Objective: To evaluate how continuous glucose monitoring (CGM)-derived metrics relate to severe hypoglycemia (SH) events in individuals with type 1 diabetes by utilizing a multistep machine-learning approach to generate virtual CGM profiles from glycemic data in the Diabetes Control and Complications Trial (DCCT). Research Design and Methods: Virtual CGM profiles were created for each DCCT participant using previously validated methods. HbA1c values and CGM metrics were analyzed as predictors of SH events within the subsequent 90 days using Poisson regression models. Sensitivity, specificity, and positive predictive value of time-below-range (TBR) 6% demonstrated only 13% positive predictive value for SH events. Conclusion: Hypoglycemia-focused CGM metrics reproduced by virtual CGM data from the DCCT were strongly associated with SH events, although the positive predictive value was low.
Bilal et al. (Mon,) studied this question.