ESPE Abstracts

Actuality: Standard methods for determining the compensation of the disease don't always reliably reflect the level of the glycemic control of the patient, which leads to decompensation diabetes and reduce the quality and duration of life for patients. Evaluation of glycemic variability indices allows the physicians to predict the risk of developing life-threatening conditions and compensate the diabetes

Aim: To conduct a comparative analysis of glycated hemoglobin (HbA1c) and glycemic variability indexes to predict the degree of compensation for the diabetes mellitus type 1

Materials and Methods: The study included 80 patients with diabetes mellitus type 1 with insulin pump therapy with the possibility of Continuous Glucose Monitoring System (CGMS). All patients done analysis of HbA1c and transmitted data to the doctor for recommendations. As independent parameters for predicting the HbA1c, we chose the glycemic indexes calculated by using the EasyGV: standard deviation, long-term glycemic index, lability index, hypoglycemia and hyperglycemia risk index, M-value. The regression neural network model was built in the environment of statistical computing type R using the software package Neural Network Wizard 1.7 (BaseGroup Labs, Russia). Statistical analysis was performed using SPSS 23.0 (IBM SPSS Statistics, USA). Descriptive statistics for abnormally distributed quantitative parameters are represented by the median and 25; 75 percentiles Me [Q1; Q3]. Statistical significance of differences was assessed by the Mann-Whitney U-test. Differences were considered significant at P <0.05.

Results: There was a significant improvement glycemic variability indexes by the end of the study. The optimal model was based on a multilayer perceptron with three hidden layers and the number of neurons in each layer. The constructed model showed a very high value of the coefficient of determination R2 = 0.987, which indicates a high confidence in predicting the level of HbA1c. When creating a traditional model based on multiple regression, the coefficient of determination was R2 = 0.254, which indicates a low prediction accuracy of the HbA1c level and a higher residual error.

Conclusions: The neural network model with a high index of determination based on glycemic variability indexes demonstrates a significantly higher accuracy in predicting the level of HbA1c in diabetes patients, which makes it possible to assess the degree of compensation for the disease and provide a personalized approach in treating these patients