AI predicts glucose level changes in children with type 1 diabetes

Key takeaways:

• The tool predicts changes in A1C in the next 90 days and risk for hospitalization for the next 180 days.
• These predictions allow providers to intervene quicker and reduce patients’ risk for poor outcomes.

Researchers developed a tool that can collect daily data from insulin pumps and continuous glucose monitors and use AI to predict changes in glucose levels.

The tool can help providers monitor patients remotely and respond more quickly to changes that can increase children’s risk for diabetic ketoacidosis and hospitalization, according to a presentation at the AIMed24 Annual Meeting.

“If you look at continuous outcomes across pediatric to adult transition, we have young people who are having worse and worse glycemic outcomes,” Mark C. Clements, MD, PhD, CPI, FAAP, pediatric endocrinologist and professor of pediatrics at Children’s Mercy in Kansas City, Missouri said in a presentation. “The fact that kids are getting worse and worse from age 8 to 18 [years] is a real problem.”

In 2020, Clements and colleagues set out to solve this problem by founding the Rising T1DE Alliance to create AI-based solutions to these issues.

The project Clements discussed at AIMed24 was Rising T1DE’s Diabetes Data Dock, a tool that collects data from diabetes self-management devices like insulin pumps, smart bands, continuous glucose monitors and glucometers, and combines it with electronic health record data in the cloud. The platform allows providers to monitor patients remotely and implement interventions as soon as they are needed, rather than waiting for their quarterly in-person clinic visit.

“Patients only spend 0.03% of their waking hours in the clinic with us, so it’s very difficult to move the needle if we’ve only got four points of contact in a year,” Clements said.

Additionally, they ran the data through machine learning models, which can forecast risk for change in A1C up to 90 days and risk for hospitalization up to 180 days. The model can also predict changes in time spent in target glucose range for up to 14 days, Clements said.

Over the last 4 years, Clements said this technology has reduced diabetic ketoacidosis prevalence by 33%, and 30% fewer children have an A1C above 9%, which he said is considered high risk.

The models originally predicted 65% of children would have an A1C spike within the next 90 days, but since his team has begun implementing remote interventions, the proportion is less than 25%, Clements said.

Clements and colleagues also developed a patient engagement app, which sends push notifications to patients when they are eligible for an intervention. When the patient chooses to enroll, it notifies their care team to begin the intervention.

“The goal is to create a new kind of ecosystem for care — one that is more efficient, but also empathetic and reactive to patient need as they go through their journeys with diabetes,” Clements said.

References:

• Clements M. Leveraging AI for personalized management of pediatric diabetes: the Rising T1DE Alliance experience. Presented at: AIMed24 Annual Meeting; Nov. 17-19, 2024; Orlando.
• Rising T1DE Alliance. https://www.childrensmercy.org/departments-and-clinics/endocrinology/rising-t1de-alliance/. Accessed Nov. 25, 2024.