Text

theses

The increased adoption of electronic medical records (EMR) systems and emergence of clinical data warehouses to integrate data from diverse data sources energized clinical research and prompted the biomedical informatics community to envision and implement efficient and effective tools to facilitate conduct of research. Data warehousing, a valuable platform to provide clinical data for secondary use, is one tool, traditionally built using relational database models. Though relational models proved solid in data management applications across industries, the complexity and variety of clinical data require an agile technical environment that responds to evolving research data needs. A property graph model’s data connectedness, data exploration, and visualization capabilities make it a solid candidate to represent and manage clinical knowledge. This study uses acute kidney injury (AKI) disease, an important and often overlooked disease process, to represent clinical data extracted from institutional data warehouse in a graph model. The resulting AKI graph model, which consists of entities (nodes) connected through meaningful relationships (edges), provides easy access to explore and view query results in either graphical or tabular format. The AKI model, conceptually a data lake, is horizontally scalable, which can integrate with other graphbased clinical domains of knowledge. Moreover, the AKI graph schema provides the right structure for a Bayesian network, which helps implement a Bayesian inference model to estimate AKI patients’ outcomes probabilities, and also helps envision a Markov Chain transitions model to predict non-AKI patients’ probabilities of requiring dialysis within a 48-hour.

ETD_10434

Ph.D.

Includes bibliographical references

doi:10.7282/t3-ckx1-hy92

ETD doctoral

The author owns the copyright to this work.