Chronic Obstructive Pulmonary Disease — smartphones could help improve patient outcomes

We have developed a diagnostic test for acute exacerbations of chronic obstructive pulmonary disease using cough-sound analysis and simple clinical features. Patients can use the test, unaided, on their smartphone. Obtaining a rapid and reliable diagnosis is key to improving clinical outcomes.

Like Comment
Read the paper

Chronic obstructive pulmonary disease (COPD) is the third leading cause of mortality worldwide.1 COPD typically results from prolonged exposure to harmful gases, such as tobacco smoke, which trigger airway and alveolar abnormalities. These abnormalities restrict airflow to the lungs, causing affected individuals to develop persistent and progressive symptoms, including wheezing, coughing, shortness of breath and sputum production.2,3 Patients with COPD can experience periods of acute worsening of their symptoms, known as acute exacerbations of COPD (AECOPD), which require treatment with antibiotics and steroids.4

 Problems facing AECOPD diagnosis and self-management

AECOPD is diagnosed through radiology, lung function tests and clinical assessment. To guide treatment, disease severity is graded from 0 to 4 (using CRB-65 criteria). It is important for clinicians to reach a quick diagnosis as a patient’s risk of hospitalization doubles if an AECOPD diagnosis takes longer than 24 hours.5

 To reduce delays in diagnosis, patients are encouraged to self-manage their condition by keeping a written action plan. This approach helps patients to identify AECOPD symptoms and assess disease severity, which indicates when they should seek medical treatment. However, it is important to note that this process is slow as it requires patients to monitor their symptoms over several days. In addition, effective self-management strategies rely on patients being able to not only correctly interpret the medical terminology in action plans but also correctly identify AECOPD symptoms and assess disease severity.  With these points in mind, it is not surprising that two-thirds of patients struggle to identify a worsening of key AECOPD symptoms, such as breathlessness and change in sputum amount or colour.6

 As they have an increased risk of developing severe respiratory infections, individuals with COPD may avoid clinical settings;  a situation that has been exacerbated by COVID-19.  To minimize potential exposure, patients can request remote and telehealth consultations.7  These remote consultations can be complex for health practitioners to perform as diagnoses require specialist equipment and contact procedures, such as auscultation.

 In order to improve health outcomes and reduce clinical costs, we need to find accurate, timely and straightforward ways for patients with COPD to recognise acute exacerbations of their condition. 

 Our study

We report on the development of a smartphone-based diagnostic test for AECOPD. This novel test uses an automated algorithm to analyze cough-centred audio recordings and three patient-reported clinical features. We aimed to evaluate how well the smartphone-based diagnostic test agreed with the clinical diagnoses of patients with AECOPD in a prospective study.

 How does the algorithm work?

Respiratory conditions that alter airflow, such as AECOPD, generate characteristic signatures composed of sounds from the lower airway being expelled through the open glottis.8 Our diagnostic test uses an automated algorithm that analyses audio data recorded during coughs and evaluates these sounds to identify AECOPD. Following a similar approach to speech recognition technology, a time-delay neural network identifies cough event audio segments. Mel-frequency cepstral coefficients are then calculated from the extracted cough event audio segments and inputted into a logistic regression model classifier to determine if the patient is having an AECOPD.  To improve diagnostic accuracy, patients also input the answers to three simple questions about their clinical features into the algorithm.

 We trained the algorithm using datasets of cough-sound recordings with matching clinical diagnoses (1228 subjects). The algorithm was further optimized using feature selection and cross-fold validation. The optimized automated algorithm (index test) was then fixed, ready for prospective diagnostic study assessment (Figure 1).

Prospective diagnostic accuracy testing of the algorithm

To test the algorithm's accuracy, we compared it with a full clinical assessment for AECOPD diagnosis (Fig. 1). We calculated diagnostic agreement using the following: 

  1. Positive percent agreement (PPA) = percentage of patients diagnosed with AECOPD by both the algorithm and clinical assessment
  2. Negative percent agreement (NPA) = percentage of patients assessed as negative for AECOPD by both the algorithm and clinical assessment.

 We found that the algorithm had an excellent diagnostic agreement with clinical AECOPD diagnosis (PPA 82.6%; NPA 91%). In addition, the algorithm successfully diagnosed in 79.2% of mild AECOPD, and 100% of moderate exacerbations. Of note, the algorithm correctly diagnosed AECOPD in patients aged 65 years or older (PPA 85.9%; NPA 88.9%). This finding was particularly exciting as patients in this age group are more likely to have co-morbidities that can mask AECOPD symptoms, such as chronic heart failure.


Figure 1

Figure 1. Algorithm development and deployment.

Panel A: 1228 patients were recruited to train an algorithm to detect acute exacerbations of chronic obstructive pulmonary disease (AECOPD).  Audio data from five cough events was collected along with complete medical history and examination findings.  Characteristic audio features were identified and combined with three patient-reported features (patient age and fever or new cough) to produce an optimal algorithm for diagnostic accuracy testing.

Panel B:  The algorithm was prospectively tested in a cohort of 177 patients with known COPD to determine the presence or absence of an acute exacerbation.

Study outcomes

 We have used machine learning technology to develop an algorithm that shows excellent agreement with clinical diagnoses of AECOPD. This diagnostic system is quick and straightforward to use, and can be self-administered. Our test was able to identify AECOPD in patients with mild exacerbations, as well as in patients with comorbidities, both of which are typically hard to diagnose. This research forms part of the larger Breathe Easy Study (ACTRN12618001521213), in which we have developed similar algorithms to accurately diagnose respiratory diseases including COPD,9 adult pneumonia,10 asthma and acute childhood respiratory diseases.8 In all target diseases, the algorithms analyze cough-centred audio recordings as well as input from self-reported clinical features such as fever, new cough event, runny nose, productive cough and wheeze. 

 Our algorithm has the potential to aid and expedite the diagnosis of AECOPD for patients using a self-management program, and for physicians using telehealth consultations, with implications for improved health outcomes and reduced health care costs. This technology addresses digital-first health care priorities articulated by global health regulatory bodies.


  1. GBD Chronic Respiratory Disease Collaborators. Prevalence and attributable health burden of chronic respiratory diseases, 1990-2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet Respir Med. 8, 585-596 (2020).
  2. MacNee, W. Pathology, pathogenesis, and pathophysiology. BMJ. 332, 1202-1204 (2006).
  3. Miravitlles, M., Ribera, A. Understanding the impact of symptoms on the burden of COPD. Respir Res. 18, 67 (2017).
  4. Kim, V., Aaron, S. D. What is a COPD exacerbation? Current definitions, pitfalls, challenges and opportunities for improvement. Eur Respir J. 52, 1801261 (2018)
  5. Chandra, D., Tsai, C. L. & Camargo, C. A., Jr. Acute exacerbations of COPD: delay in presentation and the risk of hospitalization. COPD. 6, 95-103, (2009).
  6. Langsetmo, L., Platt, R. W., Ernst, P. & Bourbeau, J. Underreporting exacerbation of chronic obstructive pulmonary disease in a longitudinal cohort. Am J Respir Crit Care Med. 177, 396- 401, (2008).
  7. Gerayeli, F. V., Milne, S., Cheung, C. et al. COPD and the risk of poor outcomes in COVID-19: A systematic review and meta-analysis. EClinicalMedicine. 33,100789 (2021).
  8. Porter, P., Abeyratne, U., Swarnkar, V. et al. A prospective multicentre study testing the diagnostic accuracy of an automated cough sound centred analytic system for the identification of common respiratory disorders in children. Respir Res. 20, 81 (2019).
  9. Porter, P., Claxton, S., Brisbane, J. et al. Diagnosing Chronic Obstructive Airway Disease on a Smartphone Using Patient-Reported Symptoms and Cough Analysis: Diagnostic Accuracy Study. JMIR Form Res. 4, e24587 (2020).
  10. Porter, P. et al. Diagnosing Community-Acquired Pneumonia: diagnostic accuracy study of a cough-centred algorithm for use in primary care and acute-care consultations. BJGP. 71, e258-e265 (2020).

Paul Porter

Paediatrician, Curtin University