Subspace Discovery for Disease Management: A Case Study in Metabolic Syndrome

Introduction

In various healthcare applications a large amount of data is gathered for individuals. However a few key attributes may be more critical than others for the disease management. In addition different attributes may be critical for different demographics based on age, race or gender. Thus, the focus of this paper is to identify key subspaces in large healthcare datasets for better disease management for varying demographic subsets in the data. Essentially we focus on using data mining methods to discover which key factors of a disease are more relevant for particular strata of the population using bin wise clustering. We focus on a case study in Metabolic Syndrome (MetS). MetS can be described as a combination of abnormalities that occur in the body during the processing of food and nutrients (Wright, 2005). A number of definitions have been studied to classify MetS; however, there is no clear criterion that can generally fit into a single satisfactory protocol. This is primarily because this domain encompasses quite a variety of demographics in society leading to an implication that different criteria may be appropriate for different demographic strata. Our research addresses this issue and identifies the cross section of demographic strata and the disease characteristics which are critical for understanding the disease in that subset of the population. We begin by first outlining the motivation of the case study by discussing the challenges in studying Metabolic Syndrome in general and then outlining the data mining challenges.