OBJECTIVES

To assess agility of answering acute kidney injury (AKI) disease research questions using graph models. The one-size fits-all relational design of clinical data warehouses is not well-suited to deliver timely access to answers of complex AKI clinical questions. AKI is an unappreciated disease process with high morbidity and mortality in up to half of critically ill patients. Research lags in understanding the reasons why about 65% of AKI patients have unfavorable outcomes. This study demonstrates that leveraging the right data model to represent relationships between clinical variables results in time and cost saving.

METHODS

The study included patients who received dialysis and had no history of chronic kidney disease. Data extraction was performed from the University of Arkansas for Medical Sciences clinical data warehouse. The data elements included patients’ complete clinical events and dialysis-specific data points. Neo4j, a graph database, was used to represent patients’ navigation within clinical workflows. Cypher, Neo4J’s query language, was used to answer ten user-defined clinical questions, embodying patients’ various outcomes, from the AKI graph model. Efficiency and effectiveness of producing answers to outcomes questions were compared between relational and graph models.

RESULTS

Answers to patients’ outcomes questions from the relational model required several steps including data extraction and curation, importing data files into statistical software tool to produce meaningful answers. On the other hand, answering the same questions in the graph model, excluding the step of initial step of loading data into the graph model, required building simple Cypher queries to produce similar meaningful answers.

CONCLUSIONS

The study confirms relational models’ hypothesis-driven approach where queries are constrained by existing knowledge and tailored to answer well-defined questions. Conversely, the graph model is data-driven approach that allows for visual data exploration and knowledge generation. Graph models are horizontally scalable and can accelerate clinical and translational research in other clinical domains.