Interview with Health Universe CEO Dan Caron

Dan Caron takes us on an inspiring journey through the landscape of healthcare tech innovation. When Dan hacked an insulin pump to create a groundbreaking automated pancreas, he not only changed the game for diabetes management but also shined a spotlight on the  potential of open source technology to redefine our health experiences. 

From his father's ingenious barter system sparking a young Dan's tech curiosity to the evolution of his career, merging the worlds of internet business and healthcare, his story is one of relentless passion and the drive to make a tangible difference.

Healthcare is on the cusp of a revolution, and AI and open-source technology are key enablers. Dan and I dissect how cooperative platforms like Health Universe are dismantling the walls that have kept life-saving algorithms in academic limbo, instead ushering them into hands-on clinical use. Understanding how this technology evolves from concept to real-world application underscores the vital role of platforms that provide immediate access to advanced tools, and Dan's insights guide us through this new age of medical innovation.

Host David E. Williams is president of healthcare strategy consulting firm Health Business Group. Produced by Dafna Williams.

Transcript

0:00:11 - David Williams
Dan Caron co-founded a successful healthcare startup and followed that up by hacking an insulin pump to create an automated closed loop pancreas used by tens of thousands of people, and now he's running Health Universe, a collaborative platform for Health AI. Hi everyone, I'm David Williams, president of Strategy Consulting firm Health Business Group and host of the Health Biz podcast, a weekly show where I interview top healthcare leaders about their lives and careers. If you like the show, please subscribe and leave a review. Dan, welcome to the Health Biz podcast. 

0:00:42 - Dan Caron
Thanks, David, great to be here. 

0:00:44 - David Williams
You know, I know that we talked about closed loop and I see you're wearing some sort of a loop. Is that like a loop to the pancreas or is that just a regular old ear pod thing? 

0:00:53 - Dan Caron
Regular old ear pods. I find them to be more effective than all the Bluetooth stuff. 

0:01:00 - David Williams
That sounds good, excellent. Well, let's talk a little bit about your background and upbringing. What was your childhood like and any childhood influences that have stuck with you in your career? 

0:01:08 - Dan Caron
Yeah, so my father owned a floor covering business and he put in flooring, carpet and tile and when he would meet someone that was sort of into computers or technology he would say, hey, you know, I'll carpet your house for free. You know, just teach my kids some stuff. And I had a few mentors, one of them, wally Bysick. He taught me basically computer science and electrical engineering when I was really quite young. I mean, second, I couldn't drive, I was dropped off. So I was probably 13, 14-ish, a few years, and really I just love, I love technology. 

I saw Jurassic Park and I saw those dinosaurs and I was fascinated. I asked my father, how did they make those, you know? And he was like, oh, they use computers, and that was it. Like I was like I want to know how to do that and I started building objects in 3D, 3d animation, doing digital design, and that led to programming. I was sort of fascinated by reverse engineering and hacking and how do computers work? And I just spent all my time reading books in Barnes Noble. I'd stand in the computer section in Barnes Noble reading just the nerdiest stuff and I loved it. I couldn't get enough of it and that passion has carried on to this day. Really. 

0:02:38 - David Williams
That's pretty cool. So I want to ask these guys that were getting the trades with the carpet Did they get a? 

0:02:43 - Dan Caron
good end of the deal, or did they? 

0:02:44 - David Williams
you know, did their floors look good. 

0:02:47 - Dan Caron
Oh, of course. Yeah, they're Stone Cold Pro, one of the best in the business. So yeah, they get a great deal. 

0:02:54 - David Williams
That's awesome. Okay, so besides this sort of informal education and training and, you know, going not to the library but the Barnes Noble, did you go to school too? 

0:03:05 - Dan Caron
I did. I went to Northeastern University for entrepreneurship, marketing and computer science and even in high school I was really fortunate to have I went to Quabbin Regional in Central Massachusetts and Rich Zalonaritis was a fantastic networking computer teacher, really gave me a free reign to study computing and networking and I helped manage that network. So even in high school I really had quite a bit of support from teachers that were allowing me to just kind of pursue my interests and I'm super grateful for that. 

0:03:42 - David Williams
That's cool, so I guess you weren't from the part of Quabbin that they like created the reservoir on and bury the town, is that? 

0:03:48 - Dan Caron
It actually is. It is named for that region of the state, for sure, nice. 

0:03:53 - David Williams
Very good, Okay. 

0:03:56 - Dan Caron
So when you got out of school, what did you do for your first jobs? I figured that I wanted to build internet companies. When I was probably 15 or 16, I had started a few different websites and they made money and I said, wow, even in college I had websites that were making money and I realized that the internet was a really powerful platform that was going to be transformative for commerce and the way we do things. And after college I went and worked for strategic profits and what we did was we helped teach people how to build internet companies right. What are the things that you need to think about in order to have a successful internet business? What are the systems, the processes in place? It was a little bit of process, automation of hiring, of marketing, direct response marketing, product launch marketing all rolled up into one and that gave me a really good general foundation for internet business and it was a nice natural extension of what I had studied in school. 

0:05:06 - David Williams
Pretty cool and I know you from also RxReview. What is, what was RxReview and what was your role there? 

0:05:13 - Dan Caron
Yep. So I was one of three co-founders, alongside Karm Huntress and Kevin O'Brien, and we built a prescription optimization platform, really trying to find what is the highest quality medication at the lowest cost for the patient and bringing in pricing data to allow physicians to make more informed decisions. And that has been quite a successful company and I really helped build the initial tech foundations, the initial marketing foundations. I consider myself sort of an early stage startup person. I like when there's nothing and there's an idea and you have to mold it into something viable and it's challenging, but I love that active creation. 

0:06:00 - David Williams
On these early websites that you were working on. I don't know if they were healthcare related or not, but there's a long history that even predates RxReview People coming from the technology side looking at healthcare and saying what bunch of cavemen that are there and we just put technology in it, we're gonna fix everything. Were you one of those tech guys coming into healthcare? Or how did RxReview just happen to be healthcare? What's the connection? 

0:06:25 - Dan Caron
Yeah, yeah. Well, karm had called me one day and he said you know, I'm thinking about what's new, what's next, and he rattled off a list of different opportunities in front of him and he's one of my dearest friends and we were sort of brainstorming and looking at thinking about these different opportunities and one of them came from Kevin O'Brien and he had catalogued all the different permutations of ways to save money on medications and he was going to publish a book, I think, and I said no, I said we're not gonna publish a book. I said we're gonna create a database and we're gonna create an API company and this is something that healthcare can use. And so, yeah, I definitely see myself as a bit of an outsider, but you know, being a type one diabetic, I'm close to healthcare, I interface with it quite a bit, and now I sort of consider myself a little bit more of a seasoned veteran, if you will. 

0:07:26 - David Williams
Yeah, yeah seasons, if not wily. 

0:07:30 - Dan Caron
Exactly. 

0:07:31 - David Williams
Yeah, okay, how about Dark Pilot? What was that? 

0:07:34 - Dan Caron
So Dark Pilot is a consulting company that I have. Dark Pilot takes all of my experience of building internet companies and I provide consulting services around that. A lot of that is understanding how to scale internet products, how to scale internet businesses. How do you create a value proposition that makes sense, how do you do it in a way that is compelling and really you know solid business practices that are centered around technology. 

0:08:07 - David Williams
Got it Now. What about this pancreas Closer? 

0:08:12 - Dan Caron
to pancreas. 

0:08:13 - David Williams
What's the deal with hacking an insulin pump? I hope it was your own pump and not somebody else walking down the street. 

0:08:19 - Dan Caron
It was my own. You are legally allowed to reverse engineer your own medical devices. And I was at lunch with a friend of mine one day her name's Rachel and I said to Rachel she was also a type of diabetic and I said, you know, if someone just hacked the insulin pump and the sensor, we could actually automate the delivery of insulin up or down. And she kind of you know, laughed and said, well, you know, why don't you do it Smarty? Yeah, and I was like that's kind of a fair point. You know, we all talk about how we wish things were better in the world, but what are we doing really to make it better? And so I said I'll give it a go. And so I went and I bought a software defined radio which is a USB digital radio, and then I plugged that into my laptop and I started reverse engineering the wireless signal. I was the first one to do this and I started documenting that on a Slack channel and there was a group artificial pancreas group that we kind of all consolidated around and I just started posting what I was finding. This is the frequency, this is the modulation, and some other folks joined in and helped that process and eventually we were able to understand most of the wireless communications and we almost had total control of the pump. 

But there was one, basically checksum, that we couldn't figure out how it was being generated. It was kind of called a knots and as computer security folks would call it, and pretty much the project went dormant because no one could get by it. It required really nation state level hacking. It was non-trivial, the firmware was not accessible to us All, the debug ports were closed and it just so. 

I found one of the best hardware hackers in the world at the University of Oxford and I emailed him and I said hey, I'm working on this chip, I'm trying to extract this firmware, we're trying to control this pump. And I knew I peaked his interest because I knew how to talk to him as a reverse engineer and I said this is a puzzle, help me figure this out. And we had a scanning electron microscope pointed at the memory space of a small chip and we had a lab in California that was helping us and we were going to extract the electrons as bits, as ones and zeros, to use that to up, compile into pseudo-C code to figure out this function that was generating this little checksome that was on all of the insulin delivery packets. And I tell you this because we went to the ends of the earth to be able to control this pump, because of the desperation for solutions. 

Type 1 diabetes is a very burdensome illness. Every day, around the clock, the illness is there and it is difficult to maintain blood sugars that will keep you healthy. And the highs, of course, are very detrimental to your organs and every system in your body, and the lows can take you out and put you in a seizure or a coma, and it's quite burdensome. So we were looking for solutions and as a group, as an open source community, we found that solution by being able to control that pump. And once we were able to control it, we integrated that into Loop, which is an iPhone application, and it became the first open source medical device that was approved by the FDA. 

The software yeah that was carried on by a group at Tidepool. They picked up the project and maintained it. There's been a lot of work done by those folks. There's been a lot of work in the open source community by the people that have done all the documentation for this project. But it really showed that an open source community of people could come together and create a real solution for disease state. 

And last I checked, there's tens of thousands of people using this. I think 40,000 was the last number I heard and this was a while ago, and there was a Facebook group that sprang up around it and people were posting some really amazing things. They were posting pictures of their kids, their three-year-olds, five-year-olds, running Loop and saying they're crying because of how life changing it has been for them and they can sleep better at night, knowing that the system will pull back on the insulin in case they go low. And sharing their A1Cs that have dropped one or two points, sharing their blood sugar graphs that are effectively. Even One father said there's nothing that he owns that he wouldn't sell to have the system for his daughter and it was really emotional for me. 

I still get emotional telling the story because it was transformative for me and for everyone. That was part of the project to see this kind of result created in the world and it gave me the intuition to understand like what I want the rest of my life to be dedicated towards. I had built RX review and built Arrived Health into a very successful company and after seeing the real world results of this insulin pump project, I just thought that's what I want to do for the rest of my life. I want to use technology and scale that as best I can to help as many people as possible. 

0:14:15 - David Williams
So that makes sense. So, RX review, do the pump hacking and then expand your notions. And now, of course, the next company has to have the word universe in it health universe. That's fairly broad, so I'm sort of starting to see the through line here. But what was the need that you saw to found health universe? 

0:14:35 - Dan Caron
Well, one day I was looking at GitHub and looking at Huggingface and I was thinking about my experiences with the Loop project and hacking the Omnipod and thinking about open source. It sort of just hit me like a bolt of lightning. I was like there's no platform for health AI, that's an open-source, collaborative environment. And All of the talk these days around AI and healthcare a lot of it is around how do we, how do we Introduce this safely into healthcare? Because we know that GPT-4 can hallucinate. We know there are problems. We know that, that, that you know sometimes it will come up with Information that's incorrect and in healthcare you you can't have that right do no harm. It's. You have to be incredibly Cognizant of the type of information you're putting in front of a provider. And I, you know, I sort of understood that from my time at Rx review. Right, we're building prescription decision support systems and so there's a certain level of integrity that you need an infidelity in your data To make sure that that no harm is done. But at the same time, right, providers are getting crushed by Sort of the economics of the times. We're seeing more physician burnout than ever. We're seeing more physicians dropping out of practice. We're seeing less people entering into into clinical practice because of the dynamics, because it's it's difficult. 

I mean, pandemic was was obviously brutal in a lot of ways, and so we need AI and we need ML to to come in. 

We need better technology to alleviate a lot of these problems, and so health universe was sort of my, my offering to the world of saying we can Introduce the stuff in a safe way, but the way that we do that requires collaboration. Right, because if you think back to loop and this insulin pump project, the way that we were able to automate the delivery of a deadly hormone To tens of thousands of people every five minutes around the clock, is through open source, because it allows for auditing, it allows for Transparency, it allows for code review, it allows for a Much greater sort of scrutiny, and that's really what we need. If we're gonna bring health AI Into health care in a responsible way, we need to bring collaborators together and I think that there's a there's a certain Tightness or there's a certain let's let's cling to our data and let's yeah, we're gonna develop this in-house and there's a lot of sort of proprietary Methodologies because, for obvious reasons, right, I understand. 

0:17:40 - David Williams
I think in, you know in in health care there's a lot of that right and especially see it, the pharmaceutical industry and other places that are heavily dependent on patents, intellectual property, you know trade secrets and so on. What you did made sense From a, you know, open source standpoint. People can understand. You know hacking the pump and so on. What I'm interested in is is how do you then take that Idea and then make it this health universe and turn that into a business as opposed to like an interest group? 

0:18:08 - Dan Caron
Yeah. 

0:18:09 - David Williams
What turns that into a business? 

0:18:10 - Dan Caron
Yeah, exactly. Well, when I look at academia and I see all of these great research models being developed by graduate students and the academic medical institutions, and I see all these great AI models, we read about these every day. Oh, can predict, you know heart attacks, you know better than a physician, right, we're seeing this in every sort of segment of healthcare, but those tend to just be point solutions that only exist within, you know, the walls of Stanford or Harvard or UCSF or wherever, and to me it's a bit of a travesty that those algorithms and those models are not available worldwide. You know, I want to live in a world where a physician anywhere can use a browser and can run, you know, the latest and greatest breast cancer detection model. I think that we as a society should have that. I think it should exist in the world, and I think that there is a path to monetizing those algorithms as well, because right now, there is no distribution platform, the tech transfer offices at these academic medical institutions. They don't have a viable pathway to commercialize this stuff outside of their four walls. 

And Health Universe really is that platform. 

It's a platform that you know we're aligning with some of the ONC guidelines that are coming out around health AI and AI and clinical decision support, and we're also looking at, you know, what the FDA is requiring for AI algorithms as well, and we're aligning with those as well. 

And really we're building a platform that is not only open source but supports closed source as well and something that we call mixed source, because there needs to be transparency in order to have accountability and explainability with these algorithms, but then parts of a model can be private, so you can imagine the metadata around the data that an algorithm was trained on should be visible and it should be accessible to a clinician and it should be described in a way that a non-technical user can understand. And we're really trying to bring all the best practices of open source software development and AI algorithm development together into a platform that allows researchers and clinicians to meet and work with these models in a way that gives them a sandbox, a testing environment, a way to approving ground, if you will, to move a model or an algorithm from a prototype to a stable, to a clinically validated, an institutionally validated and maybe even FDA approved. But the only way you can do that is if you have sort of this proving ground sandbox environment, and that's really what Health Universe is all about? 

0:21:03 - David Williams
Got it. So I understand the example of the insulin pump hacking, which I understand is an earlier approach, but what are the sort of apps that you're seeing on the platform and where do you expect it to get to? What sort of aspiration do you have for what's possible? 

0:21:20 - Dan Caron
Yeah, that's a great question. 

So on Health Universe, we support Python applications, specifically Streamlit and Fast API, and if you're not a programmer you may not be familiar with those, but what that means is that almost a good majority of the machine learning and AI that's being created today is written in Python, because it's a very accessible language, and so we allow all of those Python applications to be deployed through Health Universe and it really only takes like three minutes to basically publish a project on Health Universe and what that means is that, instead of a researcher sort of publishing their preprint on archive and then putting their code up on GitHub, where if you're another researcher and you want to download that code, you have to figure out how to install it and all that, it's much better to publish it on Health Universe because it can be run immediately in the browser with nothing to download, nothing to install, and that gives that researcher more distribution for their work, that gives that researcher access to clinicians to interact with it, that gives that researcher the chance for more impact, more citations, and it allows for that sort of proving ground, that collaborative proving ground, to mature that algorithm. 

0:22:42 - David Williams
So if we go back to sort of the origins of RX review and talking about, someone had the idea hey, I have all these ideas, let me put them in a book. And the internet was able to kind of scale that up and make it more accessible, more flexible. You can update it over time. Maybe it's a similar concept here Researchers going to publish a paper, they're going to have some code in a Git repository, but that's quite limited in what it could do and here it can live and it can grow and it can get more distribution and more use. Is that the concept? 

0:23:09 - Dan Caron
Yeah, that's exactly the concept I find it to be. It's really just. It's unfortunate that there are some really amazing machine learning models in healthcare that only exist in small research settings and they never make their way out to help patients and help clinicians. I mean, it takes years oftentimes to move from research into the clinic and I just I want to shorten that window. I think that you know it's a really valuable thing to do for the world. 

0:23:39 - David Williams
So what sort of you know use cases or applications lended themselves to this approach? Because I think you know, maybe a year ago is chat. Gpt is being announced. People are trying it on. You know all sorts of things and there's some places that fits better than others, but where? What are some of the early applications that you are seeing or hope to see? 

0:23:59 - Dan Caron
Yeah, of course, radiology and imaging super successful. Machine learning excels there. We find machine learning we see it excelling quite a bit in oncology, right High dimensional data, in protein design and drug development high dimensional data as well. So we find that those areas are well suited for machine learning. Healthy universe also supports basic rules based models, calculators, things like that Anything that you can write in Python you can deploy on Healthy Universe. So it's not only ML and AI, but we do support all of that stuff. 

Generative AI, of course, things like GPT-4, mistral, a mixture of experts, all the open source LLMs there. I mean the fantastic tools, right. But One of the problems with them is that we don't necessarily know when they're hallucinating or when they're not hallucinating. So we have to catalog when there are bad responses and that gives us a way to quantify the risk for that model. And clinicians on Health Universe can give a thumbs up or thumbs down and provide feedback to say, you know, for this particular use case, we're seeing bad results here, right. And it brings the physician into the generative AI world and the world of health AI, which I think is exciting because I've heard over and over from clinicians that they want to be part of the development of AI. They want to co-create that world. It shouldn't just be programmers prescribing what should happen, like there's so much wisdom from clinicians that have been practicing for years and they need to be part of it and they need to be brought in and that's part of the Health Universe. Ethos is that collaboration layer. So bringing the physicians in into that process is useful because now you can start to assess the risk a little bit more than just you know hoping for the best with a black box model like OpenAI. 

To your point about what models, though, I just want to kind of answer your question there. You know, of course, administrative models, things on the more documentation side that are a little bit you know, aside from the clinical decision support tools. Of course there's tons of opportunity there. You know claims and documentation, administrative stuff. That's a little bit easier because it's much lower risk, and so of course we're seeing tools being introduced there first, before the CDS tools. But really the sky's the limit. I mean ML and AI and specifically agents are going to kind of change the way healthcare is done in the next few years. 

0:26:47 - David Williams
Got it. So you know you talked about how it could take years and it's probably sometimes more like decades from something to go from discovery to actually be used in the clinic, and I think some of it relates to what you said with RxReview is there's like not a lot of tolerance for error. You know you don't want to do any harm, and that's in contrast with the sort of you know, move fast and break things ethos from Facebook or Meta in the tech world. And yet there is, I won't say a happy medium, but it should be the case that we don't have to settle for things to be so slow and we can use other approaches to get there, and I think that's what you're saying partly about. 

you have a way to label what's the maturity of these models and to help them to say a little more about that. 

0:27:28 - Dan Caron
Yeah, exactly. So the Health Universe team participated in the Division of Clinical Informatics Working Group at Beth Israel, harvard, and what came out of that was sort of the idea of a nutrition label, and that's what you can see on the Health Universe inside the platform. You know, we're creating this nutrition label, which is sort of like a nutrition facts right. It tells the user a little bit more about the algorithm and the model, how many times it's been executed, who created it, where is it from, and we're seeing guidance from ONC on additional fields that they want to see in order to have an understanding of how a model works and how safe it is. And the reality is that we have to sort of think about AI in a different way. My friend, graham Walker, from MD Calc, I think, put it very nicely on LinkedIn the other day. He said verify, don't trust. Yeah, it's a little spin right on trust, but verify. 

0:28:26 - David Williams
That's what Ronald Reagan said. Right, I was talking about the. I think the MIRV rockets, right. 

0:28:31 - Dan Caron
Yeah, yeah. 

Make sure the Russians didn't have too many of those, yeah, verify, yeah, and that really is sort of the psychological frame that clinicians have to come in. And of course there are other things like automation bias, because clinicians will trust the AI more than they should, and so there needs to be education and labeling and there needs to be interface elements that sort of warn them and remind them that you have to verify this stuff. This can help your workflow, it can help you discover things. It's amazing for a lot of reasons, but until we have sort of perfect AI, we have to sort of think differently and we have to verify first and foremost. 

0:29:22 - David Williams
So I want to connect a couple of things you said. So I'll talk about agents in a second and you're talking about how AI can hallucinate. Some of the ways I tend to think about it is it's more like a sociopath it can talk to you and it's like it's so convincing that sounds exactly right and you look it up and it's just like the lie versus the truth. It comes across the same very smooth. Yeah, Does the agent help deal with that? 

0:29:43 - Dan Caron
Yeah, well, it's a really fascinating area of research. We've seen from the Med Prompt paper from Microsoft that system prompts can perform better on medical evaluation tests than plain vanilla GPT-4 alone. So what that means is that system prompts can increase the accuracy and the reliability of these models by having well-curated, fine-tuned, thoughtful system prompts. And what we're building at Health Universe really is a collection of medical co-pilots, because we see a world where an endocrinologist has different needs than, say, a heart doctor, and their co-pilot needs to be fine-tuned for the work that they're doing in order to reduce the chances of hallucination or generating unsafe outputs. 

0:30:37 - David Williams
Got it Okay. So you mentioned it's open source, mentioned it's for researchers. Does it make sense for just any regular person involved in healthcare to go and check out Health Universe? Is that allowed? Allowed? Do you have to have an application? 

0:30:51 - Dan Caron
Do you have to be approved. Right now. It's fairly open and we do encourage people to get on there and try out some of the models. We do have some closed programs at the moment, but, yeah, I think all your listeners are going to be more than happy for them to take a look. 

0:31:08 - David Williams
Sounds good. Well, my last question here and turning away from Health Universe is about reading. You have a chance to read any good books lately or anything that you would recommend or recommend to avoid also. 

0:31:23 - Dan Caron
I'll stay away from the avoid category, but I think Build by Tony Fidel was I mean, I've read hundreds and hundreds of books, but Build by Tony Fidel absolutely one of my favorites. He ran the iPhone team at Apple and he also helped to build Nest and really I just thought that book was a phenomenal. Every page was gold and how he thought about building a business and creating disruptive innovation was, I mean, it's a wonderful ride. Every page is great. So I highly recommend Build by Tony Fidel Great. 

0:32:06 - David Williams
Well, Dan Karen, co-founder or founder of Health Universe and CEO, I want to say thank you so much for joining me today on the Health Biz podcast. 

0:32:15 - Dan Caron
Thanks so much, david, it was great being here. 

0:32:20 - David Williams
You've been listening to the Health Biz podcast with me, David Williams, president of Health Business Group. I conduct in-depth interviews with leaders in healthcare, business and policy. If you like what you hear, go ahead and subscribe on your favorite service. While you're at it, go ahead and subscribe on your second and third favorite services as well. There's more good stuff to come and you won't want to miss an episode. If your organization is seeking strategy consulting services in healthcare, check out our website, healthbusinessgroupcom.