Abstract

Patient similarity is an emerging field of study facilitating health care analytic of big data pertaining to patients. Its major goal is to rank or cluster patients so that each cluster exhibits one aspect of tightly related patient characteristic. These characteristics include diseases, drugs take, risk factors, life styles, habits and ethical aspects. The rapid adaption of Electronic Health Record (EHR) in hospitals and other governmental heath care organizations to store such a variety of patient information provides a comprehensive source for efficient health care delivery, for data-based analytic of patient-centric individualized perspective prediction, and decision making. It is in this context that the thesis is making contributions on structuring an EHR as a vector of multifaceted components, where each component may be an aggregation of sub-components and each sub-component has a strict type. The operations on each sub-component are part of the typing scheme, and permits semantic-based similarity assessment on each component. The suggested EHR structure is both generic and extendable. The scoring functions that measure the similarity between pairs of components are rigorously defined with respect to domain semantics and user semantics. The weighted average of the scores of components, where the weights are part of user semantics, calculates the similarity between records under analysis. Several examples are shown to comprehensively explain the behaviour of functions. Drug-Drug similarity, and patient-patient similarity analysis based on it are discussed. Experimental results are given and their merits are explained.