Time spent in different glucose ranges indicate the occurrence of adverse events and measure the quality of glucose control in type one diabetes (T1D) patients. This work proposes a Compositional Data (CoDa) approach applied to glucose profiles obtained from six T1D patients using continuous glucose monitor (CGM). Glucose profiles limited to 6-h duration were analyzed at four different times of the day These glucose profiles were distributed into time spent in five glucose ranges, which determine the composition. The log-ratio coordinates of the compositions were categorized through a clustering algorithm, which later made possible the obtainment of a linear model that should be used to predict the category of a 6-h period in different times of day. Leave-one-out cross-validation was performed, achieving an average above 90% of correct classification. A probabilistic model of transition between the category of the past 6-h of glucose to the category of the future 6-h period was obtained. Results show that the CoDa approach not only works as new analysis tool and is suitable for the categorization of glucose profiles, but also is a complementary tool for the prediction of different categories of glucose control. This prediction could assist patients to take correction measures in advance to adverse situations.
