#77 – Biological Observability: Why it matters and how it will change healthcare (Josh Clemente & Sam Corcos)

We have observability for every machine we’ve ever built, but we don’t for the most complex machine that can’t fail: the human body. In this episode, Josh Clemente and Sam Corcos, co-founders of Levels, sat down to discuss the idea of bio-observability, why is it that we don’t have much data about what’s going on inside of our bodies, and the impact of real-time monitoring.

04:24 – We have observability of machines, not bodies

Josh said that we have observability for every machine we’ve ever built, but not for our bodies. We only have data on the response or recovery.

I think to your point that we know a lot about every tool we’ve ever built because we’ve designed those machines from the ground up and we’ve recognized that if we want to control them, we need to understand them. There’s a whole science around fault detection, isolation, recovery. So it’s called fitter, but this is all about essentially the underpinnings of understanding in real-time the state of a machine and detecting when something’s going wrong, isolating it and responding to it so that the machine doesn’t fail. And we have this whole broad science that covers, I would say every industrialized process that has machine elements in it, has fault detection involved that is automated and people are paying attention to it. And the human body is the most complex machine that can’t fail and we have no such thing. We actually only have the R part of that, which is the response or the recovery.

09:54 – Observability theory

Sam said observability theory is related to control theory. The more you can observe, the more you can control.

He explained to me that observability is the mathematical dual to control theory. It’s kind of obvious when you say it, but the more observable a system is, the more control you can exert on that system. And it’s often neglected because in mechanical systems like a spaceship, it’s pretty easy to observe things, you just put more sensors on it. But once you get to the further threshold of very complex distributed computer systems, you basically have to treat it like it’s a black box as opposed to a white box. Because it’s so complex, you can’t even know the state when you’re getting terabytes, exabytes, petabytes, some very large amounts of information on a daily basis. You now have a signal to noise problem that you have to figure out. So when I started digging into it was one of the first light bulbs that went off around observability was recognizing that there were fundamental data primitives in observability theory, that map pretty closely onto what it is that we’re doing.

12:58 – Why we want to have a trace in biological systems

Sam said that in engineering you have stack traces that allow you to see what went wrong and why. We want to have the same type of information for the human body.

Stack traces are super helpful when you’re debugging code and you want to figure out how did this thing just crash? You have your stack trace and you just go back until you find the thing or like, “Oh, okay, this is the thing that was wrong.” And then you could fix that. And then this problem doesn’t happen anymore. In biological systems, what you want to have is a trace. You want to know, I ate this thing, which caused these side effects and these outcomes. I ate oatmeal, which caused me to have this energy crash at 12:00. But the problem is that those two things are so time-delayed, you don’t necessarily make the connection. It certainly wasn’t obvious to me before I started measuring this stuff, that oatmeal was such a problem for me or that things like orange juice were actually just loaded with sugar.

16:01 – Observability leads to controllability

Josh said that if you can observe the cause and effect of something, you can then make changes that will create a different outcome.

So to boil down those somewhat esoteric concepts into just maybe a statement of why this observability theory maps onto why you should measure things is essentially real-time continuous measurement of the right molecules, not all molecules, but the right ones can enable us to observe these causes and effects for our health. So like you said, the breakfast, the walk, each of these has an effect which we can see. Once you know and can isolate cause and effect, that enables control over those outcomes through better decision making. And so that is the essential duality of observability and controllability. If you can observe the cause and effect, and you know the driver, you can then change it and that can change the outcome. So that is the package deal that is observing the right channels, leads to control over those channels.

21:37 – Trace and then offer better suggestions

Josh said that ideally, you’d be able to connect the dots between actions and the outcomes and then have the technology also offer better alternatives for the original action.

So you start to see and connect these dots between that news session that you did last night, elevated cortisol, meaning elevated fight or flight response. Your body’s physiologically stimulated. You then don’t settle in, you don’t get deep sleep. You wake up, poorly slept, insulin resistant and irritable and you now have a trace connecting all these physiologic moments together and indicating that you’ve stated that your goal is to lose weight and be better able mentally to perform every day. And what’s happened here is the as chain reaction caused by the nightly news, which ends up positioning you to be at a worse vantage point. And then a quick recommendation, something like reading a book and then being able to do the same exact night, except instead of sitting down and watching the news, you sit down and you read a novel for that same hour. And you then get a physiologic update that your cortisol’s half of what it was an hour ago.

24:32 – Humans need evidence to invest in better habits

Josh said that intrinsic motivation does not work for most people. There needs to be evidence of a reward path to get us to make better decisions.

The human condition is that what you’re describing is taking something from faith, literally faith-based where I believe that this is going to help me, therefore I will continue doing it. Or sometimes we call that motivation, like just having intrinsic motivation, which gets you out of bed every day to that go running in 16-degree weather or to do the meditation for an hour. Most people aren’t going to do that. Being able to show evidence that something is improving, even if incrementally, that’s a reward path which due to evolutionary reasons and many others, that’s what we’re wired for. We’re wired to have feedback, because ultimately what we are is we’re creatures that have to make selective decisions about where to focus our energies. We have been tuned to follow reward paths. And if you don’t have one, if you don’t know that the meditation is helping you, then there is no rationale, there’s no reason for you to invest in it.

31:54 – We need more markers than glucose

Sam said that without the context of multiple markers, people can see false positives in a result that are actually not bad.

This is coming back to why we need more molecules than just glucose. The body is a complex system and there’s not just going to be one stress hormone. There’s a lot of talk around cortisol monitoring and people think it’s going to be this panacea solution and they’re going to be disappointed because it’s not. If glucose solves 10% of it, maybe adding cortisol, we now can solve 11% of what’s actually going on, if we’re being generous. So the reason, some specific examples of why more than just glucose is needed is, there are so many false positives and false negatives. If you’re only measuring one metric, the false positives would be things like you exercise, you do a CrossFit workout and you see a big glucose spike. If that’s the only thing that you’re paying attention to, you would think like, “Oh, something bad just happened.” But you know as well as anybody, that’s not a bad thing, that’s a natural response from your liver dumping glycogen to keep you fueled for intense athletic activity.

34:42 – Add multiple signals

Josh said that like how GPS is more accurate if you have multiple signals to triangulate, biological metrics are more accurate with increased data points.

So this is like I think the exponential benefit of adding capability, adding multiple signals that you can start to…it is similar to GPS where if you just have one signal from one satellite, you have no idea where on the planet you are. If you add one more, all of a sudden you know where you are with pinpoint accuracy, add a third and it gets even better. So these sorts of triangulation concepts really apply here where, when you’re trying to trace down, not just if something changes, but is that change positive or negative? We can use some intuition about this, but adding simply an accelerometer to your risk so that we know that you are working out when that exercise spike happened can really dramatically improve our ability to give you feedback and that will obviously, as you start to measure molecules rather than just bulk physiologic properties, like heart rate, that gets even better.

40:27 – Our bodies don’t know how to respond to modern stress cycles

Josh said that historically our bodies were not exposed to the amount of sugar or the stress feedback loops that put them through today.

And so I think that’s the situation that we’re in now where these loops weren’t supposed to happen. You can cause such disruption that you sleep and you can’t even recover and sleep well. I think that was just outside the bounds of physiologically possible for a long time. And similarly, with some of the algorithms we’ve designed into society with 24/7 headline news and social media that has algorithms surface the most reacted to content, these sorts of loops were intended to help us. We’re tuned to respond to threats and listen for gossip to understand what our enemies are doing and things like this. And all of a sudden we have the ability to do this 24/7 and it’s destroying our stress feedback loop. So yeah, we have unlocked these vicious vortices where we get really stressed out about the news, and then we just want to binge eat crappy food and then our bodies don’t sleep and we wake up even less disposed to be able to deal with it.

43:56 – It’s very hard to measure things within the body

Josh said that there are many reasons why things are hard to measure within the body. The low concentration of some of the molecules is a major one, but that doesn’t mean we shouldn’t try.

There’s a probabilistic problem where, if you have just a very low circulating amount of a certain protein, measuring it becomes challenging because even though there might be twice as many of a certain molecule circulating in the blood, that might just mean one molecule close to you. Or there might be half as many and still there’s only one molecule close to you. It’s just like probabilistically, you run into this problem where it’s difficult to project from the measurement you’re taking to the overall concentration in the body, but that’s way in the future when you’re measuring very low concentration things. There’s a lot that I think can be done that just to be able to measure important molecules that just hasn’t been done because for the circular reasoning problem of, “Well, why would we do it? We haven’t ever done it before.”

Sam Corcos (00:06):

This is a chicken and egg problem. It’s an interesting conversations that I’ve had with a number of doctors where they say like, “Well, you shouldn’t do this because we don’t know the answer. But we don’t know because we can’t measure it.” So if your argument is we shouldn’t measure it because we don’t know what’s going to happen, then we’re never going to know. So it’s a tricky situation.

Josh Clemente (00:29):

It’s a bit a circular reason. We have to develop the core capability of just measuring these things in order that we can eventually understand what those measurements mean.

Ben Grynol (00:45):

I’m Ben [Grinnell 00:00:47], part of the early startup team here at Levels. We’re building tech that helps people to understand their metabolic health, and this is your front row seat to everything we do. This is a whole new level. It’s interesting to think about mechanical things, things like cars, things like airplanes, where we actually have more data on those machines than we do about our own bodies. It’s this idea of observability. But we can actually see the insights we can have behind data. So in our car, when the car is out of gas, we know the gas light goes on, that’s our cue to go fill up more gas. We don’t just check in the tank and say, “Hey, doesn’t look like there’s much in there.”

Ben Grynol (01:42):

When you relay that back to our bodies, when you start to think about that, it is somewhat of a bonkers, a little bit of a wild way of thinking that we don’t actually have that much data about what’s going inside of our bodies. It’s this concept of bio-observability. So Josh Clemente, founder of Levels and Sam Corcos, co-founder of Levels, the two of them sat down and they talk about this idea of bio-observability. Why is it that we don’t have much data about what’s going on inside of our bodies at all time, real time monitoring. Very fascinating conversation and a lot of insight into where things can go in the future. Here’s where they kick things off.

Sam Corcos (02:22):

So the topic is it’s on biological observability. The goal is really to explain who our team, what it is and why it matters and why it’s such an important concept for us in future. I think that one of the things that comes to mind for me is how glucose is a very interesting molecule to measure. I would even maybe argue that if you could measure all the molecules, it might even be the one that is the plurality of what is interesting, but that might only end up being 10% of what you actually want to measure because there are so many molecules. So if we’re being super generous to glucose, it’s 10% of what you want to know. It’s important. It’s the primary energy substrate of your body. Well, is the approximate cause of why people feel awful after they eat a bunch of sugar.

Sam Corcos (03:24):

So it has lifestyle that certainly affects my mental clarity. I notice that it’s a really easy thing to notice when I’m trying to get work done throughout the day. Have these wicked crashes from eating, used to just be oatmeal. So the concept of biological observability is, what are the other things you need to be able to measure to understand your body’s state? It’s remarkable how little we actually know about our bodies in general. We know so much about mechanical systems and, I would say the only things that we can measure in real time are basically glucose and heart rate. Is that basically it?

Josh Clemente (04:08):

Yeah. With the current wearables, you’ve got glucose, heart rate variability is a newer one on the scene. And then maybe you can measure pulse oxygen like my Garmin does some SpO2 stuff when it’s pretty bad.

Sam Corcos (04:22):

Yeah.

Josh Clemente (04:24):

Yeah. I think to your point that we know a lot about every tool we’ve ever built because we’ve designed those machines from the ground up and we’ve recognized that if we want to control them, we need to understand them. There’s a whole science around fault detection, isolation, recovery. So it’s called fitter, but this is all about essentially the underpinnings of understanding in real time the state of a machine and detecting when something’s going wrong, isolating it and responding to it so that the machine doesn’t fail. And we have this whole broad science that covers, I would say every industrialized process that has machine elements in it, has fault detection involved that is automated and people are paying attention to it. And the human body is the most complex machine that can’t fail and we have no such thing. We actually only have the R part of that, which is the response or the recovery.

Sam Corcos (05:19):

Yeah. And you have the response, but you don’t necessarily know what it’s a response to.

Josh Clemente (05:25):

Right. Because there’s a forensic process called where you have to do the Sherlock Holmes, connect the dots without all of the information and try and guess what the source of the fault was.

Sam Corcos (05:34):

Yeah. I wonder in the context of mechanical systems, I’ve never really worked in, is there an equation for how rapidly the usefulness of real time information degrades in terms of control systems? Imagine it’s probably something near exponential.

Josh Clemente (05:51):

Yeah, that’s a really good question. I mean, yes. The timeliness for control definitely depends on probably the dynamic environment of the system. So if it’s something that moves very quickly then, the relevance of realtime information drops off super rapidly whereas if it’s a much slower moving machine, that’s not the case. So I think it’s proportional to the environment it’s operating in, but that would be the control engineering, being able to move the system where you need it to go. But for fault detection, all historical information is I think, equally relevant. Obviously, the stuff that happens right closest to a failure is what you’re going to pay most attention to because you’re going to start there and work outwards. So figure out what the proximal cause of the failure was.

Josh Clemente (06:34):

You’re going to look at the data that is closest in time to it, but all historical information, like for example, if you have a bearing or a seal wear out on a rotating machine, your assumption shouldn’t be initially, and it’s very easy to say, “Well, that seal or bearing must have been faulty. We bought it from the manufacturer. It was bad. That’s the failure.” That’s almost never the case. It’s almost always the case that the bearing housing was coming loose and you had misalignment or something of that nature. So maybe it was a manufacturing work instruction that they didn’t torque the bolts on the housing to the right spec. And so that whole process of understanding the time series of all the information you’ve been logging is what tells you the actual root cause of that failure.

Sam Corcos (07:17):

Yeah. Yeah. The fog wise dynamic, that’s a Toyota thing. Is that right?

Josh Clemente (07:22):

Yeah. I think that’s a Toyota lean manufacturing concept.

Sam Corcos (07:25):

Yeah. I remember they were giving an example. I think it was in the lean startup. There was some failure in process, and when you ask enough questions as to why, you get to the root cause. And it’s almost never the thing that you assumed it was when you started.

Josh Clemente (07:42):

Exactly. That process can only be possible if you have records or you have some truth that you can refer to because just asking the questions or assumption or guess work, unless you have a substrate of evidence that you can then point to you to correlate your assumptions. And that’s the problem with modern medicine is that we have none of that. What we do is the gold standard is randomized controlled trials on large numbers of people. And the goal is test this thing in a closed box, this assumption in a closed box, and then average that assumption across every person or assume that it applies to every person.

Josh Clemente (08:19):

And that’s substituting for having really good information about every person, every individual. So the underpinning of why we went down this biological observability route is that we intuitively, I think most people intuitively understand that more information about themselves will be helpful in some way. And we’re now starting to color in why that is the case. It is very relevant to complex control, systems theory, or just monitoring for fault detection, which is going to be, it will drive the next wave of targeted personalized healthcare or health intervention is having an understanding of the state.

Sam Corcos (08:55):

It’s interesting you say that because everyone, especially when you use a glucose monitor for the first time, almost everyone who uses it, same with heart rate, that’s been around for much longer, has an intuition that there’s something different about this information. And people use the word real time, it’s because it’s real time that the information is different, but it wasn’t until I went on a walk with a friend of mine who works at Datadog, which is a company that, you can consider them an observability company. They work on observing very large complex, distributed computer systems and monitor them for problems. And when I was explaining the problems that we were running into, in my own lack of understanding, he said something along the lines of, “It sounds a lot like you have an observability problem. Have you read any observability theory?”

Sam Corcos (09:54):

And I had not. And he explained to me that observability is the mathematical dual to control theory. It’s kind of obvious when you say it, but the more observable a system is, the more control you can exert on that system. And it’s often neglected because in mechanical systems like a spaceship, it’s pretty easy to observe things, you just put more sensors on it. But once you get to the further threshold of very complex distributed computer systems, you basically have to treat it like it’s a black box as opposed to a white box. Because it’s so complex, you can’t even know the state when you’re getting terabytes, exabytes, petabytes, some very large amount of information on a daily basis. You now have a signal to noise problem that you have to figure out. So when I started digging into it was one of the first light bulbs that went off around observability was recognizing that there were fundamental data primitives in observability theory, that map pretty closely onto what it is that we’re doing.

Sam Corcos (11:08):

The primitives were metrics, logs and traces. And it’s a little bit wonky going into what those mean. But a metric is the one that’s most easily understandable. We’ll use two separate examples. One is in computer systems, which is like, when your computer says that your battery is almost empty, you’re at 10% battery life or your memory is 90% full. These are warnings that something about your system is going wrong. It doesn’t tell you anything about what caused it. It just tells you that something’s off. In a biological system, like the human body that would be going and getting your blood tested or your A1C is 5.8. It’s not telling you what caused that, just tells you that it is a metric that is happening. People who have a background in computer science or software development will understand what logs of traces are, because it’s a common part of a building software. You can insert logs in the middle of a function that prints something that says, “This function was called” or whatever you want it to say. You can even say, “This function was called at this time with this data.”

Sam Corcos (12:19):

And you can follow along and you can see what events are happening. Logs are effectively just events that you’re keeping track of. And a trace is something that in again, this is something that programmers know very well. A stack trace, it’s the story, it’s basically the connections between those events. It’s a linked list. It’s the series of events that are tied together that led to some outcome. So it’s new information because how those events are connected is a new piece of information.

Josh Clemente (12:50):

And that’s chronological time, those are linked in time?

Sam Corcos (12:54):

Yeah. They’re linked in terms of which one happened in what order? Yeah. Stack traces are super helpful when you’re debugging code and you want to figure out how did this thing just crash? You have your stack trace and you just go back until you find the thing or like, “Oh, okay, this is the thing that was wrong.” And then you could fix that. And then this problem doesn’t happen anymore. In biological systems, what you want to have is a trace. You want to know, I ate this thing, which caused these side effects and these outcomes. I ate oatmeal, which caused me to have this energy crash at 12:00. But the problem is that those two things are so to time delayed, you don’t necessarily make the connection. It certainly wasn’t obvious to me before I started measuring this stuff, that oatmeal was such a problem for me or that things like orange juice were actually just loaded with sugar.

Sam Corcos (13:51):

I’ve been convinced that orange juice is a healthy thing to have for breakfast. It’s really not. So what you want is a trace, the problem is that you can’t generate a trace from metrics. Traces are generated from logs. And when you look at the human body, you can’t in the most abstract way, the way that you would want to generate a log, if you’re being very strict on the definition, you’d have to figure out a way to trace each of the molecules from something that you ate through the cellular pathways, that this is probably impossible just from laws of Physics problems.

Josh Clemente (14:31):

You need to be in an MRI machine the whole time or something?

Sam Corcos (14:34):

Yeah. Well, you’d have to add radioactive traces to each molecule of a food that you ate. I don’t even know how you would go about that, and you have to do that for every food that you eat. So yeah. It’s probably just impossible. But the intuition that most people have when using something like a glucose monitor or a heart rate monitor that is correct is that, the reason why it’s different is that technically measuring glucose through a real time glucose monitor, technically that is a metric, but it is a metric that has a sufficiently high sample rate that you can infer events, which are logs. So when you eat something and you’re measuring and you see the curve and you have this hyperglycemic spike, that’s an event, that’s a log. When you have this hypoglycemic crash, that’s an event and that’s a log.

Sam Corcos (15:31):

When you start exercising when you have your heart rate monitor on, and you have this increase in heart rate, something triggered that, that’s an event. And so there is something to this real time component. Not everything will be useful to track in real time so this is, I think where the real time nature gets tripped up a little bit when people think about everything should be measured in real time. There probably aren’t that many things that you actually need to measure in real time, but figuring out what those are, is going to be an interesting science problem.

Josh Clemente (16:01):

Yeah. So to boil down those somewhat ester concepts and to just maybe a statement of why this observability theory maps onto why you should measure things is essentially real time continuous measurement of the right molecules, not all molecules, but the right ones can enable us to observe these causes and effects for our health. So like you said, the breakfast, the walk, each of these has an effect which we can see. Once you know and can isolate cause and effect, that enables control over those outcomes through better decision making. And so that is the essential duality of observability and controllability. If you can observe the cause and effect, and you know the driver, you can then change it and that can change the outcome. So that is the package deal that is observing the right channels, leads to control over those channels.

Josh Clemente (16:56):

And the tricky part is, which channels? What do we mean by a channel? It’s a probably a molecular or a physiologic metric that is happening in a sufficiently high resolution that we can measure it in a meaningful way. So there are things like pulse, that happens 60 to 120 or 180 times per minute, we can measure are that pretty consistently and you can start to derive what’s driving pulse changes. Glucose tends to fluctuate within five to 10 minute timeframes, very relevant to lifestyle decisions like eating and exercise. There are many, many others like hormones that react to our choices. And that’s the domain that we’re interested in. Not so much that we need to be measuring cellular replication across all of the tissues in the body or something like that. It’s more so we want to be looking at those molecular metrics or derive logs from them, which are responding to the choices we’re making, because ultimately this is about giving you better control over lifestyle related outcomes that you want to avoid.

Sam Corcos (17:58):

One of my fantasies for the future on how this type of tool will work is, if you can measure things like stress hormones, you can imagine a future world where people who have say, anxiety and depression, I’ve been news sober for almost nine years now, and I’m convinced that the news is poison. And I think that for a lot of people, their relentless consumption of current events and news is the reason why they are so anxious and feels so bad all the time. So one of my fantasies for the future is we have a monitor. There’s a sensor that is measuring your stress hormones in real time. And you get an alert. The monitor doesn’t even need to know what you were doing, it just needs to say, “Hey, whatever you were doing at 2:00 is physiologically contributing to your anxiety.” And that would be the wake up call.

Sam Corcos (18:59):

Because it’s one thing for people, a friend of mine who used a glucose monitor for the first time, he drinks something like five cans of soda per day, or at least he used to, and he largely thought of himself as invincible as a lot of men in their 30s do. And he started using a glucose monitor really just to see what would happen, not expecting to find very much. Discovered that these cans of soda that he was drinking every day were pushing him into numbers that you’re not supposed to be able to see as a healthy person. He texted me within one or two days after starting and said, “My mom always told me that soda is bad, but it wasn’t until I saw it in the data that it really clicked for me how bad this is.” And I don’t think he’s had soda since then.

Sam Corcos (19:49):

It’s been many months. So that’s the closing the loop. You can tell people, “No the news, it’s bad. It’s causing you stress.” But when you can see it in data, another perfect example of this would be, you can read about how caffeine after a certain time affects sleep quality. What is that number? For me, it’s roughly 12:00. Sometimes I’ll have coffee at 1:00 or 2:00, not super strict about it. I don’t really notice a difference, but it would be great if I actually could have that information. If I had caffeine after 10:00, and it impacted my sleep quality, and I could see it in some metric, if I could see it in the data that would be incredibly useful or the effect of alcohol on sleep quality and other things. Having the data to support it, instead of just reading an article and taking it at face value, not that there isn’t value in that, but it makes a huge difference in terms of behavior to actually be able to close the loop.

Josh Clemente (20:49):

Yep. That trace layer of sitting down and watching the news and then having an indicator, a physiologic response where this sensor system just alerts you to the fact that your cortisol levels are 200% of what they were 35 minutes ago. Maybe there can be a recommendation involved, do some breathing exercise, go for a walk, something like that. But, let’s say you don’t do those things. You just get that indicator cortisols through the roof and then you try and go to sleep because it’s 11:00 PM, now you’re all worked up. You try and go to sleep, turns out you don’t get any deep sleep, you’re in all light sleep modes. Wake up the next day and you see that your blood sugar is higher and you see that you get surfaced some educational insight around the fact that disturbed sleep leads to insulin resistance. This has been measured in academia.

Josh Clemente (21:37):

And so you start to see and connect these dots between that news session that you did last night, elevated cortisol, meaning elevated fight or flight response. Your body’s physiologically stimulated. You then don’t settle in, you don’t get deep sleep. You wake up, poorly slept, insulin resistant and irritable and you now have a trace connecting all these physiologic moments together and indicating that you’ve stated that your goal is to lose weight and be better able mentally to perform every day. And what’s happened here is the as chain reaction caused by the nightly news, which ends up positioning you to be at a worse vantage point. And then a quick recommendation, something like reading a book and then being able to do the same exact night, except instead of sitting down and watching the news, you sit down and you read a novel for that same hour. And you then get a physiologic update that your cortisol’s half of what it was an hour ago.

Josh Clemente (22:33):

Because reading has been shown to reduce stress levels in the body. And so starting to now see that there are actually patterns of behavior that you can select where these traits, which seem like, “Oh, I’ll either watch the news or I’ll read a book.” It seems like a very flat trade. What’s the difference there? Well, it’s actually a massive physiologic difference. And people aren’t thinking about this because we don’t have this logging and tracing yet. And I think your theory, I totally agree with, I am also news sober now for, maybe three or four years, much less time than you, but I know that I am now interceptive enough.

Josh Clemente (23:04):

So I am in tune enough, I think, with my body at this point that I can sense that stimulation when I do read something that triggers me, whether it’s scrolling through Twitter or whatever. And so now I think I completely agree with you that this is where the evidence is going to lead us. And we need to be able to give people that personal insight, because it’s not enough to say on average, people are pissed off by the news. It’s like, “Do I have an effect that is negative?” That’s what you have to show people.

Sam Corcos (23:29):

Yeah, that specifically in terms of the personal effect is, I’ve tried meditation on several different stints and it’s never really stuck for me. I haven’t noticed a difference. I tried daily meditation for a month and it could just be because I’m already a relatively calm person. So I don’t need the meditation to calm me down. Or maybe there is actually a very significant effect and I just don’t really notice it. I would love to have the data to support this. If I could see actually your effectiveness, the entire rest of the day is improved by 50% from those 10 minutes of meditation in the morning. Great. I’m doing it every day. Or for me, it just doesn’t really make a difference because I don’t have that level of stress that some people maybe need meditation in order to help them manage it. I don’t know. I would love to have some feedback mechanism on that.

Josh Clemente (24:32):

The human condition is that what you’re describing is taking something from faith, literally faith based where I believe that this is going to help me, therefore I will continue doing it. Or sometimes we call that motivation, like just having intrinsic motivation, which gets you out of bed every day to that go running in 16 degree weather or to do the meditation for an hour. Most people aren’t going to do that. Being able to show evidence that something is improving, even if incrementally, that’s a reward path which due to evolutionary reasons and many others, that’s what we’re wired to have feedback, because ultimately what we are is we’re creatures that have to make selective decisions about where to focus our energies. We have been tuned to follow reward paths. And if you don’t have one, if you don’t know that the meditation is helping you, then there is no rationale, there’s no reason for you to invest in it.

Josh Clemente (25:20):

You look at studies on meditation, there’s this big distribution and some people have these crazy high quality return on investment, some people have none, and they average it. And the results are just the mean. And so there are people on the left side of that bell curve that don’t get any benefit from meditation. And that surely shows up in evidence. And you might be one of those people. You also might be somebody on the right tail of that. You just can’t see the effects and this reward path can be both negative and positive reinforcement. And that’s the beauty of data is that when you have a metric, it’s just objective. It doesn’t have judgment involved. It’s just a measurement for you.

Josh Clemente (25:54):

And that I think is something that people respond to much differently than the, let’s see, best practices. Best practices can be based on objective data, but they feel very maternalistic. Everyone should meditate because it’s been shown in studies that it provides X benefit. That’s very different from your body saying, “You should meditate. It’s doing us major favors.” It feels much different than being coached or being spoken down to as someone who should be meditating, because that’s what the evidence says.

Sam Corcos (26:24):

Yeah. And this comes down to the idea of personalization and what personalization means. And I think people often get tripped up in definitions there because, I coincidentally just wrote a memo on this, I think yesterday or the day before to clarify some of my own thoughts on this. But people often think about personalization, meaning things like being able to change the background color of your Slack workspace, but that’s not personalization, that’s not what it means. The best analogy I would give, is your doctor gives you personalized recommendations. You come to the doctor with symptoms, and they have enough context of all of the things that they learned in medicine to give you a personal diagnosis. If you imagine a non personalized doctor visit, which is obviously absurd, you come to the doctor with a stomach ache and the doctor says, “Well, the leading cause of death is heart disease. So here’s some heart disease medicine.

Josh Clemente (27:28):

Right.

Sam Corcos (27:29):

It’s obviously absurd. So you have to have some level of personalization based on your circumstances to make good decisions.

Josh Clemente (27:39):

Yeah. So personalization is rather than, I think you draw the distinction to customization, which is not like anyone else’s thing. So my car has flames on it and other cars don’t, that’s customization. Personalization is actually, I think, distinct from generalization. So the personal versus the general solution. So if a doctor is just vending machining out, actually I would push back a little bit and say, we do have this situation where, in many cases you may have the subject matter might be personal to your condition so I have stomach pain, but the response may actually be general from your doctor which is, eat better, work out more. We get this in medicine all the time because there’s insufficient context for the doctor to make very good targeted interventions. And so when it’s not a critical care acute situation where it’s a heart attack or it’s some known disease condition with symptoms that can be treated with a known therapy, you do get a general solution because we can’t peer into the black box deep enough yet.

Sam Corcos (28:39):

And this actually ties into something that you mentioned earlier around the control context for mechanical systems, the interventional piece. There were two concepts that are useful to semantically distinguish between which is observability and monitoring. These are just arbitrarily differentiated in terms of definition, but observability is around black box systems and monitoring is about something that you already basically know the answer to. You would want to monitor something in your spaceship, because if it starts to go off course, you want to correct it. That’s not the same thing as observability, even though they’re both paying attention to something in real time, they’re solving a different type of problem. What’s interesting is if you enable better logging in biological systems, basically this concept that we have a biological observability, it also enables monitoring. There are interventional uses for something like this as well. We don’t really know what they are, but you can speculate on a lot of them.

Sam Corcos (29:45):

You can imagine if there were conditions where knowing about something in advance is helpful. And if there were molecules that appear in advance, something like a heart attack, maybe it’s troponin, is something you can measure in advance of a heart attack or if there are volatile organic compounds you can measure for a pneumonial conversion before it becomes a life-threatening condition. There are a lot of other types of things that being able to measure them in real time comes interesting. We don’t have to get into the nuance of the basium problem of measuring these in large populations. But if they are done with selective populations, it can really make a big impact even on traditional healthcare.

Josh Clemente (30:27):

Yeah. I think what’s interesting there is that you have a few channels or markers that could be generally applicable. So for example, inflammation. So having a high resolution continuous inflammation marker like high sensitivity, C-reactive protein is one that people have probably heard of before, but this rapidly changes with our inflammatory environment. And for example, COVID right now. One of the first markers to start to climb when you’re sick are inflammatory markers as part of the immune response and hs-CRP is one of them.

Josh Clemente (31:00):

So for example, when you’re trying to avoid these encounters that might spread a virus like COVID, being able to see in real time that your inflammatory markers are up, your respiratory rate is up. Your heart rate overnight was slightly elevated, it’s based on this 95+% probability that you are coming down with something. That’s a very, very effective way of understanding your illness condition. But it also could be helpful for detecting infections like blood infections or detecting that pneumonia you were talking about. So there are high leverage molecules that if you can correlate them with other orthogonal indicators like troponin or cortisol, you can start to generate really, really high probabilities of certain conditions without necessarily being able to detect some various specific protein associated with that diagnosis. But being able to essentially triangulate a condition based on multiple other signals.

Sam Corcos (31:54):

Yeah. This is coming back to why we need more molecules than just glucose. The body is a complex system and there’s not just going to be one stress hormone. There’s a lot of talk around cortisol monitoring and people think it’s going to be this panacea solution and they’re going to be disappointed because it’s not. If glucose solves 10% of it, maybe adding cortisol, we now can solve 11% of what’s actually going on, if we’re being generous. So the reason, some specific examples of why more than just glucose is needed is, there are so many false positives and false negatives. If you’re only measuring one metric, the false positives would be things like you exercise, you do a CrossFit workout and you see a big glucose spike. If that’s the only thing that you’re paying attention to, you would think like, “Oh, something bad just happened.” But you know as well as anybody, that’s not a bad thing, that’s a natural response from your liver dumping glycogen to keep you fueled for intense athletic activity.

Sam Corcos (33:04):

It’s not a bad thing at all. And the false negatives are things like, if you hound a leader of fructose, way worse for you than glucose, by the way. Rick Johnson’s new book coming out on this, I think is going to open a lot of people’s eyes to it. Obviously a lot of Rob Lustig’s work on this as well. But you would think that it was perfectly metabolically neutral, because you didn’t have a response. So you need to be able to measure more stuff otherwise it’s just so hard to contextualize all of this. You just have to do it through education sort of a lazy way when really you want to be able to measure just four molecules.

Josh Clemente (33:42):

Yeah, exactly. The exercise and glucose example. If you had another marker like insulin, that would be a holy grail, but you could start to filter for these positive and negative circumstances because obviously exercise is a very different situation than eating a sugary meal. And the delta is, exercise is training insulin sensitivity or rather insulin independent glucose uptake. So your muscles can use during exercise glucose without insulin being released whereas if you eat a giant slice of cake and a big postable, your body has to use insulin to carry that glucose out of the bloodstream. And so those two things that might look the same on a glucose trace are phenomenally different in terms of the effects on your health. And this is multiplied across many, many different pairs of metrics. So what you get is if you relative to multi-signal monitoring, if you have two channels of measurement, you actually get three because you get each of them individually and then you get the channel that is how they move relative to each other.

Josh Clemente (34:42):

So this is like I think exponential benefit of adding capability, adding multiple signals that you can start to. It is similar to GPS where if you just have one signal from one satellite, you have no idea where on the planet you are. If you add one more, all of a sudden you know where you are with pinpoint accuracy, add a third and it gets even better. So these sorts of triangulation concepts really apply here where, when you’re trying to trace down, not just if something changes, but is that change positive or negative? We can use some intuition about this, but adding simply an accelerometer to your risk so that we know that you are working out when that exercise spike happened can really dramatically improve our ability to give you feedback and that will obviously, as you start to measure molecules rather than just bulk physiologic properties, like heart rate, that gets even better.

Sam Corcos (35:26):

Yeah. And it reminds me of, I remember it was, I think it was a podcast that Casey did with Dr. Iman. He’s had some patients where something was clearly wrong, but their A1C was normal, fasting glucose was normal, oral glucose tolerance test was normal but they also have fatty liver disease. And it can take a lot of probing and testing to try to figure out why is it that this person’s metrics… And then I think they did a fasting insulin and it was super high. That’s so weird, because usually when people have super high fasting insulin, all these other things are also bad, but in this case it just wasn’t. And so this comes down to the personalization component. There’s something unique about that person’s situation that giving them the blanket diagnosis would not have solved this problem. So the multiplicative effect I think is absolutely right, which is once you’re measuring more molecules, and it’s going to be a real challenge for us as the software layer of solving these problems is we want to build a tracing layer for biological observability.

Sam Corcos (36:38):

And the challenge is that, most of these interactions are not necessarily known to science right now. And so this is going to require developing new science and understanding. What is the interaction between glucose and uric acid? Who knows? What’s the interaction between glucose, uric acid and beta-hydroxybutyrate and triglycerides. How do those interplay with each other based on your activity? I don’t think anybody knows the answers to these questions. A big part of it is, this is a chicken and egg problem. It’s an interesting conversations that I’ve had with a number of doctors where they say like, “Well, you shouldn’t do this because we don’t know the answer. But we don’t know because we can’t measure it.” So if your argument is we shouldn’t measure it because we don’t know what’s going to happen, then we’re never going to know. So it’s a tricky situation.

Josh Clemente (37:35):

It’s a bit of circular reason. Exactly. We have to develop the core capability of just measuring these things in order that we can eventually understand what those measurements mean.

Sam Corcos (37:43):

Yeah.

Josh Clemente (37:44):

We started somewhere with every single measurement that we have in healthcare. And there was the first time that somebody measured A1C and then there was a theory about what it meant, and then over time it became the standard of care. And we’re now transitioning to a point where due to many different revolutions, you could call them including the miniaturization of electronics and just better signal processing, we have the raw technical capability to measure lots of things about the individual. It’s just that it hasn’t been the case historically.

Josh Clemente (38:13):

So now we’ve got to push across that hurdle and you open up those channels and then obviously like you said, the burden falls on the deconvolution and the signal processing and the interpretation of what all of that means not just in whole, but independence data streams, they all have some modifying effect on each other. And so that’s a tremendous challenge, but it’s also a tremendous benefit to be able to break it down into the end of one so that you, the individual know how your body responds to these choices and we can determine the source, the cause and the effect.

Sam Corcos (38:44):

Yeah. It’s interesting. One of the things you mentioned before about the trace of being really stressed before bed, getting poor sleep, being really grumpy. Something that has always seemed curious to me about the human body is that a lot of systems have corrective mechanisms and it seems like our bodies are the opposite, where there’s these downward spiral mechanisms where if you get poor sleep, you wake up and you crave sugar. Eat sugar, of course, you crash later, then your sleep quality is worse. It seems like our bodies are intended to spiral out of control.

Josh Clemente (39:32):

I think that there are circumstances that have changed so tremendously that this is possible. I think historically, maybe this was less the case, but now you have the ability, in previous eras of human evolution, the most sugar you would come across in your entire life is probably a watermelon. And that thing was about the size of a grape fruit back then it wasn’t hybridized into these massive table size things. And so you’d come across a watermelon and you’d just gorge yourself because those calories, they were life or death. Every calorie you come across was life or death back then. You eat anything. You’re tuned to chase down easy, delicious calories. And then all of a sudden over the course of a few 100 years, agriculture started a long time ago, but in the past few 100 years, you now have the ability to, in a single sitting, crush more processed sugar into your body than you would’ve come across in a lifetime literally prehistorically.

Josh Clemente (40:27):

And so I think that’s the situation that we’re in now where these loops weren’t supposed to happen. You can cause such destruction that you sleep and you can’t even recover and sleep well. I think that was just outside the bounds of physiologically possible for a long time. And similarly with some of the algorithms we’ve designed into society with 24/7 headline news and social media that has algorithms surface the most reacted to content, these sorts of loops were intended to help us. We’re tuned to respond to threats and listen for gossip to understand what our enemies are doing and things like this. And all of a sudden we have the ability to do this 24/7 and it’s destroying our stress feedback loop. So yeah, we have unlocked these vicious vortices where we get really stressed out about the news, and then we just want to binge eat crappy food and then our bodies don’t sleep and we wake up even less disposed to be able to deal with it.

Josh Clemente (41:20):

And I think the only way to fight back on it is to, because we don’t have sensory feedback loops for this, we don’t have a mechanism inside that tells us, you should probably stop eating that sugar. We only have the positive reward of that delicious taste because it was life or death. So we have to supplement that, because we’re in this era of technology that can enable this, we also have to supplement with better sensory systems, which can tell you when you don’t know.

Sam Corcos (41:47):

Yeah. I wonder, you understand the underlying hardware better than I do. Why is measuring stuff so hard in the human body?

Josh Clemente (41:59):

That’s a good question. The body does a lot to preserve resources. And so it would be awesome if you could stick something on the outside of the skin and collect sweat, and measure all of these molecules in it. But one theory is that the reason measuring, for example, glucose in sweat, doesn’t actually work is that sweat levels and blood levels don’t correlate. They are non-linearly related. The reason for that is likely because if an organism were to leak its primary energy source outside of itself, it would very quickly be eliminated. You need to preserve your energy. And so your fuel source can’t leak.

Sam Corcos (42:41):

Leaking is bad.

Josh Clemente (42:41):

Leaking is bad. And so it turns out that the same holds for tears, for any external fluid. Your body does not want to lose these important molecules that are driving it. And this is both fuels, but also hormones that are just inefficient to produce molecules that are then being lost to the environment. So that’s why now you have to make something invasive that goes into the tissue and measures it, and your body doesn’t like invasive stuff. So it has like very tuned responses to wall-off invaders, whether that’s a splinter or a sensor. And so that creates disruption where your body will quickly encapsulate the invading sensor probe, and that prevents the concentrations of molecules in that barrier from being similar to what’s in the body at large.

Josh Clemente (43:31):

Some molecules are very unique. There are small molecules like glucose that have a very unique signature, but then you have bigger macro molecules like proteins and hormones that are actually just groups of a bunch of other similar building blocks. And so identifying those is really challenging, because they’re all made of the same stuff, and you have to get really, really specific. And they’re in very low concentrations. There’s a probabilistic problem where, if you have just a very low circulating amount of a certain protein, measuring it becomes challenging because even though there might be twice as many of a certain molecule circulating in the blood, that might just mean one molecule close to you.

Josh Clemente (44:16):

Or there might be half as many and still there’s only one molecule close to you. It’s just like probabilistically, you run into this problem where it’s difficult to project from the measurement you’re taking to the overall concentration in the body, but that’s way in the future when you’re measuring very low concentration things. There’s a lot that I think can be done that just to be able to measure important molecules that just hasn’t been done because for the circular reasoning problem of, “Well, why would we do it? We haven’t ever done it before.”

Sam Corcos (44:40):

Yeah. And one of the most encouraging things, I’m sure you saw the announcement, Abbott is, I guess, that CES, is that right?

Josh Clemente (44:50):

Yep. The Lingo.

Sam Corcos (44:51):

And they’re doing glucose, ketones, do you remember what it was?

Josh Clemente (44:58):

Glucose, ketones, lactate, alcohol or ethanol.

Sam Corcos (45:02):

Yeah. Yeah. So that’s a really encouraging step that these companies are taking multi-molecule very seriously. I haven’t spent a ton of time thinking about this, what are some other molecules that you think would be interesting? It sounds like cortisol would be interesting.

Josh Clemente (45:23):

Yeah. I think that the top five are probably glucose, lactate ketones or, I actually think triglycerides so like a fat molecule and cortisol and then probably inflammation so hs-CRP. Those are all in relatively high physiologic quantities. They should be pretty easily measurable and they give you such a broad spectrum. So you have the sugar, you have the fats to help you balance that equation. Because a lot of people, you see a scary glucose spike and so you start eating super low quality but high quantities of fatty foods. And depending on the quality there, you actually could be starting to generate bad triglycerides and cholesterol levels. So you want to balance that and then having cortisol for stress insight, having an inflammatory marker and then lactate is a nice blend of several things. It tells you about the efficiency of your body at processing fuels.

Josh Clemente (46:22):

Obviously, Abbott is targeting half of that range, which is really great. The others like triglycerides and cortisol and hs-CRP are a little bit further out, I think, but we’re clearly heading in a direction that is really nice. I hope that Abbott will, I think at the CES announcement they were saying that it’s going to be four individual sensors. I’m hoping that they will start to multiplex these things. And to me that’s just for a convenience factor. I’m already running at real estate, just tracking my sleep and my sugar so I think there needs to be a real concerted effort for convenience to drive all of these together and solve the problems. It’s not going to be easy, but you have to solve those problems if we’re ever going to get out of the single mode measurement.

Sam Corcos (47:02):

Yeah. I would imagine that’s on the roadmap. What about something like uric acid that seems to be certainly David Perlmutter talks a lot about uric acid and its importance.

Josh Clemente (47:14):

Yeah. That’s something that I need to learn more about. Dr. Perlmutter talks about how so few people understand uric acid unless you’ve had a bout of gout. So yeah, I’m one of those people. I know that there’s a strong connection to fructose and metabolic inefficiency. So that could totally be one that ends up emerging as extremely powerful. Like you said, of the range of molecules, I just mentioned, there’s nothing that would really indicate maybe triglycerides, but that would really indicate the effects of a fructose load.

Josh Clemente (47:43):

And we have high-fructose corn syrup in soft drinks and energy drinks and kids are getting fatty liver disease because they’re drinking all of this fructose and it’s not going to show up on a glucose curve necessarily. It may show up as increased triglycerides because that’s all your liver can do with it, but maybe uric acid can be a much higher gain indicator. So yeah, I think that’s something people should be paying attention to while developing these sensors.

Sam Corcos (48:07):

What about, was it Rick Johnson or Gerry Schulman who talks about osmolality?

Josh Clemente (48:14):

That is Rick Johnson.

Sam Corcos (48:16):

What about, that seems like blood hydration and electrolyte balance. That seems like a really important thing to keep track of too.

Josh Clemente (48:25):

Yeah, absolutely. So osmolality is something like salt balance within the cells and that should be one that’s really easy to measure. It should not be a huge challenge. So that feels like low hanging fruit to again, add another layer of intelligence around lifestyle choices. So very much I think, my understanding is that osmolality fluctuates with hydration choices, but also salt content and the effects of osmolality fluctuations are conversion of fructose to glucose and vice versa, all of these downstream consequences. So understanding your osmolality balance is, according to Dr. Johnson, something really powerful for lifestyle choices. Yeah, I think that’s another great example. The deconvolution of these and finding the highest gain levers to pull on and to invest in measuring is really interesting. There are lots of discreet studies, but which one to focus on is a bit of a forensic project as well.

Sam Corcos (49:24):

Yeah. [inaudible 00:49:25]. And the feasibility of it is, because it seems like we don’t totally know the concentrations of a lot of these either. So you have to figure that out as well.

Josh Clemente (49:36):

Yeah. There’s a whole host like a body of work to be done in quantifying the interstitial fluid. Because the easiest measurement volume in the body gives you the highest quality information but is the least invasive is the interstitial space, which is a few millimeters below the skin surface. The problem is that we don’t have gold standard metrics for what the concentrations of molecules are, and the ranges are in that interstitial space. So there’s something to be done here of creating the reference data for what happens in the interstitial space for all of these different molecules, which is an exciting body of research that some team has to take on intent.

Sam Corcos (50:13):

Yeah, I know.

Josh Clemente (50:16):

Yeah. It’s pretty amazing that much of this hasn’t been done before. The glucose monitors we use today are measuring interstitial glucose and then trying to software correct it to match blood glucose, which leads to a lot of challenges because those two things although they track the same way they don’t happen at the same time. So that’s going to be the case for all molecules. And at some point I personally think that it makes more sense to just measure what’s in the interstitial fluid and display it as such, don’t try and software correct it to blood values because actually most of the tissue in your body is exposed to that interstitial fluid, not to the arterial blood values, if that makes sense. So your blood filters out to the muscle and the brain and all that stuff. And so the interstitial values might actually closer reflect what your body overall is experiencing. And so it might be the thing that you care about more. So circumventing all those technical challenges of correcting it to the blood might not be necessary.

Sam Corcos (51:12):

Interesting. And more than hint, hint, if somebody listening to this can solve these problems, let us know. And we might help fund it.

Josh Clemente (51:21):

100%. It’s the big glaring gap in the research is like, I’ve had so many conversations with various teams like in academia and elsewhere who are thinking about measuring things. And they’re like, “Well, we could totally measure this. It’s just that we don’t know what it’s supposed to be. So we wouldn’t know if we’re right or not. You need something to compare against.” And that just hasn’t been done. So if you’ve done it or if you want to do it, let us help you.