High blood glucose index stratifies glycaemic severity and identifies occult type 2 diabetes risk in prediabetes via single‐day continuous glucose monitoring: A multi‐cohort study
Weiming Wu, Shihan Wang, Junxiang Gao, Shuoning Song, Yating Zhang, Yong Fu, Mengmeng Zhang, Tao Yuan, Weigang ZhaoAbstract
Aims
To evaluate the platform‐independent robustness of the continuous glucose monitoring (CGM)‐derived High Blood Glucose Index (HBGI) and the predictive utility of a single day for progression to type 2 diabetes.
Methods
We analysed 365 valid CGM records from three devices across four independent cohorts. Linear mixed‐effects modelling partitioned HBGI variance by device, dataset and metabolic status. Cross‐sectional analyses, utilizing full monitoring periods to ensure biological robustness, evaluated HBGI gradients across glycaemic stages. Longitudinal analyses, utilizing single‐day recordings, evaluated its predictive accuracy for progression to overt diabetes.
Results
Device type accounted for only 3.4% of total HBGI variance, compared with 31.8% for metabolic status. Cross‐sectionally, HBGI demonstrated a robust stepwise increase from healthy through prediabetes to diabetes ( p < 0.001). Longitudinally, baseline HBGI significantly differentiated metabolic progressors from non‐progressors ( p = 0.005) and showed moderate predictive accuracy (area under the curve = 0.74; 95% confidence interval: 0.61–0.85). Discriminative performance was comparable to mean glucose and superior to the M‐value, time in range (70–180 mg/dL) and coefficient of variation. Notably, among prediabetic individuals with established glycaemic indices within recommended ranges, elevated HBGI (>0.174) identified an occult high‐risk subgroup with a 5.39‐ to 6.13‐fold higher progression risk.
Conclusions
HBGI is a device‐agnostic metric that effectively stratifies glycaemic severity. A single‐day HBGI assessment provides complementary risk stratification by unmasking occult risk in prediabetes with established CGM indices within recommended ranges. However, given the limited progression events, its predictive threshold requires validation in larger, independent cohorts.