Identifying Exacerbating Cases in Chronic Diseases Based on the Cluster Analysis of Trajectory Data on Laboratory Examinations

In this paper we present a cluster analysis method for multidimensional time-series medical data and its appli- cation to finding groups of exacerbating cases in chronic hepatitis. Our method represents time series laboratory ex- amination data of a patient as a trajectory. Compaison of trajectories is done using a two-stage approach. Firstly, it compares trajectories based on their structural similar- ity and determines the best correspondence of partial seg- ments. After that, it calculates the sum of value-based dis- similarities for all pairs of the matched segments as the final dissimilarity of the two trajectories, which can be used for clustering. Experimental results on a synthetic digit-stroke data provided low error ratio of 0.016 ±0.014 for classi- fication and 0.118 ±0.057 for cluster rebuild. Results on the chronic hepatitis dataset demonstrated that the method could discover the groups of exacerbating cases based on the similarity of ALB-PLT trajectories.