Improving accuracy and precision of glucose sensor profiles: retrospective fitting by constrained deconvolution

Area of Science:

• Biomedical Engineering
• Medical Informatics
• Clinical Research

Background:

• Accurate blood glucose (BG) measurements are crucial for clinical trials but are invasive and costly.
• Continuous glucose monitoring (CGM) offers high temporal resolution but lacks sufficient precision and accuracy for reference use.
• Existing methods struggle to balance the accuracy of sparse BG references with the high-frequency data from CGM.

Purpose of the Study:

• To develop and validate an algorithm for reconstructing accurate, continuous-time BG profiles.
• To improve the reliability of CGM data for clinical trial modeling and outcome metric computation.
• To leverage both sparse, accurate BG references and high-resolution CGM data simultaneously.

Main Methods:

• A two-step constrained semiblind deconvolution algorithm was developed.
• The algorithm estimates sensor calibration and diffusion model parameters.
• It performs regularized deconvolution of CGM data, constrained by BG reference confidence intervals.

Main Results:

• Reduced mean absolute relative deviation from 15.71% to 8.84% compared to unprocessed CGM.
• Significantly decreased errors in clinical outcome metric evaluation, such as halving the error in time-in-hypoglycemia assessment.
• Demonstrated improved accuracy and precision of reconstructed BG profiles.

Conclusions:

• The developed algorithm successfully reconstructs reliable continuous-time BG profiles.
• The reconstructed BG data is suitable for clinical trial assessment, modeling, and offline applications.
• This approach enhances the utility of CGM data by improving its accuracy and precision.