Measuring the Prevalence of Nocturnal Cough in Asthmatics by Means of Smartphone-enabled Acoustic Recording and Evaluating the Potential of Nocturnal Cough Rate as a Prognostic Marker for Asthma Control: An Observational Two-Stage Study

Asthma, a chronic respiratory disease, belongs to the most prevalent chronic conditions. In Switzerland, 7-15% of all children and 6-7% of all adults suffer from it. Common symptoms are breathlessness, coughing and wheezing. The symptoms often get worse at night and often cause awakenings. Cough is a particularly important symptom in asthma because it predicts asthma severity, indicates a worse prognosis and is perceived to be a troublesome symptom. Additionally, asthma is the leading cause for chronic cough, responsible for 24-29% of cases.

However, little is known about the utility of cough tracking for self-monitoring purposes in asthmatics. A first cross-sectional study has indicated that the cough rate during both day and night might be a valid marker for asthma control, rendering it a potentially useful parameter for self-monitoring. Unfortunately, due to considerable variance of cough rates within each category of asthma control (i.e. uncontrolled, partially controlled and controlled asthma), the statistically significant relationship between cough rate and asthma control might not be clinically meaningful. Additionally, due to the cross-sectional design of existing studies, it remains unclear whether the cough rate might have any prognostic value for predicting future asthma control.

Therefore, the purpose of this study is to explore the value which cough rate might provide for asthma self-management in more detail. In This study, the focus will be put specifically on nocturnal cough rate due to the technical reasons. In general, the plan of this study is as follows: With a longitudinal study design, it is possible to investigate to which extent trends in the nocturnal cough rates might have meaningful implications for future asthma control and asthma exacerbations of patients. However, in order to analyze the predictive value of trends in nocturnal cough rate, the symptom has to persist over multiple nights. There is no research available on the prevalence of nocturnal cough in asthmatics over multiple nights. Therefore, the incidence of nocturnal cough in asthmatics will be described and visualized over the course of one month in the first stage of our study. Additionally, the aim will be to identify and model trends in nocturnal cough rates.

Measuring cough is very time-consuming. Currently, there are no cough frequency monitors available, which measure cough rates in a fully automated and unobtrusive way. Consequently, manual labeling of cough based on video or sound recordings is still considered to be the gold standard for measuring cough rates by medical guidelines. Nevertheless, a machine learning algorithm has been successfully designed to automatically detect cough in a proof of concept study. Despite using only very limited data for algorithm development (80 coughs from 5 healthy subjects), the accuracy reached 83%. However, the data were gathered in a laboratory setting, which limits the generalizability of the results and thus applicability in practice. Therefore, the aim is to develop a machine learning algorithm which is also capable to provide robust results in the field. This study will focus on cough during the night time due to the limited interfering noise, which greatly facilitates manual labeling and enables a more reliable detection rate of the machine learning algorithm. It is important to point out that the analysis of nocturnal cough prevalence described above will not be based on cough detected by an algorithm, but on the manually labeled coughs in the audio track recorded during the night by a study smartphone, which will be provided to subjects for the course of the study.

Apart from developing a machine learning algorithm for cough detection, data will be gathered for an algorithm assessing patient's sleep quality. For this purpose, sleep quality will be predicted based on data obtained from the smartphone's sensors.