UKHSA: Sleep App Cough Data Signals Flu and COVID-19 a Week Early

The UK Health Security Agency and Swedish sleep technology company Sleep Cycle on September 3, 2026 published the results of a joint research study evaluating whether cough data collected passively through a consumer sleep app could serve as an early warning signal for respiratory illness in England. The study found that rises in nighttime coughing were often observed around one week before increases in influenza and COVID-19 activity.

According to the UKHSA announcement, the passively collected data provided a robust and regionally consistent indicator of community respiratory illness, with cough activity closely reflecting levels of respiratory illness reported through the NHS 111 telephone service. UKHSA said the findings support the role of digital health data in public health surveillance alongside established systems.

Study Design and Findings

The research, published as a medRxiv preprint and authored by seven UKHSA researchers and two Sleep Cycle researchers, compared weekly nocturnal cough metrics from January 2023 to January 2026 against UKHSA surveillance indicators: NHS 111 acute respiratory infection triage calls, influenza and COVID-19 PCR positivity, and hospital admission rates for influenza, COVID-19, and respiratory syncytial virus.

The study tested three cough metrics: total cough counts, coughs per user, and coughs per hour of sleep. The strongest associations were with NHS 111 acute respiratory infection triage calls, where the population-normalised metrics showed raw national correlations of approximately 0.95 and retained prewhitened correlations above 0.55 at lag zero, meaning cough activity tracked short-term variation in an established syndromic indicator beyond shared seasonality and long-term trends. The preprint reports that coughs per hour of sleep peaked one week before influenza PCR positivity, while both coughs per user and coughs per hour of sleep peaked one week before COVID-19 PCR positivity. Hospital-based indicators showed weaker relationships, though the normalised metrics aligned contemporaneously with influenza admissions and showed short leading associations with COVID-19 admissions.

Unnormalised total cough counts produced less stable and often non-interpretable lag structures, which the authors attribute to sensitivity to changes in observation volume, such as shifts in the number of active users and recorded sleep duration. The preprint recommends population-normalised metrics over raw counts for any surveillance application.

On-Device AI and Privacy Protections

Sleep Cycle is a smartphone app that uses AI-powered sound analysis to help users understand and improve their sleep. Cough events are identified by a machine learning audio detection model that runs locally on the user’s device, performing inference on overlapping 10-second audio clips during user-initiated sleep sessions. No raw audio is transmitted to Sleep Cycle’s servers. According to the preprint, records are anonymised before transmission through removal of personal identifiers and perturbation of geographic coordinates, and Sleep Cycle shares only aggregated data protected using differential privacy.

The dataset covered an average of 3,482 daily users in England’s South West region to 11,427 in London. Among users who voluntarily provided demographic information, the average age was 37.9 years, with 59.3% male and 40.2% female.

Origins of the Collaboration

UKHSA and Sleep Cycle announced the research collaboration on January 28, 2026, describing a 12-week project that the agency said marked the first time it would systematically assess sleep app data as a potential tool for national epidemiological monitoring. Under the arrangement, no UKHSA data was shared with Sleep Cycle; analysis was conducted on UKHSA’s secure systems by a dedicated agency research team, while Sleep Cycle contributed only anonymised, privacy-preserved and aggregated insights from its user-consented data library.

Existing surveillance systems rely on people seeking care through the NHS, which UKHSA said can be influenced by public awareness, service availability and demographic or socioeconomic differences, as well as reporting and laboratory processing times. The cough signal, by contrast, is generated automatically during normal sleep and updated daily. The preprint reports that the cough data carries a reporting lag of less than one day, compared with a normal lag of several days for existing UKHSA sources.

“No single surveillance system provides a complete picture of respiratory disease activity, but this shows that passive nocturnal cough monitoring can complement other surveillance systems to provide a timely population-level signal of upcoming disease trends, without being affected by healthcare-seeking behaviour, laboratory turnaround times, backfilling and reporting delays,” said Professor Steven Riley, UKHSA’s Chief Data Officer.

Stated Limitations and Next Steps

The authors identify several limitations. The Sleep Cycle user population skews younger and is concentrated in urban areas, particularly London and southern England. The study evaluated retrospective temporal associations rather than prospective predictive performance, and the authors state that demonstrating operational utility would require prospective evaluation against forecasting baselines in a real-time setting. Results for respiratory syncytial virus were inconclusive, which the authors attribute to a shorter available time series and differences between the app’s user population and the groups most affected by severe RSV disease.

The preprint identifies prospective evaluation of nocturnal cough activity within an operational surveillance setting, including integration into existing surveillance dashboards and evaluation across future respiratory seasons, as the next steps for the research.