Abstract:

The retrospective analysis of self-monitoring blood glucose data collected by insulin-dependent diabetic patients is crucial when revising the basic therapy regimen to improve the quality of blood glucose control. The goal in developing data analysis methods is to find representations of the data that simplify the analysis and yet allow all relevant information present in the data to be preserved. Novel data analysis methods have considerable potential in making the analysis more effective and consecuently, increasing the quality of resulting clinical decisions. In this paper, a novel analysis method of blood glucose data is presented based on the Self-Organizing map algorithm. The SOM is used to find groups or clusters of typical daily blood glucose response profiles. Preliminary evaluation results show that on the patients tested, it was possible to separate the profiles into a limited number of groups with different and clear characteristics. Based on the results, it also seems to be possible to link the factors underlying the observed response profile to the observed groups to identify meaningful correlations between the factors and the typical resulting blood glucose response.