The Dose - In the Age of AI, Health Innovation Requires Human Stories

Episode Date: May 16, 2025

Data is the engine of health innovation, but too often it can’t tell the full story. On this week’s episode of The Dose, Dr. Sema Sgaier joins host Joel Bervell to talk about the future of equitable health care: how we collect data, who’s included, and what it means for clinical trials, mental health, and the role of AI. Tune in to hear Dr. Sgaier explain why solving health care’s toughest challenges starts with understanding the human side of health — and how inclusive data can lead to smarter policies, safer treatments, and better care.

The Dose is a production of the Commonwealth Fund, a foundation dedicated to healthcare for everyone. On this episode of The Dose, my guest is Dr. Seema Sighair, a supreme problem solver whose tireless work as a non-profit executive and a breakthrough scientist and entrepreneur really is unmatched. She knows the complexity of the critical health and equity issues confronting us,

and she looks at every facet of a problem from every angle with a driving mission of better healthcare for all. She's the founder and CEO of Sergo Health, a public benefit health tech corporation aiming to use data to revolutionize clinical trials and much more, which we'll talk about today. Dr. Sigeir defines her work as, at the intersection of behavior, data, and technology, drawing on her experience in policy, strategy, and management of global development programs. She previously worked at the

Bill and Melinda Gates Foundation, managing large-scale public health programs, and oversaw the introduction of innovations in India and Africa. Dr. Sigeir is also Affiliate Assistant Professor of Global Health at the University of Washington. Dr. Sigeir, thank you so much for joining me for this conversation. Thank you for having me, Joel. It's a pleasure.

Yeah, and it's so nice to see you again, because I know last time we met up at the Clinton Foundation, and so it's great to be able to have a one-on-one conversation about your work. So I really want to just dive in. And looking at the scope of your work, it really strikes me that you face some very high tech problems and also low tech problems. One challenge that you're focused on

is elevating both the diversity and the impact of clinical trials. And there are at least a couple of data issues there, including how bias enters health healthcare data and AI models. So I was hoping we could start with your perspective on some of those high tech challenges and how they're shaping the future of equitable care. Well, I'll start by saying, you know, what we're really obsessed with is how do we reveal the human side of healthcare?

So when we look at healthcare, whether it's how do we get people to participate in clinical trials or anything downstream, at the end of the day, there are a lot of factors that determine that, that oftentimes are non-clinical. So these include things like people's beliefs, their biases, where do they live, what kind of access barriers do they face, who do they trust, where do they get their information from. And so there's really a critical information gap in the ecosystem right now to understand people holistically. So for me, that is a high tech problem in the sense that how do I actually build that data infrastructure

to really be able to understand people? And when it comes to clinical trials, it's absolutely essential because clinical trials, not only are they the bloodline of our medical research, but they're also the most costly and time consuming part of bringing any drug into market. And so the high tech problem is how do I actually build that data and how do I use AI tools so that for every single person in the United States, I know the barriers. Now the low tech problem side of that is how do I take that insight and translate that

into real world interventions, things like community engagement, communication mechanisms, you know, bias training. And so it's really an incredible combination of different ways of thinking about a problem to really close that gap. How do you sell that participation, especially to communities that have historically been excluded or harmed? Because we're talking about data, but behind every data point is a person. And so how do you make the human impact of collecting inclusive, high quality data resonate?

Yeah, I mean, in everything in healthcare, we do need to sell it, right? So why should someone participate in a trial? And there are really genuine important personal reasons. Number one, many people want to see themselves or people like them represented in trials, right? If I don't see myself as part of a clinical trial when a drug is being tested, I'm not going to trust that drug. I'm not going to believe that it's been tested on my community. So I think the first thing is really, do you want yourself represented? And do you want this innovation or this miracle to actually serve your community?

So that is a really, really important motivator. And that's a really important part of the message. Oftentimes it's also personal, right? People either themselves suffer from a disease or a condition where that drug is being tested, or they know a family member that has been affected. And so really bringing that personal story to the individual and to the community becomes an important point.

And then the third thing I will say is the messenger is really important. Who is delivering this message? Is it a scientist sitting in some academic institution coming down to a community and talking to them maybe once or twice, or is it really someone they truly trust from their community, a trusted figure, like a church leader or a healthcare worker or a teacher that is really having this very deep conversation? Yeah. Do you have examples of how CIRGO Health has actually been doing that already? Yeah, absolutely. So for us, the number one, it has been really, how do we build this data infrastructure? And we've been spending now eight years in the US

going out and collecting really rich and deep data from individuals across the country in a very representative manner. And the way we do this is we use a lot of different modalities. Typically data collection efforts will use online surveys. Now, online surveys are extremely impersonal. We probably all get bombarded by them

and there are certain types of people that respond to them. We use a lot of different modalities. We use community organizations to go to speak to different community members. We use door to door. We use phone interviews. So number one is how do we actually deploy a whole set of different modalities to reach the individual

in the way they want to be reached. Now, once we have this data, how can we then translate this to something meaningful? And we've been partnering with several life science companies to actually really close the gap when it comes to clinical trials. I'll give you one example. We've been working with a very large life science companies on a drug for colorectal cancer. And one of their goals is to make sure that communities are represented, especially under-represented

communities. And typically, these clinical trial sites will be in places that are high-end academic institutions. We work very closely with them to identify sites where very diverse communities live, but more importantly to also be able to say upfront what types of barriers those communities may be facing to be able to participate in these trials so we can actually solve for them. You know things like language barriers, trust issues, transport, and they can be so different from site to site. And when we bring this intelligence,

when we empower these companies that are running these trials with this intelligence, we've actually saw 30% improved in recruitment of diverse patients, which is really phenomenal. And we also saw a much faster recruitment timeline into these clinical trials. And as you've kind of mentioned,

when it comes to both community settings and online, data collection can be incredibly labor intensive, whether that's surveys or on the ground going to people, and it can often be messy as well. When it's poor quality or fragmented data sets, those can be difficult to manage, but they're also vulnerable to human error. I'm curious, how is AI helping to address those challenges

or is it adding new ones of its own? I think both. So, you know, it can help us address the challenges. So one aspect of collecting large data is being able to do quality controls. As you're collecting the data, you don't want to collect the data and then go back, you know, and then look at the data and then say,

oh my God, there are all these quality issues. One of the things that AI can really help us, and we use this as we collect the data and then say, oh my god, there are all these quality issues. One of the things that AI can really help us and we use this as we collect the data, it can look for patterns of low quality data and it actually send us signals and saying, Hey, you know, we're seeing some unusual patterns here. That doesn't make sense. And so can you really go down and do data quality checks? So that's a very powerful way of using AI so that you are addressing the data quality issues real time. Now, AI is also used from predictive models to causal AI

to actually also analyze the data. And that can become a problem if your data is biased. And all that AI is going to do is actually really exasperate those biases. And so when you're using AI for analyzing data, you want to make sure your starting data is really unbiased, it's representative, it's comprehensive. So it's kind of both ways.

It both helps as well as it's something you need to be really aware of when you're using it as a very powerful tool to glean insights from your data. I mean, you've just talked about how important data is. Recently, the FDA withdrew its draft guidance on diversity in clinical trials, a move that a lot of people saw as part of a larger rollback of DEI efforts. As someone who's leading innovative data-driven strategies to make trials more equitable, how are you interpreting this decision? And what do you think it signals about the future of inclusive research design in a climate where these conversations about the need for diversity are increasingly

politicized? Yeah, I mean, it's definitely an issue because you want policies and regulations to be aligned with goals. One thing that gives us hope when it comes to clinical trials and making sure that clinical trials are inclusive and representative. And to be really clear, inclusive and representative means including all types of people, including people with different lifestyles as well. And what gives me hope is that, you know, when it comes to clinical trials, there's

a really critical, both health case as well as business case to be made, right? When these trials are not including different subgroups of the population, sometimes they run into regulatory issues. They also run into safety issues when they're then put forward into the population and they run into efficacy issues. Life science companies don't want that. And so there's a real incentive and I'm encouraged by seeing our partners still continue to make sure that even if they're not necessarily following, you know, guidance or submitting this, they are actually planning their trials to make sure that they're still inclusive. I think this is now part and parcel of the ecosystem,

and hopefully people are not going to walk back because I do think there's a real safety, clinical, as well as cost issue related to it. I love that you mentioned costs because that's always where my mind immediately goes. This type of research and modeling sounds expensive. How is it being supported? Or is it actually a cost containment approach that saves money on the backend? Great question.

It can be expensive, although it has incredible cost containment advantages down the line. So, you know, when you design programs, whether it's clinical trials or even say public health interventions or even healthcare solutions, you know, you really wanna make sure you are investing

in the right thing and you're being precise in the sense that you are targeting the right intervention to the right community in the right way. When we don't have information, we can't do that. And so what happens is that we are guessing. What I found in my work over the last 20 years being in the sector, investing upfront in data helps you be much smarter downstream

and helps you actually be much more efficient. I'll give you an example. We were working with a very, very large health system. This is in India, but it's absolutely relevant. In one of the largest states, they have a population of 230 million, so it's almost the size of the US. They had some of the highest maternal mortality rates in the state. And one of the reasons was that 20% of women were delivering at home. Now, it's very problematic when you're delivering at home in a very rural area when you don't have access to emergency services, let's say something goes wrong.

The government at that point, their plan was to build more facilities. So invest actually in hundreds of millions of dollars in building more healthcare facilities in these smaller communities. We came in, we actually collected a lot of data from mothers, households, families to really understand why are these women not delivering in facilities. And we actually identified two different types of mothers. A group of mothers, they just didn't trust the healthcare system.

They just didn't want to go there. So for them, it was just that they felt that they were going to be not treated well, etc. For another group of women, mothers, it was really they didn't plan well to be able to deliver in a facility. And so now the interventions became much more of a birth planning intervention and a trust building intervention with community health workers, which was tens of millions of dollars, even less versus hundreds of millions of dollars. In fact, when the government implemented this, not only did they close the gap from 20% to 12% in one year, but they saved hundreds of millions of dollars.

So this is just an example of when you invest in data collection, and then you make sure you take that and do very targeted solutions to targeted groups, you can actually really save a lot of dollars at the back end. And what I'm hearing is that it's not just about the what, but about the why of something in order to get to the key reason and the true background of what we need to be targeting. I love that example. I'm curious though, how do you identify the best access points on the ground and

online to build a robust trial population? You've mentioned the work you did in India, but for many people that are saying, how do I actually go and find the right places that I need to be asking the questions to, or the best access points to make sure I'm getting this right data? How do you go about identifying that? Well, I think you basically hit it on the nail, which is the why question is the most important question. Our data sets are filled with what people are doing.

But if you were to ask, why did someone not participate in that epilepsy trial, It would be very, very difficult to find out. So for us, foundation, the first thing we want to do is be able to answer the why question in a very comprehensive way. We don't make predetermined hypotheses. We don't say the why is because someone has a transport issue or et cetera. We actually really interrogate everything that we know could contribute from, as I mentioned, people's beliefs, biases. So first going out and collecting this information is critical. Once you're able to answer the why question, the answer actually then leads you

to those solutions, right, it tells you whether it's a issue of the messenger or it's an issue of a message or if it's an issue of a distance or a, you know, some sort of health care gap, it really actually reveals the answer in a really magical way that then really informs the intervention. So I think for us, what's really important is not to go in there with a very fixed hypothesis of this is what we need to solve for, but actually be open-minded and actually let the data reveal the answers and therefore the intervention points.

And so the maker of the model also matters, right? Like who is actually there creating it matters as much as what it's built on. How do we ensure then that equity is baked into AI, not just in data sets, but in the very questions we're asking it to solve? Involving the people that is actually going to matter too. So again, just making this real and giving you another example, we've been asking the why question

around the youth mental health crisis in the US and actually have built one of the largest and most comprehensive youth mental health and wellbeing tracker. And for us, the question there was, why do we have this crisis? Not what is the crisis, how much depression, how much anxiety, but really what is driving this and really why are we seeing this?

And really wanted to go very upstream and look at beyond the clinical. And, you know, to help us design not only the surveys, but even just the whole effort, we actually started with youth. We made sure youth are part and parcel of the design of the questionnaires, of the analysis of the data, of the interpretation. And I think that's absolutely important because their perspective is nothing like our perspective, right? Mm-hmm. And I think that's so crucial, especially that example you gave about youth mental health. You launched that AI-powered initiative at a time when many are calling it a crisis, where nearly 20% of adolescents are reporting receiving mental

health care therapy or another 20% saying that they have unmet mental health care needs. Can you tell us a little bit more about that tool you developed and beyond just what it does, why is it a powerful case study for the role of AI in public health? You've mentioned how you worked on the ground with youth, but can you tell us a little bit more about what it's doing, what you found, the data that you've gleaned from that? Yeah, we didn't wanna look at it just as a mental health crisis.

We wanted to look at it also from the perspective of well-being. Youth are human and they're not just about a crisis. They're also about everything else in their lives. So number one, we wanted to think of it more expansively and we wanted to bring the well-being angle. Most youth studies or mental health studies are very much focused on depression and anxiety, but what about resilience and forward-looking and optimism? And one of the things that we found

when we measured this stuff as well was all of the other mental health related stuff as these two things actually coexist. The absence of one is not the presence of another. So what I mean by that, you can be experiencing depression, yet you can also be hopeful about the future or you can feel that you can cope with it. And it was really, really important to expand this. So I think that's number one. The second thing is that, you know, again, where the data and AI was really powerful, we wanted to really go upstream and identify what are the critical

driving factors, again, beyond genetics and other clinical issues, what are the critical upstream driving factors? And we identified six, which I think are really, really important. Social support, social cohesion. So again, going to community network being so, so important, especially in this era where we're talking a lot about loneliness and the importance of community stress, obviously, but also physical, physical health, physical activities, sleep. And then the other one was unexpected life experiences. So trauma life experiences. So I think where the AI was really powerful is to really pinpoint on six critical themes

that now are themselves really helping us inform the policy play. Okay, what does this mean now in terms of investments toward youth? What does it mean in the era of potential Medicaid cuts, which a lot of youth are actually getting their health care services through Medicaid? And so I think it really allowed us to focus, but also importantly, in our youth mental health tracker, which is a multi-year tracker, we also are able to go very local geographically and show different areas in the United States how they're affected and which of these six themes are really, really critical.

So it helps you think about what are the interventions in that community and how should resources be deployed. How is that new kind of data being deployed right now? And does it compliment or contradict what we already knew in meaningful ways? Yeah. So it complements in the sense that a lot of datasets that are out there, so Burfus is a good example, many of the datasets are government funded, or the existing datasets, they have been very focused, as I mentioned, on mental health, so depression, anxiety, so really measuring those. So it complements them in the sense it brings a whole new wellbeing angle, as well as we

have been very focused on making sure that underrepresented groups, which are typically not captured in those other surveys, are actually captured in this effort and actually are seen and representative. So I think those two are really, really important. In the era we're in where government data sets are, we don't know what's going to happen, data sets like these become even more critical because they're going to be there to stay. They're not being funded by government. And so I think they kind of fill an important, hopefully not, but can fill an important gap.

How are they being used? So I'll give you an example. One of our key findings was the role of parents, especially in helping youth access care. And so this information has been used to actually develop a national wide campaign of parents involvement in youth mental health and in helping them be able to access care. So that's one way that's being used. But we're working with a lot of organizations on the ground in different states to actually take this information and figure out how to translate it into different interventions on the ground.

So this is very much work in progress. The tracker was just launched a few months ago, but this phase is very much about working with on the ground organizations to do the translation. Absolutely. You're working on so much. I'm curious, what are the urgencies that are shaping your your agenda today? And even looking a few years out, I know that broadly your goal is to work towards re-imagining and redesigning healthcare.

And you shared some of the things that you're working on today, but what's next? Well, for us, it is really about building a critical layer of data infrastructure in healthcare that does not exist today. As I mentioned, a lot of the data is what, where, when, it's not about the why. And so I am really, you know, obsessed with making sure that we are building that layer that doesn't exist. For me, the urgency is also what's happening as with data today, right?

And we don't know which ones are going to survive, even the ones that are surviving, we know parts of it are being removed or erased or kind of at risk. And I think if we don't have data, we don't know what's happening, and we just won't know what to do and we won't be able to measure our impact. And so I'm really keen on continuing to focus on this data infrastructure and really thinking about using this moment as an opportunity moment, because I truly believe moments of crises or however you wanna think about it are opportunities. And I actually see a huge opportunity

in leveraging new technology to reimagine how we are developing data, how we're funding it, how can we make it much more low cost, how can we make it much more expensive? And so that's kind of where I'm very much focused. And of course, thinking about how to translate it into impact.

And so how do we put it in hands of communities, of partners? It's really that translation piece that's gonna deliver the impact. I think we're very focused on the why and very focused on building that infrastructure that we believe doesn't exist at scale and needs to be built to really reimagine that healthcare that you identified and mentioned.

Well, I truly think it's so important that we continue to reimagine healthcare. As you've mentioned, it's so important to make sure that we're including all communities and unfortunately, we haven't always done that. But it's clear that the future of health equity depends on leaders like you, who are willing to ask hard questions, to build smarter tools using things like AI, which I think can help streamline the process, but at the end of the day,

not losing sight of the human stories that are behind the numbers. So Dr. Sigeir, thank you so much for being on The Dose. It's been an honor having you here. Thank you, Joel. It's been my honor as well. This episode of The Dose was produced by Jody Becker, Mickey Kapper, and Naomi Leibowitz.

Special thanks to Barry Scholl for editing, Jen Wilson and Rose Wong for art and design, and Paul Frame for web support. Our theme music is Arizona Moon by Blue Dot Sessions. If you want to check us out online visit thedose.show. There you'll be able to learn more about today's episode and explore other resources. That's it for The Dose. I'm Joelle Brevel and thank you for listening.