Challenges in Using a Graph Database to Represent and Analyze Mappings of Cancer Study Data Standards

Area of Science:

• Biomedical Informatics
• Data Standards
• Clinical Data Management

Background:

• Manual mapping to data standards hinders research data sharing.
• Semi-automated strategies and machine learning show promise but require training data.
• Existing methods face challenges in reducing the manual mapping burden effectively.

Purpose of the Study:

• To develop a novel graph database approach for semi-automating the mapping of clinical data elements (CDEs) to data standards.
• To leverage a graph database incorporating the Biomedical Research Integrated Domain Group (BRIDG) model, NCI CDEs, and NCI Thesaurus.
• To predict CDE to BRIDG class mappings using a shortest path algorithm.

Main Methods:

• Developed a graph database integrating the BRIDG model, National Cancer Institute (NCI) Common Data Elements (CDEs), and NCI Thesaurus.
• Employed a shortest path algorithm within the graph database to predict mappings between CDEs and BRIDG model classes.
• Utilized the graph database for semantic analysis and quality assurance testing of data mappings.

Main Results:

• Successfully developed a graph database providing a semantic framework for data analysis and quality assurance.
• Predicted mappings from NCI CDEs to BRIDG model classes using the shortest path algorithm.
• Identified limitations in prediction accuracy due to the subjective nature of mapping and data quality issues.

Conclusions:

• The developed graph database offers a robust semantic framework to aid in data standardization.
• Semi-automated mapping using graph databases and shortest path algorithms can reduce manual effort.
• Further refinement is needed to address subjectivity and data quality challenges for improved mapping accuracy.