Can health IT be leveraged to promote equity and reduce bias? Dr. Alison Curfman believes it can, with humans playing a crucial role. In this exploration, we discuss the importance of clean data for building AI language models and how health IT can help reduce bias while advancing health equity.
Guest:
Alison Curfman, MD, pediatric emergency physician and co-founder, Imagine Pediatrics
Host/Producer: Carol Vassar
TRANSCRIPT:
Announcer:
Welcome to Well Beyond Medicine, the world’s top-ranked children’s health podcast, produced by Nemours Children’s Health. Subscribe on any platform at NemoursWellBeyond.org or find us on YouTube.
Carol Vassar, podcast host/producer:
Each week, we’ll be joined by innovators and experts from around the world, exploring anything and everything related to the 85% of child health impacts that occur outside the doctor’s office. I’m your host, Carol Vassar, and now that you’re here, let’s go.
MUSIC:
Let’s go-o-o well beyond medicine.
Alison Curfman, MD, Imagine Pediatrics:
Technology can only do as much as the humans that inform it.
Carol Vassar, podcast host/producer:
Is it possible to leverage health IT to promote equity and reduce bias? According to Dr. Alison Curfman, whose voice you just heard, the answer is yes, and humans are the key. Dr. Curfman is a pediatric emergency physician and co-founder of Imagine Pediatrics, which provides virtual first care in the home for children with special healthcare needs. She has spent much of her career focused on building alternative care delivery models for children and helping expand access to pediatric care using technology. And she’s worked closely with her pediatric colleagues across the nation on health IT development through the American Academy of Pediatrics, especially in the area of telehealth.
So in her estimation, what is the role of health IT in pediatrics generally, and in helping promote equity and reduce bias? Here’s Dr. Alison Curfman.
Alison Curfman, MD, Imagine Pediatrics:
I think technology gives us a lot of opportunities to find new ways to care for kids and high quality ways, but also leads to some situations that we need to critically appraise and assess is the technology designed the way it needs to be to deliver safe and equitable care? And also what sort of data is it built on? And is there any chance that whether it’s an EMR clinical decision support tool or if we’re looking down the road to AI language models or predictive analytics, whether it’s built on data that could perpetuate bias or if we can actually use the design of the technology and the basis of the data to actually help reduce inequities in health care.
Carol Vassar, podcast host/producer:
As we look at the landscape today of health IT, would you say that what is now in place is addressing health inequities and reducing bias in any way, shape, or form?
Alison Curfman, MD, Imagine Pediatrics:
Yeah, in a lot of ways when we think about pediatric care across, I’ll use telehealth as an example because that’s what my background is in, is really in helping determine how we can use telehealth, virtual care, digital care to reach populations that previously may not have had access to some of these pediatric resources. So we obviously saw an explosion in use of telehealth at the time of the pandemic. But when we look at the geographic distribution of pediatric health services, they are really consolidated in major metropolitan areas and there’s a lot of access issues even in those metropolitan areas to receiving needed pediatric services. And then particularly when you go outside of those metropolitan areas, the pediatric specialty availability or even just general pediatric availability is not necessarily meeting the demand for… Children live everywhere and children’s hospitals where all the services tend to be located are not everywhere.
So by expanding our mindset about ways to deliver care safely and taking an active leadership role in that development as pediatricians, I saw so many pediatricians stand together and help develop the AAP’s policy, which I helped write a few years ago on how we use telehealth, and other standards on how will we use this technology to use it as a tool for us to deliver high quality care? So I definitely see that children who maybe didn’t have access to resources before now have greater access.
But then there’s the flip side of that are we actually worsening any health care inequities? Are there children that don’t have access to the technology because of their socioeconomic status and therefore are actually worsening gaps in care? So I think we have to be very thoughtful about how we implement any of these solutions and to think critically about what sorts of inequities may be built in if we’re not designing it right.
Carol Vassar, podcast host/producer:
It sounds like the AAP does have a pulse on all of this, you mentioned a policy. Speaking generally, what is the policy? What’s AAP’s view on all of this?
Alison Curfman, MD, Imagine Pediatrics:
Well, I was part of the group that wrote the AAP policy statement that came out in 2020 on telehealth and the use of telehealth. And really puts a lot of emphasis on the medical home model and continuity of care and not fragmenting care and not letting a new technology have a separate set of quality standards. So things that are required for quality care at an in-person visit, we should hold virtual care or telehealth to the same standard and use it in the appropriate scenarios. And then we do talk a lot about inequities and how do we ensure that we’re not worsening any sort of inequities?
And I know at my company, which I started with my co-founders a few years ago, we will provide devices to any patient that cannot afford one or doesn’t have one because we work with a completely Medicaid population. So we’ve definitely designed around trying to look ahead on where could there be issues or problems? And language barriers is another thing. If we have an app and it’s like, okay, well, how many languages is the app available in? And now, all of a sudden, we’re doing messaging and texting and how are we integrating interpretation and translation services in the different services that we provide so that we’re reaching all of our most vulnerable patients in the right way?
Carol Vassar, podcast host/producer:
As we talk about technology, we can’t leave to the side artificial intelligence, which has really made an explosion all over practically every field, but especially in healthcare. What can we do with regard to creating AI models and algorithms to make certain that there isn’t bias there, to make certain that we aren’t leaving people out who should be included and including people who maybe are on the outside?
Alison Curfman, MD, Imagine Pediatrics:
So we can’t make certain of that, and that’s the problem. It will have bias because bias is present in every layer of the fabric of society around us. But we, as pediatricians, can absolutely be advocate for our patients.
So there’s a few examples that I can think of, but one of the things that we’re doing at AAP with the Council on Clinical Information Technology is around foundational data that a model is built on. So when you think about clinical practice guidelines, there’s definitely been studies where a model maybe has been built on a biased assumption that’s not based in fact. And I can think of there was examples of, for instance, glomerular filtration rate being different standard in Black patients than in Caucasian patients in foundational models. And this was in the past, I don’t think it’s still an issue. But if you have a different metric that you’re measuring for different populations that is assumed to be evidence-based, and then that evidence is actually kind of debunked, that that standard shouldn’t be different or there’s not a statistically significant result that supports that difference. And you’ve built a clinical practice guideline that uses, maybe directs people to make clinical decisions or clinical decision support tool where people may not even see the underlying data that is informing the technology to tell you to think about doing something.
So all of those things, the foundational data, that is a very deep level approach that everyone is going to want to get involved in advocacy around foundational data. But that is something that I do and I work with companies that are building AI models and understanding how do we, as pediatricians, embrace change, recognize that it’s coming, recognize that I don’t like AI, I don’t want it in my practice, isn’t going to prevent new things from being developed.
But taking, and I saw this in telehealth, I saw that people years ago were like, “We cannot do that virtually.” And I was really pushing people to think outside the box of like, well, what can you do? There’s some great use cases for telehealth for behavioral health, and there’s all sorts of… For follow-up checks or things like that. And I would encourage pediatricians, you may not go work for an AI company, but you can have an open mind and try and learn as much as you can about it. Maybe even use a lot of AI tools in your day-to-day life where the stakes are lower and you can get more familiar with what it can and can’t do. But we need to have these clinical voices at the table for technology that’s being developed so we can help advocate and inform.
Carol Vassar, podcast host/producer:
It sounds like AI is already in the clinical decision support, which means it’s supporting the electronic health records, which means it’s influencing decisions that are being made right there at the bedside. Is that a correct path?
Alison Curfman, MD, Imagine Pediatrics:
It may not be AI based, but we have a lot of technology that helps remind me not to forget to order a certain lab when I’m seeing a patient with a certain condition or I know that there’s order sets and there’s smart sets and another use of AI that’s really happening, taking on a lot of adoption in the clinical setting right now today is in documentation.
I’m a pediatric ER doctor and I’ve personally worked some places where we had scribes and they’re amazing. And it’s just like it totally changes your ability to sit down and look at the patient and be present when you’re not trying to remember or write down everything that they’re saying. So that’s an example of a human supporting a function that really helped me to, first off, have better documentation and second, that the scribe doesn’t sign the note, they prepare the note and I review and edit and sign. So that’s very accepted in today’s practice that human could be filling that role. But there’s a lot of solutions coming out now that are being used now for AI to fill that role.
And language models are very advanced. I use ChatGPT for all sorts of things. So being able to capture the core information in a patient visit and actually convert it to appropriate documentation, which is kind of the bane of all of our existence, is a great idea. But now if you take that and say, “Oh, and I want ChatGPT to then recommend an assessment and a diagnosis and a treatment plan and maybe pre-fill my orders,” that’s where we start getting scared, like, whoa, whoa, whoa, whoa. You can write a paragraph to summarize the conversation I just had, but it is not advanced or trained to actually take those next step. We’re not there yet, but it’s going to happen quickly. And I think that is an example of something that a lot of health systems are going to have to deal with soon.
And then also, what does that mean for our learners? Are they going to be developing those robust assessment skills that we need them to and to do that if there’s going to be robots doing it for us?
Carol Vassar, podcast host/producer:
So as we talk about AI, we talk about telehealth, we talk about technology, health IT, how can that be used to prevent bias, improve access? Are there examples of that happening right now?
Alison Curfman, MD, Imagine Pediatrics:
Yeah, so I think that I can’t speak a ton to AI models in pediatrics that are built reduce bias. I don’t know of anything that robust at this point. I do know of other interesting AI that I find it really interesting. I read an article about an AI-based chatbot that is meant to converse with people about beliefs that they may hold deeply, like against vaccinations or something. And seeing how effective this chatbot can be at helping provide a really empathetic approach and a lot of evidence and responding to the person’s kind of conversation style. And it was a really cool study because it actually made some impactful mindset shifts, which I think that is cool. But I don’t see those sorts of things robustly present in our current practice.
I can speak to other technology, in particularly digital and virtual care based on my experience with Imagine Pediatrics. So we are a medical group that contracts directly with Medicaid and COs to care for populations of children with medical complexity. So we’re a value-based company and we have our own technology platform where we can, first off, integrate all of their health data from their claims data and their current, all their other data to help us really categorize these patients and understand where their needs or their gaps may be. And then also enroll them in our intervention program, which we aim to help keep these kids healthy and out of the hospital. And a lot of that goes with a lot of upstream work that has to be done to make sure they have all their equipment and all their supplies and all their medications and their parents are supported and understand how to care for their child. And they have acute care support for if they have questions in the middle of the night.
So we built a model that is highly technology-based, but it’s also very, very human-focused. So it is a relationship-based program. We cannot make an impact on this population without earning trust with their parents. And I think that’s an example of technology being a tool that can actually help us with a human connection. And it’s not replacing a human connection because we have definitely had really amazing results in our population. We care for about, we have about 23,000 patients that we are at risk for, and we have engaged a very large number of that population and have been able to demonstrate that they have less hospitalizations and less ER visits and less utilization of the inpatient setting.
Carol Vassar, podcast host/producer:
So when it comes to building platforms like yours or building telehealth platforms or building AI, what I’m hearing is the foundational data needs to be as good as it possibly can be, and everyone needs to be at the table, including doctors, nurses, healthcare representatives, maybe even community members and patients. Talk about whether or not that’s happening now.
Alison Curfman, MD, Imagine Pediatrics:
I think it’s early. I think you’re talking about the use of digital tools, yeah, I absolutely see all sorts of children’s hospitals that have large multidisciplinary committees on how they’re implementing their tools that they’re building and getting the feedback of the right people within their organization. And a lot of them have parent and family advisory boards. I know we have that at Imagine and really want all of that input.
It’s not so much that people are necessarily informing the foundational data. What I see more of is the design of clinical models. And that’s what I specialize in now, is helping companies as they’re building out their clinical product to have an eye for things that could lead to inequity and assessing how you’re going to address that before you even start.
And I think that there’s definitely a multidisciplinary approach to a lot of new things in healthcare, which is great, but the technology initiatives need more than just a tech team. It needs that strong clinical judgment built in because technology can only do as much as the humans that inform it.
Whenever I am working on a clinical model that’s going to be tech-enabled or tech-informed, it’s really important for us to start with things that work in clinical care or even maybe things that don’t work. That’s where the complex care program came from is that our traditional healthcare system is not serving these complex kids well and they are experiencing a lot of challenges because of this. And it was almost like let’s draw a future for what it would look like if these kids were getting everything that they needed and their parents and families were getting the support they need, and then let’s build that.
But when it comes to other things like pieces of the model, you want to find things that work in clinical practice today that humans are doing and figure out a way to have the technology mirror that. So that’s similar to my example of having a scribe versus having HIPAA-compliant ChatGPT. So you want to mirror human processes that are working well. And then you also, when you’re designing or mapping out a new clinical product or model, you want to take a sweep of it and take a lens of where are the potential health equity implications here?
And whether that’s going all the way to the beginning and say like, “Oh, okay. Well, if I call a patient to enroll in the program and they don’t have a smartphone or they don’t have technology, that’s going to be something we head off then by making sure we provide that at no cost to them.” Versus things that are more further down around now they’re in the program and we have this messaging app, but they only speak some language other than English or Spanish. How are we building solutions that support that family and so they don’t get left behind?
Carol Vassar, podcast host/producer:
What happens long-term if we don’t build these tools with that kind of human element in mind that is addressing equity? If we fail to address equity, what’s the result?
Alison Curfman, MD, Imagine Pediatrics:
The current well-known gaps in care expand. So the difference between the care that different populations have access to will worsen. And I think this is a huge advocacy point at all levels of all things of the AAP, I feel like that’s one of our core beliefs as pediatricians, because quite frankly, children are marginalized populations. They are vulnerable, they are folks that don’t get nearly the resources that other patient populations get. And we see that across healthcare, and we, as pediatricians, are pretty good at having a loud voice to stand up for them.
So yeah, just inherently, if we don’t take that viewpoint, inequities will worsen. We have to understand what they are. We have to understand what to look for. It’s more than just race and socioeconomic status and language. There are many layers of this, and there are many very brilliant experts who have devoted their life work to really making sure that we take an equity approach.
And I think the AAP does this very well and puts a huge emphasis on this, and health equity is probably mentioned in every single one of their policy statements. So I think that’s just a part of being a pediatrician. But whether you’re advocating for something that’s more focused on the traditional delivery of healthcare and equity for children and marginalized populations, or a new technology, the same pediatrician principles apply.
Carol Vassar, podcast host/producer:
What do you see moving forward? What’s the future of health IT in helping to address health equity? And what innovations do you see coming along the pike?
Alison Curfman, MD, Imagine Pediatrics:
What I can say is that it will move at a pace that is not comfortable for us, and I think that’s kind of inevitable. I think things will develop too quickly that we’re not necessarily prepared or ready for. We didn’t have a lot of time to study it, and there’s not going to be a randomized control trial before someone starts using it in a clinical setting. So I think that we’re approaching a pace that things are really going to happen quickly and it’s going to be hard to throw the brakes on sometimes.
But what I’d see as far as really promising innovations that are coming is something that I see a lot more advanced in the adult world, which is along the lines of population health and predictive analytics and large data sets to be able to understand rising risk in a population and in individuals that are being cared for in that population. I find the big data to be really exciting when we’re looking at the population that we serve at Imagine of 23,000 children with chronic conditions. We’re able to really come up with some intelligent assessments of the population and actually target interventions.
And things like in the past may have been reflected as quality metrics that we know are associated with health. So we know that if a child has their well-child exam, that is associated with good positive outcomes. But it’s not specific, it’s not targeted. It’s not customized to what the patient needs. So I think there’s a lot of indicators in people’s healthcare utilization and their other data that can help us really target, and when I say interventions, it’s really things surrounding that family with support.
So we have different appointment cadences for patients that are at different levels of risk within our population. And some children may not have a ton of chronic diagnoses, but they actually have a lot of needs and they need that additional support. And then you may have another patient that has like 20 complex conditions and a trach and event and a G tube, but the parents are like, “Hey, I’ve got this. We’re in a good place right now and thanks for your help,” but they’re kind of plugging along. So understanding populations and being able to provide people with the support they need and making sure we’re using data to target our interventions in an intelligent way, I think that’s really exciting.
Carol Vassar, podcast host/producer:
What role, if any, do policymakers play in all of this?
Alison Curfman, MD, Imagine Pediatrics:
I think they obviously make the policy around what, in short, is focused on healthcare payment. That’s where policy really is kind of the end all be all in healthcare, because we saw this in telehealth, we saw that there was no policy around the use of telehealth in children. This was back in 2013 or ’14 when I started getting involved in this work. And there was no policy because there was no evidence base. And there was no evidence base because no one was doing it. And no one was doing it because they wouldn’t get paid for it. So our approach to that, I worked with three other co-founders and we created this kind of grassroots effort to create a multi-center national research network on pediatric telehealth with the goal of informing policy. And it’s called SPROUT, it’s now part of the AAP.
So I think policymakers are very eager to hear from the experts. They want to be informed, they want to be making policy decisions that support the needs of the people they’re trying to support, the constituents they’re trying to support, but they don’t know what they don’t know. And we, as physicians and pediatricians and other pediatric healthcare providers, have that background and that expertise and that knowledge and that lived experience. So policymakers want your input.
Carol Vassar, podcast host/producer:
When it comes to the intersection of health IT and health equity, what have we not covered that you would like to share?
Alison Curfman, MD, Imagine Pediatrics:
I think that really the call to action here is that pediatricians can have a role. Even if you think that, “Oh, I don’t have a technology background.” Guess what? I don’t either. I started a digital health company and I don’t have a technology background. I learned a lot about developing technology and a very, very different mindset compared to how we work in traditional healthcare. But I’m not even a health informaticist, but I have really strong opinions about what my patients need and what children need and what supports they need and where the problems are. And it’s like those are the viewpoints that inform technology development to support the current problem.
So if you know of a problem, if you’ve ever been like, “Oh, I wish we could just do this instead, it would be so amazing if a world existed where this could happen for these patients,” that’s where I started. I started with a blank piece of paper. Give me a whiteboard and I will fill it with ideas. That’s how Imagine started is my co-founders and I just mapping it all out. And we can have a role in this. It doesn’t matter if your background’s not in tech, you are an expert in child health.
Carol Vassar, podcast host/producer:
Dr. Alison Curfman is a pediatric emergency physician and co-founder of Imagine Pediatrics.
MUSIC:
Well beyond medicine.
Carol Vassar, podcast host/producer:
Thanks to Dr. Curfman for sharing her time with us, and thanks to you for listening. Here at the Well Beyond Medicine Podcast, we’re always open to your ideas for upcoming episodes. Just email us at [email protected] with your ideas and suggestions. You can also head over to our website NemoursWellBeyond.org to leave us a voicemail with your concept. That’s the place where you can also subscribe to the podcast, like 37,000 of your friends and neighbors and leave a review. That’s NemoursWellBeyond.org.
Our production team for this episode includes Cheryl Munn, Susan Masucci, Lauren Teta, and Steve Savino. Join us next time as we explore the importance of early access to early intervention services for children with disabilities. I’m Carol Vassar. Until then, remember, we can change children’s health for good, well beyond medicine.
MUSIC:
Let’s go-o-o well beyond medicine.
