Hierarchical and Partitional Cluster Analysis of Glucose and Insulin Data from the Oral Glucose Tolerance Test

Abstract

The body's ability to regulate glucose homeostasis is commonly assessed through the oral glucose tolerance test (OGTT). Several variations of OGTT exists, but the most used in clinical practice is the 2-sample 2-hour OGTT, in which glucose is measured in fasting and two hours after a glucose load. In the 5-sample 2-hour OGTT, glucose is measured in fasting and every 30 minutes after a glucose load, during two hours. In these tests, besides glucose, insulin level can also be measured from the blood samples, increasing thus the number of variables to analyze and perform a better metabolic assessment. In this paper, a cluster analysis is carried using the levels of glucose and insulin from the 2-sample 2-hour OGTT and from the 5-sample 2-hour OGTT, from subjects with metabolic syndrome and professional marathon runners. Different configurations of k-means and agglomerative hierarchical clustering were used to perform the clustering of data and analyze the relationships between clusters with the study groups. Results show that the k-means clustering algorithm performs better than the agglomerative hierarchical clustering, and, with the Manhattan distance measure, k-means perfectly groups subjects using the ten variables from the 5-sample 2-hour OGTT.