Data mining based on HL7 reference information model

Abstract

The Health Level Seven (IlL 7) organization has developed a powerful abstract model of patient care called the Reference Inforn1ation Model (RlM), which is intended to serve as a unified framework for the integration and sharing of information and the usage of data across different healthcare domains. There are a number of exciting research challenges posed by health care data that make them different from data in other industries: data sparseness, high dimensionality, schema change, continuously valued data, complex data modeling features and performance. Entity Attribute Value (EAV) is a widely used solution to handle these above challenges of medical data but EAV is not a search efficient data model for knowledge discovery. The thesis presents two search efficient open schema data models: Optimized Entity Attribute Value (OEAV) and Positional Bitmap Approach (PBA) to handle data sparseness, schema change and high dimensionality of medical data as alternatives of widely used EAV data model. It has been shown in both analytically and experimentally that the proposed open schema data models are dramatically efficient in knowledge discovery operations and occupy less storage space compared to EAV. We have transformed HL7 RIM healthcare data into EAV, OEAV and PBA data models and applied the proposed data mining algorithms. New data mining algorithms have been proposed to discover knowledge from healthcare data stored in the above models. We have evaluated the performance of the proposed algorithms experimentally by using synthetic datasets. The experimental results show-in all the new developed data mining algorithms, OEAV data model outperforms all the others. Next comes PBA which performs better than EAV and in EAV these algorithms are quite slow.