A visual interactive analytic tool for filtering and summarizing large health data sets coded with hierarchical terminologies (VIADS)

Hierarchical terminologies, such as the International Classification of Diseases 9th Revision-Clinical Modification (ICD9-CM) [1], ICD10-CM [2], SNOMED CT [3], Logical Observation Identifiers Names and Codes (LOINC) [4], RxNorm [5], and Gene Ontology (GO) [6], have been used in biomedicine for a long time. Enormous volumes of data, coded through hierarchical terminologies, are generated continuously within electronic health record (EHR) systems, in biomedical literature databases (such as Medical Subject Headings [MeSH] [7] in PubMed), and in other sources of information. The EHR adoption rate in the United States had reached 96% [8] in hospitals and 87% [9] in office-based physician practices by 2015, and these rates are continually increasing. EHRs contain a wealth of clinical data represented in hierarchical structures, and the large and continually generated data sets in EHRs can be used to explore new patterns or to reveal unknown facts about disease and health as well as to optimize care delivery operations. The large size of these health data sets and their complex terminology structure make them difficult for clinicians, clinical researchers, or administrators to understand. Further processing and analysis of these health data sets are needed to make them manageable and comprehensible.

One way to analyze the large health data sets coded by hierarchical terminologies is to use graphical representations to demonstrate the relationships among the terms. Graphs have long been used to represent hierarchical information as a means to facilitate comprehension. Notably, the human brain can comprehend graphs when the data set is not too large, empirically, and contains no more than 120 nodes. A graph with thousands of nodes is challenging for both the presentation and comprehension of the information. The sizes of the terminology range from dozens of thousands (e.g., ICD9-CM) to over a million terms (e.g., SNOMED CT). A visualization of that scale would not be comprehensive to the human brain. Further filtering to reduce the sizes of the data sets is necessary. Neol [10] and Homer [11] each developed methodologies to reduce the complexity of the graphs and to aggregate hierarchical structures to analyze information security networks. Gu [12, 13] developed methodologies to partition large terminologies for further usage. None, however, developed methods specifically to provide a summary or comparison of data sets based on the semantic relationships of the terminologies and analytic results. Our group has developed the core algorithms to aggregate, filter, summarize, and compare large data sets coded by hierarchical terminologies [14] and methods to set thresholds, demonstrated case studies [15], and published the preliminary results of the comparison of two solutions to develop an online tool for the algorithms [16]. To make the algorithms more accessible to broader audiences, we believe that it is necessary to develop an online, publicly accessible tool to implement the algorithms and to facilitate the decision-making process by clinicians, clinical researchers, and healthcare administrators more conveniently.

We used the software, Graphviz [17] during the algorithm development and case study demonstration. Graphviz is a powerful tool that provides many different types of graphs, such as hierarchical (i.e., dot). Graphviz, however, cannot typically be used by users who lack programming skills. Further, the hierarchical graphs created by Graphviz are static and are not efficient in terms of the use of available screen space. We are developing a Web version of a visual interactive analysis tool for filtering and summarizing large data sets coded with hierarchical terminologies (VIADS) to make the algorithms publicly accessible. VIADS provides more interactive visual features to facilitate users’ comprehension of the data sets and can be used for educational and research purposes at no cost.

The rapid adoption of EHR systems in both office-based practices and hospitals has led to an increasing number of available coded data sets. These coded data sets are becoming increasingly common not only on the administrative side (e.g., for billing purposes) but also on the clinical side (e.g., to generate a problem list). The development of a publicly accessible tool will help users to achieve a summary view, secondary analysis, and visualization of their health data sets with minimal technical effort. VIADS shows more efficient screen space usage in a graph display and more interactive features. Being able to view and interpret larger health data sets with ease is a great advantage of VIADS. Furthermore, the interactive features provide a level of convenience that may facilitate users in interpreting the results. VIADS, along with all these features can be used to facilitate clinicians, clinical researchers, and healthcare administrators to make data-driven decisions more conveniently.

Project homepage: https://​www.​viads.​info