Interactive Cohort Analysis and Hypothesis Discovery by Exploring Temporal Patterns in Population-Level Health Records

Abstract

It is challenging to visualize temporal patterns in electronic health records (EHRs) due to the high volume and high dimensionality of EHRs. In this paper, we conduct a formative study with three clinical researchers to understand their needs of exploring temporal patterns in EHRs. Based on those insights, we develop a new visualization interface that renders medical event trajectories in a holistic timeline view and guides users towards interesting patterns using an information scent based method. We demonstrate how a clinical researcher can use our tool to discover interesting sub-cohorts with unique disease progression and treatment trajectories in a case study.