US bodies employ machine learning to tackle diabetes
June 20, 2017
US institutions have joined forces to use machine learning to improve care for people with type one diabetes.
Children's Mercy Kansas City and Joslin Diabetes Center announced the project with Cyft to apply predictive analytics to identify and manage at-risk individuals. They are working with the Leona M and Harry B Helmsley Charitable Trust to create a new learning health system to improve the care of individuals diagnosed with type one diabetes (T1D).
Starting this summer, Children's Mercy and Joslin will deploy machine learning to manage health outcomes proactively in patients with T1D at two independent diabetes clinics. Using technology pioneered by Massachusetts-based Cyft, the project will work to optimise aspects of diabetes management by supplying novel information to clinical staff at the point of treatment.
Four industry leaders in care and innovation will guide the project as they seek to create, evaluate and deploy predictive models at the two selected clinics. They are: Mark Clements (pictured), medical director for the paediatric clinical research unit at Children's Mercy Kansas City; Sanjeev Mehta, Joslin Diabetes Center's chief medical information officer; Leonard D'Avolio, founder of Cyft and assistant professor at Harvard Medical School and Brigham & Women's Hospital; and Susana Patton, associate professor of paediatrics at the University of Kansas Medical Center.
Machine learning is an entirely new approach to health analytics because it can generate robust insights from unstructured and imperfect data, such as the free text notes found throughout electronic health records. Validated by over a decade of research and clinical applications, Cyft technology will employ machine learning and natural language processing as well as device signal processing to analyse multiple data sources and create predictive models for use by health professionals.
These models will detect and alert caregivers to opportunities to intervene in the care of patients at risk for deterioration in their health outcomes.
The three-year project is funded by a grant from the Helmsley Charitable Trust, a foundation that seeks to improve lives by supporting efforts in the USA and around the world.
"Advancing care for type one diabetes has traditionally been difficult as we are working to better understand the impact of clinical and socio-demographic risk factors on outcomes, while also incorporating these insights into patient management strategies," said Clements. "Due to the development of machine learning technologies, we can now make these data points immediately useful to individuals who are delivering care, not just those conducting research. This project aims to not only prove we can generate accurate type one diabetes learning models, but also use this information to proactively improve health outcomes and impact the wider type one diabetes community."
T1D affects roughly 1.5 million people in the USA, with more than 18,000 new cases diagnosed among youth in the country from 2008 to 2009, according to the American Diabetes Association. T1D is the second most prevalent chronic disease of childhood after asthma. Studies show that proactive care for glycaemic control early in the course of the disease has a persistent influence on long-term clinical outcomes, making management of the disease during childhood paramount to reducing life-long risks for those living with T1D.
"For individuals living with T1D, we have learned much about risk factors for suboptimal health outcomes, but there remain significant opportunities to proactively identify and engage our patients who are at risk for future deterioration," said Mehta. “Predictive analytics holds promise in this area as well as in the identification of novel clusters of patient factors that could identify high-risk patients. This learning health system will further our goal of leveraging the power of our data to identify and proactively support patients who are at risk for clinical deterioration to positively impact their health and general well-being."
Cyft partners with care organisations to identify opportunities to optimise care management efforts for chronic and complex, behavioural health, and dual-eligible populations.
"We're increasingly seeing the effective use of predictive analytics to solve challenges in our healthcare system and this project represents the next step in that evolution," said D'Avolio. "We can no longer be a wait-and-see industry. Instead we're pulling real insights from disparate data sources and using these to inform clinical care. We're thrilled to partner with these leading institutions to serve such a critical patient population, and believe that the work this new learning health system will accomplish could fundamentally change how we care for people with T1D and their families."
Founded in 1897, Children's Mercy is one of the nation's top paediatric medical centres. With not-for-profit hospitals in Missouri and Kansas, and numerous specialty clinics in both states, it provides care for children from birth through the age of 21.
Joslin Diabetes Center is renowned for its expertise in diabetes treatment and research. It is dedicated to finding a cure for diabetes and ensuring that people with diabetes live long, healthy lives. Joslin is an independent, non-profit institution affiliated with Harvard Medical School, and one of 11 NIH-designated diabetes research centres in the USA.
The Leona M and Harry B Helmsley Charitable Trust aspires to improve lives by supporting effective organisations in health and select place-based initiatives. Since beginning its active grant making in 2008, Helmsley has committed more than $1.8bn for a wide range of charitable purposes.