Objective versus subjective assessment of cough burden – proof-of-concept of an automated cough recorder

<bold>Introduction:</bold> The burden of cough can be assessed through subjective or objective measures. We present as proof-of-concept a novel AI-based continuous cough recording system. <bold>Methods:</bold> The cough recording system NELA (Carepath Technologies GmbH) consisting of 3 interconnected devices was installed at-home for 3 consecutive days. The data has not been used for training of NELAs algorithms. Postprocessing consisted of removing duplicates and retaining only cough sounds that belong to the respective patient. <bold>Results:</bold> The cohort consist of 241 patients with chronic cough (>8 weeks) (M_age= 59.2, SD 20.5, male= 50.6); COPD (75), Asthma (83), RCC (45), chronic bronchitis (18) and ‘other disease’ (20), confirmed diagnosis in 67%. Performance of NELA for cough detection increased with the additional algorithm to assign cough to the individual patient (sensitivity 92.7 to 93.7%, specificity 95.8 to 96.7%). Average cough rate was 5.9 coughs/h. Cough rates differed significantly between disease types (M_asthma=3.8 c/h, M_COPD= 7.9 c/h, M_RCC=6.1 c/h), Tukey's HSD post-hoc test showed significantly higher mean hourly cough rates in COPD patients compared to Asthma (p<.001, 95%CI [1.62, 6.69]). Self-reported cough frequency, reported as ‘often’ [n=168, range 7-686 total coughs] or ‘very often’ [n=68, 18-3905 coughs], did not result in significant different means for total cough event recorded (|t| (≈ 79.94) = -1.53, p=.13). <bold>Conclusion:</bold> This cohort illustrates the discrepancy between subjective and objective assessment of burden of cough and serves as a proof-of-concept of NELAs continuous, at-home, individual cough detection with high performance, across different respiratory diseases.