Coughing vs Clearing Throat: An Interview With Hyfe Team Member, Dr. Carlos Chaccour

For researchers, the ability to define and identify cough sounds in respiratory diseases should be diagnostically valuable. To properly leverage the diagnostic information conveyed by cough sounds, it is fundamental to develop strategies for measuring their characteristics.

Coughing and throat clearing are examples of cough-like sounds that are familiar to many health professionals and researchers.

To explore more on the research done on the differences between coughing and clearing throat, this week we had a virtual interview with Dr. Carlos Chaccour, M.D., who is an internist and assistant research professor at the ISGlobal Barcelona Institute for Global Health and the Universidad de Navarra.

He supports Hyfe’s medical team. His tasks include study design, linking labeling efforts with clinical usage, and contributing to the usage of cough in different clinical settings.

Marion Sereti (MS): Thank you for accepting this interview. Please brief us about your present roles and their scope.

Carlos Chaccour (CC): I am a researcher and my focus is global health. My largest projects revolve around malaria, but in early 2020, at the start of the pandemic, I decided to diversify my portfolio to include COVID-19 and started to help Hyfe with the mining of the valuable data available through cough acoustics. I have gained early clinical and research experience with Hyfe as a tool conducted a large cohort study in Navarra and have ongoing data collection in one other.

Part of my work with Hyfe has been the labeling of coughs captured through the app. The only way to obtain a gold standard of reference is to have trained humans to listen to the captured sounds (and sometimes “see” them using special software) and determine whether they fit within the definition of cough or not. For this, we have developed an internal standard operating procedure (SOP), which is nothing but a highly elaborate instruction on how to do this.

MS: How can coughing and throat clearing be differentiated? What is a cough, and what is a throat clear?

CC: A cough is the body’s reaction to irritation in the airway. Air is taken in, pressure builds within the ribcage, and, finally, the air is expelled abruptly, sometimes carrying the materials causing the irritation. This abrupt expulsion of air causes the characteristic sound of a cough.

A throat clear is the voluntary or involuntary vibration of the larynx, vocal cords, and palate that can sometimes help expel materials or mucus from the upper respiratory tract.

From a labeling point of view, a throat clear can be very difficult to differentiate from a cough, and, in some cases, the subjective nature of the sound may make agreement impossible. In our SOP, we use abrupt air expulsion, which only occurs in coughs, as a key marker to differentiate these from throat clears. We also rely on the shape of soundwaves produced by both processes.

MS: You are part of the research team at Hyfe that aims at quantitatively characterizing cough sounds using digital signal processing techniques. What has this research process been like?

CC: When I came on board, the computer algorithm that differentiates coughs from non-cough sounds was already trained against thousands of human-labeled sounds. As the dataset grew, one of our biggest challenges was to go beyond just training and provide a “ground truth” of which sounds are undoubtedly a cough, which ones are undoubtedly not, and which ones are in a subjective grey zone.

That is when developing standard procedures for hearing, seeing, and labeling each sound became critical. It is very laborious and can be tedious, but listening to hours of sounds and identifying specific patterns has greatly helped us to better understand the coughing process from an acoustic perspective and improve the detection and classification algorithms.

MS: What is in the grey zone and why is it important?

CC: We use acoustic data and its visualization to try to discern coughs from non-coughs. Even with clearly established SOPs, some sounds will cause disagreement between labelers. Since human classification is the standard against which we measure artificial intelligence, disagreement between humans can be a problem on a small scale. Once you have enough labelers, this problem is minimized.

MS: What challenges have you come across during labeling processes in clinical settings?

CC: Some sounds are captured from afar, are faint, or partly overlap with others. All this can hamper labeling The case of throat clears vs coughs is already difficult, and factors like distance and overlap can take differentiation difficulty to the extreme. Here again, we will need to settle for a threshold of agreement considered appropriate, as human nature makes 100% concordance unlikely. In the future, the standard may well be A.I. itself.

MS: From the labeling study, which is the biggest challenge in differentiating between a cough and a throat clear sound when listening to continuous audio recordings?

CC: After labeling thousands of sounds, I can confidently say that everything is more complex than we think. Just listening to an audio stream and saying “that is a cough” and “that one is not” may seem very straightforward. It is not. Early in our efforts, we also thought it was simple, but we must watch out for the bias known as the Dunning-Kruger effect that teaches us that early in the learning path our confidence grows faster than our knowledge. It took a vast collective effort to develop, iterate, and improve our labeling procedures until reaching a state in which we feel confident with the results and agreement between labelers.

MS: What are your findings on the differences in the frequency of coughing vs throat clearing?

CC: Along the lines of the description above, there is no fixed pattern for the frequency of coughs or throat clears. Most people have a dominance of one or another and “pure” coughers or throat clearers are rare or non-existent.

MS: What are the limitations of current methods of cough quantitative analysis?

CC: In one of our studies, we recruited more than 900 volunteers, from the University of Navarra in Spain, to use Hyfe and try to compare the coughs in that cohort with the local incidence of COVID-19.

Some of those participants used the app for months. In this period, some of them experienced big changes in their cough frequency or pattern. One of our key findings is that, similarly to what happens with steps taken or physical activity (think Fitbit), to understand changes in cough changes a snapshot sample cannot produce accurate results. We need longitudinal data to understand relevant clinical changes in cough patterns. These findings are currently in press and will be disseminated appropriately.

MS: What are the implications of using digital signals to aid in quantitative measures for cough in the context of healthcare?

CC: We have highly developed technology that can aid in the interpretation of many physiological parameters. Blood pressure, pulse, EKG, imaging, all of them have evolved tremendously in the last century. Yet our approach to cough, one of the most common symptoms prompting care-seeking, remains based on a simple question from the provider and verbal report from patients — “Is your cough better?” “Is it productive?”. Of course, there are elaborations on this process and patient-reported outcomes now feature importantly in many studies and the clinical practice, but it is more than fair to expect that currently available technology be used for this important symptom.

MS: Do you have any ideas or research on what the medical advice would be when it comes to cough or throat clear?

CC: I think broad use of cough monitoring may help differentiate better between patients with more dominant throat clears and those with more cough. The former may benefit from speech therapy and drug-based approaches that can greatly differ from those used for cough.

MS: Thank you for your time. Is there anything else you’d like to add to the conversation?

CC: Cough is an important symptom and we are just beginning a journey. Tools such as Hyfe are the equivalent of traveling back in time to the 1600s and handling out a precision thermometer to doctors assessing fever with the back of their hands. It opens a whole new world of diagnostic and therapeutic possibilities.

Conclusion

New advanced technology for recording and analyzing the acoustic features of cough sounds has been introduced as a result of huge progress in digital systems. But as this discussion highlights, one of the challenges in cough interpretation is in classifying cough sounds into different ‘types’ because these sounds vary in many ways.

However, if captured and properly evaluated, the acoustic variations of cough may have fundamental clinical benefits. The latter can be achieved through the broad use of cough monitoring.