We investigated prediction abilities for a risk of hypoglycemia (R-hypo) in intermediate-term glycemic variability metrics (GVM) using continuous glucose monitoring (CGM) data. We cross-sectionally investigated CGM (FreeStyle Libre Pro) data of 104 patients with type 2 diabetes whose 24 h glucose levels were measured continuously for 13 days during hospitalization. Values for 13 days in all GVM were evaluated. We have proposed GVM as follows: mean of daily difference 1 (MODD1) ÷ mean glucose level (MGL) × 100 (MODD1/M), mean absolute glucose (MAG) ÷ MGL × 100 (MAG/M), and glycemic variability percentage (GVP) ÷ MGL × 100 (GVP/M). The standard deviation (SD), MODD1, MAG, GVP, and MGL were significantly negatively correlated to the low blood glucose index (LBGI) (r=-0.34∼-0.68, p<0.001). The coefficient of variation (CV), MODD1/M, MAG/M and GVP/M were significantly positively correlated to the LBGI (r=+0.22∼+0.53, p=0.03∼<0.001). The prediction ability for LBGI>2.5 in MAG/M was significantly higher than that in CV and GVP/M. The prediction ability for LBGI>5.0 in MAG/M was significantly higher than that in CV, MODD1/M and GVP/M. The prediction ability for LBGI>2.5 was higher than that for LBGI>5.0 in MODD1/M (Table). The GVM which are divided by MGL should be used to predict a R-hypo. Among them, MAG/M has the highest prediction ability for a R-hypo. In addition, MODD1/M reflects a midium R-hypo more strongly than a high R-hypo.
S. Takeishi: None. H. Tsuboi: None.
