Accuracy of Glucose Meter Technologies in Low vs High Blood Sugar Ranges: A Bayesian Meta-Analysis

Background: Self-monitoring of blood glucose (SMBG) is essential for diabetes management, but the comparative accuracy of monitoring devices across different glycemic ranges remains poorly understood. Objective: To quantify and compare the accuracy of blood glucose monitoring systems in hypoglycemic versus hyperglycemic ranges using Bayesian meta-analysis. Methods: A systematic review following PRISMA guidelines was conducted across five databases. Studies reporting paired Mean Absolute Relative Difference (MARD) values for both hypoglycemic (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$&lt;70 \text{mg} / \text{dL}$</tex>) and hyperglycemic (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\geq 180 \text{mg} / \text{dL}$</tex>) ranges were included. Bayesian hierarchical modeling was implemented using PyMC to quantify uncertainty and provide probability statements about accuracy differences. Results: The analysis included 26 devices from 4 studies. The posterior mean MARD for the low glucose range was 18.42 % (95 % credible interval: <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$14.92 \%-22.33 \%$</tex>), compared to 9.05 % (95 % credible interval: <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$7.53 \%-10.74 \%$</tex>) for the high glucose range. The mean difference in MARD between ranges was 9.37 % (95 % credible interval: <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$5.54 \%-13.50 \%$</tex>), with posterior probability <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$&gt;$</tex> 99.9 % that devices are less accurate in hypoglycemic ranges. Conclusions: This Bayesian approach to meta-analysis demonstrates that the posterior probability-reflecting updated parameter beliefs-confirms reduced hypoglycemia detection accuracy in glucose monitoring devices. The systematic accuracy disparity underscores an urgent need for technological advancements to enhance low-glucose measurement reliability.