Digital Twins in Healthcare: Can AI Reverse Disease? | Twin Health CTO Terry Poon
Terry Poon, chief technology officer of Twin Health, explains how his company applies digital twin technology to metabolic disease such as diabetes and obesity. He describes how wearable sensors, machine learning and a human care team are combined to give members personalized recommendations, and what results the company reports.
Transcript
Welcome to Agents of Tech, the podcast where we dive into the innovations shaping our future. I'm Autria Godfrey, coming to you from Washington, DC, and today we are exploring how digital twins, which are virtual replicas of physical systems, are revolutionizing health care, from personalized treatment plans to early disease detection. This technology is unlocking new possibilities in patient care. Digital twins were once the stuff of science fiction, but today, scientists can replicate organs and even complex metabolic systems with incredible accuracy.
Well now, we are diving deeper into how this groundbreaking technology is revolutionizing personalized health care and disease management. Today, we'll be joined by Terry Poon, CTO at Twin Health.
We've developed a lot of technologies, to, to really achieve, deep understanding of biology and behavior. So using sensors that we can understand, you know, what makes a person's biology unique and what makes their metabolic responses unique.
If you are enjoying our show, please make sure to like and subscribe to Agents of Tech as this really helps develop the show and shine a light on cutting edge research. Thank you. Now it's my pleasure to introduce Laila Rizvi, the Director of Strategy and Research here at WebsEdge the company behind this podcast. Laila. Pleasure to have you today.
Great to be here. As always.
Absolutely. Okay. I love this topic that we're diving into today. You know, digital twins were once a concept used by NASA to simulate space missions. But now we're taking that same technology, applying it to healthcare and really changing how we're able to address personalized health care for people. It's fascinating.
Yeah. And not only is it fascinating, Autria, I think it's the potential that's here that the biological and biomedicine community hasn't had for a very, very long time. Like you mentioned, NASA was using this stuff, you know, many, many years ago. And people who work with kind of physics based systems and physical systems have had the joy of being able to use, kind of digital twins and data driven models more than people dealing with biological systems. And that's because biological systems are really confusing and they're at multiple scales, they're stochastic, they're dynamic. And so they're really, really complicated. And we've just not had the tools to to achieve kind of personalized medicine previously. We've had data which has been kind of choppy and it's given us an indication, but we've never really had something like AI to do this.
I mean, when we talk about, you know, personalized medicine, you know, you have, you know, clinical trials or you have drug trials that the results can be applied to kind of a general population, but not necessarily to a personalized individual utilizing all of their biometrics. So what's interesting about digital twins is that it's not just kind of like a copy and paste digital replica, which I think some people think it is, but because we can utilize AI and machine learning, these digital twins can actually simulate treatments, predict disease progression, and help support clinical decision making. So like when you're thinking about fully personalized medicine, I think you can't get more specific and targeted, you know, than this.
Exactly. And I think you touched upon such an important point about digital twins, which is that it's holistic, that it's from start to finish. It's not just this software that tells you to take the correct drugs, it's the whole person. It's nutrition, it's lifestyle. It's what might happen in the future, preventative care. And I think it's all encompassing and holistic also, in terms of kind of the life cycle of what is being offered here, where it's not only the software or the model, but also the collection of the data, the storage of the data, the feedback loop and how that all operates, the hardware that's been developed, the sensors that have been developed. So there's a lot more here than just a little kind of online digital twin model that you click on your computer to see. I think it would be really interesting to dive more deeply into that entire process with Terry.
Right. And you know, as you mentioned, this technology continues to advance. We're looking at everything from being able to predict heart attacks before they happen to even like optimizing hospital workflows. The possibilities are endless when it comes to the different areas where this type of technology can be applied. But of course, Laila, with all new technology, with, you know, every new frontier that we reach, there could be some challenges, there could be some setbacks, in bringing this type of technology to mainstream medicine.
But first, before we hear from Terry, we want to make this disclaimer. The content of this podcast is not intended to be a substitute for professional medical advice, diagnosis or treatment. Listeners or viewers should not disregard or delay in obtaining medical advice for any medical condition they may have, and should seek the assistance of their health care professionals for any such conditions. Here now is our incredible conversation with Terry Poon, CTO of Twin Health.
Terry, thanks so much for your time today.
Thank you for having me. It's great to meet you.
Certainly. Okay. So let's first dive right in with what exactly is a digital twin in health care? And how is that different from, say, traditional data analysis or monitoring?
So for us what we focus on at Twin Health is on applying, digital twin technology to improve metabolism and really address the biggest epidemics in the US and around the world, which is metabolic disease, including diabetes, obesity and other conditions. For digital twin technology, it's a software model of complex physical systems. It was originally developed for use in the Apollo missions to send rockets to space in a safe way. And, we're applying the same technology towards the human body to be able to understand the complexity of the human body and metabolism and the various conditions that happen and how to improve that over time using very precise recommendations.
But Twin Health takes a different approach to many healthcare AI companies focusing on reversing diseases, rather than just monitoring and managing them. What successes have you seen?
At Twin Health we really measure ourselves on the outcomes. So it's about health outcomes, for our members as well as cost savings for employers. And what we have seen is that we've been able to really help our members reverse their diabetes, as well as lose weight over time. The American Diabetes Association published a paper where they recognize Twin for achieving the highest rate of diabetes remission, across all of the solutions out there. And, we see that, we're able to help our members reduce blood sugar, reducing by 2.3 points on average of their A1C, as well as losing an average of 13 pounds.
Terry, I want to move on to just the astounding complexity of metabolism and metabolic systems, which are incredibly kind of stochastic and uncertain and dynamic. And I'm from a medical background myself. And I was always told that, you know, the possible combinations and interactions that various cells can have in biological systems at multiple scales as well, is just so difficult to capture. So how are you guys capturing that complexity?
Absolutely. Metabolism is very, very complex. And, what we see is that with the sensors now, the wearable sensors, you capture so many aspects of your metabolism, in real time, continuously. Using the sensors we capture about 3000 data points per person per day. So this is everything from your heart rate to your blood sugar to your different activities every day. What you're eating, how that affects your blood sugar. As well as activities like your exercise, your stress, your sleep. Using all of these data points, we're able to form a picture of how your metabolism is uniquely responding to these different actions that you take every day.
At Twin, we have helped many members improve their metabolism. Overall, we have about 10 billion data points, across all of our members. And that has helped us learn about different types of metabolism and different responses that people have tied to these different activities. And, that's where we're able to understand using machine learning and AI, the complexity of metabolism, what drives different, reactions to activities. And then we've been able to generate recommendations based on that.
Wow. That sheer volume of data is just astounding. And I think that actually brings me quite nicely onto my next question, which is about the scale of data, where I guess in other areas of AI and particularly large language models, these models have shown resilience to kind of low quality data by compensating for scale.
Yes.
And would you say this is something that you also see in healthcare when it comes to the models used in healthcare? And in addition to this, obviously we've seen kind of new approaches which don't necessarily require scale or don't immediately look to scale to improve the models. So are there any new techniques that you've kind of looked into?
We combine data quality with data scale. To achieve high accuracy in our AI and our machine learning predictions. On the scale side, you know, we spoke about the 10 billion data points. On the data quality side, we have a patented technique. So overall at Twin we have 10 patents. And one of our patents is on achieving high TAC. That's an acronym that we have at Twin for timeliness, accuracy and completeness of data. And we have developed mechanisms to achieve this TAC by analyzing all the data that we get from sensors and ensuring the timeliness, accuracy and completeness.
We have correction techniques for data, ways to check for issues with that data. And that really goes end to end. So we're measuring not only the sensors themselves, but also understanding our members behaviors. For example, if we, see that a member has a blood sugar spike, a pretty significant one, but, you know, it's meal time and they didn't report a meal. So that's one of the examples where we're able to detect a member behavior and prompt the member to log in that meal they may have missed.
Yeah, that's really cool. And I also imagine that your models probably are able to work with missing data or incomplete data. And probably your sensor data, I imagine, is collecting data which is analytical as opposed to self-reporting, which I think plagues the field of healthcare quite a bit, that a lot of data is actually just self-reported, which doesn't give a definitive kind of, picture of what's going on.
So I do want to talk, though, about the data that is collected by Twin Health, and then it's interpreted by a combination of machine learning models, and then human health experts called these Twin Health coaches. Do you think that this combination approach is critical? And what can AI do well, where, as Laila just mentioned, humans are still a little essential.
So at Twin, we really see the combination of three things as the key to achieving improvements in metabolic health. And those three things are biology, behavior and empathy. The first two, we've developed a lot of technologies, to really achieve, deep understanding of biology and behavior. So using sensors, we can understand, you know, what makes a person's biology unique and what makes their metabolic responses unique, with behavior also, we're able to detect a lot of that automatically. So we can track exercise and sleep without self-reporting as the sensors can detect the number of steps you're taking or the quality of sleep that you have.
So that deep understanding comes from our AI. But just as critical is the deep empathy that our care team provides. So, we see that, you know, every one of our members, they go through, different challenges in their work and their life, different stresses, really. That human connection is so critical, to achieving, changes in your metabolic health, through changes in behavior. So that's really where our care team plays a critical role in understanding what's going on in our members daily lives and then helping to deliver the recommendations in a way that makes sense for people, that they can really incorporate, in a sustainable way and long term.
Do you feel like this data could be extrapolated and applied to other diseases or overall public health?
Certainly we believe that digital twin technology can be applied to many different areas of health, especially ones where the conditions affect different people in different ways and you really need to personalize the approach to achieve the best results.
That rings so true, just especially with metabolic disease, where kind of behavioral management and adherence is just so important. But this also brings me on to my next question, which is about kind of medical professionals and whether they should be defining what we measure and track, or whether it should be AI which is discovering new patterns which we haven't considered. And the reason I ask this is because I guess fundamentally our knowledge in some shape or form, is inevitably wrong, because we probably know there's gaps which we will probably fill in the future when we learn more and we discover more. So should we be putting in what is potentially wrong foundational knowledge into the systems and directing it? Or just let the system discover patterns which would then lead to the correct knowledge?
Human metabolism is very complex and there is always more to learn. You know, so the way that we started is by working with medical researchers and doctors and also reviewing, all the literature that is out there on metabolism and the drivers of metabolic health. With that, we were able to have a starting point of which data points to look at, through sensors, through blood work, through different kinds of app data.
And then, using machine learning and AI, we were able to uncover additional aspects that were really discovered automatically through the data, yeah. So it's really a combination. And, you know, we find surprising connections, for example, between things like, you know, detailed aspects of how your stress can impact your visceral fat or impact different aspects of blood sugar. For some people, exercising in the afternoon is better. For others, exercising in the morning is better. So these are things that we discovered through the data and through our machine learning.
You know, many companies, organizations talk about personalized medicine, but very few have achieved it at scale. So what do you think from your perspective is the key to actually delivering personalization that works?
From our experience we see two main keys to making personalization work well. One aspect is taking a holistic approach. You know, we see that for metabolic health, it's very complex. And so the different aspects all interact with each other. So, you know, you cannot look at just blood sugar, you cannot look at just heart rate, you cannot just look at blood pressure because these things all impact each other. And then what you do every day with, food and exercise and sleep and stress, they all impact each other as well.
So, so the first key, I think, is, is to look at it holistically, you know, rather than measuring just one aspect and look at the human body holistically and try to understand the whole system as much as you can. The other key is to have a really tight feedback loop. So with behavior change, the tricky thing is, how do you know that it's working? If there's a recommendation, you know, you take that recommendation, but you don't know if it works or not for six months. It's very difficult to sustain that change. That's where we've seen, you know, traditional approaches maybe not work as well.
With sensors that measure continuously, there's an opportunity to really have a feedback loop where you can tell right away, you know, in maybe a few minutes, ideally, or at most a few days, whether your change worked or not. And you can very quickly iterate on that and keep optimizing. And with that feedback loop, then that really encourages people to keep going as well. They see that the small changes they make are working, and then they keep going. And, you know, the small changes can compound into big benefits over time.
Okay. Before we let you go, one final question from me. What advice would you give to others that are maybe trying to bring AI into health care? Is there something in particular that maybe they need to think about or understand that might not be obvious?
Healthcare is very complex. There's the human body aspect, but also the end to end aspect of healthcare delivery, and how every person's health is really quite different. One thing I would say is, you know, definitely, approach it with humility and a spirit of learning. You know, there's so much that I've learned about my own body and metabolism through working on it, coming at it with an open mind, and really understanding each person as an n of 1, helps a lot.
Yeah, absolutely. And I also just on that note, I think we'll probably look back in the future and be horrified that we treated people like generally rather than on a personalized level. Thank you so much, Terry.
Thank you.
Thank you for having me.
It's really interesting. I mean, I think with everything that he's laid out, I think your point that we will look back and look at the very generalized medicine application that we've utilized for how many decades now, like, will seem so archaic and also a little bit barbaric.
I think his point about, stress, sleep, etc. and it varies for each person. And I think that's so true in that I'm not applying the same approach to everyone. It can't be right. And I think this is the best step in the right direction to addressing that.
One of the things that, I didn't think about that he mentioned is, you know, in the three components, that they really seem to focus on the empathy component.
Yeah. But it's so true when you're talking about adherence and you're talking about like intrinsic motivation. The medicines are pretty good, the drugs are pretty good. But it's the adherence to these regimes. And like the behavioral management and the empathy, like the relationship between the physician and the patient. And I think it also speaks to the feedback loop he was talking about, which I really, really hadn't thought of when I was kind of researching Twin Health, which is when you have a much tighter feedback loop, you almost get little, encouragements much more frequently that like, yeah, this is working like incrementally and it compounds into massive effect.
Yeah, yeah. I mean that constant, feed of information that, you know, these patients are getting, it's like giving them little pieces, little glimmers and little bits of motivation. So that does it for today's show. If you enjoyed our discussion about digital twins, don't forget to like and subscribe and share, as this really helps us to improve the show for you. Thanks so much for taking the time to listen today. Until next time, goodbye.
Goodbye.
Agents of Tech is brought to you by WebsEdge. Special thanks to our studio engineers, Adam Deen and Sam Saris, and our editor, Abbie Harries. Agents of Tech is produced by Cath Sheehan.