Comment on Selvin et al. The Glucose Management Indicator: Time to Change Course? Diabetes Care 2024;47:906–914

The insightful Perspective by Selvin (1) presents compelling arguments cautioning against use of estimated HbA_1c (eHbA_1c) derived from continuous glucose monitor average glucose (AG) data. Nonetheless, both clinicians and people with diabetes will seek out such conversion, as our own personal experiences attest. The correlated nature of HbA_1c and AG not only is well-known but also is rooted in biochemistry. Yet there are myriad factors, including biological ones, that cause systematic deviations in HbA_1c and AG from the ideal (2). Crucially, the impact of these factors varies between people. We agree with Selvin (1) regarding the need for a deeper understanding of the variances involved and of the downstream consequences of imprecise interconversion. That said, we want to address a correctable, systematic error that has propagated as a resource for interconversion of AG data to eHbA_1c. The ADAG (A1C-Derived Average Glucose) study in 2008 reported a data-derived linear regression of AG on HbA_1c for the purpose of calculating estimated AG (eAG) based on measured HbA_1c (mHbA_1c) (3). The resulting original regression is where eAG is expressed in milligrams per deciliter and mHbA_1c is expressed in percent. Conversion tables based on Eq. 1 are widely disseminated. For example, the American Diabetes Association currently maintains an online conversion table and calculator based on Eq. 1 (4).

Erroneously, the reversed equation, i.e., the simple algebraic rearrangement of Eq. 1 that necessarily describes the exact same XY line as Eq. 1, is often also presented to estimate the reverse, eHbA_1c based on measured AG (mAG), using

As pointed out by Selvin (1), “least-squares linear regression is not symmetric, that is, regressing A on B will produce a different regression line than regressing B on A.” Hence, Eq. 2 produces suboptimal predictions of eHbA_1c from mAG. Unfortunately, this reversed equation (Eq. 2) (shown as an orange line in Fig. 1 in Selvin [1]) has gained widespread use (4). Importantly, correction of this error is mathematically straightforward (5), as the proper reverse regression, i.e., regression of HbA_1c on AG, can be reasonably estimated from the original regression (i.e., Eq. 1) intercept and slope parameters coupled with the squared correlation coefficient of r² = 0.84 and mean HbA_1c 6.8%, as reported by the ADAG study (3). Using these parameters, the proper regression line for best estimating HbA_1c from mAG based on the ADAG data can be reasonably derived as

Likewise, the proper reverse regression of HbA_1c on AG for the Diabetes Control and Complications Trial (DCCT) formula (blue line in Fig. 1 of Selvin [1], being also an erroneously used reversed equation, derived from DCCT data in this case) can be calculated as from the published parameters of the original AG-on-HbA_1c regression slope = 35.6, intercept = −77.3, r = 0.82, and mean HbA_1c 8.15% (6).