Feasibility and Acceptability of Mobile Methods to Assess Home and Neighborhood Environments Related to Adolescent Sleep

Abstract

Objectives:

A growing evidence base suggests home and neighborhood environmental exposures may influence adolescent sleep, but few studies have assessed these relationships using methods that account for time-varying, location-specific exposures or multiple neighborhood contexts. This study aimed to assess the feasibility and acceptability of using smartphone GPS tracking and ecological momentary assessment (EMA) to assess time-varying home and neighborhood environmental exposures hypothesized to be associated with adolescent sleep.

Methods:

Adolescents aged 15-17 years in Philadelphia completed 7 days of continuous smartphone GPS tracking, which was used to identify daily levels of exposure to geocoded neighborhood factors (e.g. crime, green space). Four daily EMA surveys assessed home sleep environment (e.g. noise, light), stress, health behaviors, and neighborhood perceptions. Feasibility and acceptability of GPS tracking and EMA were assessed, and distributions of daily environmental exposures were examined.

Results:

Among 25 teens (mean age 16, 56% male), there was a high level of GPS location data captured (median daily follow-up: 24 hours). 78% of EMA surveys were completed overall. Most participants (96%) reported no privacy concerns related to GPS tracking and minimal burden from EMA surveys. Exposures differed between participants’ home neighborhoods and locations visited outside the home neighborhood (e.g. higher crime away from home). Sleep environment disruptions were present on 29% of nights (most common: uncomfortable temperature) and were reported by 52% of adolescents.

Conclusions:

Results demonstrate the feasibility and acceptability of mobile methods for assessing time-varying home and neighborhood exposures relevant to adolescent sleep for up to one week.

A. Introduction

The social ecological model posits that environmental factors may impact health behaviors such as sleep, including exposures within the home and in neighborhoods where people live and spend time.¹ Among adolescents, home sleep environment problems like noise, light, or uncomfortable temperatures are associated with poorer sleep outcomes, including sleep-wake problems, sleepiness, and greater variability in sleep schedule.^2,3 In addition, evidence is mounting that neighborhood socioeconomic status (SES), physical features (e.g. greenspace, walkability, ambient light) and social features (e.g. crime, physical disorder, cohesion) are associated with outcomes including sleep duration, timing, and quality.^4–6 These environmental features, which vary as a result of structural factors including racially discriminatory housing policies and neighborhood disinvestment, are hypothesized to impact sleep either adversely (e.g. crime) or favorably (e.g. neighborhood greenspace), and might contribute to well-described racial and socioeconomic disparities in sleep health.⁷ For example, among adolescents, neighborhood crime or low perceived neighborhood safety are associated with self-reported sleep problems,^8,9 later bedtime,¹⁰ lower sleep efficiency,^8,11 and lower odds of obtaining sufficient sleep.¹² Other aspects of the neighborhood physical and social environments, such as social cohesion,^13–16 greenspace,^17–19 and physical disorder,^14,15 have been less frequently studied in relation to adolescent sleep outcomes, with mixed findings.⁶

A limitation of prior research on neighborhood environments and sleep among adolescents is the nearly exclusive focus on residential neighborhood exposures.⁶ Neighborhood exposures are often assessed based on the administrative units (e.g. census tract, block group) in which participants live, or within a buffer around their homes. However, adolescents often spend large portions of their day away from home²⁰ and may accrue environmental exposures in other settings that can impact sleep. In addition, sleep environment conditions within the home may vary from night to night,^21,22 but are rarely measured as time-varying exposures in studies. These limitations may result in misclassification and incomplete understanding of the processes by which home and neighborhood environments influence adolescent sleep.

The near ubiquity of smartphones offers a promising opportunity to improve exposure assessment by leveraging built-in global positioning system (GPS) sensors, which capture mobility and enable linkage with neighborhood environmental exposures across all context where adolescents spend time.²³ In addition, ecological momentary assessments (EMA) enable in-the-moment measurement of contextual factors that vary between and within days, such as sleep environment conditions or neighborhood perceptions.²³ The objective of this report is to demonstrate the feasibility and acceptability of using GPS tracking and EMA to assess environmental exposures hypothesized to impact adolescents’ sleep. Results are presented on feasibility, acceptability, and descriptive analyses of daily home and neighborhood exposure patterns from a pilot study of adolescents in Philadelphia. We also present bivariate associations of these environmental exposures with sleep outcomes as an exploratory analysis, although results should be interpreted with caution given the small sample size which limited our ability to conduct multivariable analyses.

B. Participants and Methods

B.1. Study Population and Recruitment

A convenience sample of adolescents aged 15-17 years old in Philadelphia, Pennsylvania was enrolled between November 2020 and July 2021. Adolescents were eligible to participate if they were (1) English-speaking, (2) had a smartphone with a text/data plan, and (3) lived and attended school in the city of Philadelphia. Exclusion criteria included diagnosis with a sleep disorder, use of sleep medication, diagnosis with a condition that might impact sleep or mobility (e.g., depression, musculoskeletal or neurological disorder that limits activity), and experiencing symptoms of acute illness. Because the study took place during the COVID-19 pandemic, a remote recruitment and data collection strategy was implemented using the REDCap data management platform²⁴ and WebEx videoconferencing software. Participants were primarily enrolled through email invitations sent by the Children’s Hospital of Philadelphia (CHOP’s) Recruitment Enhancement Core (REC) to parents/guardians of 15-17-year-old adolescents with addresses in Philadelphia who had received care in the CHOP network. Flyers were also posted in CHOP primary care practices and on the hospital campus. Parents of potential participants completed an online screening form and, if eligible, parents provided written informed consent and adolescents provided assent.

B.2. Data Collection

Upon enrollment, the adolescent and one parent each completed a baseline survey assessing self-reported demographics, home and neighborhood environment perceptions, and sleep patterns. Upon receipt of a mailed sleep monitor and paper sleep diary, adolescents completed three forms of continuous data collection over the course of 7 days: (1) GPS tracking, (2) ecological momentary assessment (EMA), and (3) sleep monitoring. GPS tracking was conducted using an app from the AWARE Framework system,²⁵ which participants downloaded onto their own smartphones. The app ran continuously in the background to collect and transmit the geographic location (latitude and longitude coordinates) of the participants’ phones every 60 seconds, as well as speed, bearing, altitude, and the estimated accuracy in meters of the location points. In addition, participants responded to four EMA surveys per day via text messages with embedded REDCap survey links. Surveys were sent out at random times within the following intervals on school days: 6:00-8:00, 14:00-16:00, 17:00-19:00, 20:00-22:00. On non-school days, the intervals were: 9:00-12:00, 13:00-16:00, 17:00-19:00, 20:00-22:00. The morning window was adjusted as needed based on participant’s self-reported wake times and school start times. EMA surveys assessed sleep quality, sleep environment,² location, environmental perceptions, stress, and health behaviors (see Table A.1). Finally, participants wore an Actiwatch-2 (Philips Respironics) on their non-dominant wrist for 7 days and completed a sleep diary each morning after waking up and evening before bed.

At the end of the study period, adolescents participated in a brief exit interview. Interviews consisted of open and closed-ended questions regarding feasibility and acceptability of the study procedures. Participants also viewed a map showing their GPS location data for a recent date while they were in the study and reported on the accuracy of the locations plotted. Adolescent participants received up to $90 for participation and parent participants received $10 for their participation. The Children’s Hospital of Philadelphia Institutional Review Board approved this study.

B.3. Feasibility and Acceptability Outcomes

Feasibility of GPS tracking was assessed by the total number of coordinates captured, number of minutes tracked per day, and participant review of example daily travel paths. Feasibility of EMA was assessed by the proportion of surveys completed, overall and by survey time. Acceptability of study procedures was assessed using 5-point Likert-scaled and open-ended questions in exit interviews.

B.4. Daily Activity Path-Based Exposure Assessment

Daily activity path-based measures were created to quantify adolescents’ exposure to neighborhood-level factors hypothesized to impact sleep.²⁶ GPS coordinates were excluded if they were located outside of Philadelphia or if their estimated accuracy was greater than 100 meters. To better reflect adolescents’ daily activity spaces, person-days were excluded if fewer than 360 minutes of accurate GPS data were available.²⁷ Using ArcGIS Pro 2.5.1, (Esri, Redlands, CA) the remaining points were overlaid on multiple layers of geographic data selected to represent a range of neighborhood domains hypothesized to impact sleep either adversely (social disorganization, crime, physical disorder, noise) or favorably (greenspace, neighborhood opportunity).²⁶ The tract-level measures included the social disorganization index made up of 8 variables from the 2019 American Communities Survey²⁸ and the 2015 Childhood Opportunity Index.²⁹ Proximity-specific measures, defined as a 660 foot buffer (approximate length of a city block) around each GPS point, included noise due to aviation, road and rail from the Bureau of Transportation Statistics³⁰; 2020 count of all police-recorded crimes from the City of Philadelphia³¹; and percentage tree canopy cover.³² The block-level litter index, indicating the quantity of litter on the block as assessed by trained raters from the City of Philadelphia, ranging from 1 to 4 was also included.³³ GPS point-level exposures were aggregated to the person-day level by taking the average value across all points within a given day for each participant, resulting in up to 7 values (one per day) per participant for each exposure. This approach inherently weights exposure values such that areas where adolescents spent more time made larger contributions to the daily average.

As a secondary approach to defining neighborhood exposures, the percentage of time adolescents spent in areas with a high level of exposure to each neighborhood domain was calculated. High exposures were quantified using benchmarks based on local average values for social disorganization, crime, litter, and tree canopy cover.^31,34,35 For noise due to aviation, rail, and roads, a threshold of 55 decibels (U.S. Environmental Protection Agency-recommended limit) was used.³⁶ Finally, for the Child Opportunity Index, the two highest quintiles, reflecting “high” and “very high” neighborhood opportunity were used.

B.5. Statistical Analysis

Demographic characteristics of participants were examined using descriptive statistics. Sleep measures including duration, efficiency, sleep onset latency, wake after sleep onset (WASO), and sleep onset and offset were calculated using Phillips Actiware version 6.1.1 software, and summarized as means and 95% confidence intervals using mixed effects linear models with participant random intercepts to account for clustering of nights within participants.

To understand the extent of adolescents’ mobility during follow-up, the maximum distance each participant traveled from home (within the Philadelphia city limits) was calculated using straight-line distances. Then, the distribution of daily activity path-based exposure to neighborhood stressors and supports was examined using descriptive statistics. Average daily neighborhood exposures and percentage of time spent in “high exposure” neighborhoods were calculated across teens using mixed effects linear models with participant random intercepts, yielding means and 95% confidence intervals that accounted for clustering of days within teens. Finally, participants’ average levels of exposure to neighborhood stressors and supports were compared for GPS coordinates within 1000 meters of their geocoded home address versus >1000 meters from home. This buffer was selected based on prior studies^17,37,38 in order to compare exposures in participants’ residential neighborhoods versus other neighborhoods intersected by daily travel paths. Differences were estimated using mixed effects linear regression models with each neighborhood exposure as the dependent variable, home (versus not home) location as the independent variable, and including participant random intercepts to account for clustering of GPS points within participants.

Data from the morning EMA surveys, which assessed whether adolescents’ sleep had been disrupted the previous night due to seven problems in the home sleep environment (e.g. noise, light), ² were examined descriptively. The proportion of teens ever reporting each sleep environment disruption during the study period was calculated, as well as the proportion of nights each disruption was reported, to describe variation both within and between adolescents.

Finally, as an exploratory analysis, we assessed bivariate associations of the GPS-based environmental exposures described above, as well as EMA-reported home sleep environment disruptions, with sleep outcome measures. We used mixed effects linear regression models to assess associations with the objectively measured sleep outcomes listed above, with participant random intercepts to account for clustering of nights within adolescents. We then used modified Poisson regression models to calculate prevalence ratios reflecting associations of home and neighborhood exposures with likelihood of reporting a sleep problem on the morning EMA survey, accounting for clustering of nights within adolescents using generalized estimating equations.³⁹ Neighborhood measures were transformed into z-scores for this analysis to increase interpretability, given the different units of the various measures.

C. Results

Among 25 participants, the mean age was 16.2 years and 14 (56%) were male (Table 1). Thirteen adolescents (52%) reported their race/ethnicity as non-Hispanic white, 9 (36%) as non-Hispanic Black, and 3 (12%) as other race/ethnicity categories. Two thirds (68%) had parents with a bachelor’s degree or higher, and 6 (24%) had annual household incomes below $50,000. Participants’ average sleep duration was 7.5 hours (95% CI: 7.2, 7.9) and average sleep efficiency (ratio of total sleep time to time in bed) was 84.2% (95% CI: 82.1, 86.4). Average sleep onset and offset times for overnight sleep periods were 00:54 (00:18, 01:30) and 08:24 (07:48, 08:54).

Table 1.

Participant Characteristics (n=25)

	N (%)
Teen characteristics
Age in years (mean ± SD)	16.2 ± 0.7
Gender
Male	14 (56)
Female	11 (44)
Race/Ethnicity
Non-Hispanic White	13 (52)
Non-Hispanic Black/African American	9 (36)
Other race/ethnicity	3 (12)
Parent characteristics
Age (mean ± SD)	48 ± 6.1
Relationship to Adolescent
Mother	24 (96%)
Father	1 (4%)
Marital Status
Married/Living with partner	15 (60)
Not married	10 (40)
Education Status
Bachelor’s degree or higher	17 (68)
Associate degree	3 (12)
1 or more years of college, no degree	5 (20)
Employment Status
Employed	22 (88)
Unemployed	3 (12)
Annual Household Income
<$50,000	6 (24)
$50,000 to $99,999	8 (32)
$100,000 or more	11 (44)
Teen Sleep Measures
Actigraphy-based (n=142 nights from 23 teens)a	Mean (95% CI)b
Sleep Duration (hours)	7.5 (7.2, 7.9)
Sleep Efficiency (%)	84.2 (82.1, 86.4)
Sleep Onset Latency (minutes)	8.0 (5.5, 10.5)
Wake After Sleep Onset (WASO) (minutes)	53.1 (45.1, 61.1)
Sleep Onset	00:54 (00:18, 01:30)
Sleep Offset	08:24 (07:48, 08:54)
From Morning EMA Survey (n=108 nights from 24 teens)	N (%) of days reported
Self-reported sleep problemc	33 (30.6)

EMA: Ecological Momentary Assessment

^aActigraphy-based sleep outcomes were available for 23 out of 25 teens. Sleep outcomes were missing due to failure to return the Actiwatch (n=1) and Actiwatch malfunction (n=1).

^bCalculated using mixed effects models to account for clustering of nights within participants

^cIndicates adolescent gave a positive response on any of 4 questions on sleep disturbances

C.1. GPS Tracking Feasibility

A total of 202,763 GPS coordinates were captured from the 25 participants over 7 days of follow-up. Five participants (20%) had location coordinates available for every possible minute over their 7-day follow-up period. Across participant-days, location coordinates were available for an average of 19.6 hours per day (standard deviation (SD): 8.2 hours, median: 24 hours, range 0-24 hours). Missing GPS data primarily resulted from participants turning their phone off and from an issue with Android phones where additional settings needed to be manually enabled to override limits on data collection. To confirm accuracy, participants reviewed maps showing their GPS locations on a recent day during the exit interview. Overall, participants reported that the maps accurately reflected their movements, with a few exceptions (e.g. one participant had inaccurate data due to an unexplained technical issue).

C.2. Ecological Momentary Assessment Feasibility

Overall, participants completed 78.6% of EMA surveys, with the highest completion percentages on in the late afternoon (88.0%) and the lowest completion percentages in the morning (61.7%). Across participants, the median completion rate was 86% (range 29-100%). Participants reported that the main barrier to EMA completion was not being near their phone (e.g. putting it down to complete an activity) when a notification was sent. Adolescents suggested having consistent pre-specified times for survey completion and a longer window to complete the morning surveys.

C.3. Acceptability of Study Procedures

Nine Likert-scaled acceptability questions assessed privacy concerns and participant burden (Figure 1). Only 1 participant (4%) agreed with the statement “I was worried about my privacy due to the app tracking my location.” In the open-ended portion of the interview, respondents elaborated that having their location tracked did not concern them due to using other phone applications with location tracking or having a parent track their location with an app, and that they did not change their usual behaviors in response to knowing that their locations were being tracked (Table A2). In terms of participant burden, 80% agreed/strongly agreed that it was easy to respond to EMA surveys and 84% disagreed/strongly disagreed that responding to the EMA surveys interfered with daily activities (Figure 2).

Figure 1. Acceptability of Study Methods to Participants-Summary of Likert-Scaled Responses.

All questions were assessed on a 5-point Likert scale in the post-data collection exit interview. Percentages on the left indicate the percent of participants responding “disagree” or “strongly disagree”, percentages in the middle indicate the percent of participants responding “neither agree nor disagree”, and percentages on the right indicate the percent of participants responding “agree” or “strongly agree”.

Figure 2. Map of participant GPS activity paths overlaid on census tract-level social disadvantage data in Philadelphia.

This figure shows how GPS activity paths can be overlaid on geographic data sources and linked to neighborhood characteristics. In this example, GPS points were plotted (black dots) to represent participants’ activity paths over 7 days. GPS points were aggregated across participants and randomly jittered to mask exact locations. Participants’ activity paths were overlaid on census tract-level values of social disadvantage, an index created by summing 8 proportion variables from the 2019 American Communities Survey). Darker colors indicate higher levels of social disadvantage.

C.4. Daily Neighborhood Exposures

Figure 2 displays participants’ GPS coordinates over the study period, combined and jittered randomly to protect privacy, overlaid on to of census tract-level social disorganization scores. Daily exposure to neighborhood stressors (crime, litter, noise, social disorganization) and supportive factors (greenspace, neighborhood opportunity) varied between teens (Table 2). For example, average activity path greenspace, defined as percentage tree canopy cover, ranged from 2.2% to 47.8%, with an average of 14.9%. Participants spent 41.6% of their follow-up time on average in neighborhoods with crime levels higher than the average across all Philadelphia census blocks, 34.7% in neighborhoods with higher litter than average, and 37.8% in neighborhoods with higher than average levels of social disorganization. They spent 14.7% of their time on average in neighborhoods with transportation noise levels above the EPA-recommended threshold of 55 decibels. Finally, participants spent 22.4% of their time in neighborhoods with greater tree canopy coverage than the overall city-level average, and 20.0% in neighborhoods that were “high” or “very” high in overall neighborhood opportunity according to the Childhood Opportunity Index.

Table 2.

Distribution of Adolescents’ Daily GPS-based Neighborhood Exposures to Stressors and Supportive Factors (n=135 days from 21 adolescents)

	Daily activity path-based average exposurea		Percentage of adolescents’ daily activity path spent in “high exposure” neighborhoodsb
Neighborhood stressors	Mean (95% CI)c	Range	Mean (95% CI)c
Number of police-recorded crime incidents (higher=more crime)	98.7 (63.3, 134.1)	4.3-537.0	41.6% (23.1, 60.0)
Litter index score (possible range 1-4, higher=more litter)	1.8 (1.7, 1.9)	1.3-2.5	34.7% (16.2, 53.1)
Noise from aviation, road, rail (in decibels, higher=more noise)	51.1 (50.1, 52.2)	45.1-56.0	14.5% (4.9, 24.2)
Social disorganization index: (possible range 0-8, higher=more disorganization)	1.3 (1.2, 1.5)	0.5-2.0	37.8% (19.2, 56.5)
Neighborhood supports
Percentage tree canopy cover (possible range 0-100%, higher=more trees)	14.9 (10.8, 19.0)	2.2-47.8	22.4% (6.3, 38.5)
Child Opportunity Index score (range 0-100, higher=greater Opportunity)	31.2 (20.4. 42.0)	1.8-87.5	20.0% (5.0. 35.0)

^aDaily activity path neighborhood exposures were calculated by overlaying each teen’s GPS points on layers of geospatial data (aggregated at the census tract or block level) and calculating teens’ average values for each day for each neighborhood measure. Example interpretation: average activity path tree canopy cover ranged from 2.2% to 47.8%, with an average of 14.9%.

^bReflects percentage of GPS points across a participant’s follow-up period that were located within a neighborhood classified as “high exposure” to a given neighborhood feature based on benchmarks created based on local data or national norms. Benchmarks were as follows: crime ≥90.7 incidents (average across Philadelphia blocks), litter score ≥1.92 (average across Philadelphia blocks), noise ≥55 decibels (EPA recommended limit for a 24 hour period), social disorganization score ≥1.42 (average across Philadelphia tracts), tree canopy ≥20% (Philadelphia average), Child Opportunity Score in top 40% nationally (high or very high opportunity). Example interpretation: on average, 41.6% of adolescents’ daily activity paths were in neighborhoods with higher crime levels than the average for Philadelphia as a whole.

^cCalculated from “empty” mixed effects linear regression model with a participant-level random intercept to calculate means and 95% confidence intervals of neighborhood exposure values that account for clustering of days within participants.

Adolescents’ average maximum distance travelled (within the Philadelphia city limits) was 8,122 meters (SD: 6,548, range 17-20,581). Exposure to several neighborhood stressors (crime, noise, and social disorganization) were higher at locations >1000 meters from participants’ homes compared to locations within 1000 meters of home, while litter was higher on average in their home neighborhoods (Table A3). For neighborhood supportive features, percentage tree canopy cover was higher in participants’ home neighborhoods while neighborhood opportunity was higher away from home (p<0.001 for all comparisons) (Table A3).

C.5. Home Sleep Environment Problems

Sleep environment disruptions were reported on 28.7% of morning EMA surveys overall, and 13 adolescents (52%) ever reported a sleep environment disruption during the study period (Table 3). Across participants, the median percentage of nights experiencing a sleep environment disruption was 27.7% (range 0-100%). The most reported sleep environment disruption was uncomfortable temperature (36% of teens), followed by an uncomfortable bed and noise inside (16% each).

Table 3.

Frequency of Home Sleep Environment Problems among Adolescents^a

Home sleep environment problems	N (%) of teens ever reporting	% of days teen reported
		Mean (SD)	Range
Any problem	13 (52%)	27.7 (35.4)	0-100
Specific problems:
Noise outside	3 (12%)	2.3 (6.6)	0-25
Noise inside	4 (16%)	4.6 (12.2)	0-50
Someone snoring	1 (4%)	0.7 (3.3)	0-17
Uncomfortable bed	4 (16%)	6.3 (17.2)	0-75
Temperature of the room	9 (36%)	16.1 (27.9)	0-100
TV, radio, computer	2 (8%)	4.0 (14.7)	0-67
Too much light	1 (4%)	4.0 (20.0)	0-100

^aBased on 108 morning ecological momentary assessments completed (out of 175 possible, 25 teens x 7 days). Data were collected between November 2020 and July 2021.

C.6. Bivariate Associations of Neighborhood and Home Sleep Environment Measures with Sleep Outcomes

Table A4 presents bivariate associations of daily activity path-based neighborhood exposures and EMA-reported home sleep environment problems with sleep outcomes. A standard deviation (SD) higher daily exposure to neighborhood crime was associated with a 25.4-minute earlier sleep offset timing (95% CI: 1.4, 49.4) and with higher prevalence of self-reported sleep problems (prevalence ratio (PR): 1.56, 95% CI: 1.07, 2.28). A SD higher exposure to neighborhood litter was associated with a 40.2-minute earlier sleep offset timing (95% CI: 3.5, 77.0), while an SD higher exposure to neighborhood social disorganization was associated with a 27.5-minute longer sleep duration (95% CI: 1.5, 53.5). Finally, participants who reported ≥1 home sleep environment disruption were more likely to report sleep problems on the same night (PR: 2.82, 95% CI: 1.57, 5.06) compared to participants who did not report any home sleep environment disruptions. However, in general, point estimates were small and confidence intervals were wide (Table A4).

D. Discussion

This paper demonstrates the use of smartphone GPS tracking and EMA to measure spatially and temporally varying environmental features hypothesized to impact sleep among adolescents. These methods have been applied to other important adolescent health behaviors, in particular substance use,^27,28,40,41 but are thus far underutilized in the sleep research field. These methods were found to be feasible and acceptable to adolescents in this pilot study, with few participants reporting privacy concerns or feelings of high burden.

The availability of smartphone location tracking enables sleep researchers to move beyond the home neighborhood and capture exposures from all places where teens spend time. The findings of this study suggest that exposure to neighborhood physical and social conditions may differ between home and other settings. This is consistent with prior studies indicating that daily activity path-based exposures differ from home exposures for a variety of neighborhood environmental domains,^42,43 and that adolescents may spend significant amounts of time away from home.^20,41 This approach also enables examination of variability in exposure to features like crime or greenspace between teens who live within the same neighborhood. The predominant approach of focusing on the residential neighborhood assumes the same level of exposure for all individuals living in the same neighborhood – however, past research suggests a majority of the variation in certain exposures (e.g., neighborhood violent crime) occurs at the day and person-level rather than the neighborhood level.⁴⁴ There may also be value in measuring environmental exposures in specific non-home contexts, such schools. Feeling unsafe at school was associated with higher odds of insufficient sleep among a large sample of adolescents in Florida, and associations were larger when adolescents reported feeling unsafe both at school and in their neighborhood.¹² School environment data were not available for our study.

Smartphone GPS tracking enabled participants in this study to use their own devices, which might improve compliance, is more scalable than providing participants with separate research grade GPS tracking devices, and might avoid potential bias from introducing a new device.⁴⁵ However, several challenges inherent in this approach include ensuring that study apps work across operating systems and handling missing data due to signal loss, participants turning phones off, or interference from tall buildings. Continuous monitoring of data enables study teams to identify issues and troubleshoot with participants. In addition, statistical approaches such as imputation may be applied to handle missing GPS data.⁴⁵ While privacy is a particularly important consideration for GPS tracking, teens in this study expressed minimal concern about location tracking. However, concerns may vary across populations, and strong measures are needed to protect confidentiality of these potentially sensitive data, include encrypted data transfer protocols and certificates of confidentiality.⁴⁶

In this study, participant-reported home sleep environment problems were reported by approximately half of participants, but varied across days for many participants. As in studies with school-aged or adolescent populations that assessed sleep environment problems using a 1-week recall,^2,3,47 uncomfortable temperature was the most commonly cited sleep environment problem. Using EMA to assess home environment exposures may reduce recall bias for conditions that change nightly compared to traditional retrospective survey assessments.²³ In addition, daily measurement of the sleep environment can facilitate examination of associations at the day-level with sleep outcomes such as duration, timing, or quality and enables examination of both between and within-person variation. However, EMA has the potential to become burdensome to participants, so survey frequency must be carefully determined. The EMA completion rate in this study is comparable to other studies with adolescent populations^22,28,48; however, challenges were identified related to the timing of the morning survey, where some participants found it difficult to complete before school. EMA data on adolescents’ perceived home environment exposures could be augmented with external and objective data, such as estimates of nightly light pollution or home air quality and noise monitoring.

The current research was a small pilot study that primarily focused on assessing the feasibility and acceptability of GPS and EMA exposure assessments. It provides a foundation for future, larger scale work which will quantify associations of daily home and neighborhood exposures with between and within-person variation in adolescent sleep outcomes after adjustment for potential confounders. In the current analysis, we present bivariate associations between several environmental exposures and sleep outcome measures. For example, we found that greater exposure to neighborhood crime and litter were associated with earlier sleep offset timing, while neighborhood crime was associated with self-reported sleep problems.

Neighborhood crime and physical disorder are hypothesized to impact sleep through decreasing perceptions of personal safety, which may increase stress and hyperarousal, making it harder to fall asleep and stay asleep.⁸ In contrast to a prior study that assessed temporal associations of neighborhood violent crime with actigraphy-assessed sleep outcomes among adolescents,¹⁰ in our study neighborhood crime was not associated with later sleep onset timing. Our findings should be considered exploratory in nature and must be interpreted with caution given the small sample size of this pilot study. Point estimates in general were close to the null and confidence intervals were wide, which might be due to the small sample size, insufficient variability in exposures, unmeasured confounding, or true lack of association. Future work with larger and more diverse study populations is needed to confirm and build upon these findings.

This study has several limitations that should be noted. First, the small sample size limited power to detect smaller associations. We enrolled a convenience sample of adolescents that reflected a non-representative, relatively high SES population, which also may have underestimated challenges related to smartphone data availability and WIFI access. Second, data collection took place entirely during the COVID-19 pandemic, which may have impacted participants’ mobility and sleep. While the study took place after the earliest months of the pandemic when greater physical distancing restrictions were in place, adolescents were typically attending school remotely during data collection, which likely limited mobility. Even so, most participants did travel away from home during the data collection period, and greater variability in exposures is expected in future studies. Third, the analyses reported here are descriptive, and the methods described allow for identification of associations but not causal inference. Fourth, the 7-day monitoring period provided limited data to describe sleep patterns, EMA responses, and GPS-based neighborhood exposures by week and weekend days.

E. Conclusions

Results demonstrate the feasibility and acceptability of using mobile methods to assess time-varying home and neighborhood exposures relevant to adolescent sleep. These approaches should be considered by researchers investigating environmental determinants of sleep, given their ability to facilitate the capture of exposures across multiple contexts and variation across days.