inside—miniaturizing them onto silicon chips. There’s a whole industry out there which most people are not aware of, which is taking the technology that makes chips in your phones and manufacturing optical systems with it. And that’s been used mostly for internet communication. 

A Zoom call, for instance, gets converted to light when it goes to a data center, and that light goes through fiber, which has to be converted back into electricity to talk to other electronics and computers. That’s being done by chips with miniaturized optics on them.

Josh Clemente: Just to clarify for people, the electronics is the electricity, but the optics refer to the light. It’s moving photons instead of electrons.

Michael Dubrovsky: Exactly. It’s moving units of light that go through fibers. That’s how communication is done. Light is very fast and travels very quickly and doesn’t interact much. You can put a little bit of light into a glass tube that spans the ocean and it actually doesn’t make it all the way to the other end. You have to amplify it as you go. But it can travel a very long way without diminishing, and you can also pack a lot of data into it. 

That industry has been built up, which has enabled the miniaturization of other optical systems. We’re leveraging that, especially my co-founder, Diedrik. He has a ton of experience in it. He was part of the team that commercialized the most successful optical chip for data-center communication. There are millions of them out there. They’re transmitting maybe 50% of internet traffic.

We leverage that technology and use it to miniaturize all the optics in the blood analyzer. That has allowed us to cut the cost by a factor of a hundred to a thousand, and cut the size by a factor of a hundred to a thousand. 

You can have an Alexa-speaker-sized device on your kitchen counter that will give you five or 10 results out of a very small sample of blood. We’re focused on things it makes sense to measure at home—proteins and hormones which change relatively frequently and which are associated with either health goals people have or chronic-disease management, chronic-disease prevention, things like that. 

That might be a panel of all the female hormones for IVF, or it could be a cardiometabolic panel that covers what’s going on with your insulin, your inflammation, and your lipids—those kinds of panels that actually cover an area of interest for people.

Josh Clemente: That’s really exciting. What’s the timeframe where you feel like you’ll have something? Can you describe a little bit more about the user experience? Will this thing be used in place of a continuous wearable? Is this something you would just be measuring once every few months? What’s the interaction pattern you’re thinking about?

Michael Dubrovsky: The holy grail would be a wearable that’ll do the metabolites like glucose and proteins and hormones. We’ve talked about this a lot. Our goal is to run a pretty large study toward the end of this year, or early next year, where we recruit lots of people who are interested in measuring their markers, mostly for wellness. Then we can show both the efficacy of the device, but also the value of it so that people are able to use the data to improve their lives.

After that, we’ll submit it to the FDA for general use. That’s the current plan, and it’s really focusing on what we’ve been discussing: all of these types of use cases and building panels around those.

Josh Clemente: It’s basically making it more likely someone can measure something that’s specifically useful for them, and on a more continuous—or at least a more regular—basis.

Michael Dubrovsky: There are a lot of telemedicine use cases where you make a point measurement and you need to have, let’s say, a doctor’s visit associated with it. But one of the best use cases is giving more data to companies like Levels, honestly, because somebody has to take all this data, interpret it, and give the person actionable insights. It’s really about giving very reliable data that’s also a low barrier to entry for people to collect.

But there might be an optimistic world where, at some point, there’s enough data on enough people that you don’t need to do all of these experiments on yourself. At some point, you should be able to pool data from millions of people and start getting real insights that are forward compatible. 

If today you have to send a phlebotomist to someone’s house to take two tubes of blood and bring it to a lab, we could cut that down to something where they can do it for 10 minutes before they make their coffee. We want to get that barrier so low that it actually becomes part of normal life. We take blood tests all the time and we have our prototypes, so this is something we’re actually experiencing, and it’s very interesting and fun. We’re pretty excited to get other people using it outside the company.

Josh Clemente: Competing with myself and with other people on optimizing blood metrics is one of my goals in life. When we get to that point, we will have reached a really important milestone for healthcare and for individual health outcomes. When I care enough and there’s a competitive element of doing better with something that has such long-term leverage on my health, that’s a really good sign because today, it’s really hard to connect the dots between that PDF printout of my total cholesterol and anything negative in my life. I can walk through life and have no symptoms associated with that, but that could be the thing that puts me on my deathbed.

I’m really excited about it. I love what you guys are working on. We’ve gone deep on the technical side for a few years now as you guys have worked on it. I would recommend people check out your website. You guys have a good breakdown of the technology.

I expect more to come as you guys continue—and as we continue—to build in this next-gen space. Anything else you wanted to touch on before we jump off?

Michael Dubrovsky: I was actually curious what the transition was like for you, going from doing engineering on very complicated machines at SpaceX to dealing with biomarkers and everything related to that? 

Josh Clemente: It’s very different.

Michael Dubrovsky: What do you miss about machines?

Josh Clemente: A lot. Machines are pretty easy to fully characterize. You can have a closed-form solution for a structure or a mechanism. It can just be really well understood and designed, and the analog messiness of biology is something I don’t even know what I don’t know yet, but I know that it’s difficult. It’s very different. 

I’ve got this excitement to start to unlock that black box and learn more about what’s going on by attaching better sensing to our day-to-day lives and really starting to describe what’s going on in the body better. It was actually while I was at SpaceX that I got really frustrated by the complete lack of information I had about my health, and then found out there really isn’t any technology that can make this better. I can’t just pay more money to describe the hormone patterns in my body. It basically doesn’t exist. I look forward to us really leaning into that and new technologies, like what you’re working on. 

Relative to working on machines, it’s quite a bit more frustrating. There are a lot of hurdles to overcome to be able to build a technology and test it. If there’s anything I miss at SpaceX, it was an awesome experience, and definitely a lot of great people that are still there. 

But the thing that I miss most is being able to move really, really quickly and blow things up. It’s harder to have that mode of operations, especially when you’re dealing with people’s health. As Theranos found out, you just can’t play with that. You can’t be taking huge risks. I certainly miss being able to just work with aluminum because there’s less risk of that sort of thing. 

It’s also a really exciting space and I’d love for Levels to eventually be the SpaceX kind of innovator in healthcare: taking a completely different angle, lowering the cost of access, and increasing the actionability, usability, and scale of this sort of tech. That’s what I want to do. 

There’s carryover in a philosophical sense, in terms of how we’re going about things and the scrappy nature. And we’re definitely a small upstart, especially relative to the healthcare industry. But day to day, it’s quite a different work experience.

Michael Dubrovsky: The original insight at SpaceX was that you could use off-the-shelf electronics and things like that. What was the original insight that made them successful and what do you think is the one for Levels? What is the insight that drives the ability to get to the next level where you’re an established player in the industry? Or maybe that’s already happening now.

Josh Clemente: The insight at SpaceX is, if you just take a spreadsheet and calculate how much in raw material and manpower it takes to build a rocket, it’s orders of magnitude different than how much it would cost you to buy a launch on a rocket.

At the time that SpaceX was founded, Elon thought, This is fundamentally a blockade to us being able to increase access to space and eventually become multi-planetary. Not only are you paying orders of magnitude more than it costs, but you are then throwing that thing in single-use fashion into the ocean. Both of those factors can be solved by building it for what it costs and then also reusing it, like an airplane.

For Levels, the insight I might correlate there is that, right now, the person receiving value and the person paying for value are totally different in healthcare. I ostensibly get value from my healthcare visits. My doctor’s the one that delivers it to me, and somebody else, like an insurance provider or some unknown third party, is responsible for paying for it. 

What I consider to be valuable and what the insurance company considers to be valuable are very different, and what I’m willing to pay and what they’re paying and the value I’m getting are totally disconnected. You have this broken three-party system where what I end up with is a product experience that is not intended to make me happy. It’s intended to make the insurance company pay.

People want to participate in healthcare, but they want to get value from it. We want to shift to a model where it’s a direct, consumerized version of this sort of thing, where people just buy products that are relevant to their quality-of-life concerns and they then have to get value from them. Otherwise you don’t get a return customer and your business suffers. That’s what we need in healthcare. Certainly, I won’t say all of healthcare. That’s what we need in preventative and health optimization. 

Then there’s terminal illnesses and there are accidents and the things for which healthcare does super well. That’s a different question. But that’s how we’ll really improve the feedback loop between products that are being produced and a market existing for them. 

Opening up access to unlock traditional supply and demand is the other part we’re doing differently. We aren’t going for captive markets where there’s a moat built around it and a guaranteed long-term customer. It’s actually that we want to go to the area where the most potential users exist, and that’s people who aren’t yet sick.

Even though rates of illness are really high, there are still more people who aren’t sick than who are, and if you build a product that is exceptional to use and you get viral distribution because say, “This is the best thing ever,” and they tell all their friends about it, your healthcare product will be less like a thermometer and more like an Apple Watch. That’s what you want. 

It’s the vitamins-versus-pills scenario. People don’t take vitamins. They take pills because it solves a problem for them. That’s why pills are so valuable. There are millions of supplement providers. You want to build the pill because it provides value for people and they share that and they rave about it. 

Those are my rambles. TBD on whether or not it’s the same degree of unlock that SpaceX has in aerospace. But there are orders of magnitude in the healthcare system that are not being seen on the value side.

Michael Dubrovsky: The semiconductor industry is the exact opposite of healthcare, or what the rocket industry looked like in 2000. The semiconductor people are pretty, and this might be an old-timey expression, but they’re kind of hard-boiled. They’re not all techno-optimists or pretending to work 24 hours a day or anything. They’re just forced to go fast because of competition. They would go potentially as slowly as the healthcare industry if there was one semiconductor company, or if there was one rocket company. If there was one semiconductor company, they would maybe still be putting out a hundred nanometer transistors. 

It’s very hard to make transistors smaller. They have to gamble. Companies that are 50 years old have to gamble half the company value to get to the next note, and why would they ever do that? It’s that element of competition that forces these people. It’s not good people or bad people. It’s really the dynamics of the market that brings out the best in you—competition and this steady march of progress that they’ve had. If there was a Moore’s Law of healthcare, who knows what would be happening at this point?

Josh Clemente: Totally. It’s a great point. If there was a national semiconductor company, that’s exactly what you would see. It’s what you saw with aerospace. There were a few countries that could put something in orbit, and none of them had innovated past Apollo, with the exception of the Shuttle, which we put in a museum. And it’s arguable that was an innovation in terms of costs. 

That’s what we have with healthcare. We have this gigantic healthcare system. It’s heavily regulated. It solves certain things really well, but you do not have competition at the system scale. You have competition, you have a couple blessed providers that work inside Medicare for the same thing, but there’s a lot of price fixing, unfortunately, because of the way the thing works. Medicare sets the price and then everybody gets that price.