Leveraging past GMI vs A1c discordance markedly improves A1c prediction in a real-world cohort

ObjectiveThe glucose management indicator (GMI) is a widely used surrogate for A1c, but clinically significant GMI-A1c discordance is common. Here, we aim to develop a new GMI leveraging past discordance to improve accuracy in A1c prediction. Research Design and MethodsThis retrospective cohort study included 4,891 A1c records from 2,555 patients seen at BCDiabetes between 2021-Feb and 2025-Sep. Regression models adjusted for past discordance were trained on data from patients enrolled before 2025-Jan-01 and validated on those enrolled afterwards. Clinically significant discordance was defined as an absolute difference between GMI and A1c of [&ge;]0.5%. ResultsIn the validation cohort with 90-day CGM data, 39/256 (15.2%) patients with past discordance available showed clinically significant discordance between BCDiabetes GMI and A1c, compared to 101 (39.4%) patients with discordant Bergenstal GMI (relative risk [RR] 0.39, 95% CI 0.3--0.5, P < 0.0001). For BCDiabetes GMI, mean absolute discordance and Pearson correlation were 0.29% and 0.93, respectively, compared to 0.45% and 0.8 for the Bergenstal GMI. When including all validation patients regardless of availability of past discordance, BCDiabetes GMI showed discordance of 27.9% compared to 37% from Bergenstal GMI (RR 0.76, 0.69--0.83). Using 14-and-28-day CGM data, BCDiabetes GMI again showed reduced discordance compared to Bergenstal GMI with past discordance available (28-day RR: 0.54, 0.44--0.67; 14-day: 0.67, 0.56--0.80). ConclusionsBCDiabetes GMI substantially reduces clinically significant discordance, especially when past discordance is available. Graphical abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=114 SRC="FIGDIR/small/25342697v1_ufig1.gif" ALT="Figure 1"> View larger version (32K): org.highwire.dtl.DTLVardef@759265org.highwire.dtl.DTLVardef@de3796org.highwire.dtl.DTLVardef@1136abcorg.highwire.dtl.DTLVardef@1c472bc_HPS_FORMAT_FIGEXP M_FIG C_FIG Article HighlightsO_LIWhy did we undertake this study? GMI is a widely used surrogate for A1c, but clinically significant GMI-A1c discordance is common. C_LIO_LIWhat is the specific question we wanted to answer? Does accounting for past GMI-A1c discordance improve A1c prediction? C_LIO_LIWhat did we find? New GMI equations accounting for past GMI-A1c discordance markedly improved A1c prediction. C_LIO_LIWhat are the implications of our findings? Discordance-adjusted GMI may improve CGM-based decision-making. C_LI Twitter summaryThis work develops new GMI equations adjusted for past GMI vs A1c discordance, reducing the risk of clinically significant discordance (>0.5%) by over 60% when compared to the traditional GMI."