This abstract reports the performance of the CoughCheck™ algorithm, developed to automatically detect cough events from real-world recordings collected with the RESP® Biosensor. Using over 3,000 hours of annotated training data from 249 subjects with chronic and subacute cough, the algorithm was validated on an independent dataset of 36 individuals with subacute cough (845 hours, 14,486 annotated coughs).

Results showed strong agreement with human annotation, with an R² of 0.994 for average cough frequency and a mean absolute error of just 1.16 coughs/hour. Precision and sensitivity for absolute cough counts were 97.4% and 95.7%, respectively, with a low overcounting rate. Temporal detection achieved 91.5% precision across the sensitivity range.

These findings demonstrate that CoughCheck can accurately measure 24-hour cough frequency in ambulatory conditions, offering a scalable alternative to manual annotation and supporting future applications in cough research and clinical development.