Habitual sleep as a contributor to racial differences in cardiometabolic risk

Significance

Large differences in cardiovascular disease and diabetes prevalence exist between African American and European American adults. The US federal government has committed to reducing racial disparities in health; however, the precise mechanisms are not well understood. Sleep is one potential behavioral explanation for current racial differences in cardiometabolic conditions. We show that more than one-half of racial differences in cardiometabolic risk can be explained by sleep patterns—namely, less total sleep and lower sleep efficiency among African American than European American adults. Sleep is a malleable health behavior that is linked with characteristics of the social and physical environment and could be an effective target in national efforts to reduce racial health disparities.

Abstract

Insufficient and disrupted sleep is linked with cardiovascular and metabolic dysregulation and morbidity. The current study examines the degree to which differences in sleep between black/African American (AA) and white/European American (EA) adults explain racial differences in cardiometabolic (CMB) disease risk. Total sleep time and sleep efficiency (percent of time in bed asleep) were assessed via seven nights of wrist actigraphy among 426 participants in the Midlife in the United States Study (31% AA; 69% EA; 61% female; mean age = 56.8 y). CMB risk was indexed as a composite of seven biomarkers [blood pressure, waist circumference, hemoglobin A1c (HbA_1c), insulin resistance, triglycerides, HDL cholesterol (HDL-C), and C-reactive protein]. Covariates included sociodemographic characteristics and relevant health behaviors. Results indicated that AAs relative to EAs obtained less sleep (341 vs. 381 min) and had lower sleep efficiency (72.3 vs. 82.2%) (P values < 0.001). Further, 41% and 58% of the racial difference in CMB risk was explained by sleep time and sleep efficiency, respectively. In models stratified by sex, race was indirectly associated with CMB risk via sleep time and efficiency only among females (explaining 33% and 65% of the race difference, respectively). Indirect effects were robust to alternative model specifications that excluded participants with diabetes or heart disease. Consideration of sleep determinants and sleep health is therefore needed in efforts to reduce racial differences in CMB disease.

Black/African American adults (AAs) have disproportionately high rates of cardiovascular and metabolic diseases compared with white/European American (EAs), including elevated hypertension, diabetes, and stroke prevalence (1, 2). The reduction of these racial health disparities has become a pillar of the United States’ national health strategy (3, 4). However, given the complexity of risk exposures at multiple levels and numerous potential mechanisms (5, 6), the underlying reasons for AA–EA health differences are not well understood. For example, differences in health behaviors related to cardiometabolic (CMB) disease risk—namely, diet, physical activity, and smoking—have been considered as behavioral pathways through which AAs experience greater health risk, but adjusting for these behaviors leaves a substantial racial gap in CMB conditions unexplained (6–8).

Three recent review papers have proposed that differences in sleep patterns are one further behavioral mediator of the unequal cardiovascular disease and diabetes burden between AA and EA adults (9–11). Race differences are apparent across several sleep parameters even after adjusting for socioeconomic characteristics, so that AAs are more likely to sleep for an insufficient duration and to experience fragmented sleep relative to EAs (12, 13). In turn, sleep problems are risk factors for coronary heart disease, stroke, hypertension, and diabetes and predispose to more abdominal adiposity and low-grade systemic inflammation (14–19). Despite the expanding literature, few studies have examined sleep as a mediator of AA–EA differences in CMB diseases. Specifically, to our knowledge, sleep has been considered as a mediator of racial differences only in blood pressure (16, 20, 21), and therefore additional research is needed to examine CMB dysregulation more widely.

Therefore the primary aim of the current study was to test the degree to which differences in habitual sleep explain racial differences in CMB disease risk. Data were derived from 426 AA and EA adults from the Midlife in the United States (MIDUS) study who participated in a sleep substudy. Total sleep time and sleep efficiency (i.e., the percentage of the time in bed spent sleeping) were measured over seven nights using actigraphy—a watch-like activity monitor that has good correspondence with polysomnography (22). CMB risk was measured as a composite of ideal, borderline high, and high (coded as 0, 1, and 2, respectively) levels of waist circumference, hemoglobin A1c (HbA_1c), insulin resistance [homeostatic model assessment of insulin resistance (HOMA-IR)], blood pressure, triglycerides, HDL cholesterol (HDL-C), and C-reactive protein (CRP) (Table 1) (23–25). CMB risk scores were averaged across the biomarkers so that total scores ranged from 0 to 2. Additional models were fit to examine whether associations between sleep and CMB risk varied by demographic factors (race, sex, and age) and, given that AA–EA differences in obesity and diabetes are larger for females than for males (26, 27), whether mediation of racial differences in CMB risk varied by sex. To reduce the likelihood of reverse causality, we also tested study hypotheses among participants without heart disease or diabetes.

Table 1.

Biomarker cut points for CMB risk index

Biomarkers	Ideal, coded 0	Borderline, coded 1	High, coded 2
Waist circumference, cm	Male: <94	≥94 to <102	≥102
	Female: <80	≥80 to <88	≥88
Blood pressure, mmHG*	<120/80	120/80 to <140/90	≥140/90
HbA1c ,%	<5.7	5.7 to <6.5	6.5
HOMA-IR	Lowest tertile	Middle tertile	Highest tertile
Triglycerides, mg/dL	<150	150–199	≥200
HDL, mg/dL	≥60	Male: 40–59	<40
		Female: 50–59	<50
C-reactive protein, mg/L	<1	1–3	>3

When possible, borderline-high and high cut points are based on recommendations from the NCEP ATP III and the AHA (1–3). HOMA-IR, homeostatic model assessment of insulin resistance.

^*Blood pressure cut points were coded as a single biomarker using systolic and diastolic values.

Results

Sample descriptive statistics are listed by race in Table 2. Race differences were identified in several of the sociodemographics, health behaviors, and CMB markers. Notably, AAs obtained nearly 40 fewer minutes of sleep and had 10% lower sleep efficiency than EAs. Further, AAs relative to EAs had significantly larger waist circumference, higher diastolic blood pressure, HbA1_c, HOMA-IR, and CRP levels but had a more favorable lipid profile (lower triglycerides and higher HDL-C). In total, AAs had a higher unadjusted CMB risk than EAs including higher risk scores for four of the individual CMB risk scores.

Table 2.

Descriptives of AAs (n = 133) and EAs (n = 293) in the MIDUS study

Variables	AA, mean ± SD or %	EA, mean ± SD or %	Difference, P value
Age, y	54.9 ± 10.2	57.6 ± 11.8	0.016
Female, %	72.9	54.9	<0.001
Partnered, %	43.2	75.8	<0.001
Educational attainment, %
≤High school degree	44.4	26.0	<0.001
Some college	32.3	27.7	0.34
≥B.S. or equivalent	23.3	46.2	<0.001
Income, $103	39.4 ± 33.3	74.1 ± 56.1	<0.001
Exercise, min/d	35.5 ± 65.2	39.6 ± 46.0	0.47
Fast food consumption, %
None	13.6	18.1	0.26
Less than once a week	35.6	32.8	0.57
Once a week or more	50.8	49.2	0.76
Depressive symptoms	11.8 ± 9.6	6.34 ± 6.26	<0.001
Cigarette smoker, %	24.8	10.9	0.001
Total sleep time, min	340.6 ± 73.9	381.3 ± 62.9	<0.001
Sleep efficiency	72.3 ± 11.5	82.2 ± 8.7	<0.001
Waist circumference, cm	102.7 ± 19.4	97.1 ± 16.4	0.005
Systolic BP, mmHg	135.0 ± 21.1	131.7 ± 17.1	0.12
Diastolic BP, mmHg	78.8 ± 11.9	76.6 ± 10.2	0.046
HbA1c	6.72 ± 1.79	5.95 ± 0.76	<0.001
HOMA-IR	4.85 ± 4.47	3.49 ± 4.31	0.004
Triglycerides, mg/dL	114.7 ± 66.3	134.2 ± 81.1	0.009
HDL-C, mg/dL	59.1 ± 19.2	52.1 ± 16.2	<0.001
CRP, μg/mL	4.85 ± 5.87	2.79 ± 4.98	0.001
CBM risk	1.17 ± 0.46	1.04 ± 0.50	0.013

Data not designated as percentages are expressed as mean ± SD. Race differences were tested using independent samples t tests and χ² tests. BP, blood pressure.

Racial Differences in Sleep and CMB Risk.

After adjusting for sociodemographic characteristics, AAs obtained less sleep per night [regression coefficient (B) = −0.72 h, SE = 0.13, P < 0.001] and had lower sleep efficiency (B = −10.14 percentage points, SE = 1.12, P < 0.001) relative to EAs. Race differences were equivalent to an effect size of 0.63 and 0.95 SD units, respectively.

See Table 3 for parameter estimates for Models 1–4. In Model 1, AAs had elevated CMB risk relative to EAs (B = 0.15, 95% CI: 0.05, 0.26) after adjusting for sociodemographics, a difference equivalent to 0.32 SD units. When CMB risk was further adjusted for diet, exercise, cigarette smoker status, and depressive symptoms (Model 2), the racial gap in CMB risk was attenuated by 27% but remained significant (B = 0.11, 95% CI: 0.01, 0.22).

Table 3.

Series of regression models demonstrating racial disparities in CMB risk in the full sample (n = 426) after adjusting for sociodemographic covariates (Model 1), health behaviors (Model 2), total sleep time (Model 3), and sleep efficiency (Model 4)

Variables	Model 1		Model 2		Model 3		Model 4
	B	95% CI	B	95% CI	B	95% CI	B	95% CI
Race (African American = 1)	0.15**	[0.05, 0.26]	0.11*	[0.01, 0.22]	0.07	[−0.04, 0.18]	0.05	[−0.06, 0.16]
Age	0.01***	[0.01, 0.01]	0.01***	[0.01, 0.02]	0.01***	[0.01, 0.02]	0.01***	[0.01, 0.02]
Male	0.02	[−0.07, 0.11]	0.01	[−0.08, 0.10]	−0.03	[−0.12, 0.06]	−0.03	[−0.13, 0.06]
Partnered	0.09	[−0.01, 0.19]	0.09	[−0.01, 0.18]	0.11*	[0.01, 0.20]	0.10	[0.00, 0.19]
Education (referent is ≤high school degree)
Some college attendance	0.03	[−0.09, 0.15]	0.01	[−0.11, 0.13]	0.01	[−0.11, 0.13]	0.01	[−0.11, 0.13]
≥4 y college degree	0.03	[−0.09, 0.15]	0.03	[−0.08, 0.15]	0.03	[−0.08, 0.14]	0.03	[−0.08, 0.14]
Income	−0.01*	[−0.02, 0.00]	−0.01*	[−0.02, 0.00]	−0.01	[−0.02, 0.00]	−0.01**	[−0.02, 0.00]
Physical activity			−0.07*	[−0.14, 0.00]	−0.08*	[−0.15, −0.01]	−0.07*	[−0.14, 0.00]
Fast food consumption (referent is none)
Less than once a week			0.08	[−0.05, 0.21]	0.08	[−0.05, 0.21]	0.08	[−0.05, 0.21]
Once a week or more			0.19**	[0.06, 0.31]	0.19**	[0.06, 0.32]	0.18**	[0.05, 0.30]
Current cigarette smoker			0.09	[−0.05, 0.23]	0.07	[−0.06, 0.20]	0.07	[−0.07, 0.20]
Depressive symptoms			0.00	[0.00, 0.01]	0.00	[0.00, 0.01]	0.00	[0.00, 0.01]
Total sleep time, in hours					−0.07***	[−0.10, −0.03]
Sleep efficiency, %							−0.01**	[−0.01, 0.00]

Significance: *P < 0.05. **P < 0.01. ***P < 0.001.

Sleep Indicators Explain Differences in CMB Risk.

Total sleep time and sleep efficiency were added separately as predictors of CMB risk in Models 3 and 4, respectively. We found that shorter sleep time was significantly associated with greater CMB risk (B = −0.07, 95% CI: −0.10, −0.03). Furthermore, a significant indirect effect of race on CMB risk via sleep time was detected (B = 0.05, 95% CI: 0.02, 0.08), explaining 41% of the estimated racial gap, which was no longer significant. Nonlinear associations between sleep time and CMB risk were also tested using sleep time squared, but results indicated no significant association (P = 0.60).

In Model 4, lower sleep efficiency was associated with greater CMB risk (B = −0.007, 95% CI: −0.012, −0.002). Notably, racial differences in CMB risk were found to operate partly through sleep efficiency (indirect effect: B = 0.07, 95% CI: 0.02, 0.13), so that the racial gap in CMB risk was reduced by 58% and was nonsignificant. When sleep variables were simultaneously tested as mediators, the race estimate in CMB risk was similar to Model 4 (B = 0.04, 95% CI: −0.07, 0.15; attenuated by 64% from Model 2), but neither of the indirect effects (sleep time: B = 0.03, 95% CI: −0.01, 0.07; sleep efficiency: B = 0.04, 95% CI: −0.02, 0.11) was significant, likely because of multicollinearity.

We also examined race, age, and sex as moderators of the link between sleep and CMB risk (Table S1 for full results). No evidence of moderation by racial group or age was found for either of the sleep indicators. In contrast, sex moderated associations between both sleep indicators and CMB risk (sleep time: P = 0.005; efficiency: P = <0.001), indicating that the link between sleep and CMB risk was stronger for females than for males.

Table S1.

Regression models demonstrating whether associations between total sleep time or sleep efficiency and CMB risk vary by demographic factors (n = 426)

Variables	Model 3a		Model 4a		Model 3b		Model 4b		Model 3c		Model 4c
	B	95% CI	B	95% CI	B	95% CI	B	95% CI	B	95% CI	B	95% CI
Race (African American = 1)	0.07	[−0.04, 0.18]	0.06	[−0.06, 0.17]	0.07	[−0.04, 0.17]	0.04	[−0.07, 0.15]	0.06	[−0.05, 0.16]	0.02	[−0.09, 0.13]
Age, in years	0.01***	[0.01, 0.02]	0.01***	[0.01, 0.02]	0.01***	[0.01, 0.02]	0.01***	[0.01, 0.02]	0.01***	[0.01, 0.02]	0.01***	[0.01, 0.02]
Sex (male = 1)	−0.03	[−0.13, 0.06]	−0.04	[−0.13, 0.06]	−0.03	[−0.14, 0.06]	−0.04	[−0.14, 0.06]	−0.03	[−0.12, 0.07]	−0.04	[−0.13, 0.05]
Partnered	0.10*	[0.01, 0.21]	0.10	[0.00, 0.20]	0.11*	[0.01, 0.21]	0.10	[0.00, 0.20]	0.12*	[0.02, 0.22]	0.12*	[0.02, 0.21]
Education (≤high school degree is reference)
Some college attendance	0.00	[−0.11, 0.13]	0.00	[−0.12, 0.12]	0.01	[−0.11, 0.13]	0.01	[−0.11, 0.13]	0.01	[−0.11, 0.13]	0.01	[−0.11, 0.12]
≥4 y college degree	0.03	[−0.08, 0.14]	0.03	[−0.08, 0.14]	0.03	[−0.09, 0.14]	0.04	[−0.08, 0.15]	0.02	[−0.09, 0.13]	0.03	[−0.08, 0.13]
Income	−0.01**	[−0.02, 0.00]	−0.01*	[−0.02, 0.00]	−0.01**	[−0.02, 0.00]	−0.01*	[−0.02, 0.00]	−0.01**	[−0.02, 0.00]	−0.01*	[−0.02, 0.00]
Health behaviors
Physical activity	−0.08*	[−0.15, −0.01]	−0.07	[−0.14, 0.00]	−0.07*	[−0.15, −0.01]	−0.08*	[−0.15, 0.00]	−0.08*	[−0.15, −0.01]	−0.09*	[−0.16, 0.01]
Fast food consumption (none is reference)
Less than once a week	0.08	[−0.06, 0.20]	0.07	[−0.06, 0.20]	0.08	[−0.06, 0.20]	0.08	[−0.05, 0.21]	0.08	[−0.05, 0.21]	0.07	[−0.05, 0.20]
Once a week or more	0.19**	[0.07, 0.31]	0.18**	[0.05, 0.30]	0.19**	[0.07, 0.31]	0.17**	[0.05, 0.30]	0.19**	[0.06, 0.31]	0.17**	[0.06, 0.29]
Current cigarette smoker	0.07	[−0.07, 0.20]	0.07	[−0.06, 0.21]	0.07	[−0.07, 0.20]	0.06	[−0.08, 0.19]	0.07	[−0.06, 0.21]	0.08	[−0.05, 0.21]
Depressive symptoms	0.00	[0.00, 0.01]	0.00	[0.00, 0.01]	0.00	[0.00, 0.01]	0.00	[0.00, 0.01]	0.00	[0.00, 0.01]	0.00	[−0.01, 0.01]
Sleep indicators
Total sleep time, in hours	−0.07**	[−0.12, −0.02]			−0.07**	[−0.10, −0.02]			−0.11***	[−0.16, −0.06]
Sleep efficiency, %			−0.01**	[−0.02, −0.00]			−0.01**	[−0.01, 0.00]			−0.02***	[−0.03, −0.01]
Race × Total sleep time	0.02	[−0.07, 0.10]
Race × Sleep efficiency			0.00	[−0.01, 0.01]
Age × total Sleep time					0.00	[0.00, 0.00]
Age × Sleep efficiency							0.00	[0.00, 0.00]
Sex × total Sleep time									0.10**	[0.03, 0.17]
Sex × Sleep efficiency											0.02***	[0.01, 0.03]

Sleep efficiency and sleep time were centered to aid in the interpretation of the main effects of relevant demographic variables. Significance: *P < 0.05. **P < 0.01. ***P < 0.001.

Sex-Stratified Models.

Models 1–4 were also fit in sex-stratified subsamples. Because the sample of AA males was small, analyses in the male subsample were exploratory and are described in Supporting Information. AA females obtained 0.75 h less sleep and had 9.1% lower sleep efficiency than EA females (Ps < 0.001). Differences in CMB risk were also substantial, so that, after adjusting for demographics, AA females had 0.52 SD units higher risk than EAs (B = 0.25, 95% CI: 0.12, 0.38).

As depicted in Models 2–4 of Table S2, health behaviors, sleep time, and sleep efficiency were tested as explanations for differential CMB risk between AA and EA females. Adjusting for sleep time explained 33% of the racial gap in CMB risk, and the test of mediation was significant (indirect effect: B = 0.07, 95% CI: 0.02, 0.12). Sleep efficiency was also associated with CMB risk and explained 65% of the race difference in females (indirect effect: B = 0.14, 95% CI: 0.07, 0.22). When sleep time and sleep efficiency were considered simultaneously, the indirect effect via sleep efficiency was significant (B = 0.13, 95% CI: 0.03, 0.22), but the indirect effect via sleep time was not, and the sleep indicators together explained 73% of the race estimate (B = 0.07, 95% CI: −0.07, 0.20).

Table S2.

Regression models demonstrating differences in CMB risk between AA (n = 97) and EA (n = 161) females after adjusting for sociodemographics (Model 1), health behaviors (Model 2), total sleep time (Model 3), and sleep efficiency (Model 4)

Variables	Model 1		Model 2		Model 3		Model 4
	B	95% CI	B	95% CI	B	95% CI	B	95% CI
Race (African American = 1)	0.25***	[0.12, 0.38]	0.22***	[0.09, 0.35]	0.15*	[0.01, 0.28]	0.08	[−0.06, 0.22]
Age	0.01**	[0.00, 0.01]	0.01***	[0.01, 0.02]	0.01***	[0.01, 0.02]	0.01*	[0.00, 0.02]
Partnered	0.10	[−0.03, 0.23]	0.10	[−0.03, 0.23]	0.14*	[0.01, 0.27]	0.13*	[0.01, 0.25]
Education (referent is ≤HS degree)
Some college attendance	−0.01	[−0.07, 0.14]	−0.03	[−0.18, 0.13]	−0.02	[−0.18, 0.13]	−0.02	[−0.16, 0.13]
≥4 y college degree	−0.11	[−0.11, 0.21]	0.07	[−0.09, 0.22]	0.06	[−0.10, 0.21]	0.06	[−0.09, 0.21]
Income	−0.01*	[−0.03, 0.00]	−0.01*	[−0.03, 0.00]	−0.02*	[−0.03, 0.00]	−0.01*	[−0.02, 0.00]
Physical activity			−0.09	[−0.21, 0.02]	−0.11	[−0.22, 0.01]	−0.11	[−0.23, 0.02]
Fast food consumption (none is reference)
Less than once a week			0.03	[−0.15, 0.21]	0.03	[−0.14, 0.21]	0.04	[−0.13, 0.22]
Once a week or more			0.14	[−0.04, 0.31]	0.14	[−0.03, 0.32]	0.14	[−0.03, 0.31]
Current cigarette smoker			0.08	[−0.11, 0.27]	0.05	[−0.13, 0.23]	0.04	[−0.14, 0.22]
Depressive symptoms			0.01	[0.00, 0.01]	0.01	[0.00, 0.01]	0.00	[−0.01, 0.01]
Total sleep time, in hours					−0.10**	[−0.15, −0.04]
Sleep efficiency, %							−0.02***	[−0.03, −0.01]

Significance: *P < 0.05. **P < 0.01. ***P < 0.001.

Supplemental Analyses.

To decrease the possibility of reverse causality (i.e., CMB disease disrupting sleep), we tested mediation hypotheses after excluding participants with heart disease or diabetes (manifest by HbA_1c ≥6.5%). In this restricted sample (n = 303), race was still indirectly related to CMB risk via sleep time (B = 0.05, 95% CI: 0.02, 0.11). The indirect effect estimate of race on CMB risk via sleep efficiency was similar in magnitude but was not significant (B = 0.05, 95% CI: −0.02, 0.12). In the female subsample without heart disease or diabetes (n = 190), significant indirect effects of race on CMB risk via sleep time (B = 0.09, 95% CI: 0.03, 0.18) and sleep efficiency (B = 0.12, 95% CI: 0.03, 0.24) were also detected. In addition, to separate poor sleep patterns resulting from sleep disorders, models were also fit excluding participants with chronic sleep problems; the results are described in Supporting Information.

Models 2–4 were also fit treating the individual CMB markers as continuously coded outcomes using multivariate regression. Systolic and diastolic blood pressure were modeled separately, constituting two of the eight outcomes. Antihypertensives were included as a predictor of systolic and diastolic blood pressure, and lipid-lowering medications were included as a predictor of triglycerides and HDL-C levels. Results are shown in Table S3. In summary, relative to EAs, AAs had larger waist circumference, greater HbA_1c, HOMA-IR, and CRP (and higher HDL-C and lower triglycerides, which are not discussed further). The race estimate for each of respective biomarkers was attenuated by 24, 14, 23, and 13%, respectively, when adjusting for sleep time, and by 26, 24, 23, and 19%, respectively, when adjusting for sleep efficiency. Sex-stratified models also showed that racial health differences and the explanatory role of sleep were driven by the female subsample (Table S4). In particular, relative to EA women, AA women had larger waist circumference and greater systolic and diastolic blood pressure, HbA_1c, HOMA-IR, and CRP. These differences were attenuated by 20, 13, 13, 8, 19, and 16%, respectively, when adjusting for sleep time and by 27, 22, 19, 19, 30, and 22%, respectively, when adjusting for sleep efficiency.

Table S3.

Series of multivariate regression model results demonstrating racial differences in individual CMB indicators in full sample (n = 426) after adjusting for sociodemographics and health behaviors (Model 2), total sleep time (Model 3), and sleep efficiency (Model 4)

Variables	Outcome: Systolic blood pressure			Outcome: Diastolic blood pressure
	Model 2: B (SE)	Model 3: B (SE)	Model 4: B (SE)	Model 2: B (SE)	Model 3: B (SE)	Model 4: B (SE)
Race (African American = 1)	3.85 (2.20)	3.49 (2.23)	2.72 (2.30)	2.50 (1.33)	2.41 (1.33)	2.20 (1.43)
Total sleep time, in hours		−0.53 (0.81)			−0.13 (0.49)
Sleep efficiency, %			−0.13 (0.10)			−0.03 (0.06)
	Outcome: HbA1c			Outcome: lnHOMA-IR
Race (African American = 1)	0.76*** (0.19)	0.65*** (0.19)	0.58** (0.19)	0.22** (0.07)	0.17* (0.08)	0.17* (0.08)
Total sleep time, in hours		−0.15** (0.06)			−0.07* (0.03)
Sleep efficiency, %			−0.02* (0.01)			−0.01* (0.00)
	Outcome: Waist circumference			Outcome: lnC-reactive protein
Race (African American = 1)	6.62** (2.12)	5.03* (2.13)	4.88* (2.22)	0.32*** (0.09)	0.28** (0.09)	0.26** (0.09)
Total sleep time, in hours		−2.27** (0.82)			−0.06* (0.03)
Sleep efficiency, %			−0.20* (0.09)			−0.01 (0.00)
	Outcome: lnTriglycerides			Outcome: High density lipoproteins
Race (African American = 1)	−0.21*** (0.06)	−0.23*** (0.06)	−0.24*** (0.07)	6.05** (2.11)	6.32** 2.16	5.67* (2.28)
Total sleep time, in hours		−0.04 (0.02)			0.38 (0.67)
Sleep efficiency, %			−0.00 (0.00)			−0.04 (0.09)

Note: All models are adjusted for sex, age, partnership status, education, income, physical activity, fast food consumption, current cigarette smoker, and depressive symptoms. Systolic and diastolic blood pressure were adjusted for antihypertensives, and triglycerides and HDL-C for lipid lowering medications. Significance: *P < 0.05. **P < 0.01. ***P < 0.001.

Table S4.

Series of multivariate regression model results among female subsample (n = 258) demonstrating racial differences in individual CMB biomarkers after adjusting for sociodemographics and health behaviors (Model 2), total sleep time (Model 3), and sleep efficiency (Model 4)

Variables	Outcome: Systolic blood pressure			Outcome: Diastolic blood pressure
	Model 2: B (SE)	Model 3: B (SE)	Model 4: B (SE)	Model 2: B (SE)	Model 3: B (SE)	Model 4: B (SE)
Race (African American = 1)	7.78** (2.79)	6.76* (2.84)	6.06* (2.91)	4.94** (1.67)	4.30* (1.70)	3.99* (1.84)
Total sleep time, in hours		−1.53 (1.13)			−0.97 (0.63)
Sleep efficiency, %			−0.24 (0.19)			−0.13 (0.10)
	Outcome: HbA1c			Outcome: lnHOMA-IR
Race (African American = 1)	0.87*** (0.22)	0.81*** (0.24)	0.71** (0.24)	0.33*** (0.09)	0.27** (0.09)	0.23* (0.10)
Total sleep time, in hours		−0.09 (0.06)			−0.09* (0.04)
Sleep efficiency, %			−0.02* (0.01)			−0.01* (0.01)
	Outcome: Waist circumference			Outcome: lnC-reactive protein
Race (African American = 1)	10.61*** (2.55)	8.50** (2.65)	7.74** (2.76)	0.50*** (0.11)	0.42*** (0.12)	0.39** (0.12)
Total sleep time, in hours		−3.03** (1.10)			−0.11* (0.04)
Sleep efficiency, %			−0.36* (0.15)			−0.01* (0.01)
	Outcome: lnTriglycerides			Outcome: High density lipoproteins
Race (African American = 1)	−0.19** (0.07)	−0.24** (0.08)	−0.30*** (0.07)	5.93* (2.90)	6.10* 3.00	7.09* (3.22)
Total sleep time, in hours		−0.07* (0.03)			0.21 (1.02)
Sleep efficiency, %			−0.01*** (0.00)			0.15 (0.16)

Note: All models are adjusted for age, partnership status, education, income, physical activity, fast food consumption, current cigarette smoker, and depressive symptoms. Systolic and diastolic blood pressure were adjusted for antihypertensives, and triglycerides and HDL-C were adjusted for lipid- lowering medications. Significance: *P < 0.05. **P < 0.01. ***P < 0.001.

Discussion

Reducing racial/ethnic health disparities is highlighted as one of the four overarching goals in Healthy People 2020 (3). Given that ∼40% of the gap in premature mortality between AAs and EAs (and more than 50% of the gap between AA and EA females) stems from cardiovascular and metabolic conditions, improving the CMB health of AAs is an important step toward greater health equity (28). Sleep represents one potential intervention point (9, 10), but the relative influence of sleep on racial differences in CMB conditions is largely unknown. In a sample of middle-aged and older adults, we find support for the notion that differences in habitual sleep contribute to the unequal burden of CMB disease risk. In particular, after adjusting for sociodemographics and health behaviors, sleep time and sleep efficiency accounted for 41% and 58% of the AA–EA gap in CMB risk scores, respectively. Path analysis allowed for the simultaneous consideration of the sleep mediator and CMB outcome—each of which was adjusted for sociodemographics—and a formal test of mediation.

These findings are consistent with previous investigations of sleep as a mediator of race differences in blood pressure (16, 21, 29). Specifically, using data from the Coronary Artery Risk Development in Young Adults Study, actigraph-assessed sleep time mediated AA–EA differences in diastolic blood pressure change over a 5-y period (16). In a population-based cohort in Chicago, actigraph-assessed sleep maintenance, but not sleep duration, was associated with hypertension and explained ∼11% of the AA–EA difference in prevalence (29). Another study of middle-aged adults in the Southeast reported that lower sleep quality explained, in part, differences between AAs and EAs in nocturnal blood pressure dipping (21). In contrast, a study using data from the male subsample of the Boston Area Community Health Survey did not find evidence of short sleep duration (≤5 h) to be a mediator of racial disparities in 5-y self-reported hypertension incidence (20), but analyses were tested only for males and were limited by a dichotomous self-report sleep-duration variable. Thus, extant evidence suggests that sleep problems contribute to racial differences in hypertension; this study adds evidence regarding their influence on racial differences in the risk for CMB disease more broadly.

When examining findings by biological sex, sleep indicators were associated with CMB risk among females but not males (Table S5 for results among male sample), and sleep also explained a substantial portion of the race differences in CMB risk among females. The sex moderation of the sleep–health association is consistent with prior research documenting insufficient sleep as a risk factor for larger waist circumference, hypertension, and elevated inflammatory markers among females but not among males (30–32). However, given the small sample of AA males, the current dataset was able to offer only preliminary support for sex differences in the extent to which sleep explained racial differences in CMB risk. Additional research is needed to determine whether differences in sleep lead to elevated CMB morbidity among AA males relative to EA males, and, if so, to elucidate the mechanisms underlying sex differences in associations between sleep and physical health.

Table S5.

Regression models demonstrating differences in CMB risk between AA (n = 36) and EA (n = 132) males after adjusting for sociodemographics (Model 1), health behaviors (Model 2), total sleep time (Model 3), and sleep efficiency (Model 4)

Variables	Model 1		Model 2		Model 3		Model 4
	B	95% CI	B	95% CI	B	95% CI	B	95% CI
Race (African American = 1)	−0.05	[−0.21, 0.11]	−0.10	[0.01, 0.22]	−0.12	[−0.29, 0.05]	−0.11	[−0.29, 0.07]
Age	0.01***	[0.01, 0.02]	0.01***	[0.01, 0.02]	0.02***	[0.01, 0.02]	0.01***	[0.01, 0.02]
Partnered	0.09	[−0.07, 0.25]	0.09	[−0.01, 0.18]	0.09	[−0.06, 0.24]	0.09	[−0.06, 0.24]
Education (referent is ≤HS degree)
Some college attendance	0.04	[−0.14, 0.22]	0.01	[−0.11, 0.13]	0.00	[−0.18, 0.18]	0.01	[−0.17, 0.19]
≥4 y college degree	−0.01	[−0.17, 0.15]	−0.02	[−0.08, 0.15]	−0.03	[−0.18, 0.13]	−0.02	[−0.18, 0.13]
Income	−0.01	[−0.02, 0.01]	−0.01	[−0.02, 0.00]	−0.01	[−0.02, 0.01]	−0.01	[−0.02, 0.01]
Physical activity			−0.07	[−0.14, 0.00]	−0.07	[−0.16, 0.01]	−0.07	[−0.16, 0.01]
Fast food consumption (none is reference)
Less than once a week			0.05	[−0.05, 0.21]	0.05	[−0.15, 0.25]	0.05	[−0.15, 0.25]
Once a week or more			0.18	[0.06, 0.31]	0.18	[0.00, 0.37]	0.18	[−0.01, 0.36]
Current cigarette smoker			0.14	[−0.04, 0.33]	0.14	[−0.05, 0.32]	0.14	[−0.05, 0.33]
Depressive symptoms			0.00	[−0.01, 0.01]	0.00	[−0.01, 0.01]	0.00	[−0.01, 0.01]
Total sleep time, h					−0.03	[−0.07, 0.02]
Sleep efficiency, %							0.00	[−0.01, 0.01]

Significance: ***P < 0.001.

Examining individual biomarkers, we also found broad support for the role of sleep as an explanation of CMB risk. Specifically, relative to EAs, AA females had greater dysregulation in each of the six expected CMB markers, and these differences were attenuated by 8–29% when adjusting for sleep time and by 19–30% when adjusting for sleep efficiency. These findings are consistent with one study that tested associations between sleep and allostatic load (indexed by seven physiologic systems). When individual systems were examined in this study, inadequate sleep duration was particularly related to cardiovascular, inflammatory, and glucose metabolic systems, all of which constitute our CMB risk index, suggesting that sleep may be especially important for CMB health (33).

In addition to differences in sleep being an issue of public health, the presence of these sleep differences appears to be, in part, a result of social factors. Specifically, sleep has been robustly associated with exposure to social stress (34). Moreover, household and area-level socioeconomic conditions and experiences of discrimination help explain a portion of racial sleep differences (35, 36). Insufficient sleep therefore represents a plausible biobehavioral pathway through which disproportionate exposure to social and economic adversity leads to diverging health trajectories between AAs and EAs. Future studies will need to integrate research on stress exposure and sleep as they relate to a broad variety of racial and socioeconomic inequities in health and well-being. Furthermore, initiatives to reduce racial health disparities will need to consider the costs and benefits of varying points of intervention (e.g., upstream social determinants vs. sleep). Evidence-based sleep interventions could target AA communities and, if successful, could serve to reduce disparities in physical health, socioeconomic factors, and well-being (37, 38).

The use of sleep actigraphy is one strength of the present study. One recent review found that AA–EA differences in sleep time were larger when assessed objectively relative to self-reports (13), so that objective assessments may be key in research on racial sleep differences. Racial differences in the bias of sleep self-reports has also been documented, leading to an underestimation of differences (39). Further, findings considering racial differences in the association between sleep and CMB outcomes vary by the sleep assessment methodology. Namely, some research has shown that self-reported short duration is associated with blood pressure and diabetes to a larger degree among AAs than among EAs (19, 40), whereas one study and the present research found no racial differences in the link between actigraph-assessed sleep duration and cardiovascular markers (16). Polysomnography is also vital because it measures sleep architecture and has an advantage over actigraphy in detecting wakefulness (41). However, actigraphy is a more feasible means of measuring multiple nights of sleep and has significant cost advantages that are important in epidemiologic studies.

An additional strength of this study relates to our conceptualization of CMB risk and the utilization of several key biomarkers that predict subsequent cardiovascular and metabolic disease. One advantage of multisystem indexes of dysregulation is that the biological pathways from risk factors to disease may vary across individuals (42). Although other multisystem indexes have included a broader range of biomarkers relevant to aging, our focus was on the biomarkers of CMB risk because of their relevance to racial disparities in premature mortality and the established associations between sleep and many of the individual biomarkers. Further description of the strengths of our CMB risk index is found in Supporting Information.

Several study limitations are also important to note. Analyses used cross-sectional data, and therefore no conclusions could be drawn about temporal ordering. In particular, low sleep efficiency and insufficient sleep may result from health conditions, particularly among older adults (43). However, when participants with heart disease or diabetes were excluded, racial differences in CMB risk continued to be explained by sleep time and sleep efficiency. Moving forward, longitudinal data will be critical to demonstrate the long-term effects of habitual sleep problems.

The possibility of residual confounding, resulting from imprecise controls and lack of consideration of all third-variable explanations, also needs to be acknowledged. Specifically, although we adjusted for coarse indicators of diet and physical activity, detailed measures are needed to remove their influence on sleep fully. Similarly, in addition to controlling for education and income, subsequent research must elucidate the interdependence of a broad variety of socioeconomic factors and sleep as they relate to racial differences in health. Two additional analyses also were conducted to reduce the possibility of age and sleep disorders as confounding variables; these analyses are described in Supporting Information and are discussed here only briefly. First, given that normative changes in sleep and CMB functioning occur with aging, we tested and found no evidence of age as a moderator of the link between sleep and CMB risk. Second, we excluded participants with chronic sleep problems and found an identical pattern of results.

Generalizability is limited because the EA subsample was primarily from the Midwest and the AA subsample was from Milwaukee. Milwaukee suffers from hypersegregation (extreme segregation across several indicators) and adverse economic conditions (44), so that racial differences in heath could be larger than would be found in other contexts. Nonetheless, the majority of AAs in the Midwest reside in industrial cities that are segregated and have been adversely affected by changes in the economy resulting in the loss of manufacturing jobs (45). Thus, the sample offers an important context to consider sleep as a behavioral pathway through which an historically disadvantaged racial group may develop elevated CMB disease risk.

Conclusions

The results from the present study highlight the importance of eliminating racial differences in sleep as part of efforts to reduce racial health disparities. Importantly, sleep is a malleable behavior that can be improved through behavioral and educational interventions to influence physical health (38). Further, despite limited information regarding nonclinical samples, sleep interventions have been shown to be cost-effective (37, 46). Thus, recent public health initiatives that include monitoring sleep behaviors and society-wide messaging about adequate sleep duration and sleep hygiene represent important progress (47, 48).

Methods

Participants.

Data for the analyses were derived from a sleep substudy that was part of the second wave of the Midlife in the United States (MIDUS II) Study (49). MIDUS began in 1995 as a nationally representative survey of more than 7,000 noninstitutionalized adults. Approximately 75% of surviving respondents of the first wave participated in MIDUS II, ∼9–10 y later. As part of this new data collection, a subsample of respondents participated in clinic-based (overnight) biomarker data collection at one of three sites (Los Angeles; Madison, WI; and Washington, DC). Details of the biomarker sample and measures are provided elsewhere (49). Data are publicly available at https://www.icpsr.umich.edu/icpsrweb/ICPSR/studies/29282. To increase the participation of AAs in biomarker data collection, a new oversample from Milwaukee (n = 592) was also recruited. MIDUS staff made travel arrangements and covered travel-related costs to ensure maximum participation in this portion of the study.

We included only participants from the Madison, WI site, which drew primarily on respondents from the Midwest region and Milwaukee, because they were invited to participate in a home-based week-long sleep substudy. A total of 441 adults (83% of the Madison biomarker sample) participated. Because of our focus on AA and EA adults, 15 individuals categorized as other racial/ethnic groups were excluded. Data collection was approved by the University of Wisconsin-Madison Institutional Review Board, and all participants provided written, informed consent.

Measures.

CMB risk.

Cardiovascular and diabetes risk can be quantified from markers of abdominal adiposity, dyslipidemia, glucose metabolism, blood pressure, and systemic inflammation (50–52). These risk markers also cluster together to indicate a general state of CMB dysregulation. For the present study, CMB disease risk was indexed using clinical or recommended risk cut points for seven biomarkers: waist circumference, HOMA-IR, HbA_1c, blood pressure, CRP, triglycerides, and low HDL levels. This measure of CMB risk includes indicators of metabolic syndrome and two additional biomarkers (HOMA-IR and CRP) that add incremental validity in the prediction of cardiovascular disease (50, 53). Ideal (coded 0), borderline high (coded 1), and high (coded 2) levels for each marker were selected primarily based on recommendations from the National Cholesterol Education Program Adult Treatment Panel III (NCEP ATP III) and the American Heart Association (AHA) (23–25). Scores (0–2) for each of the seven biomarkers were averaged to create an overall index of CMB risk. Detailed descriptions of blood collection and processing procedures and information about biomarker assays and other assessments have been reported previously (54) and are reviewed briefly below. A further description of the advantages of our CMB risk index is given in Supporting Information.

Waist circumference.

Waist circumference at the minimal girth was measured by trained nurses according to a standardized protocol (32). Waist circumference risk categories were chosen according to AHA guidelines (25).

Blood pressure.

Blood pressure was measured in triplicate according to standardized procedures, and the second and third measurements were averaged (54). Prehypertension and hypertension were coded if either systolic or diastolic blood pressure levels met Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure (JNC-VII) criteria (23). The use of antihypertensives was also coded as hypertension (i.e., given a value of 2).

Glucose metabolism.

Fasting blood samples were collected in the morning of the second day of the laboratory visit. Glucose was assessed using an enzymatic assay, performed on an automated analyzer (Roche Modular Analytics P), and insulin was measured using a two-site sandwich immunoassay using direct chemiluminescent technology on a Siemens ADVIA Centaur analyzer (54). Insulin resistance was computed using the original homeostasis model assessment of insulin resistance (HOMA1-IR) equation: HOMA1-IR = fasting glucose (mmol/L) × fasting insulin (mU/L)/22.5 (55). The HOMA1-IR is an acceptable method for considering between-group differences and relative change in insulin resistance but not absolute levels of insulin resistance, because the original HOMA model was calibrated on assays that were no longer in use at the time of the study (56). Recommendations for clinical cut points for HOMA-IR are also varied (57), and therefore insulin resistance tertiles were used to indicate risk levels. Prediabetes/diabetes was indexed using HbA_1c, a measure of blood glucose over the past few months, according to recommended cutoffs (58). HbA_1c was measured from whole-blood samples using a Cobas Integra analyzer (Roche Diagnostics) (54).

Triglycerides and HDL-C.

Concentrations of triglycerides and HDL-C were measured using a Cobas Integra analyzer (Roche Diagnostics) (54). The use of lipid-modifying medications was coded as high levels of triglycerides (high-risk category).

CRP.

CRP levels were measured using a high-sensitivity assay performed on a BN II nephelometer (Dade Behring) (54). Borderline and high CRP levels were coded according to previous recommendations (59).

Sleep minutes and sleep efficiency.

Sleep characteristics were measured using wrist actigraphy and sleep logs. Detailed procedures relating to the collection and processing of actigraphy data are available at www.midus.wisc.edu/midus2/project4/ and are summarized herein. At the biomarker visit participants received a Mini-Mitter Actiwatch-64 activity monitor (Respironics, Inc.) that was to be worn continuously for 7 d and nights. Physical activity counts were collected in 30-s epochs during the week-long period, and periods of intended sleep were then demarcated in the actigraphy data using bedtimes and rise times reported on sleep logs. Epochs were scored as wake or sleep using Actiware Software (version 5.0) (Respironics, Inc.) based on a medium threshold of 40 activity counts (32).

Total sleep time was measured as the number of minutes coded as asleep between sleep onset and rise time, and sleep efficiency was measured as the percentage of epochs coded as asleep from bedtime to rise time. All participants had at least three nights of actigraphy data, with nearly all (96%) having at least six nights. In the present analyses, daily sleep parameters were averaged across all nights of available data to compute sleep-related outcomes; total sleep time and sleep efficiency showed high internal consistency (α = 0.85 and α = 0.91, respectively). Sleep time and sleep efficiency were strongly correlated (r = 0.62) and therefore were initially considered separately in mediation analyses. Actigraph-assessed sleep time and sleep efficiency show good correspondence with polysomnography in community samples and with the device used in the present study (22, 60).

Sociodemographic covariates.

Biological sex, age (in years), and relationship status (married or cohabiting = 1, otherwise = 0) were included as demographic covariates. Total household income was measured using several items assessing wages, pension, social security, and government assistance for all household members. The participant’s educational attainment was coded into dummy variables representing high school degree or less (reference category), some college attendance, and a 4-y college degree or greater. All models were adjusted for sociodemographic characteristics.

Mental health and health behaviors.

Depressive symptoms were assessed using the Center for Epidemiological Studies Depression Scale, a widely used and well-validated measure (61). One item measuring restless sleep was removed from the computation of the total score to reduce conflation between sleep and depression, and the revised scale had high reliability (α = 0.89). Habits relating to physical activity, diet, and cigarette smoking were also assessed (36, 62). Physical activity was reported on nightly sleep logs by the answer to the question “How many minutes of moderate or vigorous exercise did you get today?” The average of all available reports across the 7 d was used. Despite this single item being a coarse indicator of physical activity, the short recall and repeated measurements are considerable strengths that improve the validity of the item (63). Fast-food consumption was selected as an index of a putative unhealthy dietary habit (64) and was coded in three dummy variables: no consumption (referent), less than once per week, and once a week or more. The frequency of fast-food consumption is positively associated with high-fat, high-sugar diets and is inversely associated with the intake of whole grains, fruits and vegetables, and fiber (64). Single-item reports of the frequency of fast-food consumption also correspond closely with typical fast-food items measured using detailed month-long diet histories (64). Smoker status was coded as a dummy variable representing current smoking activity.

Previous heart disease.

Participants reported if a doctor had previously suspected or confirmed a “heart trouble” and recalled the diagnosis, broadly categorized as myocardial infarction, angina, valve disease, hole in heart, blocked artery, arrhythmia, heart murmur, or heart failure.

Analysis Plan.

Descriptive statistics were examined by race, and unadjusted differences between AAs and EAs were tested using independent-samples t tests and χ² tests. Linear regression and path models were used to test study hypotheses. An initial regression model estimated racial differences in total sleep time and sleep efficiency, adjusting for sex, age, relationship status, education, and household income. A series of regression models then was estimated to examine racial differences in CMB risk and the role of sleep indicators as mediating variables. In Model 1, differences in CMB disease risk between AAs and EAs were estimated after adjusting for sociodemographics. Model 2 then added health behaviors and mental health—specifically, diet, physical activity, cigarette smoking, and depression—as additional covariates. Two additional models were fit in which total sleep time (Model 3) and sleep efficiency (Model 4) were added, in turn, as predictors of CMB risk. Mediation was tested in path models in which each sleep indicator was simultaneously modeled as a mediator predicted by race (and sociodemographics) and as a predictor of CMB risk. The indirect effect of race on CMB risk via each sleep indicator was estimated using the product of coefficients method with bias-corrected confidence intervals (65), and the proportion of the race difference in CMB risk explained was calculated to indicate the effect size. Modeling indirect effects using path models is a widely used test of mediation that allows simultaneous controls on both the mediator and outcome variable. Further, indirect effects are easily interpreted—the estimate is in the units of the direct effect (in our case, the indirect effect indicates the absolute value of the racial gap in CMB risk explained by the sleep mediator). Study hypotheses were tested using Mplus 7.11 (66). Missing data were handled using full information maximum likelihood, a common estimation technique that uses all available data and allows a consistent sample size across models. Only 2.6% of participants had missing data for physical activity, and all other covariates had less than 1% missing data.

A few sensitivity analyses were also conducted. Given that some studies have shown associations between short sleep duration and CMB risk to vary by demographic factors (19, 40), race, age, and sex were tested as moderators of the link between sleep and CMB risk. The series of models described above was also fit among samples that were stratified by sex to determine if mediation findings were present for both male and female subsamples. In addition, because both short and long sleep duration are correlates of negative health outcomes, each of which is more common among AAs (12), linear and nonlinear associations between sleep minutes and CMB risk were estimated (i.e., sleep minutes squared was included as predictor). Furthermore, to reduce the likelihood of reverse causality, we fit models in which we excluded participants with heart disease or diabetes. Last, mediation models were fit among participants who reported no experience of chronic sleeping problems.

SI Methods

CMB Disease Risk.

The main text describes the measurement of CMB disease risk in detail and presents some discussion of its validity and utility. Herein, we expand on the potential strengths of using a cumulative index of CMB risk. In particular, compared with a dichotomous indicator of disease (e.g., metabolic syndrome), continuous measures of aggregated risk scores more fully capture the spectrum of between-person variability in CMB dysregulation. Multisystem aggregates of physiologic dysregulation have also been shown to have stronger effects on mortality than do individual biomarkers or chronic conditions (67). The influence of measurement error from a single biomarker is also reduced when modeling the co-occurrence of dysregulation across physiologic systems. This co-occurrence is evident in our index by the significant intercorrelations between biomarkers (rs ranged from 0.10 to 0.48). In addition, although there is a possibility that relations between physiologic markers can vary between racial/ethnic groups, prior research modeling allostatic load as a set of interrelated physiologic latent factors has found factor loadings to be of generally comparable magnitude across AA and EA males and females (68). Consistent with this research, our measure of CMB risk had high internal consistency for both AA (α = 0.68) and EA (α = 0.73) adults.

Combined measures of physiologic dysregulation are an important tool in population health sciences. However, because multisystem aggregate indexes are by nature less specific, additional insight can be gained by examining each biomarker independently. Therefore, supplemental analyses considering each biomarker separately as a continuous variable were also conducted. Model results from the full sample and female subsample are described in Tables S3 and S4.

Chronic Sleep Problems.

Participants self-reported if, in the past 12 mo, they had experienced or were treated for “chronic sleeping problems.” Endorsement of sleeping problems was used as an exclusionary criterion in some supplemental analyses, removing 60 participants from the analytic sample, to distinguish between poor sleep habits and more severe, chronic sleep problems.

SI Results

We also fit models in which racial differences in sleep and CMB risk were estimated only among the male sample. Similar in magnitude to the racial difference among females, we found that AA males slept 0.65 h less and had 12.5% lower sleep efficiency than EA males (P = 0.011 and P < 0.001, respectively). However, we found no significant differences in CMB risk between AA and EA males (P = 0.55). (Table S5 for results of Models 1–4.) These results indicate that neither sleep time nor sleep efficiency was significantly associated with CMB risk (P = 0.30 and P = 0.86, respectively).

When participants with chronic sleeping problems were excluded (resulting in a sample size of 366), CMB risk was elevated among AA as compared with EA adults (0.14, 95% CI: 0.03, 0.27). Next, total sleep time and sleep efficiency were considered as explanations of the difference in CMB risk between AAs and EAs. The indirect effect of race on CMB risk via sleep time was significant (0.05, 95% CI: 0.02, 0.10) and explained 36% of the racial gap in CMB risk. Similarly, the indirect effect of sleep efficiency was significant (0.07, 95% CI: 0.02, 0.13), explaining 49% of the racial difference in CMB risk. In parallel models that were fit only among the female subsample (n = 216), we detected a substantial racial difference in CMB risk (0.26, 95% CI: 0.12, 0.42). The indirect effects of total sleep time (0.08, 95% CI: 0.03, 0.16) and sleep efficiency (0.15, 95% CI: 0.07, 0.25) were significant and explained 34 and 63% of the AA–EA difference in CMB risk, respectively.

SI Discussion

Sex differences in the extent to which sleep mediates racial CMB differences are reported in the main text. These findings should be considered preliminary and should be interpreted with caution given the small sample size of AA male participants. In addition to the AA male sample being underpowered to detect statistical associations, generalizability to the Milwaukee population of AA males is also limited. For example, Milwaukee County has one of the highest rates of incarceration of AA males in the country (69), a population that was excluded in the MIDUS Study. In addition, sampling procedures intended to recruit a representative sample are biased by selective mortality among AA males in the US caused by elevated homicide rates and premature mortality from chronic diseases (28). Thus, AA males who were eligible for study participation may have been healthier than the broader AA population in Milwaukee; if so, this bias could account for the lack of a AA–EA difference in CMB risk among males in the current study.

To maximize the representativeness of the Milwaukee sample, MIDUS Study staff and the University of Wisconsin Survey Center (who carried out the recruitment and a 2-h in-home interview) used best practices in sampling and recruitment, as evidenced by a 70.7% participation rate among eligible adults (detailed at www.icpsr.umich.edu/cgi-bin/file?comp=none&study=22840&ds=1&file_id=991619&path=ICPSR). We considered whether selection effects may have arisen subsequently when agreeing to participate in the biomarker visit and sleep substudy by testing whether the Milwaukee sample of AA males in the sleep substudy (n = 35) varied from the Milwaukee sample of AA males who completed only the in-home interview (n = 177). Using self-reports of health from the in-home interview, we found no significant differences in self-reports of diagnosed heart disease (P = 0.73), hypertension (P = 0.56), diabetes (P = 0.20), waist circumference (P = 0.77), or self-rated general health (P = 0.29). These results suggest that AA males from the Milwaukee sample did not select into the sleep substudy based on their health status.

Age is another dimension that needs to be considered with respect to the CMB effects of poor sleep habits. Because of normative changes in both sleep and CMB functioning associated with aging (43), we tested the moderating role of aging on the association between sleep and CMB risk. We did not find evidence that the association between either total sleep time or efficiency and CMB risk varied as a function of age. Given that racial differences in CMB conditions emerge before or during middle adulthood and that sleep has been shown to be a moderator of AA–EA disparities in blood pressure change during this developmental period (16), future research and health initiatives should give adequate attention to the first half of the life course.

One limitation of the present study is that clinically significant sleep disorders were not assessed. Specifically, the AA–EA gap in obstructive sleep apnea prevalence is substantial, and the effects of obstructive sleep apnea on cardiovascular diseases are established (11). Thus, obstructive sleep apnea may be a key third variable in the association between sleep problems and CMB risk. Although sleep apnea was not assessed in MIDUS, chronic sleep problems over the past year were self-reported. When participants reporting chronic sleep problems were excluded, the pattern of findings was not changed. However, the assessment of sleep problems was limited by a lack of specificity about specific sleep disorders and the likely possibility that the understanding of the definition of “chronic sleep problems” varied. It is important for future studies to elucidate effects of clinically significant sleep disorders versus habitual sleep problems in community samples.