Validation and Accuracy of the Automated AI-powered Continuous Hyfe Cough Monitoring System

<bold>Background:</bold> Efficient management and research of coughing can be enhanced by passive and continuous cough monitoring. However, a clinically validated, automated, and accurate cough monitor for routine use in clinical and research settings has not been available until now. <bold>Objective:</bold> To assess the Hyfe CoughMonitor, a smartwatch device, against manual cough counts (gold standard) during daily activities in individuals with various cough etiologies. <bold>Methods:</bold> Subjects wore the CoughMonitor and an audio recorder for 20-24 hours, capturing continuous sound data. Trained annotators identified coughs from these recordings, and cough timestamps were compared with CoughMonitor detections using event-to-event and hourly rate correlations. <bold>Results:</bold> In 546 hours of monitoring across 23 subjects, 4,454 coughs were detected by the trained annotators. Hyfe's CoughMonitor sensitivity was 90.4% (95% CI: 89.5-91.2%), with a false positive rate of 1.03/hour (95% CI: 0.94-1.11). Hourly cough rates showed a high correlation between manual counts and CoughMonitor data (Pearson r = 0.99, OLS slope = 0.94, intercept = 0.68). Cumulative event-based performance is displayed in Figure 1. <fig><object-id>erj;64/suppl_68/PA788/F1</object-id><object-id>F1</object-id><object-id>F1</object-id><graphic></graphic></fig> <bold>Conclusion</bold>: The CoughMonitor's accuracy, ease of use, and scalability suggest it could significantly enhance cough management and research.