Glucose Predictions Improve Glycemic Control: A Digital Twin Evaluation

Area of Science:

• Endocrinology
• Digital Health
• Artificial Intelligence in Medicine

Background:

• Continuous glucose monitoring (CGM) users can benefit from predictive alerts to prevent adverse glycemic events.
• The Accu-Chek SmartGuide Predict app utilizes machine learning for predictive features like low glucose prediction (LGP) and 2-h glucose forecasting.

Purpose of the Study:

• To quantify the glycemic benefits of the Predict app's features in adults with type 1 diabetes (T1D).

Main Methods:

• An in silico comparative study using the University of Virginia Replay digital twin simulator.
• 204 digital twins (29,929 days of data) were used, comparing standard hypoglycemia/hyperglycemia mitigation with interventions incorporating user responses to app predictions.

Main Results:

• Simulated use of predictive features reduced time below range (<70 mg/dL) by 2.9% and increased time in range (70-180 mg/dL) by 3.6%.
• CGM hypoglycemia alarms were reduced by 67%, with 10g fast-acting carbohydrates recommended for LGP notifications to balance efficacy and prevent rebound hyperglycemia.

Conclusions:

• The Accu-Chek SmartGuide Predict app shows significant potential for improving glycemic management in adults with T1D.
• In silico evaluation provides strong evidence for the app's clinical utility in diabetes care.