Sleep patterns and risk of chronic disease as measured by long-term monitoring with commercial wearable devices in the All of Us Research Program

Abstract

Poor sleep health is associated with increased all-cause mortality and incidence of many chronic conditions. Previous studies have relied on cross-sectional and self-reported survey data or polysomnograms, which have limitations with respect to data granularity, sample size and longitudinal information. Here, using objectively measured, longitudinal sleep data from commercial wearable devices linked to electronic health record data from the All of Us Research Program, we show that sleep patterns, including sleep stages, duration and regularity, are associated with chronic disease incidence. Of the 6,785 participants included in this study, 71% were female, 84% self-identified as white and 71% had a college degree; the median age was 50.2 years (interquartile range = 35.7, 61.5) and the median sleep monitoring period was 4.5 years (2.5, 6.5). We found that rapid eye movement sleep and deep sleep were inversely associated with the odds of incident atrial fibrillation and that increased sleep irregularity was associated with increased odds of incident obesity, hyperlipidemia, hypertension, major depressive disorder and generalized anxiety disorder. Moreover, J-shaped associations were observed between average daily sleep duration and hypertension, major depressive disorder and generalized anxiety disorder. These findings show that sleep stages, duration and regularity are all important factors associated with chronic disease development and may inform evidence-based recommendations on healthy sleeping habits.

Fig. 1. Heatmap of sleep onset time and total sleep duration for all person-nights.

Heatmap of sleep duration is plotted versus sleep onset time across all person-nights. All sleep periods were flagged as ‘main sleep’ by Fitbit devices.

Fig. 2. Phenome-wide analyses to explore relation between Fitbit-derived sleep metrics and incident disease.

Phenome-wide analyses were performed to identify associations between each Fitbit-derived sleep metric and incident disease. All phenome-wide analyses were performed using multiple logistic regression models adjusted for age, sex, average daily step count across the entire monitoring period and EHR length. All reported P values were based on two-tailed probability. The Bonferroni significance line of 3.06 × 10⁻⁵ is indicated by a red line and a P value of 0.05 is indicated by the blue line. Upwards pointing triangles indicate OR >1 and downwards pointing triangles indicated OR <1. GERD, gastroesophageal reflux disease; NOS, not otherwise specified.

Fig. 3. Heatmap of significant incident phenotypes across all Fitbit-derived sleep metrics.

Heatmap of −log₁₀(P) is overlaid on a table of significant associations between all incident phenotypes and Fitbit-derived sleep metrics. OR values (95% CI) are reported within each heatmap table box. OR values for average (avg.) sleep, average restless and s.d. of sleep duration are reported per hour change. OR values for average percent sleep stage are reported per 10% change. All reported ORs are from multiple logistic regression models adjusted for age, sex, average daily step count across the entire monitoring period and EHR length. Boxes that include bold text indicate associations that were significant after Bonferroni correction line of 3.06 × 10⁻⁵. All reported P values were based on two-tailed probability. Trad., traditional.

Fig. 4. J-shaped relationship between average daily sleep duration and chronic diseases.

Cox proportional hazard models were used to compute HR values as a function of average daily sleep duration and plotted as blue curves for generalized anxiety disorder, hypertension and major depressive disorder. The median of average daily sleep (6.8 h) was used as the reference. Gray area indicates 95% CI. The red horizontal line indicates an HR of 1. All Cox proportional hazards models were adjusted for age, sex, baseline BMI, baseline systolic blood pressure, smoking status, alcohol drinking status, education status, time-varying average daily step count and previous diagnoses of cancer or coronary artery disease.

Extended Data Fig. 1. CONSORT diagram showing inclusion/exclusion criteria.

The figure shows the inclusion and exclusion criteria for participants and participants-nights for the study. Boxes on the left-hand side of the figure describe the inclusion criteria and boxes on the right-hand side of the figure describe the exclusion criteria.

Extended Data Fig. 2. Population medians of Fitbit-derived sleep metrics from 2017 to 2022.

Collection for Fitbit data for sleep duration, restless sleep duration, and sleep irregularity started prior to 2017 whereas collection for Fitbit-derived sleep stages started in 2017. Point and number indicate the population median. Errors bars indicate the 25th and 75th percentile. Sample sizes were 3,617 for 2017, 4,345 for 2018, 5,319 for 2019, 5,647 for 2020, 5,585 for 2021, and 4,626 for 2022.

Extended Data Fig. 3. Forest plots from stratified analyses of Cox proportional hazard models for associations between 75th vs. 25th percentile Fitbit-derived sleep metric and chronic disease.

The figure shows significant findings from a total of 112 stratified analyses (13 strata 9 significant associations from overall Cox proportional hazard models). The points indicate the hazard ratios and the error bars indicate the 95% confidence intervals. Strata with fewer than 20 cases were excluded to comply with All of Us Data and Statistics dissemination policies. All Cox proportional hazards models were adjusted for age, sex, baseline body mass index (BMI), baseline systolic blood pressure, smoking status, alcohol drinking status, education status, time-varying average daily step count, and prior diagnoses of cancer or coronary artery disease.

Extended Data Fig. 4. Phenome-wide analyses exploring relationship between traditional sleep onset times (8:00 PM to 2:00 AM) and incident disease.

All phenome-wide analyses were performed using multiple logistic regression models adjusted for age, sex, average daily step count across the entire monitoring period, and EHR length. All reported P-values were based on two-tailed probability. Bonferroni significance line of 3.06 × 10–5 is indicated by red line and P-value of 0.05 is indicated by the blue line. Upwards triangles indicate odds ratios > 1 and downwards triangles indicated odds ratios < 1.