ABSTRACT
Background: The potential of continuous cough monitoring is in an early stage, while almost every other clinical symptom has a way to be objectively monitored. Objective(s): Study if continuous cough monitoring is useful for early notice of an onset or worsening of respiratory conditions. Method(s): A free mobile application was used to detect and record cough sounds. Only 0.5s snippets of explosive sounds are sent to the server for AI to analyse. The 3 cases presented were identified within a study in Navarra, Spain. Result(s): Case 1: a 56-year-old was using the app with an average of 600 coughs/d, unknown cause. A trial with gabapentin was started, which within a month resulted in 150 coughs/d. With omeprazole added, coughing reduced to ~50 coughs/d. Case 2: a 70-year-old smoker was using the app with an average 52 coughs/day in over 2 months. She quit smoking and noticed improvements in cough, app showing 12 coughs/d. The next month, smoking relapsed, reaching 34 coughs/d. Data dynamics renewed her motivation to quit. Case 3: a 35-year-old was using the app at night, with an average of 4 coughs/hr (not self-perceived). Suddently, the patient felt general malaise and the app detected 12 coughs/hr (not self-perceived). Next day, she received a diagnosis of uncomplicated COVID-19. Conclusion(s): Cough patterns correlate with clinical progress and perceived improvement, accurately indicate signs of smoking cessation and relapse.
ABSTRACT
TYPE: TOPIC: Biotechnology PURPOSE: To evaluate the potential of AI-enabled syndromic surveillance tools for the detection of outbreaks of respiratory disease. METHODS: We carried out a prospective observational study (NCT04762693) between November 2020 and August 2021 in the province of Navarra, Spain. Participants recruited from the campus of the University of Navarra, in Pamplona, and neighboring towns were instructed to download a publicly available smartphone application that automatically detects and records episodes of cough, with their time and location. Clinical records from consenting participants were reviewed regularly. Aggregated cough data was compared with local incidence of covid-19, and the time of medical consultation due to respiratory symptoms, using a randomization routine. Focused group discussions were carried out to determine barriers and facilitators for the usage of the application. RESULTS: Cough rates of participants who consulted medical services were higher in the five-day period around the date of consultation, compared to the baseline (p = 0.012), suggesting that longitudinal cough records reflected changes in symptomatology. In contrast, no clear relationship between aggregated cough frequency and covid-19 incidence in the study area was observed. Interface simplicity and the possibility of observing changes of cough frequency in real time were appreciated by participants. The possibility of programming the application so it automatically starts and stops at certain times was the most common recommendation. CONCLUSIONS: Acoustic cough monitoring can capture changes in symptomatology. Adequate uptake and usage remain challenging. CLINICAL IMPLICATIONS: Upon further development, AI-enabled cough detection could offer an attractive complementary tool to traditional respiratory epidemiological surveillance. DISCLOSURE: JB is the CEO of Hyfe Inc. EK is a Hyfe employee. CCh discloses consultancy fees and equity from Hyfe. All other authors declare no conflict of interests. KEYWORD: cough