|
Written by Ametris |
With traditional clinical research methods, we’re often limited to the “tip of the iceberg” when it comes to understanding patient health and function. While we collect structured data during study visits, much of what happens in patients’ daily lives remains invisible, leaving a critical gap in our understanding of how therapies perform in the real world. Digital health technology and advanced analytics, including AI and machine learning, give us the opportunity to collect more frequent, real-world data, and we believe this is necessary to fundamentally modernize clinical research and help make better decisions in clinical trials.
Learn more about leveraging digital health technology in Respiratory R&D in our Digital Endpoint Guide.
As part of this vision, Ametris (formerly ActiGraph) designed the ActiGraph LEAP®, a multisensor wearable that can capture rich, real-world data across various domains. In our recent Digital Health Monthly webinar, we focused on the device’s microphone sensor and explored how audio data can be leveraged, specifically for cough detection. For the past two years, we’ve collaborated with Hyfe, a leader in AI-powered cough detection, to evaluate how their algorithm can be applied in clinical research. By combining Hyfe’s audio-based insights with synchronized data on physical activity and sleep, we aim to unlock a more holistic view of respiratory health and its impact on quality of life in real-world settings.
During this discussion we were joined by Dr. Fan Chung, MD, DSc, a professor of respiratory medicine at Imperial College London and consulting physician at the Royal Brompton Hospital, who has been doing research on cough monitoring and in respiratory research for several decades, and by Dr. Peter Small, MD, Chief Medical Officer at Hyfe, whose career has focused on developing and delivering innovative solutions to improve health globally as a physician and clinical investigator.
Cough as a Disease and Cough in Different Clinical Conditions
Approximately 5-10% of the population suffers from chronic cough, which is defined as cough lasting longer than 8 weeks1. Chronic cough is often a long-lasting and burdensome condition that can be difficult to diagnose. Many factors and diseases can contribute to chronic cough, such as etiological factors like infection, pollution or respiratory diseases, or indications like asthma, rhinitis, bronchiectasis, interstitial lung fibrosis, or gastroesophageal reflux disease (GERD).
However, almost half of patients with chronic cough have no identifiable cause (unexplained chronic cough, UCC), or experience symptoms even though they’ve received optimal treatment (refractory chronic cough, RCC). Experts including Dr. Chung have shared evidence that this type of chronic cough is a neuropathic disorder arising from nerve damage, and in certain cases refer to this condition as cough hypersensitivity syndrome. By identifying this mechanism of chronic cough, more effective antitussives can be developed for patients2.
Cough is also a prominent symptom in many other conditions. This symptom is important to monitor and track accurately in addition to other symptoms of these conditions because cough can be predictive of disease progression or exacerbations as well as negatively impact patients' quality of life (QoL). Dr. Chung reviewed the significance of cough in several conditions, such as -
Chronic Obstructive Pulmonary Disease (COPD): More than 70% of patients with COPD report experiencing cough. It is one of the three cardinal symptoms, and chronic cough is associated with severe exacerbations. Cough could also help identify patients at risk of progressive disease as it is an early feature of COPD. Almost half of COPD exacerbations can be predicted 3 days early by cough monitoring at home3,4,5.
Idiopathic Pulmonary Fibrosis (IPF): In the FDA’s Voice of the Patient report, coughing was identified by over three quarters of participants as being one of their most significant symptoms. Worse cough severity in patients with IPF is associated with worse health-related QoL, disease progression, reduced transplant-free survival and mortality6,7.
View the IPF Patient-Focused Digital Measure Report.
Asthma: Poor asthma control is associated with higher coughing, the presence of nighttime cough, and lower QoL. Cough is common in asthma independent of pulmonary function tests (PFTs) and poorly captured by existing methods. Cough monitoring could be used to detect asthma exacerbations 3-5 days in advance8,9.
View the Asthma Patient-Focused Digital Measure Report.
Bronchiectasis: Cough can be used as a biomarker of treatment response, to detect exacerbations early, as a data-stream to driving adherence to airway treatment protocols and for personalizing treatment10.
Congestive Heart Failure: Cough might be an early biomarker of decompensation in congestive heart failure and a treatment response biomarker.
The Importance of Continuous Cough Monitoring
The objective measurement of cough is a relatively recent advancement in clinical care, with the first cough frequency monitors introduced around 30 years ago. These early devices, which could record coughs over a 24-hour period, laid the groundwork for ongoing efforts over the past 25 years to develop ambulatory and automated cough monitoring tools. One of the most widely used tools in antitussive trials has been the VitaloJak system, which offers excellent sensitivity but is limited to single-day use and requires manual counting of cough events.
In the clinic, patient-reported outcomes like the Visual Analog Scale (VAS), where patients rate cough severity by marking a line, and the Leicester Cough Questionnaire (LCQ), which measures cough-related quality of life, are commonly used. There is some correlation between objective measures (e.g., explosive phases of cough) and subjective scores like the VAS, but this relationship is not always strong.
The push toward continuous, longer-term cough monitoring has gained importance as studies show significant day-to-day variability in cough frequency for many patients. While some individuals exhibit stable cough patterns, research involving close to 50 participants found that nearly two-thirds had highly variable cough rates, making it difficult to accurately assess their condition with short-term monitoring alone11.
Additionally, the correlation between subjective assessments and objective cough counts weakens when measured over multiple days12. These findings underscore the need for continuous cough monitoring technologies that go beyond 24-hour snapshots in order to obtain a more complete picture of cough patterns and improve the ability to evaluate treatment efficacy.
How Continuous Cough Monitoring Works with the ActiGraph LEAP and Hyfe Algorithms
The Hyfe cough detection algorithms are used on the ActiGraph LEAP device in a two-step approach to detect coughs while also preserving patient privacy. First, environmental sounds are continuously monitored by the microphone without recording any audio. When a sound resembling a cough (an "explosive-like" event) is identified by the algorithm, it captures a brief ~1-second audio snippet, which is then analyzed by a second algorithm based on an AI-driven convolutional neural network.
This model has been trained on tens of millions of labeled explosive sounds to accurately distinguish coughs from other non-cough noises. This method protects patients’ privacy because all this processing occurs locally on the ActiGraph LEAP device.
In a real-world study involving 23 individuals with problematic cough, the algorithms demonstrated 90% accuracy with only one false positive per hour when compared to human annotations13. This new solution enables longitudinal, real-time cough frequency monitoring while safeguarding patient privacy.
Download the Continuous Cough Monitoring Solution Sheet for more details.
How Continuous Cough Moniroting Works with the ActiGraph LEAP and Hyfe Algorithms
Antitussive trials have been plagued by high placebo responses that can complicate the interpretation of treatment effects in chronic cough clinical trials14.
There are 4 contributing factors to the placebo effect in cough studies -
1. Inaccurate Measurement
Inaccurate measurements not only because they introduce a lot of noise and potentially also bias, but they also accentuate regression to the mean. This is a problem because it will lead to the data looking like there is a reduction in cough in patients who actually didn’t have a reduction. For example, looking at this graph of cough/hour, it’s clear that the patient shows a cough reduction after treatments with gabapentin or omeprazole. However, if you were to take any one of these measures (black dots) as representative measures of their cough, it would actually show their cough increasing about 8% of the time15.
Potential Solution: Continuous cough monitoring provides more accurate measures of how cough patterns are changing over time to potentially limit the impact of the placebo effect.
2. Endpoint Variability
In a study of 97 individuals with persistent cough, continuous monitoring of cough rates over 30 days showed substantial variability both within a day and across days11. This variability makes it very difficult to demonstrate a change in cough rates.
Potential Solution: Continuous cough monitoring provides an opportunity to select for enrollment patients that are more consistent coughers to potentially decrease the placebo effect. While the study mentioned above monitored subjects for 30 days, statistical modeling suggests that 7 days of data collection is sufficient for an accurate measure of cough rates.
3. Internal / External Perceptions
Pain research has shown that some subjects are more in touch with their internal perception of the symptom, while in other pain patients, external factors have a larger influence on their perception of pain.
In cough research, a similar pattern has been observed. For example, in the graphs below, you can see in Case 012 that the subject’s subjective cough score and objective cough rate is closely aligned, indicating they have a relatively accurate perception of their cough severity, unlike Case 01612.
Cough is different than other symptoms; your blood pressure can be elevated, for example, but for the most part people don’t have a lot of control over their blood pressure. However, people can cough on command or even suppress cough if they try hard enough. Monitoring coughs discreetly, in their natural environment, and in a way that protects privacy is very important to collecting accurate cough data.
Potential Solution: Continuous cough monitoring can be used to select for patients where there is a tighter correlation between objective cough rates and subjective cough scores, or it can be used in some subjects to train them to be more perceptive to their actual cough levels, which can diminish the placebo effect.
4. Study Oversold
The more someone expects the drug to have an impact, the more likely it is they will have a placebo response. Sites are working hard to enroll subjects, and they may not realize that sometimes studies are sold in a way that exacerbates the placebo phenomenon.
Potential Solution: Use a consumer app or social media to make recruitment in these types of cough studies a little easier and diminish the likelihood of “overselling” a study in a way that could increase the placebo effect.
Conclusion
It's clear that cough is an important symptom for many indications. Continuous monitoring has more information and higher accuracy about cough rates than shorter term monitoring solutions, and using the ActiGraph LEAP device to monitor cough now enables researchers to integrate cough with important aspect of health such as physical activity, sleep, and vital signs to better understand how these aspects may interact.
Cough has a significant negative impact on quality of life, and developing effective treatments to address this symptom depends on being able to assess their effect with accurate measures.
References
1. Chung KF, McGarvey L, Song WJ, Chang AB, Lai K, Canning BJ, Birring SS, Smith JA, Mazzone SB. Cough hypersensitivity and chronic cough. Nat Rev Dis Primers. 2022 Jun 30;8(1):45. doi: 10.1038/s41572-022-00370-w. PMID: 35773287; PMCID: PMC9244241.
2. Chung KF. Approach to chronic cough: the neuropathic basis for cough hypersensitivity syndrome. J Thorac Dis. 2014 Oct;6(Suppl 7):S699-707. doi: 10.3978/j.issn.2072-1439.2014.08.41. PMID: 25383203; PMCID: PMC4222934.
3. Choate R, Pasquale CB, Parada NA, Prieto-Centurion V, Mularski RA, Yawn BP. The Burden of Cough and Phlegm in People With COPD: A COPD Patient-Powered Research Network Study. Chronic Obstr Pulm Dis. 2020 Jan;7(1):49-59. doi: 10.15326/jcopdf.7.1.2019.0146. PMID: 31999902; PMCID: PMC7182382.
4. Burgel PR, Nesme-Meyer P, Chanez P, Caillaud D, Carré P, Perez T, Roche N; Initiatives Bronchopneumopathie Chronique Obstructive (BPCO) Scientific Committee. Cough and sputum production are associated with frequent exacerbations and hospitalizations in COPD subjects. Chest. 2009 Apr;135(4):975-982. doi: 10.1378/chest.08-2062. Epub 2008 Nov 18. PMID: 19017866.
5. Crooks MG, den Brinker AC, Thackray-Nocera S, van Dinther R, Wright CE, Morice AH. Domiciliary Cough Monitoring for the Prediction of COPD Exacerbations. Lung. 2021 Apr;199(2):131-137. doi: 10.1007/s00408-021-00435-9. Epub 2021 Apr 7. PMID: 33829322; PMCID: PMC8053154.
6. Myall KJ, Cho PSP, Birring SS. What causes cough in pulmonary fibrosis, and how should we treat it? Curr Opin Pulm Med. 2024 Sep 1;30(5):523-529. doi: 10.1097/MCP.0000000000001087. Epub 2024 Jun 24. PMID: 38913018; PMCID: PMC11495478.
7. Ryerson CJ, Abbritti M, Ley B, Elicker BM, Jones KD, Collard HR. Cough predicts prognosis in idiopathic pulmonary fibrosis. Respirology. 2011 Aug;16(6):969-75. doi: 10.1111/j.1440-1843.2011.01996.x. PMID: 21615619.
8. Lai K, Satia I, Song WJ, Wang G, Niimi A, Pattemore P, Chang AB, Gibson PG, Chung KF. Cough and cough hypersensitivity as treatable traits of asthma. Lancet Respir Med. 2023 Jul;11(7):650-662. doi: 10.1016/S2213-2600(23)00187-X. Epub 2023 Jun 16. PMID: 37336227.
9. Rassouli F, Tinschert P, Barata F, Steurer-Stey C, Fleisch E, Puhan MA, Baty F, Kowatsch T, Brutsche MH. Characteristics of Asthma-related Nocturnal Cough: A Potential New Digital Biomarker. J Asthma Allergy. 2020 Dec 3;13:649-657. doi: 10.2147/JAA.S278119. PMID: 33299332; PMCID: PMC7721277.
10. D.E. Griffith, A.R. Levin, M. Rudd, P. Small, and C.L. Daley. Using Continuous Cough Monitoring to Assess Bronchiectasis Therapy [abstract]. Am J Respir Crit Care Med 2024;209:A7084.https://doi.org/10.1164/ajrccmconference.2024.209.1_MeetingAbstracts.A7084
11. Chung KF, Chaccour C, Jover L, Galvosas M, Song WJ, Rudd M, Small P. Longitudinal Cough Frequency Monitoring in Persistent Coughers: Daily Variability and Predictability. Lung. 2024 Oct;202(5):561-568. doi: 10.1007/s00408-024-00734-x. Epub 2024 Jul 31. PMID: 39085518; PMCID: PMC11427503.
12. Lee SE, Rudd M, Kim TH, Oh JY, Lee JH, Jover L, Small PM, Chung KF, Song WJ. Feasibility and Utility of a Smartphone Application-Based Longitudinal Cough Monitoring in Chronic Cough Patients in a Real-World Setting. Lung. 2023 Dec;201(6):555-564. doi: 10.1007/s00408-023-00647-1. Epub 2023 Oct 13. PMID: 37831232.
13. Chaccour C, Sánchez-Olivieri I, Siegel S, Megson G, Winthrop KL, Botella JB, de-Torres JP, Jover L, Brew J, Kafentzis G, Galvosas M, Rudd M, Small P. Validation and accuracy of the Hyfe cough monitoring system: a multicenter clinical study. Sci Rep. 2025 Jan 6;15(1):880. doi: 10.1038/s41598-025-85341-3. PMID: 39762316; PMCID: PMC11704278.
14. Kum E, Hassan W, Satia I. Placebo responses in clinical trials of refractory chronic cough: mechanisms, challenges, and mitigation strategies. Lancet Respir Med. 2025 Apr;13(4):297-299. doi: 10.1016/S2213-2600(25)00018-9. Epub 2025 Mar 12. PMID: 40088918.
15. Gabaldón-Figueira JC, Keen E, Rudd M, Orrilo V, Blavia I, Chaccour J, Galvosas M, Small P, Grandjean Lapierre S, Chaccour C. Longitudinal passive cough monitoring and its implications for detecting changes in clinical status. ERJ Open Res. 2022 May 16;8(2):00001-2022. doi: 10.1183/23120541.00001-2022. PMID: 35586452; PMCID: PMC9108969.