Methods of cluster analysis for detection of homogeneous groups of healthcare time series

Statistical analysis is widely used for problem solving in different fields. We present a research on Saint Petersburg morbidity rate. The aim of the work is to detect the heterogeneity in districts of the city with respect to morbidity rate, which was chosen as an indicator of population health. Methods of cluster analysis was utilized for grouping districts to homogeneous sets. Clustering can be considered as an optimization problem as the distance between elements from the same group must be as little as possible, at the same time the distance between elements from different clusters must be as great as possible. Key feature of the research is that data are time dependent so it is necessary to use special dissimilarity measures. Besides each district is characterized by three values: children, teenagers and adult morbidity that call for multidimensional time series analysis. Firstly, a multidimensional clustering analysis was made. Then we conduct the analysis of children morbidity rate and propose a new dissimilarity measure for short time series.