Original Article
Cough frequency monitors: can they discriminate patient from environmental coughs?
Abstract
Background: Objective cough frequency measurements are increasingly applied in clinical research. Technological advances enable automated detection and counting of cough events from sound recordings of many hours’ duration. A possible limitation of sound-based cough frequency measurement is the contamination of recordings by environmental coughs (coughs from persons other than the patient). This study aimed to investigate the accuracy of a sound-based cough monitor for detecting and discriminating patient cough from environmental cough.
Methods: As part of a stroke trial (ISRCTN40298220), patients on a hospital ward underwent 15-minute recordings using the Leicester Cough Monitor (LCM), a sound-based cough monitor (‘semi-automated counts’). Participants and other persons in the environment were prompted to cough. An observer present in the room recorded the number of patient and environmental coughs (‘live counts’). LCM counts were also compared against a manual cough count, the most commonly used gold standard to determine accuracy (‘manual sound counts’ from listening to recordings), by a blinded assessor who cross-referenced timed cough events from the respective methods. Data for automated, manual and live cough counts were analyzed using agreement statistics.
Results: On sound recordings from five patients, there were 65 patient coughs and 78 environmental coughs (manual counts). Absolute agreement for patient cough count between all three measurement methods (LCM automated, live, and manual sound counts) was high, with intra-class correlation coefficient of 0.94 [95% confidence intervals (CI): 0.74, 0.99]. The proportion of exact agreements for patient cough between LCM and manual count was 0.92, and kappa was 0.84 (95% CI: 0.75, 0.93). The LCM showed sensitivity of 0.94 (95% CI: 0.84, 0.98), specificity of 0.91 (95% CI: 0.82, 0.96), positive predictive value of 0.90 (95% CI: 0.79, 0.95) and negative predictive value of 0.95 (95% CI: 0.86, 0.98) for detecting patient coughs.
Conclusions: This preliminary study supports the validity of the cough monitor for detecting and discriminating patient from environmental cough. Further validation is recommended, to describe the level of accuracy with greater precision.
Methods: As part of a stroke trial (ISRCTN40298220), patients on a hospital ward underwent 15-minute recordings using the Leicester Cough Monitor (LCM), a sound-based cough monitor (‘semi-automated counts’). Participants and other persons in the environment were prompted to cough. An observer present in the room recorded the number of patient and environmental coughs (‘live counts’). LCM counts were also compared against a manual cough count, the most commonly used gold standard to determine accuracy (‘manual sound counts’ from listening to recordings), by a blinded assessor who cross-referenced timed cough events from the respective methods. Data for automated, manual and live cough counts were analyzed using agreement statistics.
Results: On sound recordings from five patients, there were 65 patient coughs and 78 environmental coughs (manual counts). Absolute agreement for patient cough count between all three measurement methods (LCM automated, live, and manual sound counts) was high, with intra-class correlation coefficient of 0.94 [95% confidence intervals (CI): 0.74, 0.99]. The proportion of exact agreements for patient cough between LCM and manual count was 0.92, and kappa was 0.84 (95% CI: 0.75, 0.93). The LCM showed sensitivity of 0.94 (95% CI: 0.84, 0.98), specificity of 0.91 (95% CI: 0.82, 0.96), positive predictive value of 0.90 (95% CI: 0.79, 0.95) and negative predictive value of 0.95 (95% CI: 0.86, 0.98) for detecting patient coughs.
Conclusions: This preliminary study supports the validity of the cough monitor for detecting and discriminating patient from environmental cough. Further validation is recommended, to describe the level of accuracy with greater precision.
