Pre-Processing Highly Sparse and Frequently Evolving Standardized Electronic Health Records for Mining

Abstract

EHRs aid in maintaining longitudinal (lifelong) health records constituting a multitude of representations in order to make health related information accessible. However, storing EHRs data is non-trivial due to the issues of semantic interoperability, sparseness, and frequent evolution. Standard-based EHRs are recommended to attain semantic interoperability. However, standard-based EHRs possess challenges (in terms of sparseness and frequent evolution) that need to be handled through a suitable data model. The traditional RDBMS is not well-suited for standardized EHRs (due to sparseness and frequent evolution). Thus, modifications to the existing relational model is required. One such widely adopted data model for EHRs is entity attribute value (EAV) model. However, EAV representation is not compatible with mining tools available in the market. To style the representation of EAV, as per the requirement of mining tools, pivoting is required. The chapter explains the architecture to organize EAV for the purpose of preparing the dataset for use by existing mining tools.