Heterogenous Knowledge Discovery from Medical Data Ontologies

Abstract

A variety of knowledge discovery applications on healthcare big data require effective medical ontologies of diseases that can abstract the healthcare record data in order to support formal reasoning. Domain specific ontologies are often created by teams of clinicians manually, partitioning the conditions present in that domain on pre-defined boundaries. However, it is often hard to determine the underlying patterns of diseases that partitioning such data. We use exploratory and confirmatory factor analyses to describe the variability in the observed patterns or groupings of two such ontologies in terms of a potentially lower number of latent factors. This gives valuable preliminary insights into the multimorbidity of conditions prevalent in these populations which can be used to better inform diagnoses and recommend preventative measures for the same.