Proactive Identification of Patients with Diabetes at Risk of Uncontrolled Outcomes during a Diabetes Management...

Area of Science:

• Diabetes management and telehealth
• Machine learning applications in healthcare
• Predictive modeling for chronic diseases

Background:

• Telehealth advancements enable identification of high-risk individuals for diabetes management.
• Predictive modeling is crucial for improving diabetes care.
• Remote diabetes monitoring programs (RDMP) offer targeted support.

Purpose of the Study:

• To develop a novel machine learning (ML) approach for proactive identification of participants at risk of uncontrolled diabetes within an RDMP.
• To predict diabetes risk at 12 months for participants in a remote monitoring program.

Main Methods:

• Utilized registry data from the Livongo for Diabetes RDMP.
• Developed dynamic predictive ML models for monthly checkpoints (month-0 to month-11).
• Incorporated participant attributes: survey data, blood glucose (BG) levels, medication fills, and health signals. Models trained using light gradient boosting machine with hyperparameter tuning.

Main Results:

• ML models demonstrated strong performance in identifying at-risk participants.
• Recall ranged from 70%-94% and precision from 40%-88% for observable at-risk individuals.
• Model performance improved over the program journey, highlighting the value of engagement data.

Conclusions:

• Proactive ML models accurately identified participants at risk of uncontrolled diabetes with high, generalizable precision.
• Personalized interventions can be implemented based on identified risk at various program stages.
• This approach advances large-scale remote monitoring, preventing complications and improving glycemic control.