Visualization of Electronic Health Record Sequences at Scale

Abstract: We present ParcoursVis, a Progressive Visual Analytics tool designed to explore electronic health record sequences of patients at scale. Existing tools process and aggregate the whole dataset upfront before showing the visualization, taking a time proportional to the data size. Therefore, to remain interactive, existing tools are limited to data sizes that can be processed in under a few seconds to meet the latency constraints of human attention. To overcome this limitation and scale to larger sizes, ParcoursVis relies on a progressive algorithm that quickly shows an approximate initial result of the aggregation, visualized as an Icicle tree, and improves it iteratively, updating the visualization until the whole computation is done. With its architecture, ParcoursVis remains interactive while visualizing the sequences of tens of millions of patients, each described with thousands of events; three to five orders of magnitude more than similar systems. Managing large datasets allows for exploring rare medical conditions or unexpected patient pathways, contributing to improving treatments. We describe the algorithms we use and our evaluation concerning their scalability, convergence, and stability. We also report on a set of guidelines to support visualization designers in developing scalable progressive systems. ParcoursVis already allows practitioners to perform analyses on two large real medical datasets. Our prototype is open-source.