Estimation of HbA1c for DMT2 risk prediction on the Mexican population based in Artificial Neural Networks

Abstract

In this paper, the main objective is to estimate the percentage of glycosylated hemoglobin through an easily accessible computational platform to estimate the risk of generating type 2 diabetes mellitus in the Mexican population. The estimation of the computational tool is developed through an artificial neural network model, which was trained and validated according to a population sample of 1120 Mexican people between 18 and 59 years old. The model inputs were gender, age, body mass index, waist circumference, weekly food consumption, family history, and whether the person suffers from any chronic degenerative disease other than T2DM. We used the percentage of glycosylated hemoglobin as output, estimated according to a dynamic glucose model. The estimation results present a coefficient of determination of 99 %, demonstrating an acceptable performance of the neural network model. The developed platform is an aid tool for health personnel, which seeks to generate a first approximation to the glycemic status of those communities with a high marginalization index for generating disease prevention strategies.