Associations of the residential built environment with adolescent sleep outcomes

Abstract

Study Objectives

Over 75% of US high school students obtain insufficient sleep, placing them at risk for adverse health outcomes. Identification of modifiable determinants of adolescent sleep is needed to inform prevention strategies, yet little is known about the influence of the built environment on adolescent sleep.

Methods

In this prospective study, actigraphy was used to assess sleep outcomes among 110 adolescents for 14 days each in eighth and ninth grades: duration (hours/night), onset and offset, and sleeping ≥8 hours. Home addresses were linked to built environment exposures: sound levels, tree canopy cover, street density, intersection density, population density, and housing density. Mixed-effects regression estimated associations of built environment measures with sleep outcomes, adjusting for sex, race, parent education, household income, household size, grade, weeknight status, and neighborhood poverty.

Results

A 1-standard deviation (SD) increase in neighborhood sound was associated with 16 minutes later sleep onset (β = 0.28; 95% confidence interval (CI): 0.06, 0.49) and 25% lower odds of sleeping for ≥8 hours (odds ratio (OR) = 0.75, 95% CI: 0.59, 0.96). A 1-SD increase in neighborhood tree canopy was associated with 18 minutes earlier sleep onset (β = −0.31, 95% CI: −0.49, −0.13) and 10 minutes earlier sleep offset (β= −0.17, 95% CI: −0.28, −0.05). No associations were observed for density-based exposures.

Conclusions

Higher neighborhood sound level was associated with lower odds of sufficient sleep, while higher tree canopy cover was associated with more favorable sleep timing. Neighborhood sound levels and tree canopy cover are potential targets for policies and interventions to support healthier sleep among adolescents.

sleep, neighborhood, noise, built environment, children, adolescent

Statement of Significance

A longitudinal cohort study of 110 adolescents examined whether the residential built environment, including measures of noise, tree canopy cover, street connectivity, and population density, associated with sleep duration and timing, measured using actigraphy. We observed that increased neighborhood noise was associated with later sleep onset and reduced odds of ≥8 hours of sleep per night. We also found increased neighborhood tree canopy cover was associated with earlier sleep onset and earlier sleep offset. This means that the neighborhood built environment, particularly noise and green space, may be important targets for interventions to support healthier sleep among adolescents.

Introduction

Insufficient sleep is associated with adverse health outcomes including obesity [1], unhealthy cardiometabolic profiles [2, 3], depression, and anxiety [4, 5]. In the United States, over 53% of middle school and 75% of high school students do not meet national guidelines for a healthy sleep duration on school nights [6, 7], increasing their risk for future chronic disease development. In addition, sleep timing metrics may be independent predictors of adverse outcomes in adolescence, including excess adiposity [8]. Given the importance of sleep to adolescent health, the identification of modifiable determinants of adolescent sleep is a clinical and public health need.

The importance of neighborhood-level factors in sleep is increasingly recognized [9, 10]. Lower neighborhood socioeconomic status (SES) has been linked with poor sleep outcomes among youth [11–14]. However, the mechanisms linking lower SES and poor sleep remain uncertain.

Residing in a socioeconomically disadvantaged neighborhood may increase residents’ exposure to adverse built environment features that could impair sleep, such as excess noise and light [15, 16]. Lower SES neighborhoods may also lack access to green space, including tree canopy cover [17]. Tree canopy cover has the potential to benefit sleep through protecting residents from light, noise [18], and excessive temperatures [19] and has also been found to be associated with lower stress [20], better mental health [20, 21], and greater psychological restoration [22].

Although it is plausible that the built environment may influence adolescent sleep health, this topic has not been extensively studied. Indeed, evidence regarding neighborhood effects on sleep largely stems from cross-sectional studies, self-reported sleep outcomes, and adult populations [9, 10]. Given the paucity of evidence regarding the relationship between the built environment and objectively measured adolescent sleep, we examined associations of the residential built environment with objectively measured sleep duration and timing during the middle-to-high school transition period. We hypothesized that adolescents residing in more densely populated and louder neighborhoods with less tree canopy cover would have shorter sleep duration and later timing of sleep onset and offset.

Methods

Participants

The Sleep and Growth Study is a longitudinal observational study that enrolled 118 adolescents in the eighth grade of middle school (2016–2018), with repeated assessments 1 year later (±1 month) when the adolescents had transitioned to the ninth grade of high school (2017–2019) [23]. Participants were recruited from the Children’s Hospital of Philadelphia (CHOP) network, which covers southeastern Pennsylvania and southern New Jersey [23]. Adolescents diagnosed with a medical or behavioral health condition that could affect their growth or sleep were excluded (e.g. cancers, sleep disorders, anxiety, depression, attention deficit hyperactivity disorder, and autism spectrum disorder). Parents/guardians provided written informed consent, and adolescents provided assent. The CHOP Institutional Review Board approved the protocol.

Sleep outcomes

As described previously, we used the Actiwatch-2 (Philips Respironics) to objectively estimate sleep patterns in the home setting [23]. The Actiwatch-2 was worn on the nondominant wrist [24, 25] for 14 days in eighth grade and for 14 days in ninth grade [23]. After each 14-day wear period, the device was returned by mail, and Actiware (v.6.0.8) was used to automatically detect sleep and rest periods using the medium wake threshold [24]. Manual adjustments to rest periods were considered when automatically generated start and/or end times were ±15 minutes different from the self-reported sleep diary [23]. All manual edits were reviewed by a second reviewer, and any disagreements were resolved by a third reviewer [23]. Sunday through Thursday nights were defined as school nights, and Friday and Saturday nights were defined as nonschool nights [23]. Sleep duration was operationalized as hours per night and a binary variable for sufficient sleep was generated (≥8 or <8 hours [26]). Sleep onset and offset timing were operationalized as hours from midnight (e.g. −1 = 11:00 pm).

Built environment exposures

Participant home addresses were geocoded using ArcMap (v.10.5.1). To link built environment exposures, we defined each participants’ residential neighborhood as: (1) the census block group in which their home was located and (2) a half-mile buffer around their home address based on Euclidean (straight line) distances. These definitions were chosen because we expected built environment features that might influence sleep to operate at close proximity and to enable comparisons with prior studies among adults [27, 28]. In a sensitivity analysis, we also used a quarter-mile buffer around home addresses to measure built environment exposures in closer proximity to home addresses.

Sound levels

Sound data were obtained from the US National Park Service Sound Map [29]. The Park Service used random forest (decision tree-based machine learning algorithm) models to predict the expected sound levels across the contiguous United States on a typical summer day with calm conditions based on empirical acoustical data from 479 locations in national parks and urban and suburban areas. The model incorporated geospatial features including topography, climate, land cover, and anthropogenic (human made) sources and was validated using leave-one-out cross validation [30]. Estimates were available at a resolution of 270 m. The specific sound metric used was L50 sound pressure level, which reflects the median sound level across all summer daytime seconds recorded for a given location after adjustment for the sensitivity of human hearing to very low and high frequencies (“A-weighting”) [31]. We calculated neighborhood-level sound levels by taking the average sound levels in A-weighted decibels (dBA) within a half-mile buffer of each participant’s residence.

Tree canopy cover

Data on tree canopy cover were obtained from the 2016 National Land Cover Database [32]. The Database was created by the US Forest Service and provides a raster file reflecting the estimated percent tree canopy coverage at a 30-m resolution for the continuous US. Estimates were produced using random forest regression models incorporating multi-spectral Landsat satellite imagery and elevation data [33]. We calculated the average percent tree canopy cover within a half-mile buffer of each participant’s home.

Street connectivity

Street connectivity was assessed by street density and intersection density using data obtained using Esri’s 2016 StreetMap Premium (Esri, Redlands, CA). Street and intersection density were calculated by dividing (1) the total linear meters of road in the half-mile buffer and (2) the total number of intersections in the buffer, by the buffer area in hectares.

Density

Block group-level population density and housing density were calculated using the data from the 2012–2016 American Community Survey by dividing each participants’ census block group population and number of housing units, respectively, by its land area in square miles.

Covariates

Participants self-reported their sex and race/ethnicity using the standard National Institutes of Health categories and the number of members in their household. The accompanying parent/guardian self-reported household income and education level. Based on data availability, we categorized participants as White, Black, and Other. The number living in the household was categorized as: ≤3, 4, 5, or ≥6. Parental household income was categorized as: <$40K, $40–69K, $70–99K, and >$99K. Parent/guardian education level was categorized as: some college or less, college degree, and graduate degree. Neighborhood poverty was assessed using the US Census Bureau’s 2012–2016 American Community Survey as the percent of residents in participants’ home census tract with household incomes below the federal poverty limit. Census data were retrieved from the IPUMS National Historical Geographic Information System [34].

Statistical analyses

As in other studies [28, 35], built environment variables were standardized to z-scores in order to facilitate comparisons across measures. Spearman correlations were calculated for the built environment variables for descriptive purposes. Separate mixed-effects models were used to test for associations between each built environment exposure and sleep outcomes. These models, which accounted for clustering of multiple days within participants (i.e. correlated, longitudinal data), adopted maximum likelihood estimation, random intercepts, robust standard errors, and an unstructured correlation matrix. We did not adjust for clustering by neighborhood because the vast majority of participants (85%) were the only participant in their block group and the number of block groups included in the study was relatively small [36]. Linear mixed models were used for continuous outcomes, and logistic mixed models were used for the sufficient sleep outcome. Based on prior work using these data [23], sleep outcomes changed during the middle-to-high school transition and the changes were different for school and nonschool nights. We, therefore, included grade, school night status, and grade-by-school night interaction terms in model 1. Race and income were also included as covariates in model 1, as Black adolescents and adolescents from lower-income households were more likely to have missing data in ninth grade (Supplementary Table 1) and lower adherence to the Actiwatch-2 protocol (Supplementary Table 2). In model 2, sex, parental education, and household size were added as covariates, and, in model 3, neighborhood poverty was added. Finally, differential associations between built environment exposures and sleep outcomes were tested by including grade and school night status interaction terms: built environment exposure-by-grade and built environment exposure-by-school night status. An alpha level of 0.05 was used to determine associations. All analyses were prespecified and completed using Stata 14.2 in 2020 (StataCorp, College Station, TX).

Results

A total of 118 adolescents in eighth grade were enrolled. As illustrated in Figure 1, sleep data from 107 participants in eighth grade (1,362 nights; average number of wear days: 13.5 [SD = 2.1]) and 84 participants in ninth grade (1,047 nights; average number of wear days 13.1 [SD = 2.3]) were captured, totaling 2,388 nights from 110 adolescents. In eighth grade, 53% of the adolescents were female, 69% reported being White, 65% reported living in a household with an income ≥$99K per year, and 72% lived in suburban neighborhoods (Table 1). Parent/guardian education levels and the number of members living in the household varied (Table 1).

Table 1.

Demographic, sleep, and neighborhood descriptive data

	Eighth grade (N = 107)	Ninth grade (N = 84)
Demographics
Age, mean (SD), y	13.9 (0.36)	14.9 (0.36)
Sex
Female, N (%)	57 (53.3)	42 (50.0)
Male, N (%)	50 (46.7)	42 (50.0)
Race
White, N (%)	74 (69.2)	65 (77.4)
Black, N (%)	27 (25.2)	14 (16.7)
Other, N (%)	6 (5.6)	5 (6.0)
Household income
$<40,000, N (%)	10 (9.5)	5 (6.0)
$40,000-69,999, N (%)	16 (15.2)	10 (11.9)
$70,000-99,999, N (%)	11 (10.5)	11 (13.1)
$>99,999, N (%)	68 (64.8)	58 (69.1)
Parent education
Some college or less, N (%)	40 (37.4)	27 (32.1)
College degree, N (%)	31 (29.0)	29 (34.5)
Graduate degree, N (%)	36 (33.6)	28 (33.3)
Census tract poverty rate, mean (SD), % below the poverty line	10.2 (8.6)	10.2 (8.4)
Neighborhood status
Urban, N (%)	30 (28.0)	22 (26.2)
Suburban, N (%)	77 (72.0)	62 (73.8)
Number living in the household:
≤3 family members, N (%)	21 (19.8)	12 (14.5)
4 family members, N (%)	48 (45.3)	42 (50.6)
5 family members, N (%)	24 (22.6)	20 (24.1)
≥6 family members, N (%)	13 (12.3)	9 (10.8)
Sleep outcomes
Sleep duration, mean (SD), hours per night	6.95 (0.66)	6.62 (0.75)
School nights	6.70 (0.79)	6.27 (0.84)
Nonschool nights	7.52 (0.91)	7.46 (0.98)
Sleep onset, mean (SD), hours from 00:00	−0.63 (0.87)	−0.29 (1.01)
School nights	−0.95 (0.87)	−0.58 (1.03)
Nonschool nights	−0.02 (1.11)	0.26 (1.10)
Sleep offset, mean (SD), hours from 00:00	7.19 (0.59)	7.16 (0.80)
School nights	6.59 (0.54)	6.45 (0.73)
Nonschool nights	8.45 (1.08)	8.72 (1.22)
Neighborhood built environment exposures
Sound level, mean (SD), A-weighted decibels‡	48.6 (2.94)	48.5 (2.90)
Tree canopy cover, mean (SD), % cover‡	17.9 (14.4)	19.2 (14.6)
Street density, mean (SD), meters per hectare‡	119.7 (52.0)	116.1 (54.4)
Intersection density, mean (SD), N per hectare‡	0.76 (0.53)	0.74 (0.59)
Population density, mean (SD), N per mile2†	7,727.3 (9,448.1)	7,950.8 (9,813.1)
Housing density, mean (SD), units per mile2†	3,348.5 (4,559.2)	3,489.7 (4,813.0)

Missing household income data in eighth grade: N = 2. Missing number living in household data in eighth grade and ninth grade, respectively: N = 1 and N = 1.

^†Within the participant’s home census block group.

^‡Within a half-mile buffer around the participant’s home address.

Figure 1.

Participants actigraphy sleep data. *Three participants did not provide actigraphy data in eighth grade but were retained and provided actigraphy data in ninth grade. The analytical sample includes 110 participants (i.e. 107 participants with at least eighth-grade actigraphy data plus three participants with only ninth-grade actigraphy data).

Means and SDs are provided in Table 1 for each built environment exposure, and additional descriptive statistics are provided in Supplemental Table 3 to provide more detailed information on the distribution of the built environment exposures. Positive correlations were observed between population, housing, street, and intersection densities (Supplemental Figure 1, ρ = 0.72 to 0.98). Spearman correlations also indicated that neighborhoods with higher population, housing, street, and intersection densities have less tree canopy cover (Supplemental Figure 1, ρ = −0.76 to −0.62) and higher sound levels (Supplemental Figure 1, ρ = 0.74 to 0.87). Neighborhood canopy cover was negatively correlated with sound levels (Supplemental Figure 1, ρ = −0.73).

Neighborhood noise and sleep patterns

Higher neighborhood sound was associated with later timing of sleep onset adjusting for grade, school night status, race, and parental income (Figure 2, model 1). This association remained after additional adjustment for sex, parental education, number living in the household, and neighborhood poverty (Figure 2, model 3: β = 0.28; 95% confidence interval (CI): 0.06, 0.49). Translated into minutes, a 1-SD higher neighborhood sound level was associated with sleep onset timing 16 minutes later per night on average in the fully adjusted model. In addition, higher neighborhood sound was associated with 25% lower odds of obtaining sufficient sleep (Figure 3, model 3: OR = 0.75; 95% CI: 0.59, 0.96). These associations remained consistent when using a quarter-mile buffer (Supplementary Figure 2).

Figure 2.

Main associations between built environment exposures and sleep-related outcomes. Model 1 (dark navy blue circles): adjusted for race and household income. Model 2 (red squares): adjusted for race, household income, sex, parent education, and number living in household. Model 3 (green triangles): adjusted for race, household income, sex, parent education, number living in household, and neighborhood poverty rate. All models include school night status, visit, and a school night status × visit interaction as fixed effects. Estimates reflect the difference in sleep duration or timing (or relative odds of obtaining sufficient sleep) associated with a 1-SD increase in each built environment measure. All error bars are 95% CIs.

Figure 3.

Neighborhood environment associations with sleep outcomes by weeknight status (gray-shaded backgrounds) and school grade level (white backgrounds). All models include weeknight status, visit, and a weeknight status × visit interaction, plus the following covariates: sex, race, parent education, household income, number living in household, and neighborhood poverty rate. Units on the x-axis are SDs for each built environment exposure, with a value of 0 indicating the mean level across the study population. All shaded areas are 95% CIs.

The neighborhood noise associations with sleep outcomes were consistent on school and nonschool nights (Supplementary Figure 3) and in eighth and ninth grades, with the exception of sleep offset (Supplementary Figure 4). A 1-SD higher neighborhood sound level was associated with later sleep offset in ninth grade only (β = 0.21 [13 minutes], 95% CI: 0.07, 0.36) (Figure 3).

Neighborhood canopy cover and sleep patterns

Higher neighborhood tree canopy cover was associated with earlier timing of sleep onset adjusting for grade, school night status, race, and household income (Figure 2, model 1). This association remained with additional adjustment for sex, parental education, number living in the household, and neighborhood poverty (Figure 2, model 3: β = −0.31; 95% CI: −0.49, −0.13). The same direction of associations was observed between neighborhood tree canopy cover and timing of sleep offset (Figure 2, model 3: β = −0.17, 95% CI: −0.28, −0.05). Translated into minutes, a 1-SD increase in neighborhood canopy cover was associated with sleep onset that was on average 18 minutes earlier per night and with sleep offset on average 10 minutes earlier per morning. When using a quarter-mile buffer, these associations were in the same direction but attenuated with CIs crossing the null (Supplementary Figure 2).

The associations between neighborhood canopy cover with sleep onset and sleep offset were modified by weeknight night status (Supplementary Figure 3). A 1-SD higher neighborhood canopy cover was more strongly associated with earlier sleep onset on nonschool nights (β = −0.40 [24 minutes], 95% CI: −0.61, −0.20) compared with school nights (β = −0.26 [16 minutes], 95% CI: −0.44, −0.09). However, a 1-SD higher neighborhood canopy cover was associated with earlier sleep offset on nonschool nights (β = −0.33 [20 minutes], 95% CI: −0.51, −0.15), but not school nights (Figure 3). Furthermore, an association between higher neighborhood tree canopy cover and sleep duration was observed in eighth grade only (Figure 3, Supplementary Figure 3). Specifically, a 1-SD higher neighborhood canopy cover was associated with longer sleep duration in eighth grade (β = 0.15 [9 minutes], 95% CI: 0.03, 0.27).

Neighborhood density-based measures and sleep patterns

We did not observe associations between density-based built environment exposures and any sleep outcome (Figure 2). However, street-density-by-grade and intersection-density-by-grade interactions were observed with respect to sleep offset (Supplementary Figure 3). In ninth grade, a 1-SD higher street density was associated with later sleep offset (β = 0.17 [10 minutes], 95% CI: 0.04, 0.31) (Figure 3); similarly, a 1-SD higher intersection density was associated with later sleep offset (β = 0.14 [8 minutes], 95% CI: 0.01, 0.27) (Figure 3).

Discussion

In a prospective cohort of adolescents, we examined the association of features of the built environment with objectively measured sleep duration and timing. Results suggest that greater exposure to neighborhood noise is related to later sleep onset and reduced likelihood of sleeping sufficiently. In contrast, greater neighborhood tree canopy cover was associated with earlier sleep timing, particularly on nonschool nights and with a longer sleep duration in eighth grade specifically. Our results provide evidence supporting the importance of several modifiable built environment exposures to adolescent sleep health.

Currently, the evidence on the role of environmental noise in adolescent sleep is limited by a lack of objective measurement of sleep outcomes. Several studies in European populations of children and adolescents have reported associations between environmental noise and more adverse parent-reported child sleep outcomes [37–40]. Only one study measured sleep objectively in a subsample and did not find associations between noise and actigraphy-assessed sleep [37]. Our finding that higher neighborhood noise was associated with reduced odds of obtaining sufficient sleep, assessed using actigraphy, provides additional support for the importance of environmental noise for adolescent sleep health. We also found higher neighborhood noise level to be associated with later objectively measured sleep timing, which is an important indicator of sleep health. Independent of sleep duration, later sleep timing among adolescents has been associated with adverse outcomes, including adiposity [8], substance use [41], emotional distress [42], and lower academic achievement [42]. A prior study reporting results from a nationally representative cross-sectional sample of urban US adolescents indicated that environmental noise was associated with later self-reported sleep onset but not with shorter sleep duration [43]. Additional research in other populations using objective estimates of sleep duration and timing are needed to verify our neighborhood noise findings.

We did not observe a main association between greater neighborhood tree canopy and objectively measured sleep duration among adolescents, but interaction analyses revealed that greater neighborhood tree canopy was associated with longer sleep duration in eighth grade only. In addition, we observed that greater neighborhood tree canopy cover was associated with earlier sleep timing in adolescents, and these associations were particularly strong on nonschool nights. Interestingly, these associations were detected using a half-mile buffer, but not a quarter-mile buffer, although results were in the same direction. This adds to a small body of evidence linking green space with better sleep outcomes among adults [27, 44–46] and a larger body of work suggesting green space promotes improved physical and mental health [47–49]. Four prior studies of adults have found greater exposure to green space to be associated with a lower risk of self-reported insufficient sleep [27, 44–46], two of which specifically examined tree canopy cover [27, 46]. Objective and self-reported sleep assessments may not correlate well [50], which may contribute to the difference in study results. However, our findings do suggest that greater tree canopy cover may help to shift adolescent sleep timing earlier without changing sleep duration. Tree canopy cover has the potential to benefit sleep through protecting residents from light, noise [18], and excessive temperatures [19]. Therefore, it is possible that tree canopy cover regulates exposure to zeitgebers, environmental cues that help to regulate circadian rhythms, so that the timing of sleep is more aligned with internal circadian clocks. Again, replication of these findings is necessary, particularly in a sample with a wider age range given that our sleep duration association with tree canopy cover finding was restricted to eighth grade. Neighborhood canopy cover may have a greater impact on sleep at younger ages, before the onset of puberty and earlier high school start times become predominant determinants of older adolescent sleep patterns. Also, given the inconsistency of associations for differing buffer sizes, future research should use more refined spatial epidemiological methods (e.g. GPS (global positioning system) tracking [51] and audits of the immediate block environment) [52].

We did not find evidence of an association between adolescent sleep and neighborhood density and street connectivity. However, interaction analyses revealed that higher street and intersection densities were associated with later timing of sleep offset in ninth grade only. In prior research, greater walkability, intersection density, population density, and density of social engagement destinations were associated with shorter sleep duration among middle-aged to older adults in the Multi-Ethnic Study of Atherosclerosis [28]. In that study, self-reported neighborhood-level noise partially explained these associations, suggesting higher noise in denser, more walkable neighborhoods as a potential mechanism for these built environment features to impair sleep. Similarly, greater urban land use was associated with shorter parent-reported infant sleep duration among a cohort of infants in Massachusetts, although no associations were observed for population density or proximity of a major roadway [53]. More dense, walkable neighborhoods may expose residents to greater levels of noise, traffic, light, and air pollution, which may disrupt sleep [9]. However, density and walkability may also promote physical activity, including walking within the neighborhood [54]. Thus, dense and walkable neighborhoods may have both sleep-promoting and sleep-impeding features. Future work is needed to disentangle specific pathways and mechanisms through which these features of the built environment may influence sleep in adult and pediatric populations. This could involve performing audits of neighborhood blocks to allow for a detailed assessment of sidewalk accessibility, walkability, and neighborhood disorder, which cannot be achieved with geocoding [52].

Our findings suggest a potential role for neighborhood tree canopy cover and sound levels in adolescent sleep health that should be explored in future research. The observational nature of our study limits our ability to make causal assertions regarding the role of the built environment in adolescent sleep health. However, future experimental research may help to establish causality of these associations by testing whether built environment changes such as vacant lot greening [55], increasing the urban tree canopy, or noise reduction policies (noise emission limits and improved acoustical design practices) [56] improve adolescent sleep outcomes. Such experiments would allow for more certainty in advocating for policy-level changes in the built environment to improve adolescent sleep health. Future experiments should be designed to investigate not only whether the built environment has a direct influence on sleep health (e.g. sound levels in the bedroom), but also whether there are upstream benefits of built environment improvements that positively impact social and psychological health, which in turn leading to better sleep health. Further, if experiments are performed in lower-income neighborhoods, this will provide an opportunity to learn if reducing noise levels and/or increasing canopy cover partly explain why residents of lower-income neighborhoods tend to have poorer sleep health.

A strength of our study is the use of objectively measured sleep. Most of the studies examining associations between neighborhood context and sleep have relied upon self-reported or parent-reported sleep measures [9, 10], which may not correlate well with objective measures [50]. However, several limitations must be noted. First, our exposure measures were assessed based on participants’ home addresses and do not account for adolescents’ exposure to the built environment in other contexts during daily activities. Second, these exposures are averages based on secondary datasets that do not account for individual-level variation in exposure. Further study is needed with a more detailed, individualized assessment of exposures such as neighborhood noise within the bedroom environment that may vary from night to night. Third, the noise measure was based on a model predicting sound levels at a 270-m resolution, which may have led to measurement error as we were unable to account for sound pressure dissipation. Fourth, we were unable to examine the role of neighborhood change, given the relatively short time frame of the study, the fact that no family relocated in ninth grade, and the lack of availability of exposure measures (e.g. noise and tree canopy) on an annual basis. Fifth, although we adjusted for potential confounders including family and neighborhood-level SES, results may be subject to residual confounding due to unmeasured variables. Sixth, results might be biased by neighborhood self-selection if parents who choose to live in neighborhoods that are quieter and/or greener are also more effective at enforcing child sleep. Finally, our study population was made up of a majority of White and high-income adolescents, which may limit external validity by not generalizing to the larger Philadelphia metro area and southern New Jersey and to other sociodemographic groups more broadly.

Conclusions

Higher neighborhood noise level was associated with insufficient sleep among adolescents, while higher tree canopy cover was associated with earlier sleep timing and longer sleep duration (in eighth grade only). The results suggest that the neighborhood built environment, particularly noise and green space, should be considered when intervening to support healthier sleep among adolescents.

Acknowledgments

We thank the families for volunteering to take part in this study. We appreciate the dedication of the research staff who helped to collect and manage the data and the Recruitment Enhancement Core and the Pediatric Research Consortium at the Children’s Hospital of Philadelphia.

Funding

This study was supported by an National Institutes of Health (NIH)/National Heart, Lund, and Blood Institute (NHLBI) Career Development Award K01HL123612 (J.A.M.) and the following NIH/National Center for Advancing Translational Sciences (NCATS) awards: UL1TR000003 and UL1TR001878. A.A.W. is supported by Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) Career Development Award K23HD094905 and by the Sleep Research Society Foundation. S.F.A.G. is supported by the Daniel B. Burke Endowed Chair for Diabetes Research and R01HL143790.

Conflict of interest statement. Financial disclosure: None. Non-financial Disclosure: None.