BrainFlux: An Integrated Data Warehousing Infrastructure for Dynamic Health Data

Abstract

Intensively sampled longitudinal data are ubiquitous in modern medicine and pose unique challenges. Future progress in medicine will exploit the fact that multiple physiological measurements can be recorded electronically at arbitrarily high temporal resolution, but novel information systems are needed to create knowledge from these data. In this paper, we present BrainFlux - a novel data warehousing technology that implements a holistic paradigm for storage, summarization and discovery of trends in dynamic data from continuously measured physiological processes. We focus on one specific example of high-resolution longitudinal data: Electroencephalographic (EEG) recordings obtained in a large cohort of comatose survivors of cardiac arrest. Post-arrest EEG conveys important clinical and prognostic information to clinicians, but the rigor of prior analyses has been limited by a lack of proper data processing infrastructure. Our EEG data are complemented by a complete set of contemporaneously recorded electronic medical record data, clinical characteristics, and patient outcomes. In this paper, we describe the architecture and performance characteristics of BrainFlux’s scalable data warehousing infrastructure that efficiently stores these data and optimizes information-preserving summarization.