Clustering Socio-Demographic and Medical Attribute Data in Cohort Studies

Longitudinal epidemiological studies like the Study of Health in Pomerania (SHIP) analyze a group of thousands of subjects (a cohort) by imposing a multitude of socio-demographic and biological factors. Epidemiological findings rest upon hypotheses which yield a selection of disease-specific cohort study parameters. They are then analyzed for significant interactions to identify risk factors. We propose an alternative approach by incorporating clustering algorithms with a Visual Analytics system to form subject groups which are the basis for an exploratory analysis of the underlying parameter interactions. We investigated three clustering techniques (k-Prototypes, DBSCAN and hierarchical clustering) for their suitability in these data sets. With our system, groups can be automatically determined to provide insights into this complex data.