Differentiating Cough Origins in Non-Contact AI Monitoring: Analyzing 'Non-User Coughs' with the Hyfe CoughMonitor Smartwatch

<bold>Background:</bold> The Hyfe CoughMonitor, AI-powered non-contact automated cough monitor, demonstrates optimal performance when positioned on the wrist and within 3 ft (~100cm) of the person coughing. Its reported accuracy includes a 90% sensitivity and an average of 1 FP/h in real-world settings. <bold>Objective:</bold> To assess the frequency of "false positives"- non-user coughs and unrelated sounds - under real-world conditions. <bold>Methods:</bold> Non-coughing healthy volunteers were asked to wear a Hyfe CoughMonitor during their daily routines. Participants were instructed to either completely avoid coughing or to record any instances of their own coughing across various sound environments. Properly detected coughs emanating from the user were removed from analysis, and the remaining detections: non-user coughs and non-cough false positives, were quantified. <bold>Results:</bold> 7 non-coughing participants submitted 47 monitoring periods of various lengths (5-30h analyzed) from diverse non-nighttime environments - Figure 1. During 105h of monitoring, CoughMonitor detected 174 false positives (1.67/h): 113 were non-user coughs (1.08/h), and 61 non-cough sounds (0.61/h). False positive rates varied significantly by environment, from ~0 in quiet to ~5/h in noisy areas. On average, non-user coughs were detected at about 1/h, with potential reductions in quieter conditions not analyzed here: sleeping and time spent alone. <fig><object-id>erj;64/suppl_68/PA789/F1</object-id><object-id>F1</object-id><object-id>F1</object-id><graphic></graphic></fig>