Abstract
Study Objectives
To examine the associations between sleep duration, continuity, timing, and mortality using actigraphy among adults.
Methods
Data were from a cohort of 88 282 adults (40–69 years) in UK Biobank that wore a wrist-worn triaxial accelerometer for 7 days. Actigraphy data were processed to generate estimates of sleep duration and other sleep characteristics including wake after sleep onset (WASO), number of 5-minute awakenings, and midpoint for sleep onset/wake-up and the least active 5 hours (L5). Data were linked to mortality outcomes with follow-up to October 31, 2021. We implemented Cox models (hazard ratio, confidence intervals [HR, 95% CI]) to quantify sleep associations with mortality. Models were adjusted for demographics, lifestyle factors, and medical conditions.
Results
Over an average of 6.8 years 2973 deaths occurred (1700 cancer, 586 CVD deaths). Overall sleep duration was significantly associated with risk for all-cause (p < 0.01), cancer (p < 0.01), and CVD (p = 0.03) mortality. For example, when compared to sleep durations of 7.0 hrs/d, durations of 5 hrs/d were associated with a 29% higher risk for all-cause mortality (HR: 1.29 [1.09, 1.52]). WASO and number of awakenings were not associated with mortality. Individuals with L5 early or late midpoints (<2:30 or ≥ 3:30) had a ~20% higher risk for all-cause mortality, compared to those with intermediate L5 midpoints (3:00–3:29; p ≤ 0.01; e.g. HR ≥ 3:30: 1.19 [1.07, 1.32]).
Conclusions
Shorter sleep duration and both early and late sleep timing were associated with a higher mortality risk. These findings reinforce the importance of public health efforts to promote healthy sleep patterns in adults.
Keywords: epidemiology, UK Biobank, adults, accelerometry
Graphical Abstract
Graphical Abstract.
Statement of Significance.
Disrupted sleep patterns including short sleep duration, insomnia, and inconsistent sleep timing resulting from nightshift work, have all been shown to be associated with an increased risk for mortality. However, it remains uncertain whether milder forms of sleep disruption as that experienced by more generalized samples of adults who do not engage in shift work may also have an increased mortality risk. Leveraging wearable data in large prospective cohorts can fill this gap and facilitate the interpretation of sleep quantities obtained from commercial wearables providing valuable information for adults to improve their health. Our study addressed these gaps by examining the associations between wearable-measured sleep characteristics and mortality in a sample of 88 000 adults.
Introduction
Professional sleep societies [1, 2] recommend adults sleep 7–9 hours per night for better health, including a lower risk of early mortality. Studies have also shown that poor sleep quality [3, 4] and inconsistent sleep timing as a result of shift work or lifestyle are also linked to an elevated risk for metabolic disease and mortality [5, 6]. However, a large body of this evidence has been generated from self-reported characterizations of sleep (e.g. questionnaires) and may not be able to be directly translated to sleep amounts recorded by consumer wearables now available to millions of adults worldwide. The mortality associations with sleep duration have also been inconsistent across populations [7, 8] and these mixed findings can be partially attributed to the subjective nature of self-reported sleep measures used in these studies [9]. Leveraging actigraphy data in large cohorts may help clarify the sleep-mortality association across populations while providing sleep amounts that can be interpreted by adults through their personal wearables and used to improve their health.
Epidemiological studies of sleep have consistently shown a U- or J-shaped dose–response with mortality [10–12]. Compared to adults who report sleeping 7–9 hours per day, individuals who report sleeping less than this amount have a 7%–37% higher risk of mortality, while those reporting ≥10 hours of sleep per day are at an even greater mortality risk (i.e. ~50% higher). However, the short sleep-mortality association is not consistent among men versus women, young versus older adults, and for all-cause versus cause-specific mortality. Meta-analysis of short sleep and mortality have reported an increased risk for mortality among women but not among men [7], among older (≥60 years) but not middle-aged adults [8], or found associations for all-cause mortality but not cardiovascular- or cancer-related mortality [13]. The evidence on sleep quality, timing, and mortality is more incomplete but studies have documented links between reports of insomnia and mortality risk [14, 15] and have also found that inconsistent sleep timing is associated with unhealthy metabolic profiles [16, 17]. Thus, it remains unknown if more discrete disturbances in maintaining sleep (i.e. sleep continuity) can increase mortality risk and whether inconsistent sleep timing can also translate to an increased risk for early mortality in more generalized samples of adults. Studies of sleep and mortality using actigraphy-based measures of sleep can help fill these gaps. Actigraphy measures have been calibrated to estimate sleep and can provide more accurate estimates of sleep duration [18, 19] and capture more discrete interruptions in sleep when compared to self-reports [19]. Actigraphy can also be used to record sleep timing over multiple days and can characterize sleep patterns over a week or more.
The few studies of actigraphy-based sleep and mortality that have started to fill some of these gaps have so far examined older adults (mean age ~76–83 years) [20–22], did not include measures of sleep continuity or timing [23, 24], and had limited number of cancer-specific mortality outcomes (i.e. 171 cancer-related deaths) [22, 25] and could not provide precise estimates of sleep and mortality due to cancer, the second leading cause of death worldwide [26]. The purpose of this study was, therefore, to address these gaps by: (1) describing the dose–response relationship between actigraphy-based sleep duration and mortality, (2) examining whether sleep duration-mortality associations are generalizable to sex and age subgroups of the populations, and (3) by examining associations between indicators of sleep continuity and timing with mortality. We hypothesized there was a U-shaped relationship between sleep duration and mortality, that sleep duration-mortality associations were generalizable to men, women, older and younger adults, and that both poor sleep continuity and late sleep timing were associated with a higher mortality risk.
Methods
Study design and population
The UK Biobank is a large prospective cohort of over 500 000 adults aged 40–69 years who were registered with the UK National Health Service (NHS) and lived within 25 miles of a UK Biobank assessment center [27, 28]. Participants were asked to complete an electronic questionnaire at baseline that collected demographic, lifestyle, and medical history information. A subset of 236 519 UK Biobank participants was also invited to complete an actigraphy measurement protocol in 2013–2015 as part of follow-up assessments conducted within the cohort (45% response rate) [29]. All participants completed an electronic informed consent to participate in the study. The current study is restricted to the sub-cohort who completed the actigraphy protocol.
A total of 103 696 adults agreed to participate in the UK Biobank accelerometry study and provided data for initial processing and quality control. After inspecting the data we excluded adults with unreliable accelerometry data (n = 5570, 2.3%), with incomplete accelerometry data (i.e. <4 days with 19 + hours/d of recorded data), or that accrued no follow-up time or asked to be withdrawn from the study (n = 27). The analytical sample included 88 282 adults with 4–5 valid days of accelerometry data (Supplementary Figure S1) and most of the adults had complete covariate data (n = 84 034, 95.2%).
Sleep duration, continuity, and timing
Measures of sleep were obtained from actigraphy using an Axivity AX3 wrist-worn triaxial accelerometer. Accelerometers were mailed to participants with a set of instructions. Participants were instructed to wear the monitor on their dominant wrist for 7 days and return the monitor after completing the protocol. Actigraphy data were recorded at 100 Hz during the 7-day period. Data were initially calibrated to local gravity and assessed for quality control using methods previously described [29] and then processed using an open-source R package GGIR v2.6 and the Van Hees sleep detection algorithm to score sleep characteristics [30]. Sleep was defined as non-movement bouts, i.e. changes in wrist rotation ≤ 5° for 5 consecutive minutes taking place within the sleep period. The sleep period is analogous to time in and out of bed and was derived using the automated Heuristic Algorithm looking at Change of Z-Angle (HDCZA) algorithm. We then calculated a variety of sleep characteristics, including: sleep duration, continuity (i.e. wake after sleep onset [WASO], sleep efficiency, number of 5 minutes awakenings, number of nights with WASO > 30 minutes), and timing (sleep onset, wake-up time, midpoint of sleep, and midpoint of the least active 5 hours in a day [L5] [31]). We also calculated social jetlag and sleep regularity indicators, i.e. standard deviation over available nights of data for wake-up time and sleep midpoint (Supplementary Table S1). Sleep characteristics were computed to replicate established definitions of sleep [32] and explore whether our hypothesis for sleep and mortality were consistent across less established actigraphy-based indicators of sleep and that have not been fully explored in previous studies. Sleep estimates generated from the HDCZA method have been shown to have acceptable agreement with a sleep diary and polysomnography estimates for sleep duration and sleep timing and are comparable to other actigraphy-based sleep detection algorithms [30, 33–35].
Assessment of covariates
Information on covariates related to demographic factors and health status were collected at study enrollment. Demographic and lifestyle covariates included age, sex, race and ethnicity, qualifications, employment status, socioeconomic status using Townsend deprivation index, body mass index (BMI), smoking status and dose, accelerometer-measured moderate-vigorous physical activity (MVPA) levels [36], daily frequency of tea, and coffee intake. Comorbidities included self-reported diagnosis of diabetes, high blood pressure, emphysema, and disability. General health was self-reported on a 1 (excellent) to 5 (poor) scale. We also used updated ICD-10 coded medical conditions obtained from primary care visits and hospital records to capture health conditions that occurred between baseline and the actigraphy protocol and that are likely to require hospitalization. These included cardiovascular disease (I21, I25, I60, and I71) and chronic obstructive pulmonary disease (J44).
Ascertainment of mortality
Follow-up was calculated from the beginning of the actigraphy sleep measurement protocol to date of death or until the end of September/October, 2021 (England/Wales or Scotland, respectively) whichever came first. Mortality outcomes included all-cause mortality, cancer-related mortality, and cardiovascular disease-related (CVD) mortality, assessed via the NHS Digital from England and Wales, and the NHS Central Register from Scotland [37]. Underlying primary causes of mortality were defined using the International Classification of Diseases, 10th Revision (ICD-10) codes for cancer (code C00-D48) and cardiovascular disease (code I00-I99).
Statistical analysis
We used Cox proportional hazards models adjusted for covariates to examine the associations between sleep characteristics and mortality with hazard ratios (HR) and 95% confidence intervals (CI). Given the small proportion of adults with incomplete covariate data, we chose to include missing values in our models as indicator categories. See Supplementary Table S2 for respective data fields.
For models of duration, we used restricted cubic spline functions with three knots (5th, 50th, and 95th percentiles) [38] corresponding to approximately 5, 7, and 8 hrs/d of sleep. The 50th percentile of sleep duration (i.e. 7.0 hrs/d) was our reference and is the lower threshold for the recommended amounts of sleep per day for adults. Sleep duration was overall associated with mortality outcomes if the linear or non-linear term for sleep duration in the adjusted models was statistically significant (p < 0.05). To assess the level of confounding, we examined models sequentially adjusted for (1) demographic, (2) lifestyle, and (3) comorbidities and general health, across quartiles of sleep duration. Given that risk estimates were moderately attenuated following these adjustments, we chose to include all the selected covariates for all analyses (Supplementary Table S3). We examined models for sleep duration and all-cause and cause-specific mortality and conducted examinations of sleep duration and all-cause mortality stratified by sex and age groups (40–49, 50–64, and ≥65 years). We considered that associations varied by sex or age if p-value for heterogeneity was < 0.05.
For models of sleep continuity, timing, and sleep regularity, we fitted approximate quartiles for each exposure with highest sleep continuity, e.g. lowest WASO, median sleep timing, and highest sleep regularity, as the reference groups, respectively. We computed overall tests of association for each continuity, timing, and regularity exposure by testing the reference group against the combined HR for the remaining three quartiles. We also examined whether associations between sleep timing and all-cause mortality were independent of sleep duration by further adjusting sleep timing-mortality models for sleep duration.
In sensitivity analyses, we assessed effect modification using Cox models stratified by sleep duration and sleep timing (i.e. L5) separately and for selected key effect modifiers that could potentially modify the association between sleep and mortality. Potential modifiers included: employment status (employed vs. retired), BMI (normal weight vs. overweight vs. obese I vs. obese II), smoking (never vs. former vs. current), physical activity level (quartiles of MVPA), and chronotype (morning vs. evening person). There was evidence of effect modification if p-value across categories of potential effect modifiers was statistically significant (i.e. pheterogeneity < 0.05). We also inspected whether sleep midpoint and L5 midpoint associations with mortality remained consistent after excluding shift workers, day sleepers (i.e. individuals with ≥1 day of recorded sleep onsets either before 7 pm or after 6 am), individuals with poor general health status, with missing covariate information, or with <2 years of follow-up.
The proportional hazards assumption was tested and verified for all sleep exposures of duration, continuity, and timing after visual inspections of the distribution of scaled Schoenfeld residuals [39] derived from our Cox models. Analyses were done using SAS version 9.4 and were two-sided and considered statistically significant when p < 0.05.
Results
During the average 6.8 years of follow-up, 2973 total deaths occurred, with 1700 attributed to cancer and 586 to CVD. When compared with adults with sleep duration of 7–7.9 hrs/d, adults with less than 6 hrs/d of sleep duration included a higher proportion of men, and adults with a higher Townsend Deprivation Index score, with more overweight/obesity, current or former smokers, with high blood pressure, with a disability, and with fair/poor general health status (Table 1).
Table 1.
Descriptive Characteristics Stratified by Amount of Sleep Duration Among Adults
| Sleep duration (hrs/day) | ||||||
|---|---|---|---|---|---|---|
| <6.0 | 6.0–6.9 | 7.0–7.9 | ≥8.0 | Total | ||
| (n = 20 526) | (n = 33 298) | (n = 27 764) | (n = 6694) | (n = 88 282) | Diffe | |
| Mean (Std) | Mean (Std) | Mean (Std) | Mean (Std) | Mean (Std) | ||
| Age (years) | 57.1 (7.7) | 55.9 (7.8) | 56.2 (7.8) | 56.8 (7.6) | 56.3 (7.8) | 0.9 |
| BMI (kg/m2) | 27.9 (4.9) | 26.6 (4.4) | 26.1 (4.2) | 26.0 (4.2) | 26.7 (4.5) | 1.8 |
| N (%) | N (%) | N (%) | N (%) | N (%) | ||
| Sex | ||||||
| Female | 9264 (45.1) | 18 754 (56.3) | 17 519 (63.1) | 4386 (65.5) | 49 923 (56.6) | −18.0 |
| Male | 11 262 (54.9) | 14 544 (43.7) | 10 245 (36.9) | 2308 (34.5) | 38 359 (43.5) | 18.0 |
| Race and Ethnicity a | ||||||
| White | 19 403 (94.5) | 32 202 (96.7) | 27 164 (97.8) | 6592 (98.5) | 85 361 (96.7) | −3.3 |
| Black | 357 (1.7) | 228 (0.7) | 109 (0.4) | 15 (0.2) | 709 (0.8) | 1.3 |
| Asian | 376 (1.8) | 400 (1.2) | 183 (0.7) | 36 (0.5) | 995 (1.1) | 1.1 |
| Other | 297 (1.5) | 363 (1.1) | 225 (0.8) | 38 (0.6) | 923 (1.1) | 0.7 |
| Missing | 93 (0.5) | 105 (0.3) | 83 (0.3) | 13 (0.2) | 294 (0.3) | 0.2 |
| Qualifications b | ||||||
| Higher Ed/Professional | 14 316 (69.8) | 23 580 (70.8) | 19 347 (69.7) | 4299 (64.2) | 61 542 (69.7) | 0.1 |
| School qualifications | 4118 (20.1) | 6807 (20.4) | 5886 (21.2) | 1617 (24.2) | 18 428 (20.9) | −1.1 |
| No qualifications | 1862 (9.1) | 2605 (7.8) | 2253 (8.1) | 701 (10.5) | 7421 (8.4) | 1.0 |
| Missing | 230 (1.1) | 306 (0.9) | 278 (1.0) | 77 (1.2) | 891 (1.0) | 0.1 |
| Employment | ||||||
| Paid employment | 12 159 (59.2) | 21 191 (63.6) | 16 793 (60.5) | 3725 (55.7) | 53 868 (61.0) | −1.3 |
| Retired | 6822 (33.2) | 9971 (29.9) | 9005 (32.4) | 2375 (35.6) | 28 173 (31.9) | 0.8 |
| Unemployed/Other | 1488 (7.3) | 2064 (6.2) | 1919 (6.9) | 578 (8.6) | 6049 (6.9) | 0.4 |
| Missing | 57 (0.3) | 72 (0.2) | 47 (0.2) | 16 (0.2) | 192 (0.2) | 0.1 |
| Deprivation index c | ||||||
| Quintile 1 (low) | 4250 (20.7) | 7663 (23.0) | 6676 (24.1) | 1594 (23.8) | 20 183 (22.9) | −3.4 |
| Quintile 2 | 4155 (20.2) | 7172 (21.5) | 6329 (22.8) | 1510 (22.6) | 19 166 (21.7) | −2.6 |
| Quintile 3 | 4101 (20.0) | 6861 (20.6) | 5876 (21.2) | 1455 (21.7) | 18 293 (20.7) | −1.2 |
| Quintile 4 | 4219 (20.6) | 6543 (19.7) | 5253 (18.9) | 1248 (18.6) | 17 263 (19.6) | 1.7 |
| Quintile 5 (high) | 3779 (18.4) | 5016 (15.1) | 3602 (13.0) | 879 (13.1) | 13 276 (15.0) | 5.4 |
| Missing | 22 (0.1) | 43 (0.1) | 28 (0.1) | 8 (0.1) | 101 (0.1) | 0.0 |
| BMI | ||||||
| Underweight | 84 (0.4) | 177 (0.5) | 152 (0.6) | 49 (0.7) | 462 (0.5) | −0.2 |
| Normal weight | 5865 (28.6) | 12 689 (38.1) | 11 799 (42.5) | 2897 (43.3) | 33 250 (37.7) | −13.9 |
| Overweight | 8638 (42.1) | 13 846 (41.6) | 11 177 (40.3) | 2654 (39.7) | 36 315 (41.1) | 1.8 |
| Obese I | 3943 (19.2) | 4588 (13.8) | 3349 (12.1) | 773 (11.2) | 12 653 (14.3) | 7.1 |
| Obese II | 1662 (8.1) | 1567 (4.7) | 939 (3.4) | 229 (3.4) | 4397 (5.0) | 4.7 |
| Missing | 334 (1.6) | 431 (1.3) | 348 (1.3) | 92 (1.4) | 1205 (1.4) | 0.3 |
| Smoking | ||||||
| Current smoker | 1829 (8.9) | 2192 (6.6) | 1574 (5.6) | 374 (5.6) | 5969 (6.8) | 3.3 |
| Former smoker | 7930 (38.6) | 11 836 (35.6) | 9644 (34.7) | 2329 (34.8) | 31 739 (36.0) | 3.9 |
| Never smoker | 10 710 (52.2) | 19 182 (57.6) | 16,478 (59.4) | 3969 (59.3) | 50 339 (57.0) | −7.2 |
| Missing | 57 (0.3) | 88 (0.3) | 68 (0.2) | 22 (0.3) | 235 (0.3) | 0.1 |
| Tea intake | ||||||
| 0 cups/d | 3927 (19.1) | 5959 (17.9) | 4616 (16.6) | 1205 (18.0) | 15 707 (17.8) | 2.5 |
| 1 to 2 cups/d | 4822 (23.5) | 7905 (23.7) | 6600 (23.8) | 1571 (23.5) | 20 898 (23.7) | −0.3 |
| 3 to 4 cups/d | 5911 (28.8) | 9856 (29.6) | 8428 (30.4) | 1971 (29.4) | 26 166 (29.6) | −1.6 |
| 5 to 6 cups/d | 3900 (19.0) | 6464 (19.4) | 5731 (20.6) | 1314 (19.6) | 17 409 (19.7) | −1.6 |
| 7 + cups/d | 1936 (9.4) | 3072 (9.2) | 2365 (8.5) | 630 (9.4) | 8003 (9.1) | 0.9 |
| Missing | 30 (0.2) | 42 (0.1) | 24 (0.1) | 3 (0.0) | 99 (0.1) | 0.1 |
| Coffee intake | ||||||
| 0 cups/d | 5630 (27.4) | 9018 (27.1) | 7629 (27.5) | 1920 (28.7) | 24 197 (27.4) | −0.1 |
| 1 to 2 cups/d | 7788 (37.9) | 13 618 (40.9) | 11 606 (41.8) | 2782 (41.6) | 35 794 (40.6) | −3.9 |
| 3 to 4 cups/d | 4603 (22.4) | 7170 (21.5) | 5912 (21.3) | 1357 (20.3) | 19 042 (21.6) | 1.1 |
| 5 to 6 cups/d | 1818 (8.9) | 2643 (7.9) | 2060 (7.4) | 491 (7.3) | 7012 (7.9) | 1.5 |
| 7 + cups/d | 662 (3.2) | 814 (2.4) | 534 (1.9) | 134 (2.0) | 2144 (2.4) | 1.3 |
| Missing | 25 (0.1) | 35 (0.1) | 23 (0.1) | 10 (0.2) | 93 (0.1) | 0.0 |
| Physical activity d | ||||||
| ≤0.8 hrs/d | 5932 (28.9) | 7439 (22.3) | 6544 (23.6) | 2153 (32.2) | 22 068 (25.0) | 5.3 |
| 0.8–1.1 hrs/d | 4952 (24.1) | 8102 (24.3) | 7251 (26.1) | 1768 (26.4) | 22 073 (25.0) | −2.0 |
| 1.2–1.5 hrs/d | 4788 (23.3) | 8623 (25.9) | 7100 (25.6) | 1560 (23.3) | 22 071 (25.0) | −2.3 |
| ≥1.6 hrs/d | 4854 (23.7) | 9134 (27.4) | 6869 (24.7) | 1213 (18.1) | 22 070 (25.0) | −1.0 |
| Diabetes mellitus | ||||||
| Yes | 1119 (5.5) | 1007 (3.0) | 744 (2.7) | 184 (2.8) | 3054 (3.5) | 2.8 |
| No | 19 356 (94.3) | 32 223 (96.8) | 26 983 (97.2) | 6500 (97.1) | 85 062 (96.4) | −2.9 |
| Missing | 51 (0.3) | 68 (0.2) | 37 (0.1) | 10 (0.2) | 166 (0.2) | 0.2 |
| High blood pressure | ||||||
| Yes | 5320 (25.9) | 6556 (19.7) | 5398 (19.4) | 1376 (20.6) | 18 650 (21.1) | 6.5 |
| No | 15 206 (74.1) | 26 742 (80.3) | 22 366 (80.6) | 5318 (79.4) | 69 632 (78.9) | −6.5 |
| Emphysema | ||||||
| Yes | 264 (1.3) | 302 (0.9) | 246 (0.9) | 44 (0.7) | 856 (1.0) | 0.4 |
| No | 20 262 (98.7) | 32 996 (99.1) | 27 518 (99.1) | 6650 (99.3) | 87 426 (99.0) | −0.4 |
| Disability | ||||||
| Yes | 6813 (33.2) | 8785 (26.4) | 7158 (25.8) | 1944 (29.0) | 24 700 (28.0) | 7.4 |
| No | 13 258 (64.6) | 23 876 (71.7) | 20 095 (72.4) | 4628 (69.1) | 61 857 (70.1) | −7.8 |
| Missing | 455 (2.2) | 637 (1.9) | 511 (1.8) | 122 (1.8) | 1725 (2.0) | 0.4 |
| Cardiovascular disease | ||||||
| Yes | 629 (3.1) | 729 (2.2) | 531 (1.9) | 146 (2.2) | 2035 (2.3) | 1.2 |
| No | 19 897 (96.9) | 32 569 (97.8) | 27 233 (98.1) | 6548 (97.8) | 86 247 (97.7) | −1.2 |
| COPD | ||||||
| Yes | 246 (1.2) | 220 (0.7) | 154 (0.6) | 52 (0.8) | 672 (0.8) | 0.6 |
| No | 20 280 (98.8) | 32 569 (99.3) | 27 610 (99.5) | 6642 (99.2) | 87 610 (99.2) | −0.7 |
| General health | ||||||
| Excellent | 3807 (18.6) | 7542 (22.7) | 6345 (22.9) | 1390 (20.8) | 19 084 (21.6) | −4.3 |
| Good | 11 778 (57.4) | 20 171 (60.6) | 17 036 (61.4) | 4014 (60.0) | 52 999 (60.0) | −4.0 |
| Fair | 4109 (20.0) | 4810 (14.5) | 3775 (13.6) | 1069 (16.0) | 13 763 (15.6) | 6.4 |
| Poor | 778 (3.8) | 694 (2.1) | 553 (2.0) | 202 (3.0) | 2227 (2.5) | 1.8 |
| Missing | 54 (0.3) | 81 (0.2) | 55 (0.2) | 19 (0.3) | 209 (0.2) | 0.1 |
aRace and ethnicity were self-reported and categorized as white, or black/black British, or Asian/Asian British, or Other (combination of more than 1 race–ethnicity, other race and ethnicity), or Missing (do not know or prefer not to answer).
bQualifications included having a Higher Education/Professional degree (University/College degree or professional degree), or school degree (school-degree equivalents defined as Advanced Levels qualifications, Ordinary Levels, Certificate of Secondary education, or an equivalent degree), or no qualifications (i.e. no higher education or school equivalent degree).
cQuintiles of Townsend deprivation index scores obtained at recruitment and defined using participants’ postal code. Scores range from low (first Quintile) to high (fifth Quintile) deprivation index scores.
dAccelerometry measured moderate-to-vigorous physical activity defined as hours/d with raw accelerations ≥125 mg.
eComputed as the difference in percentage points between sleep duration of <6 hrs/d and our referent group for main analyses, i.e. sleep duration of 7–7.9 hrs/d. Differences for age and BMI are for differences in years and BMI units, respectively.
Sleep duration
Estimated sleep duration was on average 6.7 hrs/d and was overall associated with all-cause mortality and the dose–response was curvilinear (HR: 1.03 [1.02, 1.04], p < 0.01). Sleep durations lower than 6 hrs/d were associated with a 16%–44% higher risk for mortality when compared to 7 hrs/d of sleep duration (Figure 1, Supplementary Table S4). For example, the HR for 5 hrs/d of sleep duration was 1.29 [95% CI: 1.09, 1.52] when compared to adults who had a sleep duration of 7 hrs/d. A sleep duration of 8 hrs/d or more (vs. 7 hrs/d) was not associated with a higher risk for mortality.
Figure 1.
Dose–response associations between sleep duration and all-cause, cancer, and cardiovascular disease (CVD) mortality. p-value is for the overall association between sleep duration and mortality using restricted cubic splines.
The sleep duration and cancer mortality relationship was also statistically significant with a curvilinear shape that was similar to the all-cause mortality association (HR: 1.01 [1.00, 1.02], p < 0.01, Figure 1, Supplementary Table S4).
The association between sleep duration and CVD-related mortality was also curvilinear (HR: 1.02 [1.00, 1.03], p = 0.03). Sleep durations of 5 hrs/d or less were associated with a 46%–67% higher risk for CVD-related mortality when compared to sleep durations of 7 hrs/d (Figure 1, Supplementary Table S4).
The dose–response for sleep duration and all-cause mortality did not differ by sex (p = 0.96) or age (p = 0.06) showing that sleep duration and mortality are similarly associated among men, women, younger, and older adults (Figure 2).
Figure 2.
Sleep duration and all-cause mortality by sex and age.
Sleep continuity
Adults recorded on average 48 min/d of WASO, had 90% sleep efficiency, and an average of 1.4 awakenings of 5 minutes or more recorded throughout the night. Neither greater WASO or lower sleep efficiency were associated with increased mortality risk. For example, adults with ≥60 min/d of WASO were not at increased risk for mortality (0.98 [0.88, 1.10], Table 2) compared to adults with <30 min/d of WASO. Having more nights per week with WASO > 30 min/d was also not associated with increased risk for all-cause mortality. There was no association between either indicator of sleep continuity and cancer or cardiovascular mortality, except for the number of nights per week with WASO >30 minutes that was associated with cancer-related mortality (Supplementary Table S5).
Table 2.
Hazard Ratios for the Associations Between Sleep Continuity Exposures and All-cause Mortality
| N (%) | Deaths | HR (95% CI) | ||
|---|---|---|---|---|
| Wake after sleep onset (WASO), min | <30’ | 19 535 (22.1) | 631 | 1.00 (Ref) |
| 30–44’ | 29 248 (33.1) | 968 | 1.01 (0.91, 1.12) | |
| 45–59’ | 20 839 (23.6) | 710 | 1.00 (0.90, 1.12) | |
| ≥60’ | 18 660 (21.1) | 664 | 0.98 (0.88, 1.10) | |
| p | 0.99 | |||
| Sleep efficiency, % | >93% | 21 644 (24.5) | 647 | 1.00 (Ref) |
| 91%–93% | 27 176 (30.8) | 850 | 1.01 (0.92, 1.12) | |
| 86%–90% | 27 189 (30.8) | 933 | 1.03 (0.93, 1.14) | |
| ≤85.0% | 12 273 (13.9) | 543 | 1.15 (1.02, 1.29) | |
| p | 0.31 | |||
| 5’ awakenings per night, n | <1.0 | 40 943 (46.4) | 1329 | 1.00 (Ref) |
| 1.0–1.9 | 26 679 (28.0) | 858 | 1.01 (0.93, 1.10) | |
| 2.0–2.9 | 12 240 (13.9) | 413 | 0.94 (0.84, 1.05) | |
| ≥3.0 | 10 420 (11.8) | 373 | 0.97 (0.86, 1.09) | |
| p | 0.64 | |||
| Nights per week with WASO > 30 min, n | 0 to 1 | 11 494 (13.0) | 356 | 1.00 (Ref) |
| 2 to 3 | 12 804 (14.5) | 447 | 1.12 (0.97, 1.28) | |
| 4 to 6 | 38 299 (43.4) | 1349 | 1.07 (0.95, 1.20) | |
| 7 | 25 685 (29.1) | 821 | 0.99 (0.88, 1.13) | |
| p | 0.84 |
Hazard ratios adjusted for age + sex + race-ethnicity + education + employment status + Townsend deprivation index + BMI + smoking status + moderate-vigorous physical activity + tea intake + coffee intake + diabetes + blood pressure + emphysema + general health + disability + cardiovascular disease + chronic obstructive pulmonary disease.
p-value is for the overall association between each sleep characteristic and mortality.
Bold values indicate statistically significant associations.
Sleep timing and regularity
Adults fell asleep on average at 11:45 pm and woke up at 07:10 am, with an average sleep midpoint of 03:28 am. The midpoint of the least active 5 hours (L5) was associated with all-cause mortality (HR: 1.20 [1.09, 1.32], p < 0.01). Adults with delayed midpoints for the least active 5 hours (L5) for the day (i.e. >03:30 am) had a higher risk for all-cause mortality when compared to earlier timing reference categories (Figure 3, Supplementary Table S6). Sleep midpoint and wake-up time followed a similar trend to L5 (p = 0.04). For example, adults with late (>04:00 am) sleep midpoint times had a 13% higher risk for mortality (1.13 [1.02, 1.26]) compared to adults with sleep midpoint times between 03:30 and 03:59 am. Sleep onset was not associated with mortality (Figure 3, Supplementary Table S6).
Figure 3.
Associations between indicators of sleep timing and mortality. p-value for the overall association between each sleep characteristic and mortality.
There were no clear associations between sleep timing indicators and cancer and CVD mortality except for L5 and cancer mortality (HR: 1.17 [1.03, 1.33], p = 0.02, Supplementary Table S6). Adults with delayed L5 midpoint times ≥3:30 am had a 1.21 [1.05, 1.39] higher risk for cancer-related mortality when compared to adults with L5 midpoint times ranging between 3:00 and 3:29 am (Figure 3, Supplementary Table S6). Social jetlag and the regularity of sleep timing (i.e. standard deviation for wake-up time and sleep midpoint) were not associated with all-cause, cancer, or CVD mortality (Supplementary Table S7).
When adjusting the association between L5 midpoint and mortality for sleep duration, we found the results to be virtually unchanged (Table 3).
Table 3.
Hazard Ratios for Sleep Timing Exposures Unadjusted Versus Adjusted for Sleep Duration
| Unadjusted for sleep durationa | Adjusted for sleep durationb | ||
|---|---|---|---|
| HR (95% CI) | HR (95% CI) | ||
| Sleep onset | <23:00 | 1.07 (0.95, 1.21) | 1.07 (0.95, 1.21) |
| 23:00–23:29 | 1.01 (0.90, 1.14) | 1.02 (0.90, 1.15) | |
| 23:30–23:59 | 1.00 (Ref) | 1.00 (Ref) | |
| ≥00:00 | 1.12 (1.02, 1.24) | 1.08 (0.97, 1.19) | |
| p | 0.09 | 0.22 | |
| Wake-up | <06:00 | 1.24 (1.11, 1.39) | 1.14 (1.01, 1.28) |
| 06:00–06:59 | 0.99 (0.90, 1.09) | 0.97 (0.88, 1.07) | |
| 07:00–07:59 | 1.00 (Ref) | 1.00 (Ref) | |
| ≥08:00 | 1.10 (1.00, 1.21) | 1.10 (1.00, 1.21) | |
| p | 0.04 | 0.22 | |
| Sleep midpoint | <03:00 | 1.11 (0.99, 1.22) | 1.09 (0.98, 1.17) |
| 03:00–03:29 | 1.05 (0.94, 1.17) | 1.05 (0.94, 1.17) | |
| 03:30–03:59 | 1.00 (Ref) | 1.00 (Ref) | |
| ≥04:00 | 1.13 (1.02, 1.26) | 1.11 (1.00, 1.24) | |
| p | 0.04 | 0.07 | |
| Least active 5 hours midpoint | <02:30 | 1.29 (1.15, 1.44) | 1.26 (1.13, 1.41) |
| 02:30–02:59 | 1.10 (0.97, 1.24) | 1.10 (0.97, 1.24) | |
| 03:00–03:29 | 1.00 (Ref) | 1.00 (Ref) | |
| ≥03:30 | 1.19 (1.07, 1.32) | 1.19 (1.07, 1.32) | |
| p | <0.01 | <0.01 |
aHazard ratios adjusted for age + sex + race-ethnicity + education + employment status + Townsend deprivation index + BMI + smoking status + moderate-vigorous physical activity + tea intake + coffee intake + diabetes + blood pressure + emphysema + general health + disability + cardiovascular disease + chronic obstructive pulmonary disease.
p-value is for the overall association between each sleep characteristic and mortality.
bModels adjusted for all covariates + sleep duration. Bold values indicate statistically significant associations.
Sensitivity analysis
Sleep duration and midpoint of L5 were similarly associated with all-cause mortality across employment, BMI categories, smoking status, quartiles of MVPA, and chronotype (pheterogeneity > 0.05, Supplementary Table S8 & Supplementary Table S9). Associations between sleep duration, L5 midpoint, and mortality remained similar after excluding shift workers, day sleepers, individuals with poor general health status, with missing covariate information, or with <2 years of follow-up (Supplementary Table S10).
Discussion
In this large study adults with sleep duration less than 6 hrs/d were at a higher risk for mortality and these findings were similar among men, women, younger, and older adults. Sleep timing clock times as indicated by L5 as < 2:30 or ≥ 3:30 were also associated with a 20% to 30% higher mortality risk. Objective measures of sleep continuity (e.g. WASO) and indicators of sleep regularity were not associated with mortality risk. Our dose–response for the sleep duration-mortality and findings for indicators of sleep timing can be used to interpret sleep-related recordings obtained from wearables and facilitate clinician-patient interactions. They may also help to guide public health efforts to promote healthy sleep patterns in the adult population.
The sleep-mortality association has been studied for over 50 years using established epidemiological prospective cohorts however, there is still uncertainty about the dose–response of the sleep-mortality association and whether the sleep-mortality associations can be generalizable to men, women, young, and older adults. It is also less clear whether more nuanced variations in sleep timing as experienced by the general population (e.g. late bedtimes or early morning wake-ups) may be associated with increased mortality risk. The evidence so far was built from studies that used questionnaire-based measures of sleep and while these are informative, they are known to reflect time people allocate for sleep as opposed to sleep attained [2, 11] and they lack the precision needed to advance our understanding of sleep doses and their relation to mortality. Our study addressed these concerns by using actigraphy-based measures of sleep to describe the sleep-mortality dose–response and generate more precise sleep duration estimates and their relationship with mortality risk. Our study also examined sleep-mortality associations in approximately 90 000 adults, the largest cohort to date with sleep actigraphy and evaluated whether sleep-mortality associations were consistent across men, women, and adults of different age groups.
Our findings build on the growing evidence that the timing of sleep is an important predictor of good health. Extreme sleep timing schedules that can induce circadian disruption are common among night shift workers and have been linked to metabolic dysfunction [40], a variety of chronic conditions including coronary heart disease, diabetes and cancer [41–44], and mortality [45, 46]. We found suggestive links between early and late sleep timing with both all-cause and cancer mortality. In our study, early (before 2:30 am) or late (≥3:30 am) least active 5 hours in a day were associated with an increased mortality risk. The midpoint of the least active 5 hours in a day indicates the phase of the least active period of the day, which tends to occur during sleep. The limited evidence linking early/late timing to mortality in adults and using actigraphy comes from the Outcome of Sleep Disorders in Older men (MrOS) and the Study of Osteoporotic Fractures in older women, 6000 older adults total. Data from the MrOS study showed a U-shaped association between sleep midpoint and all-cause mortality [20] but this finding was not confirmed in more updated examinations of these same data [21]. In our study, sleep midpoint was also associated with mortality and followed a similar U-shape pattern to that of L5 midpoint and wake-up time. Actigraphy-measured sleep midpoint, wake-up time, and L5 are strongly correlated with one another and hence similarly correlated with melatonin onset in dim light, the gold standard of circadian phase [47], and can be used to advance our understanding of sleep timing-mortality implications. Future epidemiological studies are needed to obtain more comprehensive descriptions of sleep timing, and circadian disruption, and their association with mortality.
We also found that mortality risk was only more noticeable for adults with sleep durations < 6 hrs/d. This amount is lower than the recommended lowest threshold amount of 7 hrs/d for sleep duration and might suggest that lower amounts of sleep (i.e. 6 to 7 hrs/d) may not be detrimental to health. In our study, adults with sleep durations of 5 hrs/d had a 29% higher risk for all-cause mortality and sleep durations lower than 5 hrs/d were associated with an even higher mortality risk, when compared to adults with sleep durations of 7 hrs/d. The sleep-mortality dose–response was similar for cancer and CVD mortality. Our findings are aligned with clinical studies demonstrating the negative impact of sleep restriction on metabolic outcomes known to be associated with premature mortality risk and support the general 7–9 hrs/d sleep recommendation for adults. Other studies have examined actigraphy-based sleep duration and mortality but had mixed findings. For example, in the MrOS study where 2887 older men were followed for 11 years (mean age = 76.3 years), sleeping ≤ 5 hrs/d was associated with 26% higher risk for all-cause mortality when compared to adults who slept 5–7.5 hrs/d [20], but not CVD or cancer-related mortality [22]. In another study of 1734 adults followed over 7 years, sleep duration was not associated with all-cause mortality [25]. In the largest study of sleep actigraphy and mortality, Chastin and colleagues in a pooled analysis of UK Biobank and Whitehall II studies (n = 100 000) found no evidence of a dose–response association between sleep and all-cause mortality [23]. These studies of actigraphy were limited to a low number of cause-specific mortality events (e.g. 171 cancer-related deaths) [22, 25] resulting in imprecise risk estimates or used partial sleep duration measures, i.e. midnight to midnight total sleep time, while adjusting for time spent in other physical behaviors [23] and did not examine if sleep-mortality associations were generalizable across sex and age in adulthood. Our study clarifies the shape of the sleep duration-mortality dose–response among men, women, young, and older adults and suggests that sleeping <6 hrs/d is associated with a higher risk for mortality due to CVD, the top leading cause of mortality in developed countries. We also found that most adults in our study got <9 hrs/d of sleep (99th percentile) and that there was no evidence that adults who slept for >7 hrs/d had a higher mortality risk. This is consistent with recent analyses of these same data [24]. Thus, while it remains unclear whether long sleep during the night is also associated with mortality, our study suggests that sleep amounts of ≥10 hrs/d accumulated during the main sleep period at night are very unlikely when measured with actigraphy and that such reported duration patterns may better reflect behaviors other than sleep (e.g. time spent awake in bed).
Contrary to other studies, in our study, we did not find any clear association between indicators of sleep continuity and mortality. In the MrOS study of older men and sleep continuity and mortality, WASO of ~90 min/d or greater was associated with a 32%–45% higher risk for all-cause and CVD, but not cancer-related mortality, when compared to older adults with WASO less than ~90 min/d [20, 22]. WASO is more prevalent among older adults and this discrepancy might be explained by the age differences between our study sample and MrOS (mean age 53 years vs. 76 years, respectively). Poor sleep continuity as indicated by lower sleep efficiency was also associated with higher mortality risk in the Study of Osteoporotic Fractures study of older women (mean age 83 years) [21].
Limitations
Our study includes some limitations. First, our study included a variety of sleep exposures and multiple examinations with mortality which can potentially increase the likelihood of type I error, i.e. rejection of a true null hypothesis. Second, the accelerometry protocol was conducted about 5 years after the study baseline i.e. between baseline and the accelerometry protocol, and we may not have been able to account for variation in comorbidities and other covariates over time. However, our models included adjustments for updated hospital conditions and previous studies in UK Biobank have also shown that covariates between baseline and accelerometry protocol remain relatively stable [48]. Third, even though we adjusted for a variety of potential confounders of the sleep-mortality association, we cannot rule out unmeasured or residual confounding. For example, we did not have valid information about participants’ exposure to light at night. Exposure to light at night is known to suppress melatonin production and interfere with the regulation of a variety of metabolic pathways leading to an increased risk for metabolic conditions including obesity and diabetes [21, 49]. Fourth, our measure of actigraphy-based sleep relies on non-movement bouts to infer sleep states and therefore it is not a direct measure of sleep. Actigraphy-based sleep accuracy can overcome self-reports and sleep logs by providing objective amounts of sleep duration and estimates of sleep timing over multiple days that are comparable to polysomnography. However, actigraphy has a limited ability to detect periods of wakefulness throughout the sleep period when compared to polysomnography [50, 51]. Fifth, sleep characteristics were obtained from a single 7-day aggregate collected early in the study. It is possible that participants changed their sleep behaviors over time and that such long-term variations in sleep characteristics may have attenuated our sleep-mortality associations. A previous study has demonstrated that sleep characteristics are moderate-to-highly stable over time [52]. Sixth, most of the adults included in our study were White and our results may not be generalizable to other adult populations.
Conclusions
Sleep durations lower than 6 hrs/d were associated with a higher risk for mortality in a large sample of UK adults. Very early or late sleep timing schedules may be associated with a higher mortality risk while other characteristics including sleep continuity and sleep regularity were not associated with mortality. Our study expanded the sleep-mortality evidence by clarifying the associations between various sleep characteristics and mortality. The findings of this study reinforce the importance of sleep for good health and open the opportunity to interpret sleep metrics generated from personal wearables when managing individuals’ health risks.
Supplementary Material
Acknowledgments
This research has been conducted using the UK Biobank Resources under Application Number 43456.
Contributor Information
Pedro F Saint-Maurice, Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health, Bethesda, MD, USA; Breast Unit, Champalimaud Clinical Center, Champalimaud Foundation, Lisbon, Portugal.
Joshua R Freeman, Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health, Bethesda, MD, USA.
Daniel Russ, Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health, Bethesda, MD, USA.
Jonas S Almeida, Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health, Bethesda, MD, USA.
Marissa M Shams-White, Division of Cancer Control and Population Sciences, National Cancer Institute, National Institutes of Health, Bethesda, MD, USA.
Shreya Patel, Department of Epidemiology and Biostatistics, Dornsife School of Public Health, Drexel University, Philadelphia, USA.
Dana L Wolff-Hughes, Division of Cancer Control and Population Sciences, National Cancer Institute, National Institutes of Health, Bethesda, MD, USA.
Eleanor L Watts, Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health, Bethesda, MD, USA.
Erikka Loftfield, Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health, Bethesda, MD, USA.
Hyokyoung G Hong, Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health, Bethesda, MD, USA.
Steven C Moore, Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health, Bethesda, MD, USA.
Charles E Matthews, Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health, Bethesda, MD, USA.
Funding
PSM, JRF, DR, JSA, ELW, EL, HGH, SCM, and CEM were supported by the National Institutes of Health’s Intramural Research Program: National Cancer Institute.
Disclosure Statement
Financial disclosure: None. Nonfinancial disclosure: None.
Data Availability
UK Biobank data used in this study are globally available to approved researchers through the UK Biobank research portal (https://www.ukbiobank.ac.uk/). Sleep exposures generated from the original data were derived using open-source software (GGIR version 2.6; https://cran.r-project.org/web/packages/GGIR/index.html) and will be returned to the UK Biobank for future distribution.
References
- 1. Watson NF, Badr MS, Belenky G, et al.; Consensus Conference Panel. Joint consensus statement of the american academy of sleep medicine and sleep research society on the recommended amount of sleep for a healthy adult: Methodology and discussion. J Clin Sleep Med. 2015;11(8):931–952. doi: 10.5664/jcsm.4950 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Hirshkowitz M, Whiton K, Albert SM, et al. National Sleep Foundation’s updated sleep duration recommendations: final report. Sleep Health. 2015;1(4):233–243. doi: 10.1016/j.sleh.2015.10.004 [DOI] [PubMed] [Google Scholar]
- 3. Chien KL, Chen PC, Hsu HC, et al. Habitual sleep duration and insomnia and the risk of cardiovascular events and all-cause death: Report from a community-based cohort. Sleep. 2010;33(2):177–184. doi: 10.1093/sleep/33.2.177 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Harvey A, Scott H, Melaku YA, et al. The mortality risk of night-time and daytime insomnia symptoms in an older population. Sci Rep. 2023;13(1):9575. doi: 10.1038/s41598-023-36016-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Shimizu I, Yoshida Y, Minamino T.. A role for circadian clock in metabolic disease. Hypertens Res. 2016;39(7):483–491. doi: 10.1038/hr.2016.12 [DOI] [PubMed] [Google Scholar]
- 6. James SM, Honn KA, Gaddameedhi S, Van Dongen HPA.. Shift work: Disrupted circadian rhythms and sleep-implications for health and well-being. Curr Sleep Med Rep. 2017;3(2):104–112. doi: 10.1007/s40675-017-0071-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Liu TZ, Xu C, Rota M, et al. Sleep duration and risk of all-cause mortality: A flexible, non-linear, meta-regression of 40 prospective cohort studies. Sleep Med Rev. 2017;32:28–36. doi: 10.1016/j.smrv.2016.02.005 [DOI] [PubMed] [Google Scholar]
- 8. Gangwisch JE, Heymsfield SB, Boden-Albala B, et al. Sleep duration associated with mortality in elderly, but not middle-aged, adults in a large US sample. Sleep. 2008;31(8):1087–1096. [PMC free article] [PubMed] [Google Scholar]
- 9. Lauderdale DS, Knutson KL, Yan LL, Liu K, Rathouz PJ.. Self-reported and measured sleep duration: How similar are they? Epidemiology. 2008;19(6):838–845. doi: 10.1097/ede.0b013e318187a7b0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Hammond EC. Some preliminary findings on physical complaints from a prospective study of 1,064,004 men and women. Am J Public Health Nations Health. 1964;54:11–23. doi: 10.2105/ajph.54.1.11 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Grandner MA, Hale L, Moore M, Patel NP.. Mortality associated with short sleep duration: The evidence, the possible mechanisms, and the future. Sleep Med Rev. 2010;14(3):191–203. doi: 10.1016/j.smrv.2009.07.006 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Cappuccio FP, D’Elia L, Strazzullo P, Miller MA.. Sleep duration and all-cause mortality: A systematic review and meta-analysis of prospective studies. Sleep. 2010;33(5):585–592. doi: 10.1093/sleep/33.5.585 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Gallicchio L, Kalesan B.. Sleep duration and mortality: A systematic review and meta-analysis. J Sleep Res. 2009;18(2):148–158. doi: 10.1111/j.1365-2869.2008.00732.x [DOI] [PubMed] [Google Scholar]
- 14. Lovato N, Lack L.. Insomnia and mortality: A meta-analysis. Sleep Med Rev. 2019;43:71–83. doi: 10.1016/j.smrv.2018.10.004 [DOI] [PubMed] [Google Scholar]
- 15. Ge L, Guyatt G, Tian J, et al. Insomnia and risk of mortality from all-cause, cardiovascular disease, and cancer: Systematic review and meta-analysis of prospective cohort studies. Sleep Med Rev. 2019;48:101215. doi: 10.1016/j.smrv.2019.101215 [DOI] [PubMed] [Google Scholar]
- 16. Rutters F, Lemmens SG, Adam TC, et al. Is social jetlag associated with an adverse endocrine, behavioral, and cardiovascular risk profile? J Biol Rhythms. 2014;29(5):377–383. doi: 10.1177/0748730414550199 [DOI] [PubMed] [Google Scholar]
- 17. Parsons MJ, Moffitt TE, Gregory AM, et al. Social jetlag, obesity and metabolic disorder: Investigation in a cohort study. Int J Obes (Lond). 2015;39(5):842–848. doi: 10.1038/ijo.2014.201 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Jackson CL, Patel SR, Jackson WB, Lutsey PL, Redline S.. Agreement between self-reported and objectively measured sleep duration among white, black, Hispanic, and Chinese adults in the United States: Multi-Ethnic Study of Atherosclerosis. Sleep. 2018;41(6):1–12. doi: 10.1093/sleep/zsy057 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Lehrer HM, Yao Z, Krafty RT, et al. Comparing polysomnography, actigraphy, and sleep diary in the home environment: The Study of Women’s Health Across the Nation (SWAN) Sleep Study. Sleep Adv. 2022;3(1):zpac001. doi: 10.1093/sleepadvances/zpac001 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Wallace ML, Stone K, Smagula SF, et al. Which sleep health characteristics predict all-cause mortality in older men? an application of flexible multivariable approaches. Sleep. 2018;41(1). doi: 10.1093/sleep/zsx189 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Wallace ML, Lee S, Stone KL, et al. Actigraphy-derived sleep health profiles and mortality in older men and women. Sleep. 2022;45(4). doi: 10.1093/sleep/zsac015 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Smagula SF, Stone KL, Redline S, et al.; Osteoporotic Fractures in Men (MrOS) Research Group. Actigraphy- and polysomnography-measured sleep disturbances, inflammation, and mortality among older men. Psychosom Med. 2016;78(6):686–696. doi: 10.1097/PSY.0000000000000312 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Chastin S, McGregor D, Palarea-Albaladejo J, et al. Joint association between accelerometry-measured daily combination of time spent in physical activity, sedentary behaviour and sleep and all-cause mortality: A pooled analysis of six prospective cohorts using compositional analysis. Br J Sports Med. 2021;55(22):1277–1285. doi: 10.1136/bjsports-2020-102345 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Liang YY, Feng H, Chen Y, et al. Joint association of physical activity and sleep duration with risk of all-cause and cause-specific mortality: A population-based cohort study using accelerometry. Eur J Prev Cardiol. 2023;30:832–843. doi: 10.1093/eurjpc/zwad060 [DOI] [PubMed] [Google Scholar]
- 25. Zuurbier LA, Luik AI, Hofman A, Franco OH, Van Someren EJ, Tiemeier H.. Fragmentation and stability of circadian activity rhythms predict mortality: The Rotterdam study. Am J Epidemiol. 2015;181(1):54–63. doi: 10.1093/aje/kwu245 [DOI] [PubMed] [Google Scholar]
- 26. World Health Organization. Cancer. https://www.who.int/health-topics/cancer#tab=tab_1Accessed June 11, 2023.
- 27. Sudlow C, Gallacher J, Allen N, et al. UK biobank: An open access resource for identifying the causes of a wide range of complex diseases of middle and old age. PLoS Med. 2015;12(3):e1001779. doi: 10.1371/journal.pmed.1001779 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Fry A, Littlejohns TJ, Sudlow C, et al. Comparison of sociodemographic and health-related characteristics of UK biobank participants with those of the general population. Am J Epidemiol. 2017;186(9):1026–1034. doi: 10.1093/aje/kwx246 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Doherty A, Jackson D, Hammerla N, et al. Large scale population assessment of physical activity using wrist worn accelerometers: The UK biobank study. PLoS One. 2017;12(2):e0169649. doi: 10.1371/journal.pone.0169649 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. van Hees VT, Sabia S, Jones SE, et al. Estimating sleep parameters using an accelerometer without sleep diary. Sci Rep. 2018;8(1):12975. doi: 10.1038/s41598-018-31266-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Goncalves BS, Adamowicz T, Louzada FM, Moreno CR, Araujo JF.. A fresh look at the use of nonparametric analysis in actimetry. Sleep Med Rev. 2015;20:84–91. doi: 10.1016/j.smrv.2014.06.002 [DOI] [PubMed] [Google Scholar]
- 32. Fekedulegn D, Andrew ME, Shi M, Violanti JM, Knox S, Innes KE.. Actigraphy-based assessment of sleep parameters. Ann Work Expo Health. 2020;64(4):350–367. doi: 10.1093/annweh/wxaa007 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. van Hees VT, Sabia S, Anderson KN, et al. A novel, open access method to assess sleep duration using a wrist-worn accelerometer. PLoS One. 2015;10(11):e0142533. doi: 10.1371/journal.pone.0142533 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Sundararajan K, Georgievska S, Te Lindert BHW, et al. Sleep classification from wrist-worn accelerometer data using random forests. Sci Rep. 2021;11(1):24. doi: 10.1038/s41598-020-79217-x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Plekhanova T, Rowlands AV, Davies MJ, Hall AP, Yates T, Edwardson CL.. Validation of an automated sleep detection algorithm using data from multiple accelerometer brands. J Sleep Res. 2022;32:e13760. doi: 10.1111/jsr.13760 [DOI] [PubMed] [Google Scholar]
- 36. White T, Westgate K, Wareham NJ, Brage S.. Estimation of physical activity energy expenditure during free-living from wrist accelerometry in UK adults. PLoS One. 2016;11(12):e0167472. doi: 10.1371/journal.pone.0167472 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37. Biobank U. Mortality data: linkage to death registries. 2020. https://biobank.ctsu.ox.ac.uk/crystal/ukb/docs/DeathLinkage.pdfAccessed March 21, 2022.
- 38. Desquilbet L, Mariotti F.. Dose-response analyses using restricted cubic spline functions in public health research. Stat Med. 2010;29(9):1037–1057. doi: 10.1002/sim.3841 [DOI] [PubMed] [Google Scholar]
- 39. Hess KR. Graphical methods for assessing violations of the proportional hazards assumption in Cox regression. Stat Med. 1995;14(15):1707–1723. doi: 10.1002/sim.4780141510 [DOI] [PubMed] [Google Scholar]
- 40. McHill AW, WrightKP, Jr. Role of sleep and circadian disruption on energy expenditure and in metabolic predisposition to human obesity and metabolic disease. Obes Rev. 2017;18(suppl 1):15–24. doi: 10.1111/obr.12503 [DOI] [PubMed] [Google Scholar]
- 41. Shan Z, Li Y, Zong G, et al. Rotating night shift work and adherence to unhealthy lifestyle in predicting risk of type 2 diabetes: Results from two large US cohorts of female nurses. BMJ. 2018;363:k4641. doi: 10.1136/bmj.k4641 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42. Vetter C, Devore EE, Wegrzyn LR, et al. Association between rotating night shift work and risk of coronary heart disease among women. JAMA. 2016;315(16):1726–1734. doi: 10.1001/jama.2016.4454 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43. group IMV. Carcinogenicity of night shift work. Lancet Oncol. 2019;20(8):1058–1059. doi: 10.1016/S1470-2045(19)30455-3 [DOI] [PubMed] [Google Scholar]
- 44. Shi H, Huang T, Schernhammer ES, Sun Q, Wang M.. Rotating night shift work and healthy aging after 24 years of follow-up in the nurses’ health study. JAMA Netw Open. 2022;5(5):e2210450. doi: 10.1001/jamanetworkopen.2022.10450 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45. Su F, Huang D, Wang H, Yang Z.. Associations of shift work and night work with risk of all-cause, cardiovascular and cancer mortality: A meta-analysis of cohort studies. Sleep Med. 2021;86:90–98. doi: 10.1016/j.sleep.2021.08.017 [DOI] [PubMed] [Google Scholar]
- 46. Niedhammer I, Coutrot T, Geoffroy-Perez B, Chastang JF.. Shift and night work and all-cause and cause-specific mortality: Prospective results from the STRESSJEM study. J Biol Rhythms. 2022;37(3):249–259. doi: 10.1177/07487304221092103 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 47. Reiter AM, Sargent C, Roach GD.. Finding DLMO: Estimating dim light melatonin onset from sleep markers derived from questionnaires, diaries and actigraphy. Chronobiol Int. 2020;37(9–10):1412–1424. doi: 10.1080/07420528.2020.1809443 [DOI] [PubMed] [Google Scholar]
- 48. Strain T, Wijndaele K, Dempsey PC, et al. Wearable-device-measured physical activity and future health risk. Nat Med. 2020;26(9):1385–1391. doi: 10.1038/s41591-020-1012-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 49. Park YM, White AJ, Jackson CL, Weinberg CR, Sandler DP.. Association of exposure to artificial light at night while sleeping with risk of obesity in women. JAMA Int Med. 2019;179(8):1061–1071. doi: 10.1001/jamainternmed.2019.0571 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 50. Marino M, Li Y, Rueschman MN, et al. Measuring sleep: Accuracy, sensitivity, and specificity of wrist actigraphy compared to polysomnography. Sleep. 2013;36(11):1747–1755. doi: 10.5665/sleep.3142 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 51. de Zambotti M, Cellini N, Goldstone A, Colrain IM, Baker FC.. Wearable sleep technology in clinical and research settings. Med Sci Sports Exerc. 2019;51(7):1538–1557. doi: 10.1249/MSS.0000000000001947 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 52. Knutson KL, Rathouz PJ, Yan LL, Liu K, Lauderdale DS.. Intra-individual daily and yearly variability in actigraphically recorded sleep measures: the CARDIA study. Sleep. 2007;30(6):793–796. doi: 10.1093/sleep/30.6.793 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
UK Biobank data used in this study are globally available to approved researchers through the UK Biobank research portal (https://www.ukbiobank.ac.uk/). Sleep exposures generated from the original data were derived using open-source software (GGIR version 2.6; https://cran.r-project.org/web/packages/GGIR/index.html) and will be returned to the UK Biobank for future distribution.