Figure 2.

Errors associated with estimates of mean glycemia derived from CGM, HbA_1c, and a combination of both. A: Mean absolute error (wider line) and the 95th percentile (fine line) of the distribution of absolute errors for eAG_CGM calculated over different periods of time, with all ending at the same point, as compared with eAG₉₀. Distinct monitoring periods of the same duration within the same 90-day period will have different calculated eAG_CGM because of factors including healthy physiological glycemic fluctuations and sensor bias. B: Relative average glucose (AG) variation (%) between different 14-day monitoring periods. C: Comparison of theoretical constant M_RBC regression lines with the ADAG study empirical regression line. D: Variation in eAG_A1c for each fixed HbA_1c as a result of nonglycemic factors like M_RBC, with both the point estimate for eAG_A1c (black dot) and its error distribution. The black vertical bars show the theoretical 95% CIs for eAG₉₀ associated with each HbA_1c value, as a result of modeled variation in M_RBC. The theoretical error bars were calculated under the assumption that all nonglycemic variation in HbA_1c is caused by changes in M_RBC, with a population variation of 10–12% in M_RBC (∼5 days), yielding the following CIs: For HbA_1c: $1.96\times \sqrt{5^2\cdot {((1-{\text{HbA}}_{1c0})\cdot k_g\cdot \mathrm{A}{\mathrm{G}}_{90}\cdot 100)}^2}$ and for eAG_A1c: $\text{28}.\text{8}\times 1.96\times \sqrt{5^2\cdot {(\left(1-{\text{HbA}}_{1\mathrm{c}0}\right)\cdot k_g\cdot \mathrm{A}{\mathrm{G}}_{90}\cdot 100)}^2}$. The shaded areas show empirical CIs for eAG_A1c obtained in the ADAG study, ${\mathrm{CI}}_{\mathit{ADAG}}=1.96*\sqrt{4.81\cdot {{\text{HbA}}_{\mathit{1}\mathit{c}}}^{\mathit{2}\mathit{.}\mathit{03}}}$ (14), and in the study by Beck et al. (16). The similarity between the theoretical and empirical CIs supports the use of modeled M_RBC as a proxy for all nonglycemic effects on HbA_1c. See Supplementary Methods for details. E: The error expected when eAG_A1c is averaged with eAG_CGM. The mean absolute error of the average (left panel, black line) is lower than that for either eAG_A1c or eAG_CGM alone if <26 days of CGM data are available, and extreme errors are less frequent (right panel). AE, absolute error; MAE, mean absolute error.