An aide diagnosis system based on k-means for insulin resistance assessment in eldery people from the Ecuadorian highlands

Abstract

The lack of standardized cut-off values for the surrogate methods to diagnose Insulin resistance (IR) and the fact that the sensitivity of these methods have been studied in specific populations have limited their use in clinical routine. We developed a system that could aide to diagnosis IR in elderly people, analyzing four surrogate methods of IR estimation using a k-means clustering algorithm. Study subjects included 119 nondiabetic participants over 65 year old from Ecuadorian highlands. Blood tests included a two-point oral glucose test tolerance. The k-means clustering algorithm, was applied in one-dimensional experiments for the Homa-IR, Quicki, Avignon and Matsuda. The population was divided into three clusters: C N with normal values, C IR with IR and C a with values in between. The number of individuals classified in each C Ir was very different according to each method. With the cut-off values obtained, for each method, the system for the evaluation of IR in elderly people was developed. Our work is intended to aid physicians in the early detection of IR by using information from diverse methods.