Who Gets Counted in Health Data?

Chris Hutchins: Welcome back to The Signal Room. I'm Chris Hutchins, the host of this show, and I am really excited to have a special guest with me today. We're going to talk about some really fun things from a very different perspective. My guest today is Lisa Marceau. Did I say that right?

Lisa Marceau: Yeah, yep.

Chris Hutchins: Okay. Lisa's the founder and president of Alpha Millennial Health and the co-founder of Joyuus, a digital health company focused on postpartum care. Lisa has worked across clinical research, health technology, entrepreneurship, and executive leadership. She's an author and co-author of more than 50 peer-reviewed publications. Her work begins beneath the usual AI bias conversations, in the infrastructure, governance, and patient questions that really determine whether a model has trustworthy foundations at all. Lisa, welcome. I'm so glad to have you on the show.

Lisa Marceau: Thanks for inviting me. I'm really happy to have this conversation. So thank you for taking it on.

Chris Hutchins: It's a pleasure. So, to get people grounded a little bit, maybe tell us about your background and the path that has led up to the clinical research you're working on and the advisory services you provide. You've got some really interesting perspective, and I think it would be phenomenal just to understand how all that came together and how you find yourself where you are today.

Lisa Marceau: Well, it's a great question. It might seem that I had a clearer pathway to get where I am, but really the short version of my career is that it was a squiggly line. I ended up in healthcare almost by accident, just by where my summer job landed me, and it snowballed from there. It was interesting because my work has really led across clinical research. I worked in a clinical research organization early on, and for 23 years we evolved that work. And that work led to both work within the healthcare system and within entrepreneurship. As part of that work, I did clinical research as a primary researcher, but I also led a business unit where we developed product to fill a gap between clinical best practice and real world interventions. And then I ended up being an executive leader of that company for over a decade. We were ultimately acquired into a Fortune 25 insurer, which actually changes the perspective a little when you come from the outside, doing the clinical research, to how the healthcare system functions. That added a layer of information for me, which was really interesting. And during that time it really started to create the questions in my mind, based on all of this work, that it was time to step out and do something different, to really support the entrepreneurs who are changing the way health is developed, consumed, and used.

Chris Hutchins: You said something interesting just now. There's apparently a difference between best practices and real world medicine. Can you talk about that a little bit, so people understand what you're referring to?

Lisa Marceau: I was fortunate during my career to be able to look at both sides of that equation. Clinical best practice is the gold standard, and evidence-based research is really important to make sure we understand what those best practices are. But the reality of our healthcare environment is that many, many people can't achieve best practice. They don't have the resources, they don't have the access. And we did a lot of work studying those challenges as well. What was really interesting was that we were able to start to look at it this way: if this is the best practice and someone can't do those best practices, how do we create interventions that are evidence based but still allow them to do something that has evidence as an underpinning? That was really important work that we ended up doing.

Chris Hutchins: Fascinating. There's a ton of conversation around models and what trains them, whether it's good, whether it's bad. I go back to the concept of bias pretty consistently because there are so many different angles to it. You've argued that many clinical models are trained on data that does not adequately represent the patients they will treat. Can you make that a little more concrete for us? What kinds of scenarios are we talking about?

Lisa Marceau: So I think that's a really important question to have around data models, and I love this topic of conversation. When you do clinical research, the fundamental truth is that what goes in is what comes out. You've heard "garbage in, garbage out," but it's more than that. For some time we have known, and I've done this research for decades, and this is a really good example that I like to use, that when we look at some of the research data that we are importing, and this becomes even more true for AI models, the clinical and claims data stored in an EMR isn't always accurate. In the work that we did in our research, which has translated into the work that I do now, we did research studies of different patients with the same symptoms, the same exact case presentation. What we would see was that the diagnoses for different patients, by different providers within different systems, would differ, and so would the treatment pathways. That may seem kind of obvious, or maybe it doesn't seem super important, but when you think about the diagnostic differences for a condition like cardiovascular disease and the implications for long-term health, it becomes really real. So, for example, in those studies we looked at cardiovascular disease, and women were more often diagnosed with depression than men with the exact same symptoms, the exact same presentation. What that means is that women will go home with similar disease but not get the treatment early enough. Then they come back with a worse disease later in the process, which leads to higher case fatality rates. So women more often die from heart attacks than men because they are not caught earlier. When you see these differences, you see that the impact is real in real terms.

Chris Hutchins: That's really fascinating, but also a little alarming. As far as we've come in the advancement of medicine over the years, there are still some pretty significant gaps. What you're talking about just seems like something people would be connecting the dots on more often.

Lisa Marceau: Yeah, a little bit. Right. I think that's part of the challenge. We really have to understand that data, however we want to call it, data bias, data disparities, data differences, we have to start looking at the whole picture of data, because when we see the entire data set, we can actually make different decisions about health, not just for women in this example, but for women and men. There are some really good examples of that, particularly during COVID. What I find really interesting about that particular example is that more men actually died during COVID than women from heart failure. What if we had understood the biophysiology, the hormonal and immunologic impacts and differences between those two populations? We might have been able to create differential treatment pathways based on that broader knowledge. So it's really about whole data. Bringing in these different populations is critical, because we don't have the whole picture.

Chris Hutchins: I definitely want to make sure we hit on some fundamental things that people have started to realize in the last few years. I don't know that they really understand it across the board. But just the example you gave, that there are conditions that can lead two different individuals with the same diagnosis, who have basically the same demographics for that matter, to have different characteristics that shouldn't be material but often can be. What I'm getting at is, you've done some work in the postpartum space. There's a fundamental difference in a lot of ways between men and women. It's not just an emotional thing, there are physical differences, and medicine hasn't made the directional corrections it should have made a long time ago. What's your thought about that?

Lisa Marceau: I think there's a real logistical challenge that has happened. We're starting to see that shift, but it's early and we really need to be doing more. When I use cardiovascular disease, it's a really clear example of a chronic or serious condition where you can clearly see the differential effects. The other example I like to give is the difference between ovarian and prostate cancer. Why this is such an interesting example to me is that 70 to 80% of ovarian cancer is diagnosed in the later stages, three or four. 80 to 85% of prostate cancer is diagnosed early, at pre-cancer or stage one. This is directly correlated to your five-year survival rate. They are both notably hormone-related cancers. They are both notably really hard to diagnose early from symptoms. So why do we see such a difference in the way these conditions are diagnosed? The simple answer is there was an eight-year period where prostate cancer research was the priority and ovarian cancer was not. The way I look at it, if you don't look, you can't find it, and if you don't find it, there's no data to represent those outcomes. So this is another example of the problem you're asking about. We need to start looking at those data. We actually now have broader tools. We can talk a lot about the AI value and challenges, but we have more tools. We should be looking at this more broadly.

Chris Hutchins: Yeah, I think it's important. You mentioned differentiated health and the historic approach to research, which for better or worse is more often than not aggregated. It doesn't necessarily have all the detail underneath it that we would want if we're a clinician. What are some of the things that get masked in that aggregation that would be really important from a research perspective, perhaps for early detection, or for things that just don't end up getting diagnosed until a late stage?

Lisa Marceau: I think a lot of the data challenges really come down to asking the right questions. Number one, whenever you do research, and I think this applies to everything we do, but especially when you're building a product or doing research, the fundamental is asking the right question. What are we trying to solve? Then you have to start to ask the data quality and data governance questions that really impact how those data are brought in, captured, and used. Who owns it? How is it used? How does it inform what we're trying to solve? Those questions really have to be asked early on. In research, it's your hypothesis. You don't create a hypothesis after you collect the data. You create the hypothesis and then you collect the data and answer it yes or no. So when we talk about what's missing and what's masked, we have to start with the right questions, and then we have to figure out: do we have the right data sets? Or do we need to find them? Or do we need to create them? I think that's where we are right now in some of these conversations.

Chris Hutchins: So this really goes to the whole idea of underrepresentation, rather than just a fairness issue. Because the context these conversations have existed in over the last several years is health equity and things like that. Of course there's that, but there are a lot of other underlying reasons this kind of problem exists.

Lisa Marceau: Well, when we talk about the broader system problems, we can unpack that a lot, because there are some really big challenges in the way the system has reinforced those variations, and challenges with accessing care. There are books and books written on that. But what's really important, which is critical to the equity and ethics question, is that it goes back to the data governance and data quality question. You can't actually have quality representative data if you're missing half the population, for example, or a portion of the population where disease is affected differently. When you don't include them, we talk about it in terms of equity and ethics, but I'd like to reframe that to governance and quality, because we solve equity and ethics when we actually start to solve data quality and data governance problems. Bringing in the right data, making sure we are considering all of that, thinking about how we're using it, who we're protecting, how we're protecting it, is really critical to that conversation, because that starts to solve some of those system problems you're talking about. If there are better data in the system, for example, if the data is more transparent, if it's collected more consistently, then clinicians have better information. If those data systems are integrated, for example, clinicians have better tools and resources. A lot of the work that I do is around integration. How do you make sure different systems, different provider groups, community versus clinic versus other resources, are talking to each other? When we start to solve these problems at a fundamental level, it flows across the entire system.

Chris Hutchins: This has historically been a conversation around things like interoperability, when there are data silos that no one even knows about and the data's not there. But there are other things too. I think about the differences in population centers, where they are geographically. The weather can have an impact, the local industry can. There are a lot of nuances, and then there's also the human factor, where there could be cultural things, religious beliefs, behavioral things passed down for generations. There are just a number of factors that may influence whether the data is actually there for you or not. For example, there's this massive gap. I don't know the specific details, but I imagine somewhere between 18 and 40, men are just, we're a little bit thick sometimes, I'll just put it that way. We should go for routine things that we just don't bother going for. There are massive gaps if you really wanted to flip the script and be proactive and prevent later disease, which eventually costs a lot of money. That would have to flip. But that's a gap right now, right?

Lisa Marceau: Well, those gaps are across the system right now. For example, 50% of our counties in the United States are maternal healthcare deserts. That means there are limited or no maternal healthcare services. We live in healthcare deserts writ large across this country, urban centers and rural centers, and it doesn't discriminate against who lives in those populations. Clearly there are population centers where there are lacks of services, which create real challenges. One of the problems we're really facing is how we broaden the tools we have to help with prevention, with treatment, with diagnosis. I think that's one place where we are starting to see technology, innovation, and entrepreneurship addressing the challenges that the system has been unable and unwilling to address. A clear point right now is that we are seeing hospitals close in rural areas where there are underrepresented populations. We are seeing Medicaid being cut. We are seeing these challenges happening right now. But we're also seeing the entrepreneurial space really blossoming in those areas. How do we start to solve real problems in those real gaps?

Chris Hutchins: There are a lot of things from a historical perspective that changed the way we do things. I guess in 1977 or so, FDA guidance generally excluded women of childbearing potential from early-phase drug studies. Then federal policy changed substantially back, I think around 1993. But the impacts of these things seem more grounded in old knowledge than in the reality that clinicians are probably seeing face to face. Talk about some of the challenges. We're trying to bridge that gap, so how do you get to the place where people are really heavily focused on the evidence versus how we've always done things, so to speak?

Lisa Marceau: Well, my excitement is where we're going with innovation, technology, and entrepreneurship, and the way we're addressing health going forward. Take the example of women who were not included in clinical research for almost two decades. A lot of data is missing because of that choice. We've really made inroads in putting women's health on the map, and women's health is now one of the largest growing markets, for example. One of the things you mentioned is that those restrictions were put in place because of the reproductive health years. Well, that was the reason given for the exclusions. The reality is that our reproductive health is 44% of our lifespan. We have 56% of our lifespan that is broader than reproductive health. Reproductive health is critically important. It tells a story of the hormonal journey that impacts so many things, from immunology to heart disease to brain health, you name it. So really the goal right now is looking at women's health as a second market, not secondary to men's health. It's not a primary, secondary order of things. It's that most of the research that's been done has been established on male models. What we're really starting to see is that there is an entire market where women's health is actually different. In the example of cardiovascular disease, in the example of Alzheimer's disease, there are a lot of diseases where women experience the disease differently. When we consider this entire lifespan of health, we can create a much broader data set that informs not just women's health but men's health as well. So what you're talking about is how we don't lean so much on the existing evidence. It's not unimportant, but the question is how we take that evidence and look at what needs to be done moving forward to change and improve and broaden those questions and those answers.

Chris Hutchins: Yeah, it's interesting. I was talking to a physician a couple of months ago, and one of the things that impressed me when I discovered it was that Gen AI models would help me have better recall of information I would never have been able to think about, something I wrote or read ten years ago. He said, very simply, that may be good in terms of recall, but the risk is that you're still basing any kind of decision making on old understanding that very likely has been updated over a period of time. Talk about some of the challenges that exist there, because with some of these data gaps, and not thinking about things the way you do, people might not be basing their decisions on the latest information they could have at their disposal, or even techniques or protocols they could put in place to help a patient.

Lisa Marceau: Yeah, there are a couple of things to unpack there. One of them is that these data gaps do exist, and a lot of this really is about the underpinning data infrastructure we are relying on to make decisions. But this also taps into the idea that AI can make everything faster, better, and more efficient. What's really interesting about that is there is a yes and a no version of that statement. Some of that is probably in the conversation about where those reference points are. When we look at data historically, there are data that were researched and studied, and new research has been done. Those data sets have lived on a server somewhere. They have never been properly archived. That kind of stale data is now being pulled into AI models to answer the questions you may be asking. These are things we have to be aware of as we continue to broaden our data models. That's why I always go back to: what is the question you are asking, what type of data do you need, and do you have it? Asking the right question up front always leads to asking those questions about your data quality, your data integrity, and your data governance. The second part is that AI also creates ghost data, where there are fragments of data still sitting in the system that may otherwise have been deleted. Those are also being pulled into some of the answers we're getting from our data. So we have to be really thoughtful and really conscious as we move forward. I will always hold the position that we need a human in the loop when we are looking at data models, because you can't create the right data outcomes until you ask the right questions up front and understand the implications of pulling data sets, or asking generative AI to create models, when you aren't clear on where your data is coming from.

Chris Hutchins: Yeah, and I don't know that we really solve for it at the level all of us would prefer. The reality is we have to get better at asking the right questions, to your point. I'm curious, how do you think about pinpointing what you have versus what's missing? Because I think the age-old question a doctor is always going to ask is, what don't I know, and is it important? AI is not going to solve for that.

Lisa Marceau: I think that's what's exciting about where we're going with the entrepreneurship, the technologies, and the innovations. Instead of relying on the questions the system has either asked or supplied, they're starting to create their own questions and solve for those gaps. A lot of the new technology is solving for the gaps that exist in the system, and by doing so they're creating new data infrastructures that answer some of those questions. That's where there's a lot of real excitement, and that's why I talk about the paradigm shift the way I do. We are actually seeing real change in how we think about data as a core component of how we understand prevention, disease diagnosis, and treatment across all of those barriers and challenges we have with healthcare deserts and things like that. When we create the tools that solve those gaps, we start to be able to answer those questions better. We don't rely on the structures and the systems that have been in place. We actually start to evolve them to new places and new data.

Chris Hutchins: I love where you took that, because it gets to the next set of questions I want to ask you about. Talk about some of the examples you're seeing that really highlight some real possibilities. I think that's what you've been setting out to do with Joyuus. I'm not sure, am I saying that properly?

Lisa Marceau: Yeah, exactly.

Chris Hutchins: I love the meaning of that and what it indicates. There's definitely an optimism. I heard your voice change when you started to talk about what's exciting and what's possible. Talk about some of the examples of what you're seeing in your work there, where you realize the default infrastructure is not capturing it, but you're able to start doing some different things through the approach you're taking.

Lisa Marceau: Absolutely. I appreciate the point you just made, because for me, all of this is a story of hope. Joyuus was actually built from that perspective. We started this company when I met with an obstetrics colleague of mine, Dr. Allison Bryant from Mass General Brigham. We were talking about the challenge that they faced: what happens to moms when they leave obstetrics care? The answer is, we're frustrated because we can't help them, and they go into the primary prevention system and they get lost. That's a pretty significant gap. But the bigger question was, what are moms doing in that space? Because we know a lot of moms return to emergency care. There are really negative outcomes that happen during what I call the 12 months postpartum. So what we ended up wanting to do, starting again with asking the question we wanted to ask, was go straight to our moms and to our potential customers. We asked providers, what are you doing and what is missing? We had assumptions, but what we wanted to know is how many of those assumptions were right and how many were wrong. Moms told us, these are my pain points, this is where I'm looking for information, these are the parts that are missing, this is what I'm afraid of. Ultimately they were asking, am I okay? So what we wanted to do was help them define that through an evidence-based software as a service platform that covers mental, physical, social, and real world health. The reason we did that goes to the point we've been discussing during this conversation: when you start to integrate things that affect you outside of the clinical encounter, how does that information impact how you are experiencing physical or mental symptoms? The goal was to create a platform that helped integrate and connect those dots, so that we could identify what is self-care, but also identify when that self-care becomes seeking care, with really clear guidance: these are the symptoms, this is what you do when you experience these symptoms, and how to advocate for yourself. It was a great opportunity to fill a data gap by asking the people who are experiencing the data gap and helping them build a tool that's going to support that gap, on the user side and on the clinician, the provider side.

Chris Hutchins: Right. So you're tackling this by augmenting some historical data structure, and probably dealing with a lot more gaps than anyone wants to know about, to be honest. This work becomes even more important when you realize those gaps are there. What are some of the things you had to start thinking about differently? What kind of data were you collecting, given there's so much that occurs outside the clinical care setting? What are some of the things you were able to do that really started to move the needle?

Lisa Marceau: Well, one of the things we started with was depression as our first use case, because it is the biggest risk for moms postpartum. So what was important, number one, was that we wanted to know whether we could identify risk earlier. One of the biggest challenges is that moms get lost in the system, and it's not a three-month or six-month problem. You can experience it over the entire 12 months of postpartum, because as things change and evolve, some differences in how your life is moving forward may trigger depression that is not in the first six weeks, for example. This is where we were trying to identify some of these risks: if housing insecurity, or financial challenges, or cultural challenges within your own family, or social isolation are contributing to symptoms of depression, what if we could help identify those and help reduce those stressors and burdens? That may improve depressive symptoms, but it also helps a provider, when you interact with the provider system, to have more information about what might be contributing. The solution may not be, we know there's a shortage of mental health providers, but if the solution can be a combination of things that may use resources in the community or resources external to the clinical system, we might be able to make more efficient use of the providers that we have and also solve a broader set of problems in the process.

Chris Hutchins: We're starting to move, I think, in the right direction in thinking about mental health care as part of whole-person care. But there are still a lot of reasons why people don't feel comfortable talking about their mental health. I would assume it's probably the same in almost any condition, whether it's postpartum or whatever conditions people might be dealing with. Are they comfortable even talking about it, and are they getting the help they need? What's been your experience, and are we really making some movement in that space?

Lisa Marceau: It's an interesting question because there are a number of layers on that particular health challenge, for sure. One is, and I don't have the date at the top of my head, that mental health wasn't even brought into clinical care until much later than a lot of the other care. So in terms of decades of time, we have only relatively recently been paying attention to it. There's also been a long history of stigma associated with mental health. So being able to, to your point, in any population, really talk about it and feel comfortable saying it has been a challenge. The opportunity, and some of the excitement I see in this space, is that we are seeing a lot more willingness in the digital generations to actually talk about it, to bring it up as a core component of health and sometimes the core component, because the research is really, really clear that our mental health and our physical health are connected. And it's bidirectional. A diagnosis of cardiovascular disease or autoimmune disease or diabetes can lead to depression. Depression can physically affect us in so many other physical ways. So it's really important that it becomes so central to how we think about our overall health. I think we're seeing that shift, and I think in the next decade we're going to continue to see growth in that area. For example, with Joyuus, just being able to identify moms at risk, we had a safety protocol where we had to contact people who scored a certain level on the risk score. And here I am calling people out of the blue, because that was my job as the study lead and founder of the company. I was the one who contacted these moms to follow up on this safety requirement. And to a person, I'm thinking they're going to be like, why am I talking to this person? To a person, they were actually, thank you for contacting me, thank you for caring, thank you for reaching out. They just wanted to be heard. I think fundamentally we have to start there. Second of all, having a pathway to treatment or access to some sort of service is critical. I know that's a challenge providers have actually expressed to me: the more people we identify, the more people we need to see in the system, and we don't have those tools. Joyuus is trying to solve this, but there are a lot of other companies in this space, which I really respect, who are trying to integrate these data points as well, to say that if we can help solve some of the problems over here, we can also help solve some of the problems over here, like I was just talking about. We can actually help reduce the burden of some of the other factors that contribute to stress, anxiety, and depression, take the burden off the system, and improve mental health across the board. These are the exciting things that are happening that are filling some of the gaps our system has, quite frankly, failed to address.

Chris Hutchins: Right. It's interesting. I remember a period of time I was dealing with myself. It was just a really stressful thing I was trying to wrestle with, and it was giving me migraines. I didn't connect the dots myself. I went to see a neurologist, a couple of different specialists even, because it was constant for several days. And it turned out my parents called the pastor, who came and visited, and he looked at me and said, you've got a lot on your mind. You're worried about stuff. Like, yeah. Because, you're not wired to carry that, you know that, right? Just having that conversation and coming to a realization that there was something I was stressing about that I couldn't do anything about, and having somebody walk me through letting go of it, was an amazing impact. But there's not a distance between mental health concerns and the physical manifestation they can cause, oftentimes. I'm sure your research has shown a lot of different scenarios where these connected things are there.

Lisa Marceau: There's an enormous amount of research that shows this connection between physical and mental health. Those are things we know. The question is, what do we do about it? The reason I find innovation and entrepreneurship so critical, not just valuable but actually critical to the future of our whole-person health journey, is that we are taking those questions outside of the system. We are starting to connect the dots between these things outside of the system. We're starting to use everything from how we spend our free time, from dance to running to whatever physical activity gives you joy, to how you use your time to relax and find space, whether it's knitting or yoga or whatever it is for you. Social connection. Social isolation is a huge connection to mental health. So how do we think about our social connections, and understand that social connection can be through technology or through in-person, real face-to-face connection? Those are all valuable. What's really happening in some of the entrepreneurial journeys I get to support and help grow into viable businesses is that we look at all of these elements, connect those dots, and integrate these systems in ways they've never been done before, so that we can start to answer these really core questions with new data infrastructures, new ways of protecting patient data, new ways of keeping and collecting your own data to say, this is my picture of my health across my entire life journey. And then we can start to see where those data connections are, to share more about disease states or early symptom ontology.

Chris Hutchins: Yeah, this brings up a point that I think is really important, and I know we could probably talk about this for hours. I just want to re-highlight a lot of the stuff you've been working on in this space, and why you're having some pretty exciting outcomes: people are getting behind it, and you're doing the work you have to do to get the NIH funding. There's always a debate about that, what should be funded and what shouldn't be. But this is an exciting space for me. If there's one or two things you could highlight that would never have been possible without that kind of support, I'd love to hear them, if there's one or two that stand out.

Lisa Marceau: With NIH funding in particular? Well, NIH funding for me is such an exciting story. It's the NIH Small Business Innovation Research program, which I have been involved with for over two decades, and it's been around since the 1940s, I believe. It is the largest seed funder in the world. Four billion dollars goes to seed funding. So, to your question about identifying an example or two, that's really hard, because they have funded some of our most significant innovations over the course of time and are continuing to do so. The funding stories have ebbed and flowed, and I've been in this business for a long time. There have always been challenges to funding, and there have always been shifts in what's important, often for good reasons. We've seen, for example, a moonshot in Alzheimer's and dementia research, which is critically necessary. The most important thing I take away from that funding is how the SBIR federal funding program helps companies build the components of a successful company along with a really good idea. They really force structure and organization. They force you to understand who your market is, so you know how you're going to commercialize to that market. And you are beholden to a very strict financial infrastructure, so you have to understand how to run a company based on those dollars and what you're going to do with every dollar they give you. To me, these things are as valuable to making companies successful as the really good ideas underpinning them, because both are necessary in order to succeed.

Chris Hutchins: I want to pivot just a little bit. Some of the things you've talked about get to where things are going beyond inequitable training data. This term is being used a lot from a variety of perspectives, the idea of shadow AI, or call it a black box or whatever. But there's a problem where AI begins referring to AI-generated material. I think I heard recently that over 50% of what you can find online is AI generated at this point. What happens when that generated output starts masquerading as source data? Because I think people are already probably trusting it a little bit too much.

Lisa Marceau: It's funny. A few years ago, when OpenAI first came on the scene, I was driving my sons back from their first year at college and they were having a debate over whether the black box AI problem was real or not. Fast forward a few years, and it's really interesting, because we're seeing the same challenge in the AI boom as we did in the late nineties into 2000 with the dot-com bubble. AI has great potential. It has a lot of features that create opportunity. It is efficient in many ways. But to your point, AI is creating challenges for data that really require all of the things we were talking about earlier in this conversation: data quality, data governance, and most importantly a human in the loop asking the right questions of the data from the start. Because to your point, you're right. There's shadow data, there's ghost data, and AI is referring back onto itself. It's gotten really good at scraping data that should have been archived, right? So we're using data that is actually wrong. There are challenges with AI scraping data that is risking security, risking personalized information, and HIPAA violations in a lot of ways. So we have to think about it in that context and go back to the fundamental question. I believe we've seen this before. History repeats itself. This crest will have its slope. What happens after that is, I think, the most important question. What have we learned from that? And what are we going to do to make sure the data we are ingesting and the models we are building are built on strong data quality and data governance? Because that's how we get data equity and ethical data. When we are inclusive, we consider very clearly, if we don't have the data, are we going to run the model anyway? Or do we need to go create the data? Do we need to do the work to develop the data? There are a lot of companies developing very specifically female-specific data sets and data-generated models for that reason, so that we can connect the dots to broader data sets that haven't included those data. So data quality and data governance will create data ethics and data equity. I think that's where we have to keep our focus, because AI is another invention that has really skyrocketed and will also have its fallout.

Chris Hutchins: Right. Now, you've used the term infrastructure to describe how AI should be treated in the healthcare sector. Talk a little bit about what you mean by that, and what are some of the things people typically wouldn't be thinking about if they're doing things by traditional means, where they're just consuming data and something's not quite right, so you tweak it and make an adjustment, or change your programming or a calculation. AI is not that. Maybe talk about what the major differences are from your perspective and why it's so important that we think about it.

Lisa Marceau: The way I talk about data as infrastructure really comes down to this concept of AI as innovation. Where I spend a lot of time talking about this is that AI isn't innovation. It is not innovative. It's actually been around for decades. The LLMs have been around, the way we've tried to build large models, predictive modeling, isn't new. It just required a lot of human input early on. With the birth of OpenAI and these rapidly evolving AI models, we lost sight of that fact. AI is and has been infrastructure. It is the foundation that helps us do something better, faster, more efficiently. The question is, now that we can do things a lot faster and pull a lot more data, what does that infrastructure actually do? To the question you're asking, what is the purpose of the infrastructure? When I work with entrepreneurs, and even when I'm thinking about my own work, we start with: what is the problem we're trying to solve? What is the gap we're trying to fill? How are we going to get there? What do we need to do it? Once we answer those questions, we ask, does AI make this better? Does it make it faster? Does it make it more efficient? If the answer is yes, then we start to ask how. If the answer is no, then there's no reason to layer it on. So it's like any other infrastructure. You're not going to plug into an EMR if you don't need an EMR. You're not going to plug in an AI infrastructure model if you don't need AI to help your product work better. That's where we really have to start thinking about it as the infrastructure that it is, because it's always been infrastructure.

Chris Hutchins: Right. Yeah, believe it or not, I did learn something in the last year or so from some of the social media stuff. I found out what FOMO means. There does tend to be a bit of that when you're dealing with things at an executive level, in terms of the pressure to use the latest and greatest technology. I've actually heard people talk about AI like it's a verb. Like, you need to AI this thing. Well, what does that even mean? But to your point, it's got to have a purpose. It's not your default posture when you're deciding on a solution. It's like the pry bar or something. I've used those for so many different things, but it's not that. It's a very unique set of technology that, when used properly like any other tool, can really do some great stuff, but it's not something you want to do for the sake of checking a box, I guess is the point there. Talk about some of the adverse impacts when that is a default, whether it's AI or any other innovation. Sometimes that becomes a problem too and generates more issues than it's solving.

Lisa Marceau: I think this goes back to the idea that if you know what question you're asking in the first place, and you build the pathway to answer that question, then negative outcomes happen when you go down a path blindly, whether it's the AI path or, to your point, any other pathway you're going down. This is in mature companies, it's in startups, it's everywhere, as we move forward and evolve our work, our product, our services. If we are following a path blindly and we are never asking why we are on that path, then when we deviate from it, we don't know how far to go down that deviation. We don't know how much time, energy, effort, and money we're going to spend deviating before we turn around and go back, or whether that deviation was actually a very viable pathway that we should take and continue to take. It's something I talk about on repeat because it's really important. If we don't understand the reason we're doing it, and the people we're bringing in to do it, and we don't document all of the parts we need to get to our end goal, and we don't have that roadmap, we can't understand why we're on the path or whether we're getting closer to the outcome. It doesn't matter what your product or service is, or where you are in your company maturity. If you're not understanding those four components of your business, you are potentially spending resources, effort, and time, hiring people, and building platforms or architectures that you may not need or that may not be the right ones. I think that's where the biggest risk is, to the question you're asking. When we don't pay attention to all of these components, we build without knowledge and we don't necessarily get closer to our outcome.

Chris Hutchins: Yeah, I think that gets to one of the things I was hoping to ask. If you're talking to a health system executive, typically a CIO is somebody who has to guide the discovery and evaluation of technologies. What are some of the things they should be asking? What criteria should they put in place to really test, first of all, are they asking the right question? And second of all, does the partner they're evaluating have really solid answers you can trust? What are the things where, if they have this covered and the answers to these types of questions are good, great, and if they can't answer them, we probably need to move on? It's probably different in every organization, I would guess, but I would love to hear your experience.

Lisa Marceau: Well, in terms of adopting new technology, or bringing in a company with a new technology, what it really boils down to is: what problem are you trying to solve by bringing in this technology? I've worked in this environment where this has happened. We want to bring this company on board, or we want to bring this technology on board. Why? What's the value of that? So when you start to express, this is the value, this is what this company can bring to us, do we understand how this can impact our users? Or do we understand how this impacts our customers? How does this company intersect with what we are doing? What value do we bring and what value do they bring? What are the risks they bring? All of these things actually have nothing to do with technology, which is really interesting. These are standard questions we should be asking ourselves whenever we're moving forward to make a big change or investment, whether it's bringing in or integrating a new company, bringing in a new product, or identifying a new service. We have to understand where we are going with this and why, what the value is to our end user, our end customer, what the risk is to us, and what the risk is to them. When we do that, we actually can make better decisions. And that's really organizational management at its base.

Chris Hutchins: I always learn so much having these kinds of conversations. Research is something I've been really fascinated with. The first time for me was about a decade ago, when I went to an organization in New York. In my prior experience, research was separated out with firewalls from the traditional IT, data, and analytics work, so I wasn't really familiar with what they were doing. But when I was in New York I got to see a lot of the things that were possible with access to the right levels of data and starting to ask much better questions. I love what you're doing, first and foremost. And you couldn't have picked a better name. Joyuus, I think, is phenomenal. I love that. There's a lot of reason for optimism, and I know we tend to talk about some of the risk, and we should, but I do hear in your voice some excitement. I really appreciate what you're doing. So if people are listening to this today and they're not quite sure where they're at, they have some questions or they feel like you've identified some things they should work on, how should they find you, and where can they find out about the work you're doing at Joyuus and Alpha Millennial Health, and hear about your book? Maybe buy the book. I think that would probably be a good idea for some people who are curious to understand more about what you do.

Lisa Marceau: Well, I appreciate that. Breaking the System was written because I see the story of hope, and I see the story of hope for everyone who's involved in the healthcare system. That's one thing I would encourage anyone listening to really think about. The book is about starting a conversation, not ending it. It's meant to be kind of an airplane read where you can flag the pages and put your tabs in and highlight and underline it. Start the conversation with me or with anyone else, because we all have a role to play, and it doesn't matter where we sit. We have a piece of the healthcare system that we are intersecting with and contributing to. For me, why I started Joyuus was to solve a problem and a gap that the system wasn't solving. The reason I started Alpha Millennial was because the entrepreneurs and innovators are actually leading this charge in ways that are super exciting. So for me, all of these things combined are really why I do what I do. I appreciate this conversation, and your willingness to ask some of these really interesting and often hard questions. Anyone can find me by going to my website or connecting with me on LinkedIn. The book is on Amazon or at Itasca Books, and it's called Breaking the System. I would appreciate anybody's thoughts, and I appreciate very much this conversation.

Chris Hutchins: Well, thank you again, Lisa. For me personally, it's been phenomenal, because I feel like I've learned a lot from you today. One of the things I'm going to step away from this with is understanding that AI is not the solution to a missing foundation when it comes to data. And trust is probably something that needs to stay on that side of the equation, where it has to be earned with the data. So, phenomenal. I could probably talk to you about this for three or four more hours, there are so many places my head's going. Thank you so much for taking the time and coming on this show. I hope you all will take a look in the show notes. I will have all the information Lisa just talked about and more, so if you're curious, want to reach out, and have the conversation, you can. Better yet, if you're looking to put some resources behind research, please definitely look in the show notes and reach out to Lisa and her team. From The Signal Room, that's going to be it for this time, and I look forward to seeing you all next time.