Neighborhood Social Cohesion and Sleep Health by Age, Sex/Gender, and Race/Ethnicity in the United States

Although low neighborhood social cohesion (nSC) has been linked with poor sleep, studies of racially/ethnically diverse participants using multiple sleep dimensions remain sparse. Using National Health Interview Survey data, we examined overall, age, sex/gender, and racial/ethnic-specific associations between nSC and sleep health among 167,153 adults. Self-reported nSC was categorized into low, medium, and high. Very short sleep duration was defined as <6 h; short as <7 h, recommended as 7–9 h, and long as ≥9 h. Sleep disturbances were assessed based on trouble falling and staying asleep, waking up feeling unrested, and using sleep medication (all ≥3 days/times in the previous week). Adjusting for sociodemographics and other confounders, we used Poisson regression with robust variance to estimate prevalence ratios (PRs) and 95% confidence intervals (CIs) for sleep dimensions by low and medium vs. high nSC. The mean age of the sample was 47 ± 0.1 years, 52% of those included were women, and 69% were Non-Hispanic (NH)-White. Low vs. high nSC was associated with a higher prevalence of very short sleep (PR = 1.29; (95% CI = 1.23–1.36)). After adjustment, low vs. high nSC was associated with very short sleep duration among NH-White (PR = 1.34 (95% CI = 1.26–1.43)) and NH-Black (PR = 1.14 (95% CI = 1.02–1.28)) adults. Low nSC was associated with shorter sleep duration and sleep disturbances.


Introduction
Short sleep duration and sleep disturbances, such as trouble falling asleep, are highly prevalent among the United States (U.S.) population [1]. For instance, it is estimated that one-third of adults habitually obtain less than the recommended amount of at least seven hours of sleep [2]. Groups even more likely to experience short sleep and sleep disturbances include those ≥50 years old compared to 18-30 years old [3], women compared to men [4], and non-Hispanic (NH)-Black [2,5,6] Hispanic/Latinx [5,7], and Asian [8] compared to NH-White adults. For example, one study reported that 43.4% of NH-Black, 31.5% of Hispanic/Latinx, and 27.1% of Asians obtained less than seven hours of sleep, compared to 19.4% of NH-White adults [5]. It is important to identify factors that may influence sleep health because a short sleep duration and sleep disturbances have been shown to Participant data for this study came from the National Health Interview Survey (NHIS), from which survey years from 2013 to 2018 were retrieved by the Integrated Health Interview Series [29]. The NHIS is a series of annual, cross-sectional, household surveys conducted via computer-assisted in-person interviews among the non-institutionalized U.S. adult population. Trained interviewers obtained information regarding medical conditions, health care access, and health behaviors for each member of the sampled household. The NHIS uses a three-stage stratified cluster probability sampling design to obtain a nationally representative sample. A randomly selected adult and child (if present, although not included in the current analysis) from each household provided more specific health-related information. A detailed description of the NHIS procedures has been previously published [30]. The response rate for sample adults was 56.1% (range: 61.2% (2013) to 53.1% (2018)). Sampling weights were used to account for the survey's complex sampling design, non-response, and the oversampling of certain groups (e.g., racial/ethnic minorities; older adults), which resulted in unequal probabilities of selection. Each study participant provided informed consent to the NHIS, and the National Institute of Environmental Health Sciences' Institutional Review Board waived approval for publicly available, secondary data with no identifiable information.

Study Population
Participants (≥18 years of age) from all 50 states and the District of Columbia were included in the sample. Of the 190,113 participants, those with missing or implausible data for key variables, such as race/ethnicity (n = 4348), sleep duration (n = 5986), sleep disturbances (n = 1618), and nSC (n = 9675), were excluded, and those of Native American race (n = 1333) were excluded due to the small sample size. The final analytical sample size was 167,153 participants (Supplemental Figure S1).

Exposure Assessment: Neighborhood Social Cohesion
nSC was measured using a modified version of a four-item scale developed by the Project on Human Development in Chicago Neighborhoods Community Survey [31]. Participants responded on a Likert scale (1 = definitely agree; 2 = somewhat agree; 3 = somewhat disagree; and 4 = definitely disagree) to the following four statements: (1) "People in this neighborhood help each other out"; (2) "There are people I can count on in this neighborhood"; (3) "People in this neighborhood can be trusted"; and (4) "This is a close-knit neighborhood". Responses were reverse-coded, and the nSC variable was calculated as the sum of the four response options. Scores ranged from 4-16, with higher scores indicating greater perceived levels of nSC. nSC scores were further categorized into three groups based on previous literature [32]: low (<12), medium (12)(13)(14), and high (≥15).

Outcome Assessment: Self-Reported Sleep Duration and Sleep Disturbances
Sleep duration was measured by asking participants, "On average, how many hours of sleep do you get in a 24 h period?". Interviewers reported sleep hours in whole numbers as well as rounded values of ≥30 min up to the nearest hour and rounded values <30 min down to the nearest hour. Responses were categorized as very short (<6 h), short (<7 h), recommended (7-9 h) or long (>9 h), based upon the National Sleep Foundation categories [33]. Very short and short sleep were not mutually exclusive categories.
Sleep disturbances were measured by asking participants the following four questions: (1) "In the past week, how many times did you have trouble falling asleep?"; (2) "In the past week, how many times did you have trouble staying asleep?"; (3) "In the past week, on how many days did you wake up feeling well rested?"; and (4) "In the past week, how many times did you take medication to help you fall asleep or stay asleep?". If participants reported experiencing a sleep disturbance for ≥3 days/times per week (i.e., trouble falling asleep, trouble staying asleep, waking up feeling unrested, or taking medication to help fall asleep), they were considered as having a sleep disturbance. Assessed in combination and separately, insomnia symptoms included reports of either trouble falling asleep ≥3 vs. <3 times/week and/or difficulty maintaining sleep ≥3 vs. <3 times/week.

Potential Modifiers: Age, Sex/Gender, and Race/Ethnicity
Participants self-identified their age, sex/gender, and race/ethnicity. Age was categorized as 18-30, 31-49, and ≥50 years. Sex/gender, assessed in a binary vs. non-binary manner, was dichotomized as women versus men. Race/ethnicity was categorized as NH-White alone, NH-Black alone, Hispanic/Latinx (of any race), and Asian.

Statistical Analyses
Descriptive statistics were computed; continuous variables were presented as means ± standard errors (S.E.), and categorical variables were presented as weighted percentages after applying direct standardization using the 2010 U.S. Census population. We compared the three levels of nSC across sociodemographic, health behavior, and clinical characteristics for all participants.
To test associations between nSC and sleep dimensions, we used Poisson regression with robust variance to directly estimate prevalence ratios [37] (PRs) and 95% confidence intervals (CIs) of nSC for each sleep dimension overall, by age, sex/gender, and race/ethnicity, and age-sex/genderrace/ethnicity. This model with adjusted variances has been shown to provide accurate point and interval estimates using either count or binary data during one cross-sectional point in time. Furthermore, this model directly estimates PRs, unlike a logistic regression model, which provides estimated odds ratios that overestimate associations with outcomes of high prevalence, such as poor sleep health. PRs are also easier to communicate and are more interpretable than odds ratios. The overall model was statistically adjusted for the following confounders: age, sex/gender, race/ethnicity, educational attainment, annual household income, employment status, occupational class, region of residence, marital status, alcohol consumption, health status, serious mental illness, and "ideal" cardiovascular health. To test for differences by age, sex/gender, and race/ethnicity, separately and together, respective interaction terms (e.g., nSC * age) were added to the overall model. Analyses were conducted in SAS version 9.4 for Windows (Cary, North Carolina), and a twosided p-value of 0.05 was used to determine statistical significance.
A higher percentage of those aged 18-30 years old (18.7%) lived in a neighborhood with low compared to medium (16.2%) and high (13.7%) social cohesion, while a higher percentage of those aged 31-49 years old (26.0%) lived in a neighborhood with high compared to medium (23.5%) and low (21.0%) social cohesion. A higher percentage of NH-Black (14.1%) and Hispanic/Latinx (19.0%) participants lived in a neighborhood with low social cohesion compared to medium (11.7% and 13.9%, respectively) and high (7.9% and 10.2%, respectively) social cohesion, while a higher percentage of NH-White respondents lived in neighborhood with high (77.0%) compared to medium (68.1%) and low (61.7%) social cohesion. Overall, the prevalence of very short sleep and short sleep was higher among those who reported living in a neighborhood with low (11.8% and 36.2%, respectively) compared to medium (7.9% and 30.5%) and high (7.5% and 27.9%) social cohesion. The prevalence of sleep disturbances was also higher among those living in a neighborhood with low compared to medium and high social cohesion. For example, the prevalence of trouble staying asleep was also higher among those who reported living in a neighborhood with low (32.2%) compared to medium (26.3%) and high (24.4%) social cohesion ( Table 1). Additional sociodemographic characteristics by age, sex/gender, and race/ethnicity are described in Supplemental Tables S1-S4.

Discussion
In this large sample of U.S. adults, we found that perceived neighborhood social cohesion was associated with sleep health. Consistent with our hypothesis, we found that participants who reported living in a neighborhood with low vs. high social cohesion generally experienced shorter sleep duration and more sleep disturbances. Furthermore, also consistent with our hypothesis, we found important modifications of the nSC-sleep relationship by age, sex/gender, and race/ethnicity separately as well as together, although the strength of the interactions was not always in the expected direction. For instance, the interactions between nSC, sleep disturbances (e.g., trouble staying asleep) and age were stronger among younger (18-30 years of age) than older adults (≥50 years of age), except waking up feeling unrested. We observed stronger associations of sleep duration (e.g., very short sleep) among women living in areas with low vs. high nSC compared to men, which corresponded with our hypothesis, although no other notable differences were observed between sex/gender except for the fact that waking up feeling unrested was stronger among men than women. We also observed stronger associations between low vs. high nSC and very short sleep duration in NH-White adults compared to NH-Black adults, while associations for sleep disturbances (e.g., insomnia) were stronger among Asian adults, except for the fact that the use of sleep medications was stronger among NH-Black adults.
Our overall findings that those living in a neighborhood with lower social cohesion experienced shorter sleep duration and more sleep disturbances are supported by previous studies [10,20,32], including studies with objective sleep measures [17]. As an example, our finding that participants ≥50 years old living in a neighborhood with lower social cohesion experienced shorter sleep duration is supported by previous studies [10,17,20]. For instance, a study using data from the Health Retirement Survey found that lower nSC was associated with higher odds of trouble falling asleep among those ≥50 years old [20]. Older adults may be more influenced by the social neighborhood environment compared to middle-aged and young adults considering a potentially reduced network size due, for instance, to retirement, death of loved ones, and compromised health [19]. Despite these consistent findings with prior work, an interesting, novel finding we observed was stronger associations between low vs. high nSC and sleep disturbances, such as that difficulties staying asleep were stronger in younger (18-30 years old) compared to older adults (≥50 years old). While there are no other studies to compare these results with, these findings may indicate the need to intervene in sleep earlier in life, which will likely benefit overall health.
We also found that correlations between low vs. high nSC and shorter sleep duration were stronger in women than men. This is consistent with the idea that women may be more influenced by their neighborhood environment compared to men [38]. Based on a socioecological theory, prior work suggests that women's greater vulnerability to the social neighborhood environment is due to differences in how women are impacted by support networks, how they perceive their environment, and the types of stressors women face on a daily basis, particularly in terms of the social roles women occupy [22,39,40]. Women may leverage social cohesion when engaging in physical activity and other healthy behaviors that can positively influence sleep. Men, on the other hand, may engage in social activities that do not require social cohesion. In fact, prior work examining other aspects of the social neighborhood environment, such as safety, and sleep dimensions also found stronger associations among women compared to men, which is consistent with our findings [38,41,42]. Nonetheless, prior work specifically examining nSC and sleep dimensions did not find modification by sex/gender [10,43]. We also did not observe variations in associations with sleep disturbances by sex/gender, except for the fact that waking up feeling unrested was stronger among men than women. Given these mixed findings, further studies are needed to determine if these effect modification results are replicable and, if so, to assess potential drivers.
Another finding of our work was the potential effect modification of low vs. high nSC and sleep dimensions by race/ethnicity. Although these findings need to be interpreted with caution due to the substantial overlap in the CIs observed, our results suggest that the impact of lower nSC on multiple sleep dimensions may impact racial/ethnic groups differently. For example, the impact of lower nSC on short sleep duration could be larger among NH-White adults, the impact of lower nSC on trouble staying asleep could be larger among Asian adults, and the impact of lower nSC on use of sleep medication could be larger among NH-Black adults. While the potential determinants driving these differences is unclear, these results add to a growing body of literature demonstrating the relationship between low vs. high nSC and poor sleep health among racial/ethnic minority groups (i.e., NH-Black, Hispanic/Latinx, and Asian adults) [10,17,32,44]. It is noteworthy, however, that our findings were not consistent with two previous studies, which may be attributed to a difference in the operationalization and/or modeling of sleep dimensions. For example, findings from the Jackson Heart Study did not find an association between nSC and sleep disturbances after adjustment among a NH-Black population [43]. Their dichotomization of sleep disturbances compared to our ordinal operationalization may not capture meaningful differences in the average number of days of sleep disturbances. Another study that did not find an association between nSC and sleep health among a NH-Black population modeled sleep duration in a linear regression [45]; this does not account for the non-normal distribution of sleep duration. Rather, non-parametric methods, such as a Poisson regression, can better model the natural logarithm of average hours of sleep.
Our study is the first, to our knowledge, to examine the relationship between nSC and sleep health by age-sex/gender-race/ethnic groups. Our findings suggest that multiple social categories intersect to influence sleep health, although the findings were inconsistent with our hypotheses that associations would be stronger among racial/ethnic minority women and men ≥50 years old compared to NH-White women and men ≥50 years old. For example, we observed that NH-White women ≥50 years old and NH-Black women 18-30 years old who lived in neighborhoods with lower social cohesion experienced shorter sleep duration. We also observed that NH-Black women ≥50 years old who lived in neighborhoods with lower social cohesion experienced more insomnia symptoms, while Hispanic/Latinx women 18-30 years old who lived in neighborhoods with lower social cohesion experienced more trouble falling asleep. Additionally, we observed NH-Black 18-30year-old men living in low vs. high nSC experienced, on average, greater short sleep duration and more sleep disturbances (e.g., insomnia symptoms). Our findings are similar to another study that found nSC-sex/gender-race interactions with inflammatory biomarkers [46], which are associated with more sleep disturbances [47]. While our observed measures of association were not strikingly different, perhaps due to our large sample size, our findings suggest that sleep disparities may be explained by the impact of neighborhood environments across multiple identities and that age, sex/gender, and race/ethnicity impact each other in such a way that one identity alone cannot explain the sleep disparities without the intersection of the other identities.
It is hypothesized that nSC, and neighborhood environments in general, influence sleep health through different mechanisms including psychosocial, physiological, and social engagement pathways. Residing in neighborhoods with lower social cohesion and adverse environments characterized by discrimination, environmental hazards, and violence may increase anxiety, depression, and stress [15]. This may then lead to the dysregulation of the hypothalamic-pituitaryadrenal axis that impacts biological rhythms and sleep [15]. Similarly, adverse environments may also increase allostatic load and inflammatory biomarkers [12], which in turn impact sleep health. Neighborhood environments may also influence sleep health via social engagement, such as sharing resources, facilitating access to health related information, providing tangible support (e.g., transportation), reinforcing social norms for behaviors, and enhancing self-efficacy [48]. These hypothesized mechanisms may also differ by social categories. Older adults with limited mobility spend more time in their immediate neighborhood environments and thus rely more on their surroundings [18]. Women are thought to be more impacted by their social environment compared to men [22], and women may perceive neighbors' connectedness differently and utilize social support more than men. Even in neighborhoods that are limited in resources, linguistic and cultural similarities may allow for racial/ethnic enclaves to supplement smaller social networks to address the ongoing needs of those experiencing poor sleep [23]. Without a doubt, the intersection of these social categories, such as older NH-Black women, will be impacted by nSC via different mechanisms.
The limitations of this study include its cross-sectional design, which prevents causal inference regarding nSC and sleep health. As such, reverse causation is possible, where those with more sleep disturbances may be more likely to report low social cohesion and have negative perceptions about their neighborhood environment [49]. The use of both self-reported nSC and sleep dimensions may introduce measurement error. Although we used multiple sleep dimensions, the use of self-reported sleep measures tends to overestimate sleep duration compared to objective measures [50], and the degree as well as direction of measurement error tends to differ depending on the question asked [51]. Additionally, this study did not account for residential history, and changes in the neighborhood environment and perceptions of nSC may impact sleep health. Given the cross-sectional design of the study, it is suggested that future, longitudinal studies should account for the cumulative effect of living in one place, moving neighborhoods, and changes in the neighborhood environment. Another limitation includes unmeasured confounders, especially since sleep disorders, such as sleep apnea, may confound the relationship examined in our study; however, these data were not collected. Finally, the NHIS used a binary (e.g., man/woman) as opposed to a non-binary (man/woman/transgender) definition of sex/gender.
Despite these limitations, our study has strengths. For instance, we expanded upon the prior literature by investigating multiple sleep dimensions beyond duration as well as the potential modification of the nSC-sleep relationship by age, sex/gender, and race/ethnicity. Another strength of this study includes the use of a nationally representative sample, which enhances the generalizability of our results to the U.S. population of NH-White, NH-Black, Hispanic/Latinx, and Asian adults. The use of the most recent available data collected over multiple years decreased the potential influence attributable to single-year collection periods and increased the sample size. The large sample size allowed robust stratification by three variables both separately and together: age, sex/gender, and race/ethnicity. Another strength of the data includes the NHIS's quality control procedures, which increase the validity of these findings. Further, the use of a perceived measure to capture nSC is important, because perceived measures are more reflective of the impact that neighborhoods may have on health [52]. Within-neighborhood variations in perceptions of nSC are likely as important as individual-level characteristics; for example, dispositional affects can influence an individual's perception of their neighborhood's level of cohesion. Therefore, perceptions of nSC could differentially impact the sleep outcomes of individuals residing in the same neighborhood, and future research, such as assessing how nSC scores cluster among neighbors, is warranted to better understand the implications of this work. Finally, the use of the nonparametric statistic for sleep dimensions can serve as an example of how to model the natural logarithm of average hours of sleep duration and other sleep dimensions. The estimation of prevalence ratios rather than odds ratios is important when outcomes are not rare to avoid the overestimation of prevalence.

Conclusions
The current research adds knowledge regarding the important role that nSC may have on sleep health. nSC may serve as a key, modifiable neighborhood factor in health promotion programs that are focused on improving sleep. Social-cohesion oriented interventions may potentially mitigate the effect of stress on sleep by enhancing safety, trust, and social support. Research suggests that interventions that improve perceived nSC and other aspects of the social environment may result in improvements in older adults' and women's sleep health. Investments in improving the social and cultural qualities of local environments may not benefit all population subgroups uniformly, because we observed that those living in a neighborhood with lower cohesion experienced shorter sleep duration and more sleep disturbances. Future investigation of pathways linking neighborhood factors and sleep is warranted. Specifically, studies may benefit from including other variables of the social neighborhood environment, particularly those related to social resources (e.g., frequency of speaking with others), in order to determine their influence on sleep [53]. Future studies will likely benefit from also examining physical environment variables (e.g., noise, housing density, green space) in relation to sleep, as there may be competing pathways that have different effects on sleep [53]. Future studies should employ an intersectional perspective (or multiple intersecting identities) with the neighborhood environment to understand its influence on health. This approach can help disentangle the complex ways that identities intersect with the neighborhood environment to create social inequality in health. In conclusion, our findings suggest that the social neighborhood environment is associated with sleep duration and disturbances. Our findings underscore the importance of this upstream determinant of sleep health disparities and that the neighborhood environment may be a point of intervention for improving sleep health.