Quantifying Patterns and Predictors of Respiratory Distress Using Time Series Methods

• Activity Number: 192 - Study of Health Outcomes Using Large Cohort Data
• Type: Contributed
• Date/Time: Tuesday, August 4, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Biometrics Section
• Author(s): Camille Hochheimer* and Sarah Ratcliffe and Ronald Williams and Shrirang Gadrey and William Ashe
• Companies: The University of Virginia and The University of Virginia and The University of Virginia and The University of Virginia Health System and The University of Virginia
• Keywords: signal processing; streaming data; time series; respiratory monitoring

Abstract

By the time a patient’s carbon dioxide levels are high enough to alarm medical providers they are already experiencing acute respiratory failure. An early warning system that predicts approaching respiratory distress could improve clinical outcomes. This research serves as the first step towards developing an early warning system to detect the physiological effects of imminent respiratory distress. In this paper, we show how raw respiratory time series, or streaming, data from a proprietary multi-sensor system can be transformed into useful metrics for predicting distress. A banded filter was applied to remove background noise and the most appropriate epoch of time from which to make predictions was determined based on clinically interpretable physiological effects. From the processed data, we quantified the magnitude, phase, cross-correlation and other potential predictors of respiratory stage. Classification trees were then used to quantify how well these metrics predict different stages of respiratory distress across patients. Finally, we discuss the need for additional work using functional data analysis techniques.