Abstract P158: Hypoglycemia Prediction From The Coefficient Of Variation For Each Hba1c Value

It has also been reported that the coefficient of variation (CV) corresponding to the number of hypoglycemia cases decreases as HbA1c values decrease. Therefore, when an association between the CV and hypoglycemia is considered, the possibility that the CV varies according to HbA1c values should be considered. If hypoglycemia can be predicted from both HbA1c values and the CV, the relationship between glycemic variability and hypoglycemia can be assessed in detail. Thus, we studied a formula that could predict hypoglycemia based on HbA1c values and the CV. Materials and Methods This was a prospective observational study. One hundred and one outpatients with type 2 diabetes mellitus underwent HbA1c testing, wore a flash glucose monitor (FGM: FreeStyle Libre Pro, Abbott Diabetes Care, Alameda, CA, USA), and did not change diabetic treatment at the hospital visit. The CV and mean glucose levels were calculated using the FGM data over 24-h х 13 days. The glucose management indicator (GMI) was calculated using the mean glucose levels, and we compared the GMI to the HbA1c values mainly to detect differences between sensor glucose levels (SG) and blood glucose levels. It has been previously reported that the HbA1c value minus (-) GMI >0.5% is associated with the risk of hypoglycemia, and the GMI - HbA1c value >0.5% is associated with the risk of hyperglycemia. Therefore, we assessed the difference between the GMI and HbA1c value with real numbers. We calculated the “percentage of mean absolute deviation to mean glucose levels” (Metric1) as a new metric. Hypoglycemia (<70 mg/dL) absence could be significantly predicted from the CV, HbA1c value, and “glucose management indicator (GMI) minus HbA1c value” (ΔA1c) (Nagelkerke=0.68, p<0.001). When a response variable was hypoglycemia absence and explanatory variables were the CV, HbA1c value, and ΔA1c, the optimal predicted value for the logistic regression analysis was 0.35 (sensitivity: 90%, specificity: 84%; area under the curve: 0.93, p<0.001) (predicted values=1÷(1+e -(-0.28хCV+2.18хHbA1c+1.85хΔA1c-7.48) ) [e: Napier's constant]), where CVs corresponding to HbA1c values of 6%, 7%, 8%, 9%, and 10% were 22.0%, 29.8%, 37.5%, 45.3%, and 53.0%, respectively. The CV correlated with Metric1 (r=0.99, p<0.001, Metric1=0.82хCV-0.006 [formula1]). The CV should be reduced more to prevent hypoglycemia as HbA1c values decrease. For avoiding hypoglycemia, an “alarm threshold using ‘mean glucose levels (Mean) corresponding to the HbA1c values’, and ‘Metric 1 corresponding to the CV calculated using formula 1’ ” (Mean ± Mean х Metric 1 ÷ 100) should be used for personal continuous glucose monitoring, and all glucose levels should be kept within the alarm threshold.