Analysis of Respiratory Time Series Data for Breathing Discomfort Detection Prior to Sleep Onset During APAP Therapy

Abstract

Discomfort during treatment continues to be a major barrier to adherence to positive airway pressure (PAP) therapy. Thus, a key pillar of ResMed’s business strategy is to deliver intelligent tools that assist healthcare providers in identifying which patients may be struggling with therapy, and why, to enable more effective interventions and personalized patient education. One potential cause of discomfort is perceived stuffiness from pressure levels that is lower than tolerable for some patient preferences. This thesis seeks to explore which patterns in the high-resolution breathing data from ResMed devices may be used to identify patients who are experiencing breathing discomfort at low pressures at the beginning of their therapy sessions. Specifically, time-series clustering is performed on sequential respiratory data to identify groups of patients with similar breathing patterns. The independence between clusters and variables pertaining to patients’ demographic characteristics, therapy settings, usage habits, respiratory characteristics, and self-reported comfort levels are evaluated via statistical testing. Based on the results, features in breathing data are identified that may be meaningful indicators for whether a patient is experiencing discomfort or breathlessness. Additionally, opportunities for additional data collection that would enable further analysis and more accurate modelling are discussed.