Nighttime coughs rose with daily fine-particle pollution across 32 cities

The discovery

The researchers asked whether day-to-day changes in ambient fine particulate matter are followed by changes in nighttime cough frequency at population scale, and whether a commercial sleep app could provide privacy-preserving respiratory surveillance.

The research question and why it matters

Controlled and epidemiological studies already connect PM2.5 with respiratory irritation, asthma exacerbation and broader health risks, but symptom surveillance is usually based on surveys, clinics or delayed records. Phone-based audio research has shown that coughs can be detected computationally. This study extended that approach across cities and consecutive days while keeping audio processing on the device.

What researchers found

Across the city-specific meta-analysis, each 10 micrograms per cubic meter increase in daily PM2.5 was associated with a 2.4% increase in detected nighttime cough frequency: relative risk 1.024, with a 95% confidence interval from 1.013 to 1.035. Pooled models found a nonlinear relationship, including an association at comparatively low pollution concentrations. Cough frequency also rose during the Los Angeles wildfire as ambient PM2.5 increased.

Results at a glance

Key results from the tested systems

cities

The analysis spanned 12 countries.

~500 days

observation window

Consecutive city-day measurements supported time-series analysis.

+2.4%

cough frequency

Estimated change for each 10 µg/m³ higher daily PM2.5.

1.013–1.035

95% confidence interval

The relative-risk estimate was 1.024.

How the research worked

A cough-detection model ran on phones using nighttime audio captured by the Sleep Cycle app; raw recordings did not leave the devices. Counts were aggregated by city and day, then linked to daily PM2.5, meteorological conditions and influenza activity in 32 cities. Researchers fitted city-specific time-series models, combined them with random-effects meta-analysis, ran pooled multi-city models and separately examined the January 2025 Los Angeles wildfire episode.

Subjects or systemHuman

Research designMultinational ecological time-series study using on-device cough detection, city-day aggregation, random-effects meta-analysis and a wildfire event analysis

Evidence baseTens of thousands of Sleep Cycle app users across 32 cities in 12 countries over approximately 500 consecutive days; exact user count and demographic composition were not available in the public institutional and repository records reviewed for this page

How to interpret this design

Random assignment is an important strength because it reduces systematic differences between comparison groups at the start. It does not eliminate problems caused by missing follow-up, imperfect blinding, protocol deviations, short duration or selective outcome reporting.

Because the research involved people, it speaks directly to the participants and outcomes measured. It may still apply differently to people outside the eligibility criteria, age range, clinical setting, geography or follow-up period.

What strengthens or limits the finding?

The analysis covered many cities and consecutive days, adjusted for weather and influenza, combined city-specific estimates and examined a major wildfire episode. Exposure and symptoms were aggregated at city-day level, users were self-selected and the observational design cannot establish individual exposure or causation.

The result is meaningfully informative, but identifiable limitations could alter the size, reach or causal interpretation of the finding.

Funding and disclosure context

The recorded funding source is: UK Engineering and Physical Sciences Research Council award EP/Z53447X/1 and European Research Council Horizon 2020 award 833296. The complete conflict-of-interest declaration should be checked in the original publication rather than inferred. Funding or a disclosed relationship does not by itself invalidate a result, but it is relevant when judging design choices, analysis and the need for independent replication.

What it means

Nighttime coughing may respond quickly enough to pollution changes to supplement slower health-reporting systems. Aggregated on-device sensing could help public-health researchers identify population-level respiratory signals while reducing transfer of raw audio. It is not a personal pollution monitor, medical diagnosis or proof that a particular night’s cough was caused by outdoor particles.

Beyond the abstract

Deeper analysis

A large stream can still be ecological

Tens of thousands of users and hundreds of days create statistical power, but aggregation removes the ability to say which person inhaled how much pollution or produced which cough. Scale improves detection of a city signal without converting it into individual causal evidence.

The effect is relative, not a cough count

A 2.4% increase describes the modeled population-level rate ratio for a 10-microgram PM2.5 difference. Its practical impact depends on the baseline cough frequency, exposure distribution and vulnerability of the people represented.

On-device processing changes the privacy tradeoff

Running the classifier locally means raw bedroom audio need not be uploaded for the analysis. Aggregated outputs can still carry governance and re-identification concerns, so privacy protection requires more than a technical architecture.

Wildfire timing is supportive, not experimental

A sharp pollution episode provides a useful real-world stress test and the cough increase followed the same period. Wildfire smoke also changes behavior, indoor air and other exposures, so it does not function like randomized assignment.

Keep the claim in proportion

What it does NOT prove

It does not show that PM2.5 caused each detected cough or that no exposure level is safe for every person.
It does not identify asthma, infection, allergy, reflux or any other cause of coughing.
It does not measure an individual user’s indoor or personal particle exposure.
It does not show that a 2.4% population-level change is a clinically important change for an individual.
It does not validate the app as a medical device or recommend using it for diagnosis or treatment.

Important limitations

The ecological design linked city-level pollution with aggregated cough counts, so the exposure and symptom cannot be paired within an individual.
Sleep-app users are self-selected and may differ by age, income, phone access, health and sleep habits from each city’s population.
Outdoor city monitors do not capture indoor air, ventilation, travel or personal exposure, all of which can differ sharply within a city.
Microphone placement, phone model, background noise and cough-classifier errors may vary across users and nights.
Adjusting for weather and influenza does not remove all potential confounding from allergens, other infections, smoke sources or seasonal behavior.
A commercial app provided the sensing platform, and public records did not disclose the exact user count or full demographic composition.

How this fits with previous research

Questions still unanswered

Do the associations persist when personal pollution sensors and individual symptom histories are available?
How accurately does the cough model perform across phone types, languages, ages and background-noise conditions?
Which particle sources and chemical mixtures are most strongly related to the short-term signal?
Can prospective public-health use detect harmful episodes earlier than conventional surveillance without generating false alarms?
How should consent, data governance and commercial-platform dependence be managed if this approach scales?

Government verification and context

Relevant U.S. government resources

These resources serve different purposes. A registry can verify what researchers planned, a repository can locate government-funded work, and an agency page can supply authoritative background. None automatically proves that this paper's conclusion is correct.

Government repositoryU.S. Geological Survey

USGS Publications Warehouse ↗

The authoritative catalog of USGS scientific publications, used to check related government research and long-term observational context.

Authoritative contextNational Oceanic and Atmospheric Administration

NOAA research and data ↗

Federal observations and research on climate, oceans, atmosphere and ecosystems provide context for environmental claims. They do not automatically validate a separate model or paper.

Reuse note: Facts and discoveries are summarized here in original language. We link to government material instead of copying it wholesale, and we do not reuse agency logos, photographs, charts or third-party material unless the specific reuse rights are verified.

Sources and provenance