New Graphical Approach for Visualization of EMR Data with Application to Biomarker Studies

• Activity Number: 569
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Christine Duarte*+ and Ivette Emery and Andrew Prueser and Volkhard Lindner
• Companies: Maine Medical Center and MMCRI and MMCRI and MMCRI
• Keywords: EMR ; biomarker ; network ; visualization ; big data ; data mining

Abstract

The increasing use of large Electronic Medical Record (EMR) systems offers new opportunities for data mining of large clinical data sets. One application is the association of biomarkers with clinical outcomes. Standard methods do not model the complex associations among clinical outcomes. In order to visualize biomarker association with outcome in the context of correlated outcomes from an EMR, we present a novel method for visualization that is implemented in R using the igraph package with custom routines for network visualization. We build a "comorbidity network" with nodes that are ICD-9 diagnoses connected based on the rate of co-occurrence, whose size is given by the strength of the association of that diagnosis with the proposed biomarker. This visualization method allows clusters of highly associated diagnoses to be quickly ascertained among related diagnoses. We demonstrate this new approach using analysis of plasma levels of Cthrc1, and show association with Leukemia, Diabetes, Colitis, and other conditions, and further show that anemia and fever are central nodes. We anticipate that this method will improve interpretability of biomarker associations in EMR data.