Assessing chronic obstructive pulmonary disease risk based on exhalation and cough sounds.

Background and objective: Chronic obstructive pulmonary disease (COPD), a progressively worsening respiratory condition, severely impacts patient quality of life. Early risk assessment can improve treatment outcomes and lessen healthcare burdens. However, current early assessment methods are limited. This study seeks to develop innovative approaches for the early detection and evaluation of COPD.

Methods: This study employed a cross-sectional design. Initially, we created a dedicated recording application deployed on smartphones to gather audio data from participants. Following this, each individual completed pulmonary function tests and participated in questionnaire surveys. COPD risk was defined as a pre-bronchodilator FEV₁/FVC ratio < 0.7 combined with a history of exposure to risk factors like smoking or biomass fuel. Ultimately, we assessed the feasibility of utilizing smartphones to capture exhalation and cough sounds for the identification of COPD risks through the application of machine learning algorithms.

Results: We gathered valid data from 530 adults, of whom 171 met the criteria for being at risk of COPD. Utilizing the XGBoost algorithm, we achieved a precision of 0.98 and a recall of 0.89.

Conclusions: Our study demonstrates that cough audio signals provide valuable insights for identifying COPD risk, effectively complementing exhalation signals in assessments. This approach is not only feasible and practical for real-world applications, but also offers an affordable and accessible solution, especially beneficial in resource-limited settings.