Abstract
Designing a database system for both efficient data
management and data services has been one of the enduring
challenges in the healthcare domain. In many healthcare systems,
data services and data management are often viewed as two
orthogonal tasks; data services refer to retrieval and analytic
queries such as search, joins, statistical data extraction, and
simple data mining algorithms, while data management refers
to building error-tolerant and non-redundant database systems.
The gap between service and management has resulted in rigid
database systems and schemas that do not support effective
analytics. We compose a rich graph structure from an abstracted
healthcare RDBMS to illustrate how we can fill this gap in
practice. We show how a healthcare graph can be automatically
constructed from a normalized relational database using the proposed
“3NF Equivalent Graph” (3EG) transformation.We discuss
a set of real world graph queries such as finding self-referrals,
shared providers, and collaborative filtering, and evaluate their
performance over a relational database and its 3EG-transformed
graph. Experimental results show that the graph representation
serves as multiple de-normalized tables, thus reducing complexity
in a database and enhancing data accessibility of users. Based
on this finding, we propose an ensemble framework of databases
for healthcare applications.
