Improved individual and population-level HbA1c estimation using CGM data and patient characteristics

Area of Science:

• Biomedical data science
• Clinical informatics
• Diabetes technology

Background:

• Continuous glucose monitoring (CGM) provides valuable data for diabetes management.
• Accurate HbA1c estimation is crucial for assessing long-term glycemic control.
• Existing methods like the Glucose Management Indicator (GMI) formula have limitations.

Purpose of the Study:

• To evaluate the performance of machine learning (ML) and linear regression models for HbA1c estimation using CGM data.
• To compare the accuracy of these models against the established GMI formula.
• To assess the utility of ML models in remote clinical trial settings.

Main Methods:

• Utilized CGM data and participant-specific information.
• Developed and applied machine learning and linear regression models.
• Compared model predictions against the GMI formula for HbA1c estimation.
• Evaluated performance at both individual and cohort levels.

Main Results:

• ML and linear regression models reduced HbA1c estimation error by up to 26% compared to the GMI formula.
• Models demonstrated superior performance in estimating the median HbA1c at the cohort level.
• The findings suggest enhanced accuracy and reliability in glycemic assessment.

Conclusions:

• Machine learning and linear regression models offer a significant improvement over the GMI formula for HbA1c estimation.
• These models show promise for supporting remote clinical trials, especially those affected by disruptions like COVID-19.
• The enhanced accuracy can improve diabetes management and research outcomes.