Diagnosis of Type 2 Diabetes and Pre-diabetes Using Machine Learning

Abstract

The type-2 diabetes (T2D) is a multifactorial chronic disease that reduces the quality of lifestyle and produces the death of a large percentage of the population worldwide. Before the development of T2D a series of symptoms are presented even years before T2D diagnosis. This condition that appears before the development of T2D is called prediabetes. Prediabetes and T2D are diagnosed from the oral glucose tolerance test (OGTT). The OGTT consists in the measurement of glucose and insulin in five-time intervals, the first after 8 h of fasting (0 min) and the other four measurements after taking 75 g of oral glucose in 30-minutes intervals (30, 60, 90 and 120 min). Some parameters have been used to improve the efficiency in the diagnosis of prediabetes and T2D, for example: the area under the glucose (AUCG) and insulin (AUCI) curve during OGTT has been used as a parameter for the diagnosis of prediabetes, T2D and obesity. The aim of this study is to assess the k-means clustering algorithm in the classification of subjects with prediabetes and T2D using the AUCG and AUCI. A database of 188 subjects (male = 88 subjects, age = 42.11 ± 14.91 years old) with values of plasma glucose and insulin during OGTT was used. The k-means clustering performed for AUCG presents acceptable results since the silhouette coefficient is above 0.6 in all cases. The findings in this study indicate that the k-means applied in the AUCG classify subjects with T2D, prediabetes and control. Furthermore, it could even predict those subjects with high probabilities of developing T2D.