Dr Andrew Wilner (Host): This is Transformational Pediatrics with Children's Mercy Kansas City. I'm your host, Dr. Andrew Willner. Today, I have the pleasure of speaking with Dr. Mark Clements, professor at the University of Missouri Kansas City. Dr. Clements is a pediatric endocrinologist and an expert in diabetes care and diabetes data analytics. He currently serves as a faculty leader in the T1D Exchange Quality Improvement Collaborative. I invite you to listen in as we discuss population health management and pediatric type 1 diabetes. Dr. Clements, thanks for joining us.

Dr Mark Clements: Thank you for having me.

Host: Dr. Clements, before we get deep into our topic, I was reviewing your extensive bio. And since I'm a neurologist, I couldn't help but notice that you have a PhD in Developmental Neuroscience. So, how did that happen and how did that lead into population health?

Dr Mark Clements: Yes. Well, when I was attending medical school, I attended the Medical Scientist Training Program at Washington University in St. Louis, and I was a bench researcher, very much involved in molecular biology in model genetic organisms and was studying transcription factors that were involved in development of the nervous system in the laboratory of Dr. Jeffrey Milbrandt. And as I was going through my clinical training In pediatrics and, subsequently, in pediatric endocrine and diabetes, I really realized that my calling was to be a little closer to the bedside in the research that I did. And so, I began to pursue training on my own in health outcomes research and clinical trials, and that has eventually led me to population health management.

Host: All right. So, what is, from your standpoint, the biggest issue regarding population health and type 1 diabetes in children?

Dr Mark Clements: I think the biggest issue that I see is that the care in the US and elsewhere in the world is really not achieving the desired effect. In the US, we only have approximately 20% of youth meeting targets for blood glucose control and I think we can do better and we must do better. Part of the issue is that our current standards of care, which rely on four visits a year in clinic, really don't account for the incredible daily burden of self-management and the type of support that the medical community could be offering young people and their families.

Host: What is the Rising T1DE Alliance?

Dr Mark Clements: Yes, the Rising T1DE Alliance is our initiative to tackle population health and type 1 diabetes in young people and eventually, we hope, in adults. It is an initiative that Is grant-funded. It is led by Children's Mercy. We have a number of collaborators at other institutions, and the goal is really to get in front of clinically important problems by predicting them, to be able to segment the population by risk so that we understand who's struggling the most and then, to be able to differentiate our care so that we are not delivering one-size-fits-all care, which is closer to what the model looks like today, and to leverage all the tools we've gathered from psychology that are evidence-based to teach us how to nudge human behavior towards positive health-promoting behaviors.

Host: Yeah. You know, I want to get back for a second to what you told me in the beginning, that children with type 1 diabetes, I guess they're allocated, or the standard of care is four visits a year. Now, I remember when the problem was we didn't have technology, we didn't have instant glucometers. You know, you had to do a blood sample. We didn't have insulin pumps. We didn't have a sophisticated understanding of diet. You know, this goes back during my training, more than 40 years ago. But now, we have all those things. So, it seems like the problem has changed from, "Gee, we don't know how to do these things" to, "We know how to do them, but we're not doing them."

Dr Mark Clements: That's correct. And I would also point out that the observation that we have all these incredible advanced technologies. We actually have insulin pumps that have artificial intelligence embedded in them to adjust the insulin dosing every few minutes. The fact that our young people are still not achieving treatment targets means that it is only about 10% of the recipe for success if you treat the disease with a drug or a device. The other 90% of success in treating a chronic disease like diabetes is treating behavior; behavior of the child, of the family, of the adult with diabetes, and frankly the behavior of the healthcare team.

Host: Right. Well, I'm an epileptologist and I treat people with epilepsy and other chronic disease that requires daily treatment. And many, many patients, we have the drug that works, but getting the drug into the patient on a regular basis is a challenge for all those reasons, for a lot of social reasons, financial reasons, access. And of course, children are growing too, so I imagine that they're a little harder to keep up with in terms of adjusting their doses and looking after their general health. So, it's an even bigger problem.

Dr Mark Clements: You're absolutely right. One of the things I've noted for a long time about the schedule of four visits a year is that it's really not adequate in a young person who's growing so quickly and especially in a teenager, where there's the development of significant resistance to insulin, the major drug that we use to treat diabetes, just to keep up with the rapid changes developmentally.

Host: Okay. So, I was perusing these materials and there's something called the Rapid Learning Lab Concept. What's that?

Dr Mark Clements: Yeah. So, I mentioned that we are trying to get in front of clinically important problems by predicting them. And we are also using non-predicted biomarkers, you know, like your average glucose level or your hemoglobin A1c to identify patients at risk. We're also using behavioral biomarkers, like how you take your insulin at meals each day.

And the Rapid Learning Lab is on the receiving end of those risk stratifications. The idea is that we can curate, from all of my really smart colleagues across the field, the best behavioral and care delivery interventions, often delivered through a digital vehicle, and we can learn how to match patient need to the right behavioral intervention using the seven pillars of diabetes self-management as our guidepost, so that we essentially end up at the end of the day with a toolbox of interventions that we can call upon that's organized and curated in a way that we haven't done before in diabetes care.

Host: All right. Now, is that linked to the All-Data-On-Hand Deep Learning Model?

Dr Mark Clements: So, one of our very early experiences was to develop a predictive model to identify those young people who were likely to be hospitalized with diabetic ketoacidosis in the next six months. And of course, diabetic ketoacidosis is associated with some morbidity, some mortality, unfortunately. And it's also preventable. And from the perspective of the healthcare system, we're actually having to use beds in the hospital that are for preventable hospital admissions that might better be utilized in other ways. So if we can keep young people out of the hospital, that's a worthy goal. So, that all-data-on-hand model was one of our first experiences building a machine learning model, in fact, a deep learning model, to identify youth who were going to be hospitalized. And it's a pretty high-performing model. We've since revised it. We've made it better and more understandable with the goal of achieving explainable AI and, hopefully, more disseminable to other centers.

Host: Now, I'm going to guess here, but it sounds like all-data-on-hand means that you need the data from the patient more often than four times a year to feed the model. So, the model can be monitoring, you know, like my car is telling me that it's time for service. It's got some little computer in there that knows. How are you getting the data from the patients?

Dr Mark Clements: So, two comments in response to that question. So, the first is that that model that we published was based on electronic health record data. It was not just data collected four times a year, it was actually data collected at any encounter with the health system at any time. not just discrete data in the medical record, but also all of the free text notes that clinicians generate. It's what is called a very wide model. there are some problems with a model like that being highly disseminable. People are concerned about its performance when you pick it up from the training population and move it into broader populations, which is why we've really done some work to create a more explainable model that uses on the order of a dozen to a dozen and a half data features. And we are now working to incorporate the patient proximal diabetes device data, like their glucose data and their insulin delivery data. And the goal for that is so that we can revise predictions every week or even every day. instead of every month.

Host: One sort of obvious, at least seems obvious, way of collecting data would be to receive a daily Bluetooth-generated glucose number that auto-populates and on an Excel spreadsheet. And when it's too high or too low, a little alarm goes off and says, "Hey, the nurse better call the patient because it's too high or too low." I mean, the technology for that is pretty basic. Does that ever happen?

Dr Mark Clements: You'd be surprised at how little it happens currently. So, there are very few centers that are engaged in what I would term remote patient monitoring, where they are receiving glucose and insulin data on a daily basis and then are able to proactively respond to the patterns in those data. We have a few examples. So, our center has actually created a population health management software in the cloud that allows us to bring the data in from most of the devices that are used in daily self-care. And we are in fact assessing those data now for patterns to identify risk. We have colleagues at Stanford that we're partnered with who have a very interesting project called 4T, and we are collaborating with them on being the first implementation site for 4T outside of Stanford. And we are in fact, helping them to innovate on what they call the TIDE platform, which is a complete coincidence. It's unrelated to the Rising T1DE, name of our initiative. And we are, for instance, seeking to add predictive analytics that will forecast change in blood glucose on people's glucose sensors in the next few weeks.

Host: What's the biggest obstacle for you to control type 1 diabetes in children and what are you going to do about it?

Dr Mark Clements: Frankly, the biggest obstacle is that all of the data that are essential to daily self-management and essential for the care team to be empowered to provide proactive care, including between clinic visits, is that the data are all locked up in silos. You know, the manufacturers of diabetes self-management devices tend to think that they are the owners of the data. The health systems would like to think that they're the owners of the data. And of course, the people with diabetes would like to believe that everybody will allow them to be the owners of their own data. So, we really need to resolve these issues of data mobility and safe, secure, and private sharing of data to allow seamless support of care processes and care innovation.

Host: Well, I want to thank my guest, Dr. Mark Clements, for this fascinating discussion on population health and type 1 diabetes. Dr. Clements, thanks for joining us.

Dr Mark Clements: Thank you so much, Andrew.

Host: To refer your patient or for more information, please visit childrensmercy.org to get connected with one of our providers. This has been Transformational Pediatrics with Children's Mercy Kansas City. Please remember to subscribe, rate and review this podcast and all the other Children's Mercy podcasts. I'm your host, Dr. Andrew Willner. Thanks for listening.