Heterogeneity of continuous glucose monitoring features and their clinical associations in a type 2 diabetes population

Abstract

Objective: Data from continuous glucose monitors (CGM) enable the extraction of fea-tures descriptive of glycemic dynamics that may provide insight into underlying healthstatus. In this work, we analyse CGM data from a large population of individuals withtype 2 diabetes (T2D) and study the association of features with clinical covariates.Methods: We retrospectively analysed CGM and electronic health record data froma large population of individuals with T2D. We extracted 25 daily CGM features foreach individual over a 30-day period and performed statistical association tests onthe features and clinical findings from medical claims data and laboratory records.Results: Our final analysis was performed on 6533 individuals. When clustering theCGM features across the population of individuals with T2D, four distinct clusters offeatures emerged. Further, the CGM features had heterogeneous discriminatorypower with clinical covariates, including laboratory values and the presence of claimsfor diabetic complications. Features related to glycemic variability, such as coefficientof variation, showed markedly lower p-values in many association tests for the pres-ence of diabetic complications than mean glucose.Conclusions: In examining the characteristics of different features extracted fromCGM data in a large population of individuals with T2D, we found that the featureswere heterogeneously associated with different clinical comorbidities related to dia-betes. This work motivates further research to investigate the relationship betweenCGM features and health outcomes in T2D to enable precision medicine.