A streaming data visualization framework for supporting decision-making in the Intensive Care Unit

Abstract

The number of reporting activities in real time has increased over the last years. This situation has pushed the need for providing real time analysis and visualizations to support decision-making. We propose a visualization framework for exploratory data analysis of multivariate data streams that relies on dimensionality reduction and machine learning techniques for plotting the data in two dimensions. Users can demarcate regions of interest for their study, and use them to make predictions or to decide when to train a new model. The knowledge gained from these visualizations allows users to: (i) characterize the data stream scenario; (ii) track the evolution of a case of interest; and (iii) configure and raise alarms according to the user-defined regions. We illustrate the effectiveness of our proposal through a case study analyzing real-world streaming data to identify patients with multi-drug resistant bacteria when they are in a hospital intensive care unit. Our visualization framework enables the patient follow-up which can allow clinicians to support decisions about the health status evolution of a particular patient. This could provide information for deciding on a particular treatment or whether to isolate patients with a high risk of having multi-drug resistant bacteria since their presence boosts infections in intensive care units.