A Chinese study in March 2026 evaluates Cough Search for chronic obstructive bronchopneumopathy, or COPD, based on smartphone-recorded coughs. The external test involves 722 people in four hospitals in Shanghai.
Cough Search: what a positive result means
Among 105 participants with COPD, the 97 model is missing eight. Among 617 others, it correctly ranks 528 people, but produces 89 false positives. Sensitivity is 92.4% and specificity 85.6%. Of the 186 positive results, 52.2% actually correspond to COPD.
Sensitivity answers a question: among the patients, how much does the test detect? The positive predictive value poses another: among alerted people, how many are sick? Consolidating these two indicators would give a misleading impression of certainty. A device may lack few patients while pointing towards a complementary examination many people ultimately unharmed.
Performance includes adults at least 40 years of age registered according to a controlled protocol. Quality control excluded 90 of the 812 eligible participants. At the lightest stage, the model detects 11 out of 13 cases, which are too small to conclude widely. Generalization outside Shanghai remains to be checked.
A measurement of the breath remains necessary
The GOLD international initiative retains spirometry in the diagnosis of COPD. This examination measures the breath during a forced expiration. The criterion of obstruction is based on the relationship between the volume expired during the first second and the forced vital capacity after administration of a bronchodilator. The threshold is below 0.7.
A sound classification does not directly measure these volumes. It should therefore not be confused with the respiratory examination or with the entire medical assessment. The result of a tracking tool is used to guide a care path; It does not authorize a person to declare himself or herself ill or to change his or her treatment.
From experimental to daily use
A systematic review of 2026 of 89 studies on cough sounds highlights the general limitations of the field. Only 11.2% had external validation. But his research stopped in June 2025: this figure is not a particular evaluation of Cough Search, which was then published. It recalls why results obtained outside the learning data count as much.
The test log NCT06082791 describes a prospective observational auxiliary diagnostic study, not a trial demonstrating improved care. His file, whose last update was reported in October 2023, does not contain any submitted results. It therefore does not allow the performance published in 2026 to be verified independently of the scientific article.
The transition to common use still requires knowing how alerts will be confirmed, how unusable recordings will be processed and who will accompany those who are oriented. A high level of sensitivity does not answer these organizational and security questions.
Illustration photography from archives (2017): a smartphone. This image shows neither Cough Search nor a participant in the study. Jonas Svidras / Pexels.





Comments