Wearable设备显示睡眠规律性与抑郁、焦虑及孤独感相关联:TemPredict队列研究
Associations between sleep regularity indexed from wearable device data and mental health symptoms in the TemPredict longitudinal cohort study
TemPredict前瞻性队列研究纳66,664名通过Oura Ring采集数据的成年人,采用7天及最长约7个月的可穿戴数据计算睡眠规律指数(SRI),检验其与抑郁症状、焦虑症状及孤独感的关联。初步分析提示,睡眠规律性降低与抑郁症状、焦虑症状及孤独感程度增加相关,结果发表于Nature旗下期刊。该观察性研究揭示了可干预的睡眠节律指标与心理健康的关联。
Introduction
Mental health disorders, such as depression and anxiety, carry profound health and financial burdens, with prevalence rates rising worldwide1 despite increases in treatment utilization2. Additionally, loneliness is a risk factor for developing depressive symptoms, anxiety symptoms, and other mental health problems3,4,5. Thus, identifying and intervening on modifiable risk factors for mental health disorders is a public health priority. Sleep represents one such modifiable risk factor that correlates with mental health disorders and is responsive to behavioral and lifestyle interventions6,7.
Research linking sleep and mental health has historically focused on sleep duration, sleep fragmentation, and perceived sleep quality as risk factors for mental health disorders, such as depression and anxiety8,9. Several datasets suggest that poor sleep precipitates mental health disorders10; however, data also suggest that poor sleep and mental health have bidirectional associations11. For example, anxiety and rumination can disrupt sleep timing and continuity, while disrupted sleep can worsen mood regulation12,13.
Sleep regularity, an index of day-to-day consistency in sleep-wake times, may be an equally or more powerful risk factor for mental health disorders14,15. Sleep regularity, often operationalized as the sleep regularity index (SRI), is an increasingly common metric that quantifies the probability that an individual is in the same sleep-wake state at any two time points 24 h apart16. The SRI thus assesses the stability of circadian rhythms, without incorporating a sleep duration component. Recent large-scale studies suggest that sleep regularity may provide mental health information beyond sleep duration. In accelerometer-based studies, lower sleep regularity has predicted incident depression and anxiety and has shown associations with mental health even after accounting for sleep duration14,15. Most studies, however, have estimated sleep regularity over relatively short periods (e.g., 7 days). Although a 7-day window captures weekday-weekend differences, it may also reflect an atypical week, an acute stressor, travel, illness, or other short-term disruptions.
Commercially available wearable devices offer a way to assess sleep regularity over longer periods of time in real-world settings with minimal burden. Wearable devices have increasingly been established as capable of reliably distinguishing between wakefulness and sleep17,18, and thus hold promise as widely available tools to assess longer-term sleep regularity. Indeed, researchers have begun using data from widely available off-the-shelf wearable devices to study associations between sleep regularity and health. For example, using data from 6,785 adults using commercially available wearable devices, analyses using 4.5 years of data showed that lower sleep regularity, defined as the standard deviation of daily sleep duration, was associated with increased odds of incident major depressive disorder and generalized anxiety disorder, and this finding was independent of sleep duration19. Of note, no prior studies, however, have reported on associations between loneliness, a key risk factor for poor mental health that is substantially more prevalent than mental health disorders20, and sleep regularity. These results support the utility of commercial wearable data, while also leaving open questions about SRI-based sleep regularity metrics computed across longer monitoring periods.
The present analyses addressed gaps left by prior research using data from the TemPredict Study. We first examined associations between sleep regularity and depression symptoms, anxiety symptoms, and loneliness using a 7-day SRI metric designed to align with prior sleep regularity studies14,15. We then examined whether sleep regularity metrics computed from all available wearable data, spanning up to approximately 7 months, showed similar or stronger associations with mental health symptoms. We included loneliness because loneliness represents a clinically relevant correlate and risk marker for mental health problems5 yet has not received attention in sleep regularity research. We hypothesized that lower sleep regularity would be associated with greater depression symptoms, anxiety symptoms, and loneliness.
Methods
Participants
We recruited participants through invitations delivered within the Oura App; these invitations contained embedded weblinks for prospective participants to review study details and to begin the study enrollment process21. Eligible participants were at least 18 years of age, used a smartphone that could pair with the Oura Ring, and could communicate in English. Eligible participants either already had an Oura Ring, or the study provided them with an Oura Ring. We excluded individuals who identified as shift workers (i.e., reported working night or “graveyard” shifts, or rotating shifts that periodically occurred during the night), as they represent a distinct (sleep) sample with previously identified increased risk for depression and anxiety22,23,24. We excluded individuals who did not provide biological sex information or reported biological sex as “other” (n = 25). All participants (N = 66,664) provided electronic informed consent. We did not pay participants for their participation.
Procedures
The TemPredict Study was a prospective, worldwide, cohort study conducted from January 2020 to November 202025,26. While enrolled in the study, participants wore an off-the-shelf wearable device, which assesses several physiological metrics (see wearable device sleep assessment, below), and completed self-report surveys at enrollment and throughout the study period (see measures, below). The University of California, San Francisco, Institutional Review Board and the U.S. Department of Defense, Human Research Protection Office approved all study procedures. This research was performed in accordance with the Declaration of Helsinki. We followed the Strengthening the Reporting of Observational Studies in Epidemiology reporting guidelines27.
Measures
Mental health symptom severity
Depression and anxiety symptoms
We assessed depression and anxiety symptoms in monthly surveys (sent via email) using the Patient Reported Outcomes Measurement Information System (PROMIS) Adult Health Profile instruments28 for depression (v1.0 Form 4a) and anxiety (v1.0 Form 4a), respectively29. Recall periods for these measures reflected the past month, rather than the past 7 days, as we have previously reported30 and consistent with prior work31. Respondents answer items on a 5-point Likert scale ranging from 1 (never) to 5 (always). We summed these scores to obtain raw total scores, which we then converted to T-scores, as described within the PROMIS scoring manuals (healthmeasures.net). PROMIS T-scores are standardized to the general adult population (M = 50, SD = 10), with higher values indicating greater symptom severity.
Loneliness
We assessed loneliness in monthly surveys (sent via email) using the 3-item UCLA Loneliness Scale32 with a recall period of the past month. Respondents answer items on a 3-point Likert scale ranging from 1 (hardly ever) to 3 (often). We summed these scores to obtain raw total scores, with higher values indicating greater loneliness.
Wearable device sleep assessment
Participants wore the Oura Ring Gen2 (Oura Health, Oulu, Finland), an off-the-shelf wearable device that measures a range of physiological metrics, on a finger of their choice. The Oura Ring connects to the Oura App via Bluetooth, and the App is available on the Google Play and Apple App Stores.
The Oura Ring generates physiological metrics, including sleep metrics, by aggregating data gathered from on-device sensors, and has been described extensively elsewhere26,33,34. Of specific relevance for analyses reported here, the device outputs several sleep metrics, including times of sleep onset and awakening (sleep-wake states), and prior research has demonstrated that the Oura Ring Gen2 can reliably distinguish between states of being asleep/awake17,18. We used the time of sleep onset and the time of final awakening for each night in the analyses reported here. We assigned all minutes between these timestamps as asleep when constructing the minute-level sleep-wake states. We assigned minutes outside this main sleep interval as awake, including daytime naps, unless minute-level data were missing. Pairs containing missing sleep-wake states did not contribute to the SRI calculation.
Variable preparation
Mental health symptom severity
Depression and anxiety symptoms
We computed the mean across each PROMIS depression and anxiety T-score from all available monthly surveys to yield one depression and anxiety PROMIS T-score per participant. We used established thresholds for PROMIS depression and anxiety T-scores to categorize symptom severity as within normal limits (T-score < 55; WNL), mild (T-score 55–59.99), moderate (T-score 60–69.99), and severe (T-score ≥ 70)35.
Loneliness
We computed the mean across each UCLA Loneliness Scale score from all available monthly surveys to yield one raw total loneliness score per participant. We used a threshold of 5, as done in prior research36, to categorize raw total scores into categories of within normal limits (raw total score < 5) and elevated (raw total score ≥ 5).
Sleep regularity index (SRI)
We computed SRI using Eq. (1), as described in Fischer et al.37:
$$\:SRI=-100+\frac{200}{M\left(N-1\right)}\sum\:_{j=1}^{M}\sum\:_{i=1}^{N-1}\delta\:\left({s}_{i,j},{s}_{i+1,j}\right)$$
(1)
M is the number of one-minute intervals per day, N is the number of monitored days, i indexes days, j indexes minutes, \(\:{s}_{i,j}\) is the sleep-wake state at minute j on day i (i.e., 0 = sleep; 1 = wake), and \(\:\delta\:\left({s}_{i,j},{s}_{i+1,j}\right)\) equals 1 when the states at the same minute on consecutive days match and 0 otherwise. Only pairs of minute-level sleep-wake states exactly 24 h apart in which both states were observed (i.e., not missing) contributed to the SRI calculation. Pairs containing missing data were excluded from both the numerator and denominator, and missing states were not imputed. As done in prior research14,38, the SRI calculation did not use sleep episodes occurring outside of the main daily sleep episode.
The SRI typically ranges from 0 to 100, with 0 indicating 50% agreement between the timing of sleep and wake states across days and 100 indicating 100% agreement between sleep and wake states. Nearly all SRI values fall within the range of 0-100; however, the full theoretical range is -100 to 100, as SRI values less than 0 are theoretically possible if one were to stay awake for a full 24-hour period and then sleep for a full 24-hour period. That is, complete disagreement (inversion) of the sleep-wake schedule would generate an SRI of -100.
A lower SRI indicates greater misalignment in sleep-wake states across days, meaning that the sleep-wake state at a given clock time is more likely to differ from one day to the next. A higher SRI indicates a more consistent sleep pattern.
SRI metrics
We computed two SRI metrics: “All Days,” which included all available days with at least 16 h of non-missing data, and “Typical Week,” which comprised the first 7-day period before March 15th, 2020 (i.e., before COVID-related restrictions), including at least 3 weekdays and 2 weekend days with at least 16 h of non-missing data, as done in prior work38. Similarly, we ensured that the All Days metric included at least 3 weekdays and 2 weekend days with at least 16 h of non-missing data. We computed a 7-day metric to replicate methods performed in prior analyses of associations between SRI and health outcomes15,38,39.
SRI variability
We used all available days of data (All Days) to compute a rolling 7-day SRI using a 1-day forward step that included all 7-day periods that met criteria for inclusion for Typical Week analyses (i.e., a minimum of 3 weekdays and 2 weekend days). For any 7-day period not meeting criteria, we moved forward by 1 day at a time until reaching the next 7-day period that met criteria. For each participant, we computed the standard deviation (SD) across all SRI values computed using the 7-day periods as a measure of SRI variability.
Analytic plan
We used Python 3.12.840 for all statistical analyses and to create figures.
Participant characteristics
We computed means with SD, and frequencies with percentages, to describe the analytic sample.
Analyses by SRI metric
For inclusion in All Days analyses, we required at least one contiguous seven-day period containing at least three weekdays and two weekend days with ≥ 16 h of non-missing data per day. The qualifying days did not need to be consecutive but had to fall within the same seven-day period. We computed All Days SRI using all eligible days across participation, restricting comparisons to minute-level sleep-wake states exactly 24 h apart; nonadjacent days were not compared across gaps.
For inclusion in Typical Week analyses, we required that each participant’s 7-day period include at least 3 weekdays and 2 weekend days. Thus, the calendar dates defining the period were consecutive, but not all 7 days were required to meet the 16-hour non-missing data threshold. Days failing to meet this 16-hour threshold and pairs of minute-level sleep-wake states containing missing data within otherwise valid days, did not contribute to the SRI calculation.
To assess SRI variability, we computed and plotted the SRI SD.
SRI and mental health
SRI values by mental health symptom severity category
We plotted and assessed the associations between the SRI and mental health symptom severity for each outcome (depression, anxiety, and loneliness). We used Welch’s t-tests to compare SRI between the WNL level for each mental health outcome and each level of symptom severity (i.e., mild, moderate, severe). We report means (SD) for each category, and mean differences with 95% CI, and Cohen’s d. To facilitate comparison with previously published research on SRI41,42, we also repeated these analyses using non-parametric statistical testing (i.e., Mann-Whitney U tests) where we reported median (interquartile ranges [IQR]) with common language effect sizes (CLES).
Odds of mental health symptom severity by SRI
Although t-tests provide direct comparisons of mean SRI across symptom severity categories, logistic regression permits covariate adjustment and provides odds ratios commonly used in clinical research. In these logistic analyses only, we reversed SRI scores (i.e., 100 - SRI) to produce odds ratios > 1 for ease of interpretation. Values > 1 indicate greater odds of worse mental health symptom severity for each 5-point decrease in SRI. We fitted maximum-likelihood logistic regression models with categorical mental health severity as the outcome (WNL vs. mild, moderate, or severe for depression and anxiety; WNL loneliness vs. elevated loneliness) for each SRI metric. We fitted both unadjusted models and adjusted models, which statistically accounted for age (in years), biological sex (binary), and mean sleep duration (in hours) across all of the days used in the SRI metric. We excluded participants with missing data for any model covariate from the binary logistic regression analyses; consequently, we fit all models using complete covariate data. We assessed assumptions regarding independent observations, multicollinearity, and linearity in the logit.
We further compared the All Days and Typical Week metrics on a common scale and within the same sample: we restricted the All Days analysis to participants eligible for the Typical Week analysis (n = 11,895), standardized each SRI metric within this cohort, and repeated the unadjusted and adjusted logistic regression analyses with odds ratios expressed per 1-SD decrease in SRI. We reported odds ratios with 95% CIs and visualized this in forest plots.
SRI variability
In addition to plotting the overall SRI variability using the rolling 7-day SRI in the All Days sample, we also plotted SRI variability stratified by mental health domain and mental health symptom severity category. We used Mann-Whitney U tests to compare SRI variability between the WNL level for each mental health outcome and each level of symptom severity (i.e., mild, moderate, severe). We report medians [IQR] for each category and CLES.
Odds of mental health symptom severity by SRI variability
We used the same binary logistic regressions described above to evaluate whether associations with SRI variability remained after accounting for average SRI and the number of observations contributing to each variability estimate. We estimated the odds of each mental health symptom severity category relative to WNL per 1-SD increase in rolling SRI SD. Unadjusted models included rolling SRI SD alone. We included the mean rolling SRI and the number of valid rolling windows in the adjusted models, along with the covariates described above. We reported odds ratios with 95% CIs and visualized this in forest plots.
Results
Participant characteristics
Participant flow into the study is shown in Fig. 1, and participant characteristics appear in Table 1. Survey assessments spanned April 4th, 2020, to November 25th, 2020, and wearable data spanned January 1st, 2020, to October 24th, 2020.
SRI characteristics
Participants contributed an average of 153.8 days with at least 16 h of data and 128.0 valid 7-day windows. The mean (SD) SRI for All Days was 83.1 (7.3) (Fig. 2A). The mean (SD) SRI for the Typical Week was 80.8 (11.0) (Fig. 2B). The mean of the distribution of SD using 7-day rolling periods was 6.6 (Fig. 3A). Sample sizes stratified by mental health outcomes and SRI metric are described in Supplementary Table 1.
Sleep regularity index and mental health
SRI values by mental health symptom severity category
T-tests generally showed significant associations between lower mean SRI values and greater mental health outcome symptom severity in a graded fashion, and we observed this pattern in both All Days and Typical Week analyses (Table 2). For example, participants with severe depression or anxiety symptoms, respectively, had lower SRI values relative to participants with levels WNL (Figs. 2C-F). This pattern was similar for loneliness (Fig. 2G and H). Exceptions were the Typical Week comparisons of mild and severe anxiety with WNL (p = 0.062 and p = 0.059, respectively). Cohen’s d values ranged from -1.32 to -0.13 for All Days and from -0.80 to -0.04 for Typical Week, with the largest difference observed for severe depressive symptoms using All Days (mean difference, -8.81 SRI points; 95% CI, -11.50 to -6.12; d = -1.32). Non-parametric statistical testing and CLES yielded similar results (Supplementary Table 2).
Odds of mental health symptom severity by SRI
Logistic regressions showed that for each 5-point decrease in SRI, individuals had greater odds of membership in a more severe mental health symptom category (Figs. 4 and 5). Although we observed this pattern in both All Days and Typical Week analyses, analyses using more data (All Days) demonstrate stronger associations. As shown in Figs. 4 and 5, these associations remained similar after adjusting for covariates: Across adjusted models, odds ratios ranged from 1.091 (95% CI, 1.067–1.116) to 1.457 (95% CI, 1.339–1.587) for All Days and from 1.018 (95% CI, 0.998–1.039) to 1.196 (95% CI, 1.074–1.332) for Typical Week.
In analyses restricted to the Typical Week cohort and expressed per 1-SD decrease in SRI, odds ratios remained generally larger for the All Days metric than for Typical Week (Figs. 6 and 7). Model assumptions regarding independent observations and lack of meaningful multicollinearity were satisfied. Linearity in the logit was not uniformly supported, with evidence of departures primarily among the All Days models.
Participation flow.
Distributions of overall SRI (A, B), and SRI by mental health outcomes (C-H) by SRI metric (All Days and Typical Week). *All Days and Typical Week are defined in “Methods”. Dashed lines denote distribution means.
Distribution of standard deviation of SRI (dashed lines represent distribution means) using rolling 7-day periods from all available days of data (~ 7 months), stratified by outcome metric severity.
Forest plots depicting odds of mental health symptom severity compared to mental health symptom scores WNL per each 5-point decrease in SRI computed using All Days metric. Note: We reversed SRI scores (i.e., 100 - SRI) to produce odds ratios > 1 for ease of interpretation. Values > 1 indicate greater odds of worse mental health symptom severity, compared to mental health symptom scores WNL, per each 5-point decrease in SRI. Adjusted models account for biological sex, age, and mean sleep duration across all days used within the All Days SRI metric.
Odds of mental health symptom severity per each 5-point decrease in SRI computed using Typical Week metric. See Fig. 4 note. Adjusted models account for biological sex, age, and mean sleep duration across all days used within the Typical Week SRI metric.
Odds of mental health symptom severity per each 1-standard deviation decrease in SRI computed using All Days metric among only participants eligible for the Typical Week analysis. See Fig. 4 note. Values > 1 indicate greater odds of worse mental health symptom severity, compared to mental health symptom scores WNL, per each 1-standard deviation decrease in SRI. Adjusted models account for biological sex, age, and mean sleep duration across all days used within the All Days SRI metric.
Odds of mental health symptom severity per each 1-standard deviation decrease in SRI computed using Typical Week metric among only participants eligible for the Typical Week analysis. See Fig. 4 note. Values > 1 indicate greater odds of worse mental health symptom severity, compared to mental health symptom scores WNL, per each 1-standard deviation decrease in SRI. Adjusted models account for biological sex, age, and mean sleep duration across all days used within the Typical Week SRI metric.
Odds of mental health symptom severity per each 1-standard deviation increase in rolling 7-day SRI variability computed using All Days metric. Values > 1 indicate greater odds of worse mental health symptom severity, compared with mental health symptom scores WNL, per each 1-standard deviation increase in rolling 7-day SRI SD. Adjusted models account for participant-level mean rolling SRI, the number of valid rolling windows, biological sex, age, and mean sleep duration.
SRI variability
A plot of SD using the rolling 7-day SRI showed that across mental health outcomes (i.e., depression, anxiety, and loneliness), greater variability in SRI was associated with more severe mental health symptoms (Fig. 3B, C and D). CLES ranged from 0.52 to 0.68 across the rolling SRI SD comparisons, with the largest difference observed for severe depressive symptoms (Supplementary Table 3). These unadjusted comparisons should be interpreted alongside the adjusted binary logistic regression models below, in which the positive associations did not persist after accounting for mean rolling SRI and other covariates.
Odds of mental health symptom severity by SRI variability
In unadjusted logistic regression models, each 1-SD increase in rolling SRI SD was associated with greater odds of membership in each mental health symptom severity category relative to WNL (Fig. 8). Results were similar when the number of valid rolling windows was added to models adjusted for age, biological sex, and mean sleep duration. However, after additionally adjusting for mean rolling SRI, none of the positive associations remained: Greater rolling SRI SD was associated with lower odds of mild depressive symptoms (OR = 0.926; 95% CI, 0.881–0.974), moderate depressive symptoms (OR = 0.911; 95% CI, 0.858–0.967), and moderate anxiety symptoms (OR = 0.890; 95% CI, 0.839–0.943). Rolling SRI SD was not associated with severe depressive symptoms, mild or severe anxiety symptoms, or elevated loneliness. Mean rolling SRI and rolling SRI SD were strongly inversely correlated (Spearman’s ρ = -0.742).
Discussion
In this large wearable-device cohort of 22,894 adults assessed over 7 months, we found that lower sleep regularity correlated with greater depression, anxiety, and loneliness. Additionally, we observed progressively lower mean SRI values with increasing depression and anxiety severity (i.e., mild, moderate, severe). This graded pattern underscores the potential that sleep regularity carries meaningful information about mental health symptom burden. We observed this pattern across both SRI computations: SRI computed using all available days of data (All Days) and computed using a conventional 7-day SRI metric (Typical Week). Logistic regression models showed a similar pattern: Each 5-point decrease in SRI corresponded to greater odds of membership in worse mental health symptom severity categories in both unadjusted and adjusted models. Effect sizes varied across analyses but were generally larger for more severe symptom categories and for All Days than for Typical Week. Taken together, these results support the hypothesis that lower sleep regularity correlates with greater depression, anxiety, and loneliness.
Results add to a growing literature suggesting that sleep regularity, regardless of sleep duration, correlates with mental health. We used both a conventional 7-day SRI metric16 as well as SRI metrics based on longer time periods and using rolling windows, which may provide more stable estimates of sleep regularity than a given week in time. The All Days metric characterizes sleep regularity over months rather than days. The All Days analyses generally showed stronger associations with mental health metrics than did the Typical Week analyses. The Typical Week metric facilitated comparisons with prior 7-day actigraphy studies (e.g.14,15), and provided a snapshot of regularity prior to COVID-related societal disruptions. Finally, the rolling 7-day analyses initially suggested that greater SRI variability was associated with more severe mental health symptoms in unadjusted comparisons, but these positive associations did not persist after adjustment for mean rolling SRI, the number of valid windows, and other covariates. Participants with lower mean rolling SRI tended to have greater SRI variability (ρ = -0.742), making it difficult to separate their associations with mental health symptoms. The reversed estimates should therefore not be interpreted as evidence that greater SRI variability is protective. Although All Days analyses generally showed stronger associations between SRI and mental health symptoms, the metrics differed in calendar coverage and temporal overlap with symptom assessments. Overall, these associations between longer-term sleep regularity metrics and mental health provided further support for those previously observed with shorter-term sleep regularity metrics (e.g.14,15).
Although prior large-scale studies quantifying sleep regularity using research-grade actigraphy and accelerometry devices have linked lower sleep regularity with subsequent depression and anxiety14,15, results presented here used data from a commercially available wearable device to assess sleep regularity. Using widely available wearable devices that are less expensive and more sustainably used than research-grade devices holds the potential for widespread assessment of, and intervention on, sleep regularity. Commercially available wearable devices pair with smartphone apps that provide behavioral advice to users, and therefore could both gather sleep regularity information and provide users with data-driven recommendations to increase sleep regularity, such as ideal bed and wake times. Additionally, rather than solely assessing correlations with depression and anxiety, we also assessed correlations with loneliness, which has received less attention in sleep regularity research despite its status as a risk factor for poor mental health5,43. Using wearable devices, which passively collect data, to identify behavioral risk factors for mental health states (e.g., loneliness) that confer risk for mental health symptoms (e.g., depression and/or anxiety symptoms) may allow for earlier intervention.
Several plausible pathways may link irregular sleep-wake timing and worse mental health. Irregular sleep-wake timing may weaken circadian alignment, which has been associated with elevated pro-inflammatory signaling44; such inflammatory signaling has been correlated with depression and other neuropsychiatric disorders45,46. Irregular sleep-wake patterns may also expose individuals to inconsistent patterns of environmental stimuli (e.g., light) and inconsistent behavioral routines, such as eating meals, socializing, and engaging in physical activity. Such irregularity can disrupt circadian rhythms47. At the same time, mental health factors can disrupt sleep regularity: For example, anxiety and rumination may contribute to difficulties with sleep onset or middle-of-the-night awakening12,13. Taken together, the results presented here align with prior data showing that improving sleep quality can improve mental health7 and highlight that the meaningful metric of sleep regularity can be easily accessed via widely available technologies (i.e., wearable devices).
These findings have clinical and public health relevance, as people can understand and potentially modify their sleep regularity. In the context of behavioral sleep medicine, clinicians often encourage patients to maintain a consistent wake time and to protect a consistent sleep schedule by engaging in wind-down routines before a target bedtime48. These recommendations represent measurable behavioral targets that commercially available wearable devices can track over long periods. Future work should test whether feedback about sleep regularity parameters, such as wake times and bedtimes, helps people make sustainable changes in sleep-wake routines and whether those changes improve mental health.
These results should be considered in light of several limitations. Because adjacent rolling windows shared data, their SRI estimates were correlated and may have smoothed short-term changes in SRI; however, using overlapping windows yielded a greater number of weekly SRI estimates per participant and allowed us to better characterize changes across the monitoring period. Our cross-sectional analyses did not determine whether lower sleep regularity contributes to worse mental health, whether mental health symptoms disrupt sleep regularity, or whether both processes occur. The TemPredict Study cohort comprised primarily Oura Ring users, who tended to be White, middle-aged adults from North America and Europe, which may limit generalizability. Although the Oura Ring is a commercially available wearable device and not a medical device, which may be considered a limitation, we believe that the sensitivity of such a device to detect these associations, which have been previously reported in studies using medical devices14,15,49, is actually a notable strength. As has been done in other studies of sleep regularity and health, we computed the SRI from the main sleep episode and did not include other sleep episodes (e.g., naps). We treated the interval from sleep onset to the final awakening as sleep, without accounting for wake after sleep onset. Consequently, our SRI estimates did not capture day-to-day differences in the timing or duration of awakenings within the main sleep episode and may underestimate irregularity among individuals with fragmented or polyphasic sleep. Additionally, we did not statistically account for a variety of potential confounders, such as chronotype, living situation (e.g., alone versus with others), relationship status, and other potential confounders. Finally, some statistically significant differences, especially in mild symptom categories, showed small magnitudes, so future work should determine whether specific SRI differences have clinical meaning.
Taken together, these findings indicate that sleep regularity, as measured passively in the real world using a commercially available wearable device, is meaningfully associated with mental health. These exploratory analyses found that greater SRI variability was associated with greater mental health symptom severity in unadjusted models, but these positive associations did not persist after adjustment for mean rolling SRI and other covariates, and some estimates reversed direction. These findings suggest that lower SRI is a more consistent correlate of more severe mental health symptoms. These analyses contribute to a fast-growing body of literature linking sleep regularity to mental health outcomes.
Data availability
Oura’s data use policy does not permit us to make wearable device data (collected via the Oura Ring) available to third parties. Access to anonymized and privacy-protected data may be granted to a qualified academic investigator upon completing agreements with Oura Health Oy and the investigators. Please contact Ashley E. Mason and Benjamin L. Smarr to obtain an application to obtain these data.
References
Zhang, Z. et al. Global, regional and national burden of anxiety and depression disorders from 1990 to 2021, and forecasts up to 2040. J. Affect. Disord. 393 (Pt A), 120299. https://doi.org/10.1016/j.jad.2025.120299 (2026).
Zhao, S. et al. National trends in depressive symptoms, antidepressant use, and mental health consultations among US adults, 2005–2020. Public Health 247, 105916. https://doi.org/10.1016/j.puhe.2025.105916 (2025).
Weziak-Bialowolska, D. et al. Prospective associations between social connectedness and mental health. Evidence from a longitudinal survey and health insurance claims data. Int. J. Public Health 67, 1604710. https://doi.org/10.3389/ijph.2022.1604710 (2022).
Liu, Q. et al. Social isolation, loneliness, and the onset and progression of depression and anxiety: The mediating role of peripheral immune biomarkers. Psychoneuroendocrinology 192, 107977. https://doi.org/10.1016/j.psyneuen.2026.107977 (2026).
Mann, F. et al. Loneliness and the onset of new mental health problems in the general population. Soc. Psychiatry Psychiatr. Epidemiol. 57 (11), 2161–2178. https://doi.org/10.1007/s00127-022-02261-7 (2022).
Firth, J. et al. A meta-review of lifestyle psychiatry: the role of exercise, smoking, diet and sleep in the prevention and treatment of mental disorders. World Psychiatry Off. J. World Psychiatr. Assoc. 19 (3), 360–380. https://doi.org/10.1002/wps.20773 (2020).
Scott, A. J., Webb, T. L., Martyn-St James, M., Rowse, G. & Weich, S. Improving sleep quality leads to better mental health: A meta-analysis of randomised controlled trials. Sleep. Med. Rev. 60, 101556. https://doi.org/10.1016/j.smrv.2021.101556 (2021).
Freeman, D., Sheaves, B., Waite, F., Harvey, A. G. & Harrison, P. J. Sleep disturbance and psychiatric disorders. Lancet Psychiatry. 7 (7), 628–637. https://doi.org/10.1016/S2215-0366(20)30136-X (2020).
van Mill, J. G., Hoogendijk, W. J. G., Vogelzangs, N., van Dyck, R. & Penninx, B. W. J. H. Insomnia and sleep duration in a large cohort of patients with major depressive disorder and anxiety disorders. J. Clin. Psychiatry. 71 (3), 239–246. https://doi.org/10.4088/JCP.09m05218gry (2010).
Zhang, M. M. et al. Sleep disorders and non-sleep circadian disorders predict depression: A systematic review and meta-analysis of longitudinal studies. Neurosci. Biobehav Rev. 134, 104532. https://doi.org/10.1016/j.neubiorev.2022.104532 (2022).
Alvaro, P. K., Roberts, R. M. & Harris, J. K. A systematic review assessing bidirectionality between sleep disturbances, anxiety, and depression. Sleep 36 (7), 1059–1068 (2013).
Tomaso, C. C., Johnson, A. B. & Nelson, T. D. The effect of sleep deprivation and restriction on mood, emotion, and emotion regulation: Three meta-analyses in one. Sleep 44 (6), zsaa289. https://doi.org/10.1093/sleep/zsaa289 (2021).
Clancy, F., Prestwich, A., Caperon, L., Tsipa, A. & O’Connor, D. B. The association between worry and rumination with sleep in non-clinical populations: A systematic review and meta-analysis. Health Psychol. Rev. 14 (4), 427–448. https://doi.org/10.1080/17437199.2019.1700819 (2020).
Li, D. R. et al. Regular sleep patterns, not just duration, critical for mental health: Association of accelerometer-derived sleep regularity with incident depression and anxiety. Psychol. Med. 55, e239. https://doi.org/10.1017/S0033291725101281 (2025).
Moebus, M. & Holz, C. Beyond the hours slept: Inconsistent sleep routines threaten mental health in 100,000 UK Biobank participants. BMC Public Health 25 (1), 4009. https://doi.org/10.1186/s12889-025-24794-7 (2025).
Phillips, A. J. K. et al. Irregular sleep/wake patterns are associated with poorer academic performance and delayed circadian and sleep/wake timing. Sci. Rep. 7 (1), 3216. https://doi.org/10.1038/s41598-017-03171-4 (2017).
Miller, D. J., Sargent, C. & Roach, G. D. A validation of six wearable devices for estimating sleep, heart rate and heart rate variability in healthy adults. Sensors 22 (16), 6317. https://doi.org/10.3390/s22166317 (2022).
Roberts, D. M., Schade, M. M., Mathew, G. M., Gartenberg, D. & Buxton, O. M. Detecting sleep using heart rate and motion data from multisensor consumer-grade wearables, relative to wrist actigraphy and polysomnography. Sleep 43 (7), zsaa045. https://doi.org/10.1093/sleep/zsaa045 (2020).
Zheng, N. S. et al. Sleep patterns and risk of chronic disease as measured by long-term monitoring with commercial wearable devices in the All of Us Research Program. Nat. Med. 30 (9), 2648–2656. https://doi.org/10.1038/s41591-024-03155-8 (2024).
Abdalla, S. M. et al. Loneliness, depression, and generalized anxiety across eight countries. Soc. Psychiatry Psychiatr. Epidemiol. 61 (5), 909–924. https://doi.org/10.1007/s00127-025-03029-5 (2026).
Qualtrics, X. M. Experience Management Software. https://www.qualtrics.com/.
Sweeney, E. et al. The association between mental health and shift work: Findings from the Atlantic PATH study. Prev. Med. 150, 106697. https://doi.org/10.1016/j.ypmed.2021.106697 (2021).
Torquati, L., Mielke, G. I., Brown, W. J., Burton, N. W. & Kolbe-Alexander, T. L. Shift work and poor mental health: A meta-analysis of longitudinal studies. Am. J. Public Health. 109 (11), e13–e20. https://doi.org/10.2105/AJPH.2019.305278 (2019).
Xu, M., Yin, X. & Gong, Y. Lifestyle factors in the association of shift work and depression and anxiety. JAMA Netw. Open 6 (8), e2328798. https://doi.org/10.1001/jamanetworkopen.2023.28798 (2023).
Mason, A. E. et al. Detection of COVID-19 using multimodal data from a wearable device: Results from the first TemPredict study. Sci. Rep. 12 (1), 3463. https://doi.org/10.1038/s41598-022-07314-0 (2022).
Smarr, B. L. et al. Feasibility of continuous fever monitoring using wearable devices. Sci. Rep. 10 (1), 1. https://doi.org/10.1038/s41598-020-78355-6 (2020).
von Erik, M. et al. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: Guidelines for reporting observational studies. Ann. Intern. Med. 147 (8), 573 (2007).
Cella, D. et al. The patient-reported outcomes measurement information system (PROMIS) developed and tested its first wave of adult self-reported health outcome item banks: 2005–2008. J. Clin. Epidemiol. 63 (11), 1179–1194. https://doi.org/10.1016/j.jclinepi.2010.04.011 (2010).
Pilkonis, P. A. et al. Item banks for measuring emotional distress from the Patient-Reported Outcomes Measurement Information System (PROMIS®): Depression, anxiety, and anger. Assessment 18 (3), 263–283. https://doi.org/10.1177/1073191111411667 (2011).
Mason, A. E. et al. Elevated body temperature is associated with depressive symptoms: Results from the TemPredict study. Sci. Rep. 14 (1), 1. https://doi.org/10.1038/s41598-024-51567-w (2024).
Batterham, P. J., Sunderland, M., Carragher, N. & Calear, A. L. Psychometric properties of 7- and 30-day versions of the PROMIS emotional distress item banks in an Australian adult sample. Assessment 26 (2), 249–259. https://doi.org/10.1177/1073191116685809 (2019).
Hughes, M. E., Waite, L. J., Hawkley, L. C. & Cacioppo, J. T. A short scale for measuring loneliness in large surveys: Results from two population-based studies. Res. Aging 26 (6), 655–672 (2004).
de Zambotti, M., Rosas, L., Colrain, I. M. & Baker, F. C. The sleep of the ring: Comparison of the ŌURA sleep tracker against polysomnography. Behav. Sleep Med. 17 (2), 124–136 (2019).
Altini, M. & Kinnunen, H. The promise of sleep: A multi-sensor approach for accurate sleep stage detection using the oura ring. Sensors 21 (13), 13. https://doi.org/10.3390/s21134302 (2021).
Kroenke, K. et al. Minimally important differences and severity thresholds are estimated for the PROMIS depression scales from three randomized clinical trials. J. Affect. Disord. 266, 100–108. https://doi.org/10.1016/j.jad.2020.01.101 (2020).
Matthews, T. et al. Using a loneliness measure to screen for risk of mental health problems: A replication in two nationally representative cohorts. Int. J. Environ. Res. Public Health 19 (3). https://doi.org/10.3390/ijerph19031641 (2022).
Fischer, D., Klerman, E. B. & Phillips, A. J. K. Measuring sleep regularity: Theoretical properties and practical usage of existing metrics. Sleep 44 (10), zsab103. https://doi.org/10.1093/sleep/zsab103 (2021).
Chaput, J. P. et al. Sleep regularity and major adverse cardiovascular events: a device-based prospective study in 72 269 UK adults. J. Epidemiol. Commun. Health 79 (4), 257–264 (2025).
Windred, D. P. et al. Sleep regularity is a stronger predictor of mortality risk than sleep duration: A prospective cohort study. Sleep 47 (1), zsad253. https://doi.org/10.1093/sleep/zsad253 (2024).
Van Rossum, G. & Drake Jr, F. L. Python Tutorial. Vol. 620 (Centrum voor Wiskunde en Informatica Amsterdam, 1995).
Lunsford-Avery, J. R., Engelhard, M. M., Navar, A. M. & Kollins, S. H. Validation of the sleep regularity index in older adults and associations with cardiometabolic risk. Sci. Rep. 8 (1), 14158. https://doi.org/10.1038/s41598-018-32402-5 (2018).
Castiglione-Fontanellaz, C. E. G. et al. Sleep regularity in healthy adolescents: Associations with sleep duration, sleep quality, and mental health. J. Sleep Res. 32 (4), e13865. https://doi.org/10.1111/jsr.13865 (2023).
Buecker, S. & Neuber, A. Loneliness as health risk: A narrative review. Bundesgesundheitsblatt Gesundheitsforschung Gesundheitsschutz 67 (10), 1095–1102. https://doi.org/10.1007/s00103-024-03939-w (2024).
Kalkanis, A., Lenkens, D., Steiropoulos, P. & Testelmans, D. Sleep regularity as an important component of sleep hygiene: A systematic review. Sleep Med. Rev. 84, 102203. https://doi.org/10.1016/j.smrv.2025.102203 (2025).
Upthegrove, R., Corsi-Zuelli, F., Couch, A. C. M., Barnes, N. M. & Vernon, A. C. Current position and future direction of inflammation in neuropsychiatric disorders: A review. JAMA Psychiatry 82 (10), 1030–1046. https://doi.org/10.1001/jamapsychiatry.2025.1369 (2025).
Malhi, G. S. & Mann, J. J. Depression. Lancet 392(10161), 2299–2312. https://doi.org/10.1016/S0140-6736(18)31948-2 (2018).
Vetter, C. Circadian disruption: What do we actually mean? Eur. J. Neurosci. 51 (1), 531–550. https://doi.org/10.1111/ejn.14255 (2020).
Edinger, J. D. et al. Behavioral and psychological treatments for chronic insomnia disorder in adults: An American Academy of Sleep Medicine clinical practice guideline. J. Clin. Sleep. Med. JCSM Off. Publ. Am. Acad. Sleep. Med. 17 (2), 255–262. https://doi.org/10.5664/jcsm.8986 (2021).
Difrancesco, S. et al. Sleep, circadian rhythm, and physical activity patterns in depressive and anxiety disorders: A 2-week ambulatory assessment study. Depress. Anxiety. 36 (10), 975–986. https://doi.org/10.1002/da.22949 (2019).
Funding
This effort was funded under Medical Technology Enterprise Consortium (MTEC) solicitation MTEC-20-12-Diagnostics-023 and the USAMRDC under the Department of Defense (#MTEC-20-12-COVID19-D.-023). The #StartSmall Foundation (#7029991) and Oura Health Oy (#134650) also provided funding for this work. The views expressed in this manuscript are those of the author(s) and do not necessarily reflect the official policy of the Department of Defense or the U.S. Government.
Author information
Author notes
Benjamin L. Smarr and Ashley E. Mason contributed equally to this work.
Authors and Affiliations
Bioinformatics and Systems Biology Graduate Program, University of California San Diego, La Jolla, USA
Severine Soltani
Shu Chien-Gene Lay Department of Bioengineering, University of California San Diego, La Jolla, USA
Severine Soltani & Benjamin L. Smarr
Osher Center for Integrative Health, University of California San Francisco, 1545 Divisadero St, 4th Floor, San Francisco, CA, 94115, USA
Wendy Hartogensis, Frederick M. Hecht & Ashley E. Mason
Zicklin School of Business, Baruch College, The City University of New York, New York, USA
Stephan Dilchert
Halıcıoğlu Data Science Institute, University of California San Diego, La Jolla, USA
Benjamin L. Smarr
Department of Psychiatry and Behavioral Sciences, University of California San Francisco, San Francisco, USA
Ashley E. Mason
Authors
- Severine Soltani
- Wendy Hartogensis
- Stephan Dilchert
- Frederick M. Hecht
- Benjamin L. Smarr
- Ashley E. Mason
Contributions
Conceptualization: A.E.M., S.S., W.H.; Methodology: A.E.M., S.S., W.H., F.M.H.; Formal analysis: S.S., W.H., B.L.S.; Data curation: S.S., W.H., S.D., B.L.S.; Visualization: S.S., W.H., B.L.S.; Writing – original draft: A.E.M., S.S.; Writing – review & editing: all authors; Supervision: A.E.M.; Project administration: A.E.M.; Funding acquisition: A.E.M. All authors reviewed and approved the final manuscript.
Corresponding author
Correspondence to Ashley E. Mason.
Ethics declarations
Competing interests
A.E.M. has received remuneration for consulting work from Ouraring Inc. but declares no non-financial competing interests. B.L.S. has received remuneration for consulting work from, and has a financial interest in, Ouraring Inc. but declares no other non-financial competing interests.
Additional information
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Soltani, S., Hartogensis, W., Dilchert, S. et al. Associations between sleep regularity indexed from wearable device data and mental health symptoms in the TemPredict longitudinal cohort study. Sci Rep 16, 36770 (2026). https://doi.org/10.1038/s41598-026-77545-y
Received: 03 July 2026
Accepted: 21 October 2026
Published: 12 November 2026
Version of record: 12 November 2026
DOI: https://doi.org/10.1038/s41598-026-77545-y
Keywords
来源:Nature 健康护理主题 · nature.com