This study evaluated aggregated, privacy-preserving smartphone-detected nocturnal cough activity collected via the Sleep Cycle application as a potential syndromic surveillance signal across England. The analysis used weekly cough metrics from January 2023 through January 2026. Sleep Cycle provided only aggregated data protected using differential privacy; no personal identifiers were shared. The authors report that users had accepted the app terms and applicable consents, and that appropriate institutional approvals and exemptions were in place for processing non-identifiable aggregate data.
Three cough measures were compared in the study. The first was total cough counts aggregated weekly. Two population-normalised metrics were also calculated to account for changes in observation volume: coughs per user and coughs per hour of sleep. The paper reports that the normalised metrics were more stable and interpretable than unnormalised total cough counts, which were sensitive to variation in the number of observations.
Cough metrics were compared with routine surveillance indicators published by the UK Health Security Agency (UKHSA). These reference indicators included NHS 111 acute respiratory infection (ARI) triage calls, influenza and COVID-19 PCR positivity, and hospital admission rates for influenza, COVID-19, and respiratory syncytial virus (RSV). The comparison covered the same weekly period from January 2023 to January 2026.
Temporal relationships between cough metrics and each surveillance indicator were assessed nationally and regionally using cross-correlation analysis with prewhitening. Prewhitening was applied to remove shared seasonality, long-term trends, and autocorrelation so that short-term temporal alignment and lead/lag patterns could be more reliably interpreted.
The strongest and most consistent associations were observed between the population-normalised cough metrics and NHS 111 ARI triage calls. National raw correlations for the normalised metrics were reported to be approximately 0.95. After prewhitening, correlations at lag 0 remained above 0.55, indicating that nocturnal cough activity tracked short-term variation in an established syndromic indicator contemporaneously and beyond shared seasonal patterns.
Population-normalised cough metrics also showed epidemiologically plausible leading relationships with pathogen-specific PCR positivity. Specifically, 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. These findings suggest the normalised nocturnal cough signal may precede increases in laboratory-confirmed infection for these pathogens by about one week in the observed period.
Hospital-based indicators showed weaker and more heterogeneous relationships with the cough metrics than syndromic call data. Nevertheless, normalised cough metrics demonstrated plausible temporal alignment with influenza and COVID-19 hospital admissions in some comparisons: contemporaneous associations with influenza admissions and short leading associations with COVID-19 admissions were reported. The study notes hospital indicators are influenced by different care-seeking and reporting pathways, which may explain heterogeneity.
Similar near-contemporaneous patterns between population-normalised cough metrics and NHS 111 ARI triage calls were observed across regions of England. The authors emphasize that regional results showed consistent alignment, supporting the potential utility of aggregated nocturnal cough monitoring at subnational scales.
The article is a preprint and has not undergone peer review; the authors caution that findings should not directly guide clinical practice without further validation. Two authors are employees of Sleep Cycle AB, a declared competing interest. The study relied on app users and passive data collection, which introduces potential biases related to app uptake and user demographics; the source report notes that unnormalised total cough counts were particularly sensitive to variation in observation volume. Details on user representativeness, demographic breakdowns, or absolute user counts were not reported in the source. The authors state that data sharing used differential privacy and that appropriate ethical approvals or exemptions applied.
Within the limits reported, the study concludes that passive, near-real-time monitoring of nocturnal cough via a widely used smartphone application can provide a population-level signal of respiratory symptom burden. The strongest evidence of utility was as a broad syndromic indicator that closely tracks NHS 111 ARI triage calls and shows plausible short leads before pathogen-specific PCR positivity for influenza and COVID-19. The authors suggest such a signal could complement established surveillance systems that are affected by healthcare-seeking behaviour, laboratory turnaround times, backfilling, and reporting delays. Further validation, peer review, and assessment of representativeness would be needed before operational adoption.
This work is a medRxiv preprint posted July 21, 2026, and has not been certified by peer review.