Digital Acoustic Surveillance for Early Detection of Respiratory Disease Outbreaks

This is a single-center prospective observational study that pretends to evaluate the accuracy of an acoustic surveillance mobile app to detect individual episodes of cough among a monitored population, as well as the barriers and facilitators that might affect uptake of similar platforms at a population level.

The app in question, Hyfe cough tracker, runs in the background of smartphones, and records short snippets (<0.5 seconds) of explosive, putative cough sounds. These are then classified as cough or non-cough, using a convolutional neural network (CNN) model, and matched to GPS and time data collected by the smartphone.

The night-time cough of participants will be monitored for a 30-day period, and their clinical records will be reviewed regularly, specifically looking for diagnoses of cough-producing diseases, and with special emphasis on COVID-19.

Cough data will be used to create a heatmap of cough density and geographic distribution. Aggregated cough registries will be used to calculate the coughs per person-hour registered in the cohort. These data will be used to carry out an ARIMA analysis on three parallel time series at the community level: The incidence of respiratory disease in the monitored cohort, in the entire study area (including the Universidad de Navarra, and the neighbouring Cendea de Cizur), and the cough frequency per monitored hours.

Changes in cough frequency will also be compared to other environmental variables such as temperature and pollution level registered in the study area.