Poster Presentation
European Respiratory Society Congress 2025
Automated Cough Identification in Chronic Cough from Real-World Recordings by a Custom Machine Learning Algorithm
Tom deLaubenfels, Richard Powers, Jason Kroh, Adrian Marinovich, MD
What’s Inside?
This abstract presents performance results of the CoughCheck™ algorithm, designed to automatically identify cough events from continuous, real-world respiratory recordings captured by the RESP® Biosensor.
The algorithm was trained on over 3,000 hours of annotated recordings from 249 subjects and tested on an independent dataset of 938 hours from 40 individuals with chronic cough. CoughCheck demonstrated strong agreement with human annotation (R² = 0.950) and minimal bias (0.032 coughs/hour). The algorithm achieved an average precision of 94.3% across the full sensitivity range, and 85.2% precision at 90% sensitivity, confirming high accuracy in real-world ambulatory conditions.
These findings establish CoughCheck as a reliable, automated approach for objective cough monitoring, enabling faster, lower-cost, and large-scale analysis of cough frequency in clinical research and digital health applications.
