Three experts discussed the use of artificial intelligence (AI) and machine learning to improve diabetes care during the Scientific Sessions symposium Artificial Intelligence, Machine Learning, and Diabetes.
The presentation can be viewed by registered meeting attendees at ADA2020.org through September 10, 2020. If you haven’t registered for the Virtual 80th Scientific Sessions, register today to access all of the valuable meeting content.
Care models using AI can help predict and mitigate risks associated with glycemic control, said Josep Vehi, PhD, Professor of Control and Biomedical Engineering, and Associate Researcher in the Institute of Biomedical Research at the University de Girona in Spain.
Continuous glucose monitoring systems and smart-pen therapies benefit most from AI-based prediction algorithms and risk mitigation tools, he said. Risk prediction and condition assessment algorithms can also help reduce manual intervention in artificial pancreas systems and increase time in auto mode, leading to better glycemic control.
Adrian Aguilera, PhD, Associate Professor in the School of Social Welfare at the University of California, Berkeley, and Assistant Adjunct Professor in the Department of Psychiatry at the University of California, San Francisco, discussed how AI can be used to improve precision medicine and tailor treatments for type 2 diabetes patients.
Vulnerable populations—low-income, low health literacy, and ethnic minority groups—experience higher prevalence and worse outcomes in diabetes, said Dr. Aguilera, adding that mobile apps can be used to improve outcomes by helping patients engage in healthy behaviors, including physical activity.
“However, these mobile health apps are not typically designed for these populations. And if these populations have the highest need, we feel we can make the most impact by including them in the development,” he said. “It comes down to making sure that the big data that we utilize to develop our algorithms is inclusive of a broad swath of the population.”
Dr. Aguilera also discussed findings from the Diabetes & Mental Health Adaptive Notification Texting Study, which was designed to determine how to leverage reinforcement learning to increase physical activity. The study involves 200 low-income diabetes patients from San Francisco General Hospital who have symptoms of depression.
In the study, artificial intelligence was used to predict the most effective content to provide personalized, motivational text messages to study participants and increase their physical activity.
“Text messages are widely accessible, cost-efficient, and easy to use, with high engagement rates and a greater sense of personal connection,” Dr. Aguilera said.
Arianna Dagliati, PhD, from the Department of Electrical, Computer and Biomedical Engineering at the University of Pavia, Italy, and the Center for Health Informatics at the University of Manchester, England, discussed how AI can be used to screen for and predict diabetes complications. AI supports decision-making by physicians and succeeds when its results translate into transparent and accessible tools, she said.
Dr. Dagliati described the concept of learning health systems, where continuous study, data, and analytics are used in daily practice to continually improve care. Think of the components of learning health systems—patient history, transferring data to knowledge, and transferring knowledge back to the patient—as a continuous cycle that continually learns and evolves, she said.
Dr. Dagliati also discussed the MOSAIC Project, a European Union-financed initiative that is developing models and algorithms to improve the evaluation and identification of those at risk for developing type 2 diabetes and related complications.