Abstract
Study Objectives
To describe racial/ethnic differences in sleep duration, continuity, and perceived sleep quality in postmenopausal women and to identify statistical mediators of differences in sleep characteristics.
Methods
Recruited from the observational Study of Women’s Health Across the Nation (SWAN), 1,203 (548 white, 303 black, 147 Chinese, 132 Japanese, and 73 Hispanic; mean age 65 years, 97% postmenopausal) women participated in a week-long actigraphy and daily diary study in 2013–2015. Actigraphic measures of sleep duration and wake after sleep onset (WASO), and diary-rated sleep quality were averaged across the week. Candidate mediators included health-related variables; stress; and emotional well-being assessed up to 13 times across 18 years from baseline to sleep study.
Results
Whites slept longer than other groups; the significant mediators were concurrent financial hardship and increasing number of stressors for Hispanics or Japanese versus whites. Whites had less WASO than blacks and Hispanics; significant mediators were concurrent number of health problems, physical inactivity, waist circumference, vasomotor symptoms, number of life stressors, and financial hardship, and increasing number of health problems from baseline to sleep study. Whites reported better sleep quality than blacks, Chinese, and Japanese; significant mediators were concurrent physical inactivity, vasomotor symptoms, positive affect, and depressive symptoms.
Conclusions
Sleep differences between blacks or Hispanics versus whites were mediated by health problems, number of stressors, and financial hardship, whereas sleep differences between Chinese or Japanese versus whites were mediated by emotional well-being. This is the first study using formal mediational approaches.
Keywords: sleep continuity, duration, quality, actigraphy, race, ethnicity, women postmenopause
Statement of Significance.
Ethnic/racial groups vary substantially in sleep characteristics but reasons for the variation have not been evaluated through formal statistical modeling. The analysis reveals that more psychosocial and health variables facilitate understanding ethnic differences in sleep continuity measured by actigraphy and subjective sleep quality by diary than in sleep duration in postmenopausal women. Financial hardship was a strong mediator for white versus black and Hispanic comparisons. A unique feature of this prospective study was that increasing health problems and number of stressors across 18 years were related to subsequent sleep characteristics. Clinical interventions in midlife women may prevent sleep difficulties in postmenopausal years.
Introduction
Poor sleep is a major correlate and determinant of deteriorating health. Poor self-reported sleep quality leads to daytime sleepiness, fatigue, impaired memory, and adverse mood [1]. Beyond these subjective complaints, short sleep duration as well as long sleep duration have adverse effects on risk for cardiovascular disease (CVD), and early mortality [2, 3]. Insomnia or problems with sleep continuity as well as sleep disordered breathing are associated with subjective complaints, CVD risk, and poor functioning [4, 5].
Key dimensions of sleep differ by racial/ethnicity group. Meta-analyses show that blacks have shorter sleep duration, less efficient sleep, and more sleep disordered breathing diagnoses, relative to whites [6–8]. In contrast, blacks report fewer sleep complaints, primarily of problems of sleep maintenance. Recent epidemiological studies extended to other racial/ethnic groups show that Hispanics, Asians, and blacks have shorter sleep duration than whites [9, 10]; blacks, Hispanics, and Asians report poorer sleep quality than do whites [10]; Hispanic males have shorter sleep than non-Hispanic males [11]; blacks have lower sleep efficiency than whites [12]; and older Hispanics have greater age-related increases in severity of insomnia symptoms than non-Hispanic whites, with non-Hispanic whites and Hispanics reporting more insomnia symptoms over time than blacks [13].
Given that key dimensions of sleep vary by race/ethnicity, investigators have adjusted statistically for group differences in socioeconomic status; neighborhood characteristics; health problems; stress and discrimination; and emotional distress. Usually these factors are considered as confounders or covariates and often the racial/ethnicity differences in sleep remain, albeit somewhat attenuated. For example, in the CARDIA ancillary sleep study, following statistical adjustments for age, gender, education, depressive symptoms, body mass index (BMI), hypertension and diabetes, black/white differences remained in sleep duration measured by actigraphy [10]. In the NHANES 2007–2008 sample, following statistical adjustments for age, sex, marital status, overall rated health, sociodemographic variables, and health insurance, blacks and Hispanics reported very short sleep duration, relative to whites [9]. By inference then, the confounders and covariates may account in part for differences by race/ethnicity, assuming change in the estimates of associations of sleep and race/ethnicity. However, usually multivariate models are presented such that it is not clear which covariates may be accounting for the reduction or attenuation of race/ethnic differences.
Furthermore, in some cases, the presumed covariate predicts sleep differences in one group but not another or the effect is a different direction. For example, among whites greater educational attainment was associated with longer sleep duration, whereas among blacks and Hispanics, it was associated with shorter sleep duration [14]. Among whites, increasing professional responsibility within a given industry was associated with longer self-reported sleep duration whereas among blacks, increasing professional responsibility was associated with self-reported shorter sleep [15]. Reports of depression and anxiety were related to self-reported shorter sleep among blacks but not among whites [16]. Thus, from this analytic approach one cannot determine whether the covariates actually account for racial/ethnic differences but can indicate the relative importance of the covariate for one group or the other.
Another approach to elucidating the basis of race/ethnic differences in sleep characteristics is to test formal mediation models. One approach to testing mediation assumes that race/ethnicity is related to the candidate mediator; the mediator is related to the sleep outcome, independent of race/ethnicity; and the mediator significantly accounts for at least part of the association between race/ethnicity and the sleep outcome. To our knowledge no studies of racial/ethnic differences in sleep have taken this approach, despite the strong recommendation that future research is needed to evaluate mediators, rather than confounders or moderators [6].
The objectives of the present study are threefold. First, we sought to document the extent of racial/ethnic differences in sleep characteristics in women enrolled in the Study of Women’s Health Across the Nation (SWAN). Participants were white, black, Chinese, Japanese, and Hispanic almost all postmenopausal women. Second, we evaluated mediational models that tested the extent to which health-related indicators, stress, and emotional well-being measured concurrently with the sleep protocol mediated statistically significant associations. Third, because the women in SWAN had participated in an 18-year longitudinal study that began when the women were premenopausal, we evaluated mediational models that examined whether trajectories or slopes of increasing health problems and stress, and declining well-being from baseline through the sleep study statistically mediated race/ethnic differences in sleep.
We focused on three sleep characteristics: duration and minutes wake after sleep onset (WASO) measured during a week-long wrist actigraphy protocol; and subjective daily ratings of sleep quality recorded in a daily sleep diary completed concurrently with actigraphy. We were particularly interested in understanding mediators for WASO and sleep quality because midlife women increasingly complain about waking up in the middle of their sleep period, the high prevalence of women reporting vasomotor symptoms that can disturb sleep, and the menopausal transition is uniquely related to reports of insomnia symptoms [17–20].
Methods
SWAN study design and participants
SWAN is a multi-ethnic, community-based, multi-site cohort study of the menopausal transition. A total of 3,302 women were enrolled for longitudinal evaluation from a community-based survey conducted at seven sites: Boston, MA, Chicago, IL, Detroit area, MI, Los Angeles and Oakland, CA, Newark, NJ, and Pittsburgh, PA. Each site recruited a sample of Caucasian and minority women. Women self-identified at recruitment their ethnicity or race as white or Caucasian, black or African American, Japanese, Chinese, and Hispanic. Eligibility criteria were aged 42–52 years, not pregnant, not using exogenous hormones in the 3 months before interview, premenopausal or early perimenopausal with an intact uterus and at least one ovary, at least one menstrual period in the 3 months before baseline interview self-identifying as Caucasian or the site’s designated race/ethnic group. Each site’s institutional review board approved the study and all women gave written informed consent to participate [21].
SWAN sleep study participants
Starting in 2015 at the time of visit 15, 1,333 women were enrolled from seven sites, with sites a priori targeting 141–225 participants, depending on the number of potential eligible women at a given site and the racial/ethnic composition. The sites that recruited Hispanics, Chinese, and Japanese preferentially recruited those women because they represented fewer women in SWAN than blacks or whites. Exclusionary criteria were wheelchair bound, blind, use of mechanical devices to treat sleep apnea, and traveling across time zones during the sleep protocol period. Two hundred eighty-five declined participation. Of the 1,333 who enrolled, 1,217 had usable data and met study criteria of providing at least four nights of valid actigraphy data. Of these women, the 14 who were night shift workers were excluded from analysis because of their distinctive sleep patterns, characteristic of shift workers. Participants in the analytic sample did not differ from other SWAN participants at visit 15 regarding self-assessment of sleep duration (p = 0.30), insomnia symptoms in the last 2 weeks [report measure trouble falling asleep (p = 0.86), waking up several times a night (p = 0.51), waking up earlier than planned (p = 0.84)], and satisfaction with sleep (p = 0.68). By design participants in the analytic sample differed in the proportion of women by racial/ethnic group, with proportionately more Hispanic, Chinese, and Japanese women enrolled, compared with those who attended visit 15 but were not enrolled (p < 0.001). Participants in the analytic sample had lower body mass index (BMI; mean [M]s = 28.8 vs. 30.0, p = 0.001), waist circumference (Ms = 91.09 vs. 94.00, p = 0.001), and depressive symptoms scores (Ms = 6.73 vs. 8.15, p < 0.001).
Sleep study protocol
Women wore an actigraphy device (AW-2 Phillips Respironics) continuously on their nondominant wrist for a week. They completed a morning and evening diary each day of the protocol. Women were instructed via slide presentation and by the research team on how to complete the sleep diaries, to press the event-marker when they went to bed to sleep and when they got out of bed to start the day, and to wear the actigraph continuously. If they took off the actigraph, they recorded in their sleep diaries the time off and on.
The sleep diary completed at night asked about naps during the day. The diary completed in the morning asked about when they got into bed, tried to go to sleep, time to fall asleep, number of times awakening during the night, time of waking up in the morning, reports of hot flashes or night sweats during the night, that is vasomotor symptoms, whether they took “prescription or over the counter medications” to help sleep, and overall quality of sleep. Sleep quality was rated on a 5-point scale from very poor (1) to very good (5). These values were averaged over the measurement period.
The actigraphy analysis approach and coding were developed by the Sleep and Chronobiology center, University of Pittsburgh, with procedures consistent with the Society of Behavioral Sleep Medicine recommendations [22]. The AW-2 accelerometer was set at 0.05 g for 3–11 Hz. The analog signal was digitized by the digital integration method. The wake threshold was set at 40 counts per minute and data sampled in 1-min epochs and analyzed with sleep detection algorithm by Actiware 4.0.9 software. All data were sent via secure shell protocol to the coordinating center for data reduction and scoring. Actigraphy files were scored and edited with the diary in hand, with start and stop times detected by automated algorithms, followed by hand-editing to correct misidentified sleep times. In the cases where a rest interval was within 30 min of a participant primary rest interval, the rest time was incorporated into the primary rest interval. The actigraphy variables being reported herein include: sleep duration, which reflects the sum of those epochs between sleep start and sleep end scored as sleep; and WASO, which is the total number of minutes from the sleep start to sleep end when actigraphy indicated wake epochs. These data were averaged across the measurement period. Of the women in the analytic sample, 98% had at least 6 days of actigraphy sleep recordings.
Control variables
Analyses controlled for age at the time of the sleep study; educational attainment measured at baseline (high school/GED or less, some college, 4-year college degree, and more than 4-year college degree); employed at the time of the sleep study (yes/no); and diary-assessed proportion of nights reporting medications to aid sleep. In addition, natural postmenopausal status (yes/no) was considered but was nonsignificant in the univariable analyses of sleep characteristics (p > 0.40).
Potential mediators
Based on the literature, we identified candidate mediators to test in three general categories: health-related variables, stress, and emotional well-being. Health-related variables included total number of self-reports of health problems diagnosed by a health care professional out of a possible 10 reported in interview at each visit (anemia, diabetes, hypertension, high cholesterol, migraines, osteoarthritis, hyper- or hypo-active thyroid, heart attack or angina, osteoporosis, cancers); BMI measured at each visit by stadiometer; waist circumference measured at each visit by a trained staff member; leisure time physical activity reported via questionnaire at seven visits and visit 15; and proportion of nights that participant reported experiencing vasomotor symptoms in their diary. Stressful life events were measured at each visit questionnaire asking about the occurrence of 18 events and the extent to which they were somewhat to very upsetting; these included serious family problems, work-related problems, and death of significant others. The total number of events that were rated as upsetting constituted the measure of stress. To assess financial stress, women were asked about difficulty making ends meet to pay for basics, with responses of hard, somewhat hard, or not at all hard, at all visits. Emotional well-being was measured annually via the Center for Epidemiological Studies Depression (CES-D), a 20-item scale that is widely used in epidemiological studies to measure depressive symptoms in the last 2 weeks [23]. The item measuring restless sleep was removed from the total score to avoid confounding with the sleep measures. The Positive and Negative Affect Schedule (PANAS) positive affect subscale was measured starting in follow-up years 6 through 10, 12, 13, and 15 [24]. It contains 10 adjectives (e.g. excited, enthusiastic, and attentive) that women rate how they felt on a 5-point scale from very slightly or not at all to extremely.
Statistical analyses
Sleep duration and quality were normally distributed. WASO was right skewed and was square root transformed. Candidate mediators measured at the time of visit 15 when they were enrolled in the sleep study were the study’s primary focus. Slopes of the potential mediators from baseline to the time of the sleep study were calculated using the MPLUS Version 7.3 linear growth models, such that a positive slope indicated increasing exposure over time, whereas a negative slope indicated decreasing exposure over time. This method allowed for missing data at interim visits. Initial analyses by ANOVA and subsequent contrasts by post hoc tests (Tukey’s b) compared the race/ethic groups on the sleep variables, control variables, and potential mediators.
Multivariate regression models adjusting for the control variables (age, educational attainment, proportion of nights taking sleep medications while in the actigraphy protocol, and employment status) compared the sleep characteristics of each race/ethnic group separately with whites serving as the referent group, because all sites recruited whites and comprised the largest group. There are a number of approaches to testing for mediation. Tests for mediation were conducted only for those sleep characteristics that differed significantly between a specific racial/ethnic group and whites, adjusting for control variables. We calculated the relationship of each racial/ethnic group versus whites with each candidate mediator, adjusted for control variables (path a); each candidate mediator with sleep characteristic, adjusting for control variables, and each racial/ethnic group and whites (path b); and direct effect (path c′), which is the relationship between each race/ethnicity group and whites and sleep measure taking into account the candidate mediator and control variables. Path c is the total effect, that is the relationship between each racial/ethnic group and whites and sleep measure, adjusting for control variables only, that is without considering the candidate mediator. See Figure 1 for an illustration of the paths. The indirect effect was calculated as the product of path a and path b, and its confidence intervals were calculated using bootstrapping sampling methods (5,000 computations; PROCESS 3.0 for nominal predicator variables [25]). To be considered a candidate mediator, both path a and path b must be statistically significant in the expected direction, that is that the mediator results in path c′, the direct effect, being smaller than path c, the total effect. This approach assumes no misspecification of temporal ordering; that the residuals for the relationship between sleep and race and sleep with race/ethnicity, adjusting for the mediator, are independent; and that the mediator and residuals between the sleep outcome and mediator and race/ethnicity are independent [26]. To simplify presentation, we table only models for candidate mediators that met those requirements at p < 0.10.
Figure 1.
Hypothetical relationships between race/ethnicity and sleep characteristics through candidate mediators. Path “c”, total effect, represents the associations of race/ethnicity and a sleep characteristic, with adjustment for control variables. Path a represents the association between race/ethnicity and a mediator, adjusted for control variables. Path b represents the association between the mediator and a sleep characteristic, adjusted for control variables and race/ethnic comparison. Path c′, direct effect, represents the associations of race/ethnicity and a sleep characteristic, adjusted for the mediator and control variables. The difference between path c′ and path c is the estimate of the indirect effect of the mediator.
Results
Participant characteristics
Women who participated in the sleep protocol were on average in their mid-sixties (Table 1). Even though almost all were postmenopausal, nearly 1/3 reported vasomotor symptoms on at least one protocol night. It was a well-educated group, with most having some college or more. Almost half were currently working for pay and 28% reported any financial hardship. Over the course of SWAN assessments, women reported increases in number of health problems, BMI, and positive affect, and decreased in physical activity, chronic stress, financial hardship, and depressive symptoms.
Table 1.
Mean (SD or SE) N (%) of study participants
| Total | White | Black | Hispanic | Chinese | Japanese | |
|---|---|---|---|---|---|---|
| N | 1,203 | 548 | 303 | 73 | 147 | 132 |
| Age at sleep study | 64.97 (2.66) | 64.92 (2.68) | 64.79 (2.65) | 65.32 (2.92) | 65.20 (2.47) | 65.15 (2.66) |
| Education attainment, N (%)*** | ||||||
| High school or less | 247 (20.7) | 74 (13.6) | 72 (24.2) | 45 (62.5) | 36 (24.5) | 20 (15.2) |
| Some college | 369 (30.9) | 150 (27.5) | 123 (41.3) | 20 (27.8) | 30 (20.4) | 46 (34.8) |
| 4 Year college degree | 268 (22.4) | 133 (24.4) | 47 (15.8) | 5 (6.9) | 46 (31.3) | 37 (28.0) |
| Graduate degree | 311 (26.0) | 189 (34.6) | 56 (18.8) | 2 (2.8) | 35 (23.8) | 29 (22.0) |
| Currently employed, N (%)** | 560 (47.1) | 285 (52.4)c | 123 (40.9)a,b | 23 (33.8)a | 64 (43.5)a,b,c | 65 (50.4)b,c |
| Sleep medications during sleep study | ||||||
| N (%) ever taking medications*** | 253 (21.7) | 139 (25.6)b | 56 (18.9)a | 17 (34.7)b | 19 (13.0)a | 22 (16.7)a |
| Proportion of nights taking medications*** | 0.12 (0.29) | 0.16 (0.32)b,c | 0.10 (0.25)a,b | 0.23 (0.37)c | 0.05 (0.18)a | 0.08 (0.24)a,b |
| Menopausal status, N (%)** | ||||||
| Natural | 1,103 (91.7) | 508 (92.7)b | 264 (87.1)a | 65 (89.0)a,b | 139 (94.6)b | 127 (96.2)b |
| Other | 100 (8.3) | 40 (7.3) | 39 (12.9) | 8 (11.0) | 8 (5.4) | 5 (3.8) |
| Number of health problems | ||||||
| At sleep study*** | 1.82 (1.41) | 1.67 (1.38)a | 2.27 (1.41)b | 2.68 (1.47)c | 1.31 (1.25)a | 1.51 (1.24)a |
| Slope from baseline to sleep study*** | 0.08 (0.002)*** | 0.08 (0.06)b | 0.10 (0.07)c | 0.13 (0.07)d | 0.06 (0.06)a | 0.07 (0.05)a,b |
| Body mass index (kg/m2) | ||||||
| At sleep study*** | 28.80 (6.90) | 28.76 (6.36)b | 32.80 (7.04)c | 31.64 (6.44)c | 23.77 (3.72)a | 23.67 (4.28)a |
| ≥30, N (%)*** | 442 (37.1) | 199 (36.6)c | 182 (60.5)a | 38 (52.1)a | 10 (7.0)b | 13 (9.8)b |
| Slope from baseline to sleep study*** | 0.12 (0.006)*** | 0.12 (0.21)b,c | 0.13 (0.26)b,c | 0.17 (0.22)c | 0.05 (0.14)a | 0.08 (0.14)a,b |
| Waist circumference | ||||||
| At sleep study*** | 91.09 (15.22) | 91.59 (14.89)b | 97.77 (13.78)c | 101.32 (14.73)c | 81.04 (9.88)a | 78.88 (10.30)a |
| Slope from baseline to sleep study*** | 0.47 (0.014)*** | 0.48 (0.48)a | 0.16 (0.55)a | 0.76 (0.53)b | 0.35 (0.33)a | 0.40 (0.32)a |
| Physical activity | ||||||
| At sleep study*** | 2.88 (1.03) | 3.04 (1.06)c | 2.66 (0.95)b | 2.30 (0.95)a | 2.83 (0.93)b,c | 3.09 (1.08)c |
| Slope from baseline to sleep study* | 0.01 (0.001)*** | 0.01 (0.03)a | 0.005 (0.03)a | 0.02 (0.03)a | 0.01 (0.04)a | 0.01 (0.03)a |
| Vasomotor symptoms (hot flashes or night sweats) | ||||||
| N (%) ever reporting during sleep study** | 361 (30.9) | 153 (28.2)a | 117 (39.4)b | 16 (32.0)a,b | 34 (23.1)a | 41 (31.1)a,b |
| Proportion of nights*** | 0.11 (0.23) | 0.10 (0.22)a,b | 0.17 (0.28)b | 0.14 (0.27)a,b | 0.07 (0.17)a | 0.09 (0.19)a,b |
| Number of stressors | ||||||
| At sleep study*** | 2.20 (2.21) | 2.26 (2.04)b,c | 2.72 (2.52)c | 1.41 (2.02)a | 1.39 (1.75)a | 2.05 (2.36)b |
| Slope from baseline to sleep study*** | −0.03 (0.002) *** | −0.03 (0.06)a | −0.03 (0.07)a | −0.003 (0.05)b | −0.02 (0.06)a,b | −0.01 (0.06)b |
| Financial hardship at sleep study | ||||||
| At sleep study*** | 0.25 (0.49) | 0.15 (0.40)a | 0.41 (0.58)b | 0.78 (0.69)c | 0.12 (0.33)a | 0.13 (0.36)a |
| Slope from baseline to sleep study*** | −0.007 (0.001)*** | −0.005 (0.02)b | −0.009 (0.03)b | −0.020 (0.03)a | −0.004 (0.02)b | −0.009 (0.02)b |
| Depressive symptoms | ||||||
| At sleep study | 5.75 (6.72) | 5.25 (6.44) | 6.07 (7.18) | 6.18 (6.92) | 5.77 (6.41) | 6.82 (6.91) |
| Slope from baseline to sleep study*** | −0.17 (0.007)*** | −0.16 (0.22)b,c | −0.19 (0.24)b | −0.30 (0.23)a | −0.16 (0.23)b,c | −0.11 (0.20)c |
| Positive affect | ||||||
| At sleep study*** | 34.06 (8.23) | 35.29 (7.39)b | 35.20 (8.04)b | 29.67 (9.36)a | 29.97 (9.20)a | 33.15 (7.97)b |
| Slope from baseline to sleep study | 0.15 (0.008)*** | 0.14 (0.25) | 0.17 (0.27) | 0.17 (0.30) | 0.16 (0.28) | 0.14 (0.28) |
Different superscripts indicate different means from paired post hoc tests; no post hoc paired tests were conducted for categorical data.
*p < 0.05, **p < 0.01, ***p < 0.001 from ANOVA or chi-square for group differences or for slopes tests for difference from zero.
The race/ethnic groups differed in almost all study variables. With regard to the candidate mediators, blacks and Hispanics reporting more health problems at visit 15 than the other groups. BMI, waist circumference, and physical (in)activity also followed the same pattern. The proportion of nights reporting vasomotor symptoms during the sleep study varied across all groups, but post hoc paired contrasts revealed only one difference, between blacks and Chinese. Financial hardship was higher among blacks and Hispanics, and blacks reported more stressful events than other groups, except whites.
Sleep characteristics
In unadjusted analyses, sleep duration averaged 6.5 h, with almost 1/3 of the women having less than 6 h, and only 6% having 8 h or more sleep per night on average (Table 2). Women on average spent almost an hour awake after sleep onset. Sleep quality was rated as above the middle of the rating distribution. Sleep duration was longer among whites than other groups, and WASO greater among blacks and Hispanics than other groups. Sleep quality differed by group but post hoc tests did not show significant pairwise differences between pairs of racial/ethnic groups.
Table 2.
Mean (SD) or N (%) of sleep characteristics of study participants
| Total | White | Black | Hispanic | Chinese | Japanese | |
|---|---|---|---|---|---|---|
| Sleep duration (h) by actigraphy, M (SD)*** | 6.50 (0.99) | 6.78 (0.88)c | 6.29 (0.98)a,b | 6.50 (1.06)b | 6.22 (1.02)a | 6.09 (1.00)a |
| ≥8 h, N (%) | 71 (5.9) | 42 (7.7) | 12 (4.0) | 8 (11.0) | 5 (3.4) | 4 (3.0) |
| 6–7.99 h, N (%) | 766 (63.7) | 404 (73.7) | 173 (57.1) | 38 (52.1) | 80 (54.4) | 71 (53.8) |
| < 6 h, N (%) | 366 (30.4) | 102 (18.6) | 118 (38.9) | 27 (37.0) | 62 (42.2) | 57 (43.2) |
| Wake after sleep onset (min), M (SD)*** | 51.61 (23.26) | 48.57 (22.33)a | 59.60 (26.48)b | 55.89 (22.44)b | 49.19 (21.00)a | 46.22 (16.27)a |
| Diary rating of sleep quality, M (SD)*** | 3.62 (0.61) | 3.69 (0.58)a | 3.60 (0.64)a | 3.61 (0.64)a | 3.52 (0.61)a | 3.51 (0.65)a |
Higher score on diary sleep quality is better. Different superscripts indicate different means from paired post hoc tests; no paired comparisons were conducted for sleep duration categories. No covariates were included in these comparisons.
***p < 0.001 from ANOVA.
Multivariate analyses adjusting for age, education, employment status, and medications for sleep during the sleep protocol showed that whites had longer sleep than all other groups, less WASO than blacks and Hispanics; and better diary-reported sleep quality than blacks, Chinese, and Japanese (Table 3). Whites were similar to Chinese and Japanese in WASO.
Table 3.
Regression coefficients (SE) for race/ethnic groups compared with white postmenopausal women, adjusted for age, education, employment status, and proportion of nights using medications to aid sleep
| Vs. white | Black | Hispanic | Chinese | Japanese |
|---|---|---|---|---|
| Sleep duration (h) | −0.495 (0.07)*** | −0.461 (0.14)** | −0.554 (0.09)*** | −0.653 (0.09) *** |
| Wake after sleep onset (min square root) | 0.696 (0.12)*** | 0.595 (0.24)* | 0.063 (0.14) | −0.115 (0.14) |
| Diary rating of sleep quality (1–5) | −0.094 (0.05)* | −0.019 (0.09) | −0.190 (0.06)** | −0.192 (0.06)** |
*p < 0.05, **p < 0.01, ***p < 0.001.
Mediational models for race/ethnic differences in sleep characteristics
Sleep duration
The only mediator that approached statistical significance for black–white differences was financial hardship, p = 0.053 (Table 4).
Table 4.
Mediational analyses linking race/ethnicity to sleep duration in black, Hispanic, and Japanese, compared with white postmenopausal women
| Race to mediator (path a) | Mediator to duration (path b) | Total effect (path c) | Direct effect (path c′) | Indirect effect (95% CI) | |
|---|---|---|---|---|---|
| Black/white | |||||
| Financial hardship | 0.234 (0.035)**** | −0.133 (0.068)* | −0.504 (0.069)**** | −0.473 (0.070)**** | −0.031 (−0.068, 0.005) |
| Hispanic/white | |||||
| Financial hardship | 0.593 (0.070)**** | −0.390 (0.087)**** | −0.431 (0.148)*** | −0.200 (0.154) | −0.231 (−0.382, −0.103) |
| Increasing number of stressors | 0.031 (0.010)*** | −1.277 (0.599)** | −0.461 (0.144)*** | −0.422 (0.145)*** | −0.039 (−0.086, −0.002) |
| Japanese/white | |||||
| Increasing number of stressors | 0.022 (0.006)**** | −1.141 (0.572)** | −0.653 (0.089)**** | −0.627 (0.089)**** | −0.025 (−0.055, −0.001) |
Unless noted, mediators are from the same visit as the sleep protocol; increases refer to the slope of the mediator from baseline to sleep study.
*p < 0.10, **p < 0.05, ***p < 0.01, ****p < 0.001.
For the Hispanic–white differences, financial hardship and increasing number of stressors met criteria for mediation. With financial hardship in the model, the Hispanic/white difference in sleep duration was nonsignificant, with the indirect effect of financial hardship representing 53.6% reduction in the total effect. This suggests that the difference in sleep duration between Hispanics and whites was largely due to differences in financial hardship. In addition, the slope of number of stressors represented an 8.5% reduction in the total effect.
For Japanese–white differences, the slope of number of life stressors from baseline to the sleep study was a significant mediator, representing a 3.8% reduction in the total effect.
No mediator was statistically significant for the Chinese versus white comparison.
Wake after sleep onset
Black/white
Five candidate variables measured at the time of the sleep study met criteria for mediation of black/white differences in WASO: number of health problems, proportion of days reporting vasomotor symptoms, waist circumference, number of stressful events, and financial hardship (Table 5). One candidate slope variable, increasing number of health problems from baseline to sleep study, also was a significant mediator. Taken individually, the significant mediators reduced the total effect for black/white differences in WASO by 3.6%–13.2% (M = 7.9%). If all significant mediators were in one model, the overall indirect effect estimate was 0.199 (95% CI: 0.104, 0.302), which reduced the total effect by 28.6%.
Table 5.
Mediational analyses linking race/ethnicity to wake after sleep onset (WASO) in Black and Hispanic compared with white postmenopausal women
| Race to mediator (path a) | Mediator to WASO (path b) | Total effect (path c) | Direct effect (path c′) | Indirect effect (95% CI) | |
|---|---|---|---|---|---|
| Black/white | |||||
| No. health problems | 0.595 (0.104)**** | 0.119 (0.040)*** | 0.696 (0.118)**** | 0.625 (0.120)**** | 0.071 (0.022, 0.129) |
| Waist circumference | 0.321 (0.056)**** | 0.193 (0.074)*** | 0.699 (0.118)**** | 0.637 (0.120)**** | 0.062 (0.014, 0.118) |
| Proportion vasomotor symptoms | 0.049 (0.018)*** | 0.709 (0.226)*** | 0.696 (0.118)**** | 0.661 (0.118)**** | 0.035 (0.005, 0.079) |
| No. stressors | 0.152 (0.063)** | 0.162 (0.066)** | 0.696 (0.118)**** | 0.671 (0.118)**** | 0.025 (0.001, 0.061) |
| Financial hardship | 0.234 (0.035)**** | 0.390 (0.118)*** | 0.692 (0.119)**** | 0.601 (0.122)**** | 0.091 (0.027, 0.165) |
| Increasing health problems | 0.019 (0.005)**** | 2.438 (0.835)*** | 0.696 (0.118)**** | 0.649 (0.119)**** | 0.047 (0.011, 0.096) |
| Hispanic/white | |||||
| No. health problems | 0.822 (0.220)**** | 0.125 (0.045)*** | 0.595 (0.240)** | 0.492 (0.241)** | 0.103 (0.023, 0.214) |
| Waist circumference | 0.498 (0.123)**** | 0.195 (0.081)** | 0.597 (0.240)** | 0.500 (0.243)** | 0.097 (0.014, 0.205) |
| Physical activity | −0.530 (0.180)*** | −0.149 (0.060)** | 0.567 (0.255)** | 0.488 (0.256)* | 0.079 (0.010, 0.173) |
| Financial hardship | 0.593 (0.070)**** | 0.479 (0.147)*** | 0.520 (0.247)** | 0.236 (0.260) | 0.284 (0.062, 0.558) |
| Positive affect | −4.395 (1.262)*** | −0.014 (0.008)* | 0.538 (0.248)** | 0.478 (0.250)* | 0.061 (−0.015, 0.160) |
| Increasing health problems | 0.041 (0.010)**** | 2.097 (0.976)** | 0.595 (0.240)** | 0.510 (0.242)** | 0.085 (0.006, 0.189) |
Unless otherwise noted, mediators are from the same visit as the sleep protocol; increases refer to the slope of the mediator from baseline to sleep study.
*p < 0.10, **p < 0.05, ***p < 0.01, ****p < 0.001.
Hispanic/white
Four candidate variables measured at the time of the sleep study met criteria for mediation of Hispanic/white differences in WASO: number of health problems, physical activity, waist circumference, and financial hardship. One candidate slope variable, increasing number of health problems from baseline to sleep study, was a significant mediator. Taken individually, mediators reduced the total effect for Hispanic/white differences by 13.9% to 54.6% (M = 23.3%). The model for concurrent financial hardship showed full mediation, meaning that Hispanic/white difference in WASO was nonsignificant with the mediator in the model, that is path c′. If all significant mediators were in one model, the indirect effect estimate was 0.354 (95% CI: 0.116, 0.650), which reduced the total effect by 64.0%.
Perceived sleep quality
Black/white
Four candidate variables measured at the time of the sleep study met criteria for mediation testing black/white differences in sleep quality: physical activity, proportion of days reporting vasomotor symptoms, number of stressful events, and financial hardship (Table 6). Taken individually, all significant mediators reduced the total effect by 11.7%–34.4% (M = 23.4%). Models for physical activity, proportion of days reporting vasomotor symptoms, and financial hardship did not reveal a significant direct effect for black/white differences in sleep quality. Taking all significant mediators together in one model, the indirect effect was −.073 (−0.110, −0.041), which was a 79.3% reduction in the total effect.
Table 6.
Mediational analyses linking race/ethnicity to diary ratings of sleep quality in black, Chinese, and Japanese, compared with white postmenopausal women
| Race to mediator (path a) | Mediator to quality rating (path b) | Total effect (path c) | Direct effect (path c′) | Indirect effect (95% CI) | |
|---|---|---|---|---|---|
| Black/white | |||||
| Physical activity | −0.319 (0.077)**** | 0.053 (0.021)** | −0.090 (0.045)** | −0.073 (0.045) | −0.017 (−0.034, −0.004) |
| Proportion vasomotor symptoms | 0.049 (0.018)*** | −0.558 (0.084)**** | −0.094 (0.045)** | −0.066 (0.044) | −0.027 (−0.052, −0.007) |
| No. stressors | 0.152 (0.063)** | −0.074 (0.025)*** | −0.094 (0.045)** | −0.083 (0.045)* | −0.011 (−0.026, −0.001) |
| Financial hardship | 0.234 (0.035)**** | −0.140 (0.045)*** | −0.096 (0.045)** | −0.063 (0.046) | −0.033 (−0.059, −0.009) |
| Chinese/white | |||||
| Positive affect | −4.974 (0.742)**** | 0.018 (0.003)**** | −0.190 (0.055)*** | −0.099 (0.055)* | −0.091 (−0.137, −0.052) |
| Japanese/white | |||||
| Depressive symptoms | 0.338 (0.140)** | −0.096 (0.016)**** | −0.192 (0.058)*** | −0.159 (0.057)*** | −0.033 (−0.065, −0.005) |
| Positive affect | −2.204 (0.741)*** | 0.020 (0.003)**** | −0.192 (0.058)*** | −0.148 (0.056)*** | −0.043 (−0.080, −0.013) |
Unless noted, mediators are from the same visit as the sleep protocol; increases refer to the slope of the mediator from baseline to sleep study.
*p < 0.10, **p < 0.05, ***p < 0.01, ****p < 0.001.
Chinese/white
One candidate variable met criteria for mediation testing Chinese/white differences in sleep quality, that is positive affect. The indirect effect of positive affect was significant, with it accounting for 47.9% of the race/ethnic differences in sleep quality. The direct effect of race/ethnicity became marginal (p < 0.10).
Japanese/white
Candidate variables that met criteria for mediation of Japanese/white differences in sleep quality were depressive symptoms and positive affect at the time of the sleep study. The indirect effects for those variables were significant, reducing the estimate for total effects for race by 17.2%–22.4% (M = 19.8%). Combining the two significant mediators into one model, the indirect effect was −.050 (−0.089, −0.017), reducing the total effect of Japanese/white differences by 26.0%.
Discussion
This article demonstrated substantial differences in sleep duration and WASO measured by actigraphy and perceived sleep quality measured by daily diary between black, Hispanic, Chinese, and Japanese women compared with white women. Whites had longer sleep duration than all other groups, whites had less WASO than blacks and Hispanics, and whites reported better sleep quality than blacks, Chinese, and Japanese women. By and large, these findings are consistent with prior epidemiological data summarized across age and gender, and focusing on black/white differences [6–8]. The present analysis provides unique information about the sleep characteristics of a large group of women, 97% postmenopausal and who represent five different racial/ethnic groups.
Our primary objective was to identify statistical mediators of the racial/ethnic differences in sleep characteristics, with an emphasis on those sleep characteristics that become more prevalent with aging in women and increase with the menopausal transition, WASO, and poor sleep quality [17, 18]. As summarized in Figure 2, our findings reveal a number of important patterns. First, there were more mediators of racial/ethnic differences in WASO and sleep quality than in sleep duration, perhaps because those characteristics change more during the menopausal transition [17, 18]. Second, sleep differences between whites and blacks or Hispanics were mediated by health problems, number of stressors, and financial hardship, whereas sleep differences between Whites and Chinese or Japanese were mediated by emotional well-being and for Japanese the slope of number of stressors from baseline through the sleep study. Third, a number of racial/ethnic comparisons showed significant modifiable mediators. Proportion of nights experiencing vasomotor symptoms for blacks, and physical inactivity and waist circumference for blacks and Hispanics were significant mediators for sleep continuity. These mediators could be the focus of clinical interventions to improve sleep, though more research is needed to substantiate the causal nature of the observed findings and to develop intervention strategies that are efficacious within each racial/ethnic group. Fourth, financial hardship was an important mediator for all three sleep characteristics for specific racial/ethnic comparisons and reduced the direct effect to nonsignificance for the Hispanic/white differences in WASO and sleep duration, and for the black/white differences in sleep quality. As all analyses included educational attainment as a covariate, we interpret financial hardship as a particularly important source of stress for postmenopausal women, independent of typical measures of socioeconomic status, and potentially a potent source of chronic sleep disturbance. Consistent with that perspective, in the National Longitudinal Study of Mature Women, accumulation of financial strain measured prospectively predicted self-reported health, independent of occupational prestige, income, wealth, and education [27]. Nonetheless, it is also possible that poor sleep impacts financial hardship through, for example, poor work performance or disturbed interpersonal relationships. Fifth, although there were striking racial/ethnic differences in BMI and waist circumference, only waist circumference was a significant mediator of blacks and Hispanics versus whites in sleep continuity. This was particularly surprising as BMI is often correlated with short sleep, as it was in a subset of the present sample who participated in an ancillary study that measured sleep duration by in-home polysomnography [28]. Our findings suggest that regional location of fat accumulation may be a more important determinant of sleep disturbance than overall adiposity.
Figure 2.
Summary of significant mediators by racial/ethnic group compared with whites and organized by mediators. In parentheses are the sleep characteristics that showed significant indirect effects of mediators in models with race/ethnicity. WASO stands for minutes awake after sleep onset; quality of sleep rated in a daily diary; duration for sleep time; VMS for vasomotor symptoms, and N for number.
A unique feature of our study is the opportunity to examine candidate mediators that changed from the time women were premenopausal to the time of the sleep study, that is over an 18-year period. For WASO, increasing health problems were significant mediator for Hispanics or blacks, relative to whites, and for sleep duration, increasing number of stressors or a smaller decline in number of stressors were significant mediators for Hispanics or Japanese, relative to whites. These findings suggest that midlife changes in health and stress may be important precursors of postmenopausal sleep disturbance and may also be a target for prevention of poor sleep in the postmenopausal years [29–31]. From a life-span perspective, even earlier exposures to poor health and stress in adolescence and young adulthood may have set the stage for sleep disturbance in midlife and beyond.
Study limitations include the constraints of mediation modeling of observational data. Those are that statistically significant mediational result does not prove that the variable is a mediator. For example, the variable could be correlated with a third factor that is the “true” mediator. It also does not allow researcher to distinguish between alternative causal models [32]. Our study had no objective measure of sleep apnea, although women were ineligible if they reported using mechanical devices to treat apnea. The study findings cannot be generalized to understanding racial/ethnic differences in sleep characteristics in men. The racial/ethnic differences in sleep characteristics were for the most part not completely accounted for by candidate mediators, suggesting that other mediators need to be considered. These include those suggested by the existing literature, including multiple indicators of socioeconomic status, for example income, wealth, occupational prestige; neighborhood characteristics, for example exposure to violence, noise, and pollution; and discrimination and unfair treatment [33, 34]. The number of women who self-identified as Hispanic, Japanese, and Chinese were relatively small so the absence of mediators in these racial/ethnic groups should be interpreted with caution. Also, the Hispanic women were of diverse cultural origins. Study strengths include use of actigraphy to measure sleep duration and WASO, a diverse sample of postmenopausal women, a focus on key dimensions of sleep important for health and quality of life, and formal mediational testing.
In conclusion, postmenopausal women vary substantially in sleep duration, continuity, and perceived quality by race/ethnicity. Key statistical mediators of associations between race/ethnicity and WASO and sleep quality were health problems, stress, including financial hardship, and low positive affect and depressive symptoms. Our findings help understand factors that may play key roles in why women of different racial/ethnic groups differ substantially in their sleep in the postmenopausal years and identify a number of clinical targets that may improve sleep health in women.
Funding
The Study of Women’s Health Across the Nation (SWAN) has grant support from the National Institutes of Health (NIH), DHHS, through the National Institute on Aging (NIA), the National Institute of Nursing Research (NINR) and the NIH Office of Research on Women’s Health (ORWH) (U01NR004061; U01AG012505, U01AG012535, U01AG012531, U01AG012539, U01AG012546, U01AG012553, U01AG012554, and U01AG012495). The content of this article is solely the responsibility of the authors and does not necessarily represent the official views of the NIA, NINR, ORWH, or the NIH. This article also received support from R01AG053838 awarded to Hadine Joffe, PI.
Conflict of interest statement. This was not an industry-sponsored study. Dr Joffe is a recipient of grants from Merck and Pfizer, and is a consultant for Merck, MeRRe/KaNDy, and Sojourner. Her spouse is a consultant for Tango, Iomx, and Blueprint. No other authors have financial or nonfinancial disclosures.
Acknowledgments
Clinical centers: University of Michigan, Ann Arbor—Siobán Harlow, PI 2011–present, MaryFran Sowers, PI 1994–2011; Massachusetts General Hospital, Boston, MA—Joel Finkelstein, PI 1999–present; Robert Neer, PI 1994–1999; Rush University, Rush University Medical Center, Chicago, IL—Howard Kravitz, PI 2009–present; Lynda Powell, PI 1994–2009; University of California, Davis/Kaiser—Ellen Gold, PI; University of California, Los Angeles—Gail Greendale, PI; Albert Einstein College of Medicine, Bronx, NY—Carol Derby, PI 2011–present, Rachel Wildman, PI 2010–2011; Nanette Santoro, PI 2004–2010; University of Medicine and Dentistry—New Jersey Medical School, Newark—Gerson Weiss, PI 1994–2004; and the University of Pittsburgh, Pittsburgh, PA—Karen Matthews, PI.
NIH program office: National Institute on Aging, Bethesda, MD—Chhanda Dutta 2016–present; Winifred Rossi 2012–2016; Sherry Sherman 1994–2012; Marcia Ory 1994–2001; National Institute of Nursing Research, Bethesda, MD—Program Officers.
Central laboratory: University of Michigan, Ann Arbor—Daniel McConnell (Central Ligand Assay Satellite Services).
Coordinating center: University of Pittsburgh, Pittsburgh, PA—Maria Mori Brooks, PI 2012–present; Kim Sutton-Tyrrell, PI 2001–2012; New England Research Institutes, Watertown, MA—Sonja McKinlay, PI 1995–2001.
Steering committee: Susan Johnson, Current Chair; Chris Gallagher, Former Chair. We thank the study staff at each site and all the women who participated in SWAN.
References
- 1. Institute of Medicine (US) Committee on Sleep Medicine and Research. Sleep Disorders and Sleep Deprivation: An Unmet Public Health Problem. Washington, DC: National Academies Press; 2006. [PubMed] [Google Scholar]
- 2. Cappuccio FP, et al. Sleep duration predicts cardiovascular outcomes: a systematic review and meta-analysis of prospective studies. Eur Heart J. 2011;32(12):1484–1492. [DOI] [PubMed] [Google Scholar]
- 3. Grandner MA, et al. Mortality associated with short sleep duration: the evidence, the possible mechanisms, and the future. Sleep Med Rev. 2010;14(3):191–203. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Knutson KL, et al. The metabolic consequences of sleep deprivation. Sleep Med Rev. 2007;11(3):163–178. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Sofi F, et al. Insomnia and risk of cardiovascular disease: a meta-analysis. Eur J Prev Cardiol. 2014;21(1):57–64. [DOI] [PubMed] [Google Scholar]
- 6. Petrov ME, et al. Differences in sleep between black and white adults: an update and future directions. Sleep Med. 2016;18:74–81. [DOI] [PubMed] [Google Scholar]
- 7. Ruiter ME, et al. Sleep disorders in African Americans and Caucasian Americans: a meta-analysis. Behav Sleep Med. 2010;8(4):246–259. [DOI] [PubMed] [Google Scholar]
- 8. Ruiter ME, et al. Normal sleep in African-Americans and Caucasian-Americans: a meta-analysis. Sleep Med. 2011;12(3):209–214. [DOI] [PubMed] [Google Scholar]
- 9. Whinnery J, et al. Short and long sleep duration associated with race/ethnicity, sociodemographics, and socioeconomic position. Sleep. 2014;37(3):601–611. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Carnethon MR, et al. Disparities in sleep characteristics by race/ethnicity in a population-based sample: Chicago Area Sleep Study. Sleep Med. 2016;18:50–55. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Dietch JR, et al. Gender and racial/ethnic differences in sleep duration in the North Texas Heart Study. Sleep Health. 2017;3(5):324–327. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Curtis DS, et al. Habitual sleep as a contributor to racial differences in cardiometabolic risk. Proc Natl Acad Sci USA. 2017;114(33):8889–8894. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Kaufmann CN, et al. Racial/ethnic differences in insomnia trajectories among U.S. older adults. Am J Geriatr Psychiatry. 2016;24(7):575–584. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Cunningham TJ, et al. Independent and joint associations of race/ethnicity and educational attainment with sleep-related symptoms in a population-based US sample. Prev Med. 2015;77:99–105. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Jackson CL, et al. Racial disparities in short sleep duration by occupation and industry. Am J Epidemiol. 2013;178(9):1442–1451. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Seixas AA, et al. Differences in short and long sleep durations between blacks and whites attributed to emotional distress: analysis of the National Health Interview Survey in the United States. Sleep Health. 2017;3(1):28–34. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Kravitz HM, et al. Sleep disturbance during the menopausal transition in a multi-ethnic community sample of women. Sleep. 2008;31(7):979–990. [PMC free article] [PubMed] [Google Scholar]
- 18. Kravitz HM, et al. Sleep trajectories before and after the final menstrual period in the Study of Women’s Health Across the Nation (SWAN). Curr Sleep Med Rep. 2017;3(3):235–250. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Moline ML, et al. Sleep in women across the life cycle from adulthood through menopause. Sleep Med Rev. 2003;7(2):155–177. [DOI] [PubMed] [Google Scholar]
- 20. Avis NE, et al. ; Study of Women’s Health Across the Nation. Duration of menopausal vasomotor symptoms over the menopause transition. JAMA Intern Med. 2015;175(4):531–539. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Sowers M, et al. SWAN: a multi-center, multi-ethnic community-based cohort study of women and the menopausal transition. In: Lobo R, Marcus R, Kelsey J, eds. Menopause:Biology and Pathology. New York: Academic Press; 2000. [Google Scholar]
- 22. Ancoli-Israel S, et al. The SBSM guide to actigraphy monitoring: clinical and research applications. Behav Sleep Med. 2015;13 (Suppl. 1):S4–S38. [DOI] [PubMed] [Google Scholar]
- 23. Radloff LS. The CES-D scale: a self-report depression scale for research in the general population. Appl Psychol Measure. 1977;1:385–401. [Google Scholar]
- 24. Watson D, et al. Development and validation of brief measures of positive and negative affect: the PANAS scales. J Pers Soc Psychol. 1988;54(6):1063–1070. [DOI] [PubMed] [Google Scholar]
- 25. Hayes AF, et al. Statistical mediation analysis with a multicategorical independent variable. Br J Math Stat Psychol. 2014;67(3):451–470. [DOI] [PubMed] [Google Scholar]
- 26. MacKinnon DP, et al. Mediation analysis. Annu Rev Psychol. 2007;58:593–614. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Shippee TP, et al. Accumulated financial strain and women’s health over three decades. J Gerontol B Psychol Sci Soc Sci. 2012;67(5):585–594. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Appelhans BM, et al. Sleep duration and weight change in midlife women: the SWAN sleep study. Obesity (Silver Spring). 2013;21(1):77–84. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Hall MH, et al. Sleep is associated with the metabolic syndrome in a multi-ethnic cohort of midlife women: the SWAN Sleep Study. Sleep. 2012;35(6):783–790. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Matthews KA, et al. Do reports of sleep disturbance relate to coronary and aortic calcification in healthy middle-aged women? Study of Women’s Health Across the Nation. Sleep Med. 2013;14(3):282–287. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Taylor BJ, et al. Bedtime variability and metabolic health in midlife women: the SWAN sleep study. Sleep. 2016;39(2):457–465. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Fiedler K, et al. What mediation analysis can (not) do. J Exp Soc Psychol. 2011;47:1231–1236. [Google Scholar]
- 33. Owens SL, et al. Association between discrimination and objective and subjective sleep measures in the midlife in the United States study adult sample. Psychosom Med. 2017;79(4):469–478. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Johnson DA, et al. The social patterning of sleep in African Americans: associations of socioeconomic position and neighborhood characteristics with sleep in the Jackson heart study. Sleep. 2016;39(9):1749–1759. [DOI] [PMC free article] [PubMed] [Google Scholar]