Predicting the Future of Health Care: Using Big Data to Treat Type 1 Diabetes

More than 18,000 cases of type 1 diabetes are diagnosed among young people under age 20 in the U.S. each year, according to the American Diabetes Association. Studies show that poor disease control at any age significantly raises lifetime risk of serious complications, such as heart disease and kidney disease.

In 2017, Children’s Mercy Kansas City and Cyft, Inc., which leverages data science and advanced prediction methods to achieve performance improvement in health care systems, partnered on a grant-funded project to predict which youth and young adults with type 1 diabetes may be at risk for these negative health outcomes, and to evaluate new approaches to quality-improve care before those negative outcomes happen.

Led by Mark Clements, MD, PhD, pediatric endocrinologist at Children’s Mercy and Professor of Pediatrics at the University of Missouri Kansas City, and Leonard D’Avolio, PhD, Assistant Professor, Harvard University, and CEO, Cyft, Inc., the project seeks to allow clinicians to rapidly and iteratively test innovative approaches to diabetes care until they identify approaches that work. The team is using both data science and implementation science to accomplish their goal.

First, they work to “get in front of” problems by predicting negative events before they occur. This effort leverages machine learning along with information from the electronic health record, diabetes self-management devices, and patient reported outcomes surveys. Then, they use a combination of quality improvement methods and pragmatic trials to iteratively test promising behavioral interventions, Mobile Health (mHealth) interventions, and care delivery interventions. Advanced visualization tools allow the team to track outcomes for improvement. They ultimately hope to learn which innovative approaches work best for which individuals with diabetes, and then to teach other diabetes centers how to implement the same personalized approaches to care.

The project was funded by The Leona M. and Harry B. Helmsley Charitable Trust, which recently granted an additional $8.5 million to Children’s Mercy Kansas City, leveraging predictive analytics technology deployed by Cyft, to continue the project and scale it up. Also involved in the project are Sanjeev Mehta, MD, MPH (advisory committee chair; Joslin Diabetes Center), Susana Patton, PhD, CDE (director of intervention development; Nemours Children’s Health System), as well as the following Children’s Mercy staff: Ryan McDonough, DO (clinical lead, quality improvement), Emily DeWit (project coordinator), Diana Ferro, PhD (research scientist), Brent Lockee (data scientist), Mitchell Barnes (data analyst), Mark Hoffman, PhD (administrative sponsor, data science advisor), Emily Hurley, PhD, MPH (formative research), John Lantos, MD (bioethics), David Williams (biostatistics), Brian Rivera and Avinash Kollu (research compute team), Krista Nelson and Sally Guezuraga (Innovation Center), and Morgan Waller (telehealth).

The goal of the project is simple: to use big data approaches to treat diabetes while ensuring the right intervention is used for the right person at the right time.

We spoke with Dr. Clements and Dr. D’Avolio about this new paradigm in diabetes care and how data science, behavioral science, mHealth and user-centered design can be combined to change the future of health care.

What Are the Results so Far From the Work That Began in 2017?

Dr. Clements: We wanted to identify and predict our highest risk patients in a way that would allow us to intervene and prevent poor health outcomes before they happen. Predicting patients in highest need of intervention isn’t easy. We collect a lot of clinical data, but only a small amount is used to direct their care. We saw an opportunity to take all data captured during routine care and put it to work helping us understand which patients may be at risk for negative outcomes.

Working with Cyft, we initially focused on predicting which youth with diabetes would experience a rise in A1c (blood glucose) levels. Rising A1c signals that the patient’s long-term blood sugar values may be too high, which can lead to serious complications. We are also predicting hospital admissions for diabetic ketoacidosis, a life-threatening condition caused by serious insulin deficiency. Having robust models to predict these two outcomes gives us two important targets for improvement. We’ve now developed a diabetes rapid learning platform to test innovative approaches to care that might improve outcomes for these youth.

What Are Some of the Interventions You Are Testing?

Dr. Clements: We first developed a telehealth intervention to create more frequent and timely contact to address issues that youth and young adults with type 1 diabetes face.

Care delivered via telehealth provides the opportunity for more frequent contact with patients through direct-to-home video micro-visits, which may deliver more timely and personalized interactions. For example, providers can discuss with their patients what they are experiencing this week rather than 90 days ago.

The second intervention we have initiated is the PEEPS (Patients Encouraging and Engaging Peer Support) program that pairs our teen patients with young adults living with diabetes so they can experience positive mentorship. Teens can learn from the experiences of those who have gone through similar health challenges, which may reduce distress and improve coping.

We’re about to launch a third intervention in partnership with a digital health company that makes software for wrist wearables. It uses machine learning to detect when a person is eating and engage them with supportive text messages on the watch at that moment. We think this may be of interest for youth with diabetes who are predicted to have a rise in their A1c because it may help them better manage insulin at meals. This is an exciting first-of-its-kind intervention, and we are fortunate to evaluate it in the context of this project.

How Does This New Award Support the Objectives You Are Working Toward?

Dr. Clements: Our vision is to leverage data science to predict outcomes, quality improvement methods and implementation science to improve diabetes outcomes. We’re already using this as standard of care at Children’s Mercy, but we have new objectives that we can now accomplish through this new funding.

Working with Cyft, we have developed a rapid learning platform for diabetes, which allows us to efficiently make predictions and pair them with interventions.

Our model is simple: 1) select an outcome we want to improve; 2) develop a model to predict the outcome using all available data, and 3) create a “change package+.” A change package+ is designed to guide how we predict, track, and improve an outcome.

Once we have achieved those three steps, we select an intervention and put it through either a small test of change (Plan-Do-Study-Act cycle) or a pragmatic trial. Sometimes that leads to expanded testing of the intervention. Other times, we need to tweak the intervention and re-evaluate. At times, we may need to pivot and evaluate a different intervention. If we see something that looks very promising, it’s possible to plan a definitive clinical trial to help us understand with the most rigorous methods how well the intervention works.

The basic idea is that we can try early implementation of minimal-risk, novel interventions. If we see a signal for improvement, it may be an indication that we can expand it or initiate a clinical trial. If not, we can move to a different one. When we’re done, we provide a roadmap that helps other clinics adopt it and helps payers see value in it.

Dr. D’Avolio: We now can use math plus advances in computational power to detect patterns in large collections of data. This math + compute power is often referred to as machine learning or even artificial intelligence (a name I’m less fond of).

Companies in other industries have successfully harnessed machine learning to better understand consumer behavior in order to sell them more goods and services. This same approach can be used to better understand health care needs and to identify those likely to benefit from specific interventions. It’s about using data to get the right care to the right people at the right time.

This rapid learning platform gives us the opportunity to put this new capability to work helping kids and their families living with type 1 diabetes. That’s pretty exciting.

How Does This Work Support the Idea of ‘Personalized Medicine’?

Dr. Clements: People use two terms: personalized medicine and precision medicine. The traditional definition of this is that medical care can be tailored to the genetic and molecular profile of the individual. This speaks to the field of pharmacogenomics, where we identify the genetic makeup of a particular cancer and tailor chemotherapy to that cancer. Or we can identify diseases treated with drugs that have a variable effect based on a person's genetics, and tailor the dose of that medicine to their genetics. That’s what precision or personalized medicine has been about up until now.

My vision is that we are heading toward a new definition of personalized medicine, which is that medical care can be tailored to the predicted outcomes and the unique response of the individual to all available therapies. And those therapies include devices, drugs and biologics, but also mobile health, behavioral health and health care delivery interventions.

Dr. D’Avolio: There are no average patients, and it's very rare for two patients to experience the same benefit from a standardized, single approach to caring for them. Clinicians know that good health care is adaptive to individuals’ situations. The ability to use data to better understand patients, disease, treatments, and outcomes puts us in a position to tailor our approach to different individuals with different needs.

What Does the Future Hold?

Dr. Clements: We want to develop predictive models for new outcomes and then identify and test new interventions faster than we have before. More importantly, we want to learn faster than we have before.

We’re looking at other novel interventions that might help us. What if I could see 10 of my patients’ blood glucose levels continuously? What if our team could follow them every day and see what their glucose profiles look like? What if we could interact with them and say, “I notice you have a lot of high blood sugars and they're happening after lunch every day so let's talk about what you’re doing after lunch?” My goal is to take the best interventions that very talented colleagues across the field have developed, and to evaluate more of them faster as tools to prevent a problematic outcome from happening. That kind of granular, tailored, and timely interaction based on the specific needs of our patients holds so much more promise than a “one size fits all” health care system.

Dr. D’Avolio: Today there is a lot of attention around technology, and rightfully so. Telemedicine makes it possible for us to keep people healthy in their homes and even deliver care through a pandemic. Machine learning can detect patterns that allow us to personalize care to individuals. These are really important tools. However, if you’ve ever been involved in a home improvement project you know that tools alone do not build anything of value. You need a plan and a multidisciplinary team working together to build something wonderful. In the near future, these capabilities will simply be assumed and more of our attention will be adapting health care systems and practices to take full advantage of what they make possible. I think that’s exactly what we’re setting out to do with this project.