Stepwise Validation Program for Developing the Algorithm for Chronic Cough with Remote e-stethoscope Platform RESP

RATIONALE: There has been a sustained interest in chronic cough, defined as 8 to 12 weeks of daily coughing (adults) or 4 weeks (children). The interest was initially driven by discrimination of cough based on etiology and response to therapy, and the proposed diagnosis of refractory chronic cough when other major etiologies have been excluded. Perhaps not coincidentally, the convergence of cough counting as a valid method of assessing response to therapy (necessary and sufficient) with the clinical development of first generation P2X3 inhibitors has led to the emergence of a number of patient-oriented devices that claim to count coughs. The emergence of “COVID cough” and “post COVID cough” have heightened the interest in cough. METHODS: The Strados Labs RESPTM has been validated to count coughs. Presented are three methods to count coughs in a continuous recording, in order of increased automation. Each validated method has been confirmed to be within +/- 10% of the true cough count for the gold standard recordings. Each method is validated using >6 patients (>48 hours) of continuous recordings. Number of coughs are annotated by trained professionals and are then reviewed by clinical professionals. This is the gold standard cough count. Cough Counting Methods:1.Multiple trained labelers listen to the recordings and agree on if a sound is a cough or not. The number of agreed upon coughs are added together to determine a total cough count. 2.A highly sensitive machine learning algorithm highlights samples that have a probable likelihood of a cough. Multiple trained labelers listen to recordings and agree if a cough or not. 3.Fully automated machine learning algorithm. RESULTS: In comparison with the gold standard, the highly sensitive method was (>98% sensitivity, and the fully automated machine learning algorithm is >90% specific and >90% accurate and reports the number of coughs. CONCLUSIONS: The process shown here is both necessary and sufficient to validate cough data collection using the RESP. However, unlike other approaches, the RESP provides significant quantitative and qualitative data beyond counting coughs. The RESP provides spectrograms and chest wall motion as well as archival recordings so that parsing coughs into single, multiple and spasms can be reclassified as the data set and science expand. Cough architecture can be evaluated to differentiate intrathoracic from extra-thoracic cough, cough onset in inspiration versus expiration, and response to therapy in real time and with increased anatomic certainty.