Medical Concept Representation and Data Mining

Abstract

Medical data stored in clinical files and databases, such as patient histories and medical records, as well as research data collected for various clinical studies, are invaluable sources of medical knowledge. The computer-based data-mining techniques provide a tremendous opportunity for discovering patterns, relationships, trends, typical cases, and irregularities in these large volumes of data. The patterns discovered from data can be used to stimulate further research, as well as to create practical guidelines for diagnosis, prognosis, and treatment. Thus, a successful data-mining process may result in a significant improvement in the quality and efficiency of both medical research and health care services. Many studies have already demonstrated the practical values of data-mining techniques in various fields. However, in contrast with more traditional areas of data mining, such as mining of financial data or mining of purchasing records, medical data-mining presents greater challenges. These challenges arise not only from the complexity of the medical data, but more fundamentally from the difficulty of linking the medical data to medical concepts or rather medical concepts to medical data. Thus, although computerized medical equipment allows us to store increasingly large volumes of data, the problem lies in defining the meaning of the data and even more so in defining the medical concepts themselves. This paper will address issues specific to medical data and medical data mining in the context of Dr. Kazem Sadegh-Zadeh’s discussion of the typology of medical concepts. In his Handbook of Analytic Philosophy of Medicine, Dr. Sadegh-Zadeh outlines four main classes of medical concepts: individual, qualitative (classificatory), comparative, and quantitative. Moreover, he introduces a novel distinction between classical and non-classical concepts. We will explain how his typology can be utilized for conceptual modeling of medical data. Specifically we will illustrate how this typology can pertain to data used in the diagnosis and treatment of sleep disorders.