Data Visualization of Multiparameter Information in Acute Lymphoblastic Leukemia Expands the Ability To Explore Prognostic Factors.

In clinical studies and for patient care, data collections have become more and more complex. Clinical studies often include millions of data points, and even data collections about individual patients can include several thousand data points. To enable searches for meaningful relationships and patterns, and to gain understanding and knowledge of the data, state-of-the-art visualization approaches have to be adapted to the needs of clinicians and clinical researchers, in order to best reveal relevant patterns in the data. However, despite the progress that has been made in the field of information visualization, none of the currently available high dimensional visualization tools are used in clinical research or practice. The goal of this research was to develop visualization tools that allow clinical researchers to explore multidimensional datasets, as well as temporal clinical datasets. The dataset used for this presentation involves 300 pediatric patients with acute lymphoblastic leukemia diagnosed at Indiana University between 1992 and 2000. Clinical and laboratory data were extracted electronically from the Regenstrief Medical Records System. The dosages of all medications during the treatment period were extracted from the patients’ charts. This information was supplemented, for a subset of 78 patients for whom Indiana Medicaid claims data were available, with actual fill dates and quantities dispensed. Cytogenetic data were extracted from the clinical genetic database, and immunophenotype data were extracted from the pathology database at Indiana University. Temporal patient data, such as laboratory data, prescription fill dates, and medication dosage, are visualized through custom-designed multiple layer graphics. The visualization tools developed allow the user to interactively visualize and query the data. For exploratory analysis, the application offers an overview of the data through visual representations such as parallel coordinates and matrix methods. The user can interact with the data set in diverse ways, e.g, the order in which the variables are visualized can be changed; interactivity augments the insight that can be gained from visually exploring such data. The visualizations are dynamically linked, so that the user can obtain coordinate views of the data. Dynamic querying interactively filters data in all views. In addition, the user can highlight or select a subset of data elements in one view and thereby highlight data for the same subset in other views. For example, we show that patient data with a specific pattern in the parallel coordinate view can be selected, and then clinical, laboratory, and prescription data for the entire treatment period can be viewed through multiple layer graphics. In summary, the adaptation of temporal and multidimensional visualization tools to clinical data allows clinicians or clinical researchers to better explore these datasets. These tools improve understanding of the complex prognostic features of acute lymphoblastic leukemia, including type of leukemia, initial risk factors, therapy, adherence to therapy, and host factors that affect tolerance of therapy.