The phone stays on the bedside table while you sleep and, without having to wake up to fill out questionnaires, it counts your coughs. By putting together those recorded by tens of thousands of users, the researchers found a rather clear signal: when PM2.5 concentrations increase during the day, the frequency of coughing also increases during the night.
It is the result of a study published in Communications Healthconducted by researchers from the University of Cambridge, Tsinghua University, Sleep Cycle and, for Italy, the Vita-Salute San Raffaele University of Milan. The analysis spanned 32 cities in 12 countries, including Milan, Rome, Florence, Bergamo and Monza, following approximately 500 days between October 2023 and April 2025.
The most immediate data comes from the meta-analysis of individual cities: for every increase of 10 micrograms per cubic meter of PM2.5, the frequency of nocturnal cough was on average 2.4% higher. The association appeared even at relatively low concentrations and remained visible after taking into account temperature, precipitation and flu trends.
Coughing increased even when PM2.5 remained at relatively low levels
PM2.5 are particles less than 2.5 micrometers in diameter, small enough to penetrate deeply into the respiratory system. To understand how much they weighed on users’ nights, the authors crossed the app’s data with the daily concentrations detected by monitoring networks and collected through OpenAQ.
The algorithm identified coughs in the nighttime recordings and transformed them into a frequency per recorded hour of sleep. The counting took place on the devices; the researchers received anonymized and aggregated data by city and day, without individual voice recordings. Two of the authors are employees of Sleep Cycle, the company that provided the app data, a conflict of interest disclosure in the study.
In the model considering only same-day exposure, a 10 µg/m³ increase in PM2.5 was associated with approximately 1.2% more nocturnal coughing. When the authors also considered the effects spread over the following five days, the cumulative increase reached 2.4%. The day immediately following exposure showed the strongest signal.
The curve, above all, did not draw a reassuring boundary under which the particulate suddenly stopped making itself felt. The association also appeared below 15 µg/m³, the World Health Organization guideline value for the average 24-hour PM2.5 concentration. That number, however, was never conceived as a magical border between harmless air and harmful air: the WHO itself explains in its air quality guidelines that health effects have been observed even at low concentrations and recommends reducing exposure as much as possible.
From Milan to Florence, five Italian cities also ended up in the study’s large “microphone”.
In the international sample there are Milan, Rome, Florence, Bergamo and Monza. Italy also appears among the institutions involved through Marco Montagna, doctor and researcher at the Vita-Salute San Raffaele University.
The city-by-city results, however, do not all travel at the same speed. The study records strong heterogeneity among the 32 urban centers: some cities show clearer associations, others much weaker. Overall, 14 cities had relative risk estimates greater than 1 in the main analysis, while during periods of low influenza circulation positive relationships appeared in 23 cities. A global average useful for seeing the signal, therefore, does not make the cities interchangeable.
And it is also an important detail to correctly read that “+2.4%”. The authors worked on data aggregated at the city level: they knew the overall cough frequency recorded in a city and the average outdoor concentration of PM2.5 in the same period. They didn’t know how much polluted air each user had actually breathed, where they had spent the day, or whether they had other conditions that could cause coughing.
The study, therefore, identifies an association on a population scale. It does not allow us to conclude that that particular cough at three in the morning was caused by particulate matter breathed in the afternoon. And there is another necessary distinction: awakenings nor sleep quality were measured. Cough was recorded during the night; Determining how much it actually disturbed rest requires more data.
During the Los Angeles fires the signal became much stronger
The fires that hit Los Angeles in January 2025 offered researchers an extreme situation to observe almost live. During the episode, strong increases in PM2.5 were accompanied by a much more pronounced increase in nocturnal coughing: in the fire model, 10 µg/m³ more was associated with an increase close to 9.9%.
However, the authors themselves invite us to handle this result with care. Cough frequency had begun to fluctuate even before the strongest spike in particulate matter, and respiratory infections or other unmeasured environmental exposures may have contributed. The fire therefore strengthens the signal observed in the entire dataset, but does not function as an experimental proof of causality.
It remains a huge involuntary experiment: millions of hours of sleep transformed into an environmental sensor distributed among bedside tables around the world. And it shows something less spectacular, but much harder to ignore: the effects of air pollution can appear long before hospitalization or a diagnosis, through small, everyday symptoms that usually disappear amidst the noise of the city.
Cecilia Mascolo, who coordinated the research, observes that the data are compatible with the absence of a threshold completely devoid of effects, specifying however that the study remains observational. Air filters and masks on the most polluted days can reduce individual exposure; the most effective measure, on a collective scale, remains the least pocket-friendly: lowering the concentrations of particulate matter in the air we breathe.
