This cohort study examines how delaying school start time is associated with objectively assessed sleep duration, timing, and quality among adolescents from public high schools in Minnesota.
Key Points
Question
How is a delay in high school start time associated with adolescent sleep?
Findings
In this cohort study of 455 high school students, those attending schools that shifted to later starts after baseline measurements (1) got approximately 43 minutes more objectively measured sleep on school nights, (2) slept less on weekends, and (3) had similar bedtimes 2 years after the start time delay, relative to students attending comparison schools that started early throughout the observation period.
Meaning
These findings suggest that delayed school start times may be a readily deployable sleep promotion intervention that can effectively allow adolescents greater opportunity for healthy sleep.
Abstract
Importance
Sleep is a resource that has been associated with health and well-being; however, sleep insufficiency is common among adolescents.
Objective
To examine how delaying school start time is associated with objectively assessed sleep duration, timing, and quality in a cohort of adolescents.
Design, Setting, and Participants
This observational cohort study took advantage of district-initiated modifications in the starting times of 5 public high schools in the metropolitan area of Minneapolis and St Paul, Minnesota. A total of 455 students were followed up from grade 9 (May 3 to June 3, 2016) through grade 11 (March 15 to May 21, 2018). Data were analyzed from February 1 to July 24, 2019.
Exposures
All 5 participating schools started early (7:30 am or 7:45 am) at baseline (2016). At follow-up 1 (2017) and continuing through follow-up 2 (2018), 2 schools delayed their start times by 50 and 65 minutes, whereas 3 comparison schools started at 7:30 am throughout the observation period.
Main Outcomes and Measures
Wrist actigraphy was used to derive indices of sleep duration, timing, and quality. With a difference-in-difference design, linear mixed-effects models were used to estimate differences in changes in sleep time between delayed-start and comparison schools.
Results
A total of 455 students were included in the analysis (among those identifying sex, 225 girls [49.5%] and 219 boys [48.1%]; mean [SD] age at baseline, 15.2 [0.3] years). Relative to the change observed in the comparison schools, students who attended delayed-start schools had an additional mean 41 (95% CI, 25-57) objectively measured minutes of night sleep at follow-up 1 and 43 (95% CI, 25-61) at follow-up 2. Delayed start times were not associated with falling asleep later on school nights at follow-ups, and students attending these schools had a mean difference-in-differences change in weekend night sleep of −24 (95% CI, −51 to 2) minutes from baseline to follow-up 1 and −34 (95% CI, −65 to −3) minutes from baseline to follow-up 2, relative to comparison school participants. Differences in differences for school night sleep onset, weekend sleep onset latency, sleep midpoints, sleep efficiency, and the sleep fragmentation index between the 2 conditions were minimal.
Conclusions and Relevance
This study found that delaying high school start times could extend adolescent school night sleep duration and lessen their need for catch-up sleep on weekends. These findings suggest that later start times could be a durable strategy for addressing population-wide adolescent sleep deficits.
Introduction
Although the National Sleep Foundation recommends 8 to 10 hours of sleep per night for adolescents,1 more than half of US adolescents aged 16 years regularly get less than 7 hours.2 Sleep debt, the gap between biological sleep needs and a lesser amount of obtained sleep, has consequences that extend far beyond hampering day-to-day functioning. Sleep debt may be deleterious to multiple areas of physical and mental health, including being associated with risk of type 2 diabetes, hypertension, and injury throughout the life course.3
Adolescent circadian biology clashes with typical high school schedules in the United States, all but ensuring that most students will not be able to get even an adequate amount of sleep on school nights. At the start of adolescence, a neurobiological change to children’s circadian systems moves the release of nocturnal melatonin approximately 2 hours later relative to childhood timing, postponing adolescents’ sleep-wake cycles4 and making it difficult for adolescents to fall asleep before 11 pm or wake before 8 am.5 In addition, “sleep drive,” which accumulates as waking hours pass and acts as an opposing force to wakefulness in sleep/wake homeostasis, tends to build up more slowly at the onset of adolescence,6 making the expectation of sleep onset before 11 pm formidable.
Unfortunately, high schools in the United States tend to start quite early, typically earlier than primary schools, leaving adolescents with their current sleep bottleneck. In response to the discordance between adolescent biology and school day structure, the American Academy of Pediatrics released a policy statement in August 2014 recommending that high schools start at 8:30 am or later,7 and since then, multiple other public health organizations have issued similar recommendations.8 Despite this, fewer than 15% of US high schools start at 8:30 am or later, and 42% start very early, at 8 am or before.9,10 At these very early start times, adolescent sleep hours appear to be most severely curtailed.11
In reports comparing students attending high schools that had start times differing by approximately 30 to 90 minutes, adolescents attending later-start schools self-report greater sleep duration.12,13,14,15,16 Furthermore, evidence from these studies suggest that moving school start times 1 hour later has a meaningful association with varied distal behaviors and outcomes, such as improved driving safety,17,18 academic performance,19 mental health,12,13 and school attendance.20 However, this literature has been limited for determining whether delayed starts actually cause increased sleep owing to the studies’ design21 and has been primarily built around observational and cross-sectional research in which sleep has been measured subjectively. Lufi et al22 have the only report of a randomized study of the association of delayed high school start times, which was conducted in Israel. In this study, 1 class was randomly selected to start at 8:30 am for 2 weeks, and these students were compared with a class whose start time remained at 7:30 am (n = 47); sleep was measured objectively with wrist actigraphy.22 To our knowledge, no studies to date of any design have measured sleep objectively at multiple points in a cohort of adolescents over multiple years during a period when start times shifted.
In the present study, we aimed to determine whether a shift to a later start time was associated with objectively measured adolescent sleep duration, quality, and timing during an extended follow-up period. We were able to take advantage of a natural experiment in start time modification and compare students in schools that delayed their start times with students in contemporaneous comparison schools, a design that could give evidence of the causal effects of school start time on sleep. We hypothesized that relative to the cohort of participants attending comparison schools that started early throughout our study, the cohort of students in schools that adopted delayed start times after baseline (hereinafter referred to as delayed-start schools) would have longer periods of sleep on school nights. In addition, we expected that weekend night sleep duration would be shorter in delayed-start schools than in comparison schools, owing to these students having less sleep debt to pay back.
Methods
Study Design Overview and Setting
The START cohort study followed a cohort of students (n = 2426) in 5 suburban and rural high schools in the Minneapolis and St Paul, Minnesota, area, starting in the 2015-2016 academic year with annual measurements each spring (baseline in 9th grade, follow-up 1 in 10th grade, and follow-up 2 in 11th grade) through the 2017-2018 academic year. For this analysis, we used data from a subsample of students randomly selected and recruited to have their sleep measured objectively (START substudy participants [n = 455]). The University of Minnesota’s institutional review board and participating school districts approved procedures for the START study. All study participants provided written assent, and parents or guardians provided written informed consent. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.
School Start Time Change
At baseline, all schools started at either 7:30 or 7:45 am. Beginning at follow-up 1, the 2 delayed-start schools initiated a preplanned district-initiated delay in their start times, by 50 and 65 minutes, and continued these delayed start times through follow-up 2. Three comparison schools started at 7:30 am throughout the study period.
Participant Population and Substudy Recruitment
The START study sent annual information letters to the parents and guardians of all class of 2019 students in participating schools. Students who were not opted out by parents or guardians were invited to complete an in-school START survey that took approximately 15 minutes (90.3% response proportion at baseline [n = 2134] for survey). Proportional to school size, we randomly selected START participants to invite to participate in the START substudy at baseline. If an adolescent’s parents consented to their participation, the adolescent was given an opportunity to assent to participate in the research. A total incentive of $85 was offered to substudy participants for completion of several additional data collection tasks, including sleep actigraphy. Additional students were recruited to the substudy at follow-up 1 from the same stratified random invitation list generated at baseline. Students recruited at baseline or follow-up 1 were invited back for a final week of actigraphy at follow-up 2 (see eFigure in the Supplement for participant flow). Our goal, based on a power calculation that would have given us 0.87 power to detect a 15-minute difference in differences by follow-up 2, was to recruit and retain 160 participants per condition.
Data Collection
Substudy participants received the actigraph (wGT3X-BT Monitor; ActiGraph Corp) and a sleep log from May 3 to June 3, 2016, at baseline; March 28 to June 7, 2017, at follow-up 1; and March 15 to May 21, 2018, at follow-up 2. Wrist actigraphy uses an unobtrusive device similar to a wristwatch that collects objective data on movement, light, time spent sleeping vs awake, and sleeping patterns and has been shown to correlate well with polysomnography,23 which is the criterion standard of objective sleep measurement. Participants were instructed to wear the device on their nondominant wrist from the time they were given the device until the day it was returned to study staff, removing it only during contact sports or if they were going to be submerged in water. Participants entered information on time in and out of bed and napping into a sleep log. After a period of at least 7 days, study staff returned to the school to collect materials and distribute substudy participation incentives.
Before distribution to participants, the devices were initialized with 60-second epochs using the software program ActiLife, version 6 (ActiGraph Corp). Sleep logs and actigraph data were transmitted to the Brigham Sleep Reading Center, Boston, Massachusetts, for processing. Nights were scored as valid if it appeared the device was worn during the sleep period. Actigraphic data were scored using ActiLife, version 6.13, analysis software (ActiGraph Corp) using the validated Cole-Kripke algorithm.24 Using activity counts and light (ie, lux) data from the devices and sleep diary entries, a dedicated scorer masked to all other data annotated the start and end of the main sleep period.
Substudy participants were observed for 3662 school night (nights that preceded a school day) sleep periods and 1409 nonschool night (henceforth referred to as weekend) sleep periods across the observation period. Before variable creation, we removed all weekday sleep periods when nonstandard school schedules were followed, such as weather-related school cancellations or holidays. Wednesdays were also excluded from the sample because one comparison school started instruction 30 minutes later each Wednesday for staff professional development. After these exclusions, 2682 school night sleep periods had available data (73.2% of initial weeknight observations).
Measures
Outcomes
We examined several outcomes that characterized the duration and timing of the main sleep (not naps) period that preceded school days (school night sleep) and weekend days (weekend night sleep). To characterize sleep timing, we report the school night and weekend night sleep onset and wake-up (or offset) times, which bookend the main sleep. We calculated school night and weekend night sleep duration as the minutes between the recorded main sleep onset and wake-up for school and weekend nights.
To assess the quality of school night sleep, we examined several factors. Sleep midpoint was the clock time midway between sleep onset and final awakening. Sleep efficiency was the percentage of the main sleep period estimated to be spent asleep. Sleep fragmentation index was the proportion of the sleep period characterized by movement (based on movement counts and bouts of immobility). Net duration of school night sleep was calculated as the difference between sleep onset and wake-up minus any awake time (wakefulness after sleep onset) that occurred during this period.
School night sleep onset latency was the number of minutes from “lights off” to sleep onset. This variable combines information from the wrist device with the sleep log. We also compared the difference between school night and weekend night sleep midpoints to determine the degree to which sleep timing differed on these types of days (or “social jet lag”). Values for all outcomes were obtained by calculating the mean across each valid night for each individual.
Exposure
The exposure of interest is delay of school start time. At baseline, all START schools started early on regular school days (4 schools began at 7:30 am and 1 at 7:45 am). At follow-up 1, one school shifted its start time from 7:30 am to 8:20 am (a 50-minute delay relative to baseline) and another from 7:45 am to 8:50 am (a 65-minute delay). These 2 schools were pooled into the delayed-start condition. The 3 comparison schools that remained at a 7:30 am start time throughout the study period were our comparison group.
Confounders
Several demographic variables that had been hypothesized to be potentially associated with sleep and also differ by condition were reported by participants in START sleep surveys and matched to sleep actigraphy data by participant identification. We used the following demographic characteristics reported in the earliest available survey (self-report) for each participant: biological sex (male, female, or other), free and reduced-price lunch eligibility (yes, no, or do not know), Hispanic ethnicity (yes or no), race (Native American; Asian, Hawaiian, or Pacific Islander; black; white; multiple; or unknown or not reported), and each parent’s or guardian’s educational attainment (did not finish high school, finished high school or General Educational Development, some college, finished college, advanced degree, do not know, or do not have a second parent or guardian). Because day length shifted during the substudy, we also adjusted for the number of hours of daylight during the week the participant wore the actigraph.
Missing Data
Objectively measured sleep data were not available for each substudy participant at each year, owing to attrition and the recruitment of additional students to the substudy at follow-up 1. However, when objectively measured sleep duration was missing, self-reported sleep duration was often available as a proxy measure from the START survey. To estimate the effect of the missing data, we compared self-reported weeknight sleep duration for substudy participants with and without missing observations at each time point. We determined that sleep data missing owing to drop-in and drop-out appear to follow a missing-at-random pattern. We used maximum likelihood estimation in mixed-effects models because maximum likelihood can provide unbiased and asymptotically efficient estimates of outcomes when missing data are judged to be missing at random.25
Statistical Analysis
Data were analyzed from February 1 to July 24, 2019. The main effects of interest are the condition-by-time interaction terms at follow-up 1 and follow-up 2. These represent the difference in differences of sleep characteristics between students in delayed-start and comparison schools. When possible, the effects used in the mixed-effects models included a random intercept for the school and for the participant within the school as well as a participant-level random effect of time. The random school-level intercept was excluded if it caused a nonpositive definite G matrix. In these cases, school fixed effects were added to the model. Observations from 451 of the 455 START actigraphy participants were included in the analyses. The 4 excluded participants were missing demographic covariates. Of 455 participants, 230 completed a single year, 132 completed 2 years, and 93 completed 3 years of actigraphy. Two-tailed P < .05 indicated significance.
Data cleaning, univariate evaluation, and statistical analyses were completed in SAS, version 9.4 (SAS Institute, Inc) by the primary analyst (A.B.). A second analyst performed a code review to verify reported results.
Results
Sample at Baseline
A total of 455 participants were included in the analysis (among those who identified by sex, 225 girls [49.5%] and 219 boys [48.1%]; mean [SD] age at baseline, 15.2 [0.3] years). Participants at delayed-start schools were significantly more likely than those at comparison schools to report a parent having completed college (191 [89.3%] vs 179 [74.3%]) and less likely to identify as white (172 [80.4%] vs 224 [92.9%]) (Table 1). Compared with the whole START cohort, substudy participants were more likely to be white (1914 of 2422 [79.0%] vs 396 of 455 [87.0%]) and have at least 1 parent with a college degree (1771 of 2422 [73.1%] vs 370 of 455 [81.3%]) and were less likely to report eligibility for free or reduced-price meals (337 of 2422 [13.9%] vs 30 of 455 [6.6%]) (eTable 1 in the Supplement). We compared self-reported sleep duration at baseline (from the START sleep survey) between participants who entered the substudy at each time point and those who were observed at each time point regardless of when they entered the substudy. There were no significant differences for condition by wave of entry or condition by wave of observation in self-reported sleep at baseline (eTable 2 in the Supplement).
Table 1. Baseline Self-reported Demographic Characteristics of START Actigraphy Substudy Participants by Condition.
Characteristic | Study group, No. (%)a | P value | |
---|---|---|---|
Delayed-start schools (n = 214) | Comparison schools (n = 241) | ||
Biological sex | |||
Female | 108 (50.5) | 117 (48.5) | .80 |
Male | 101 (47.2) | 118 (49.0) | |
Prefer not to answer | 4 (1.9) | 3 (1.2) | |
Hispanic or Latino ethnicity | 7 (3.3) | 6 (2.5) | .63 |
Race | |||
Native American | 2 (0.9) | 1 (0.4) | <.001 |
Asian | 16 (7.5) | 4 (1.7) | |
Black | 7 (3.3) | 0 | |
White | 172 (80.4) | 224 (92.9) | |
Multiracial | 12 (5.6) | 9 (3.7) | |
Unknown or not reported | 4 (1.9) | 0 | |
Free or reduced-price lunch eligible | |||
No | 171 (79.9) | 172 (71.4) | .08 |
Yes | 14 (6.5) | 16 (6.6) | |
Do not know | 28 (13.1) | 50 (20.7) | |
≥1 Parent completed college | 191 (89.3) | 179 (74.3) | <.001 |
Percentages may not total 100 owing to missing responses. Data are from the START study, spring 2016 through spring 2018.
Association of School Start Time Delay With Sleep Characteristics
Compared with students in comparison schools, the difference-in-differences regression estimate showed that students in delayed-start schools had a mean of 41 (95% CI, 25-57) increased minutes of school night sleep time relative to the change from baseline in comparison schools at follow-up 1 and 43 (95% CI, 25-61) more minutes on school nights at follow-up 2 after 2 years of delayed start times (Table 2). Figure 1 shows how the distribution of school night sleep duration shifted right (longer) in delayed-start schools while simultaneously shifting left (shorter) in comparison schools, and weekend night sleep time correspondingly decreased in delayed-start schools. Students attending delayed-start schools had a mean difference-in-differences change in weekend night sleep of −24 (95% CI, −51 to 2) minutes from baseline to follow-up 1 and −34 (95% CI, −65 to −3) minutes from baseline to follow-up 2, relative to comparison schools. Figure 2 illustrates the changes in mean measured hours of school night sleep time for students in each condition at all points, with delayed-start school students getting a mean of 8 hours and 5 minutes of sleep on school nights at follow-up 2. Relative to the change observed in comparison schools, students in delayed-start schools had a shift to later wake-ups concluding their school night sleep after the policy change at follow-ups 1 (difference in differences, 41 [95% CI, 32-51] minutes) and 2 (difference in differences, 58 [95% CI, 46-69] minutes). Although students in both conditions had later sleep onset on school nights in 10th and 11th grades compared with 9th grade (baseline), changes in sleep onset in delayed-start schools were not substantially different from those in comparison schools, at 2 (95% CI, −13 to 16) and 16 (95% CI, −1 to 33) mean minutes later in follow-ups 1 and 2, respectively.
Table 2. Adjusted Difference-in-Differences Analysis of Changes in Sleep Duration and Timing Among High School Students Before and After a 50- and 65-Minute Delay in School Start Timesa.
Variable | Delayed-start schools | Comparison schools | Difference-in-differences analysis | |||||||
---|---|---|---|---|---|---|---|---|---|---|
Baseline (n = 122) | Follow-up 1 (n = 141) | Follow-up 2 (n = 90) | Baseline (n = 160) | Follow-up 1 (n = 162) | Follow-up 2 (n = 94) | Baseline to follow-up 1, min | P Value | Baseline to follow-up 2, min | P Value | |
School night sleep onset timeb | 22:28 (22:14 to 22:41) pm | 22:55 (22:44 to 23:06) pm | 23:09 (22:57 to 23:20) pm | 22:15 (22:06 to 22:25) pm | 22:41 (22:30 to 22:52) pm | 22:41 (22:26 to 22:55) pm | 2 (−13 to 16) | .82 | 16 (−1 to 33) | .06 |
School night wake-up timec | 6:06 (5:54 to 6:17) am | 7:01 (6:50 to 7:12) am | 7:16 (7:04 to 7:27) am | 6:14 (6:04 to 6:24) am | 6:28 (6:19 to 6:38) am | 6:26 (6:15 to 6:38) am | 41 (32 to 51) | <.001 | 58 (46 to 69) | <.001 |
School night sleep duration, h:minc | 7:38 (7:24 to 7:52) | 8:05 (7:53 to 8:17) | 8:05 (7:51 to 8:19) | 7:59 (7:47 to 8:10) | 7:45 (7:33 to 7:56) | 7:42 (7:27 to 7:57) | 41 (25 to 57) | <.001 | 43 (25 to 61) | <.001 |
Weekend night sleep onset timeb | 23:24 (23:02 to 23:45) pm | 23:50 (23:33 pm to 0:07 am) pm | 0:13 (23:54 pm to 0:31 am) am | 23:39 (23:25 to 23:53) pm | 0:08 (23:51 pm to 0:25 am) am | 23:50 (23:27 pm to 0:14 am) pm | −2 (−27 to 23) | .87 | 38 (10 to 67) | .009 |
Weekend night wake-up timec | 8:51 (8:33 to 9:09) am | 9:05 (8:49 to 9:20) am | 9:17 (9:00 to 9:35) am | 8:44 (8:30 to 8:58) am | 9:25 (9:10 to 9:40) am | 9:06 (8:45 to 9:26) am | −27 (−53 to −1) | .04 | 5 (−24 to 34) | .75 |
Weekend night sleep duration, h:minc | 9:23 (9:04 to 9:52) | 9:11 (8:54 to 9:27) | 9:00 (8:40 to 9:20) | 9:07 (8:52 to 9:22) | 9:19 (9:03 to 9:35) | 9:18 (8:55 to 9:40) | −24 (−51 to 2) | .07 | −34 (−65 to −3) | .03 |
Data are from the START study, spring 2016 through spring 2018. Baseline occurred in spring 2016 (9th grade); follow-up 1, spring 2017 (10th grade); and follow-up 2, spring 2018 (11th grade). Unless otherwise indicated, data are expressed as mean (95% CI).
Linear mixed-effects models adjusted for demographic factors and length of daylight, with school fixed effect in place of school random effect.
Linear mixed-effects models adjusted for biological sex, socioeconomic status (free or reduced lunch eligibility and parent educational attainment), race and ethnicity, and length of daylight at actigraphy, school- and student-level random effects.
Table 3 shows that relative to students at comparison schools, weekday-weekend midpoint difference, sleep efficiency, and sleep onset latency changes did not differ meaningfully at delayed-start schools by follow-up 2. Later-starting students experienced small but significant increases in sleep fragmentation compared with students in comparison schools, with a mean difference in differences of 2.2 (95% CI, 0-4.5) and 4.8 (95% CI, 2.2-7.5) on the sleep fragmentation index. Still, when wake-ups after sleep onset were subtracted from school night sleep duration, net sleep duration remained increased in the delayed-start schools relative to the comparison schools at both follow-ups, with students in the delayed-start schools netting a mean 31 (95% CI, 16-46) additional minutes at follow-up 2 (Table 3).
Table 3. Adjusted Difference-in-Differences Analysis of Changes in Sleep Variability and Continuity Among High School Students Before and After a 50- and 65-Minute Delay in School Start Timesa.
Sleep variability and continuity | Delayed-start schools | Comparison schools | Difference-in-differences analysis | |||||||
---|---|---|---|---|---|---|---|---|---|---|
Baseline (n = 122) | Follow-up 1 (n = 141) | Follow-up 2 (n = 90) | Baseline (n = 160) | Follow-up 1 (n = 162) | Follow-up 2 (n = 94) | Baseline to follow-up 1 | P Value | Baseline to follow-up 2 | P Value | |
Weekday-weekend midpoint difference, h:minb | 1:51 (1:34 to 2:08) | 1:30 (1:17 to 1:43) | 1:34 (1:20 to 1:49) | 1:57 (1:46 to 2:08) | 2:13 (2:00 to 2:27) | 1:54 (1:35 to 2:13) | −37 (−58 to −17) min | <.001 | −14 (−37 to 9) min | .22 |
School night sleep efficiency, %b | 85.9 (84.8 to 87.0) | 85.7 (84.8 to 86.6) | 84.6 (83.7 to 85.6) | 85.4 (84.6 to 86.2) | 84.8 (83.9 to 85.7) | 85.5 (84.3 to 86.7) | 0.4 (−0.9 to 1.7) | .53 | −1.4 (−2.8 to 0.0) | .054 |
School night SFIb,c | 30.1 (28.2 to 32.1) | 31.1 (29.6 to 32.7) | 32.8 (31.1 to 34.5) | 30.0 (28.7 to 31.3) | 28.8 (27.3 to 30.3) | 27.9 (25.7 to 30.0) | 2.2 (0.0 to 4.5) | .053 | 4.8 (2.2 to 7.5) | <.001 |
School night sleep onset latency, minb,d | 8 (6 to 10) | 11 (9 to 12) | 12 (11 to 14) | 9 (8 to 10) | 12 (11 to 13) | 12 (10 to 14) | 0 (−2 to 2) | .91 | 1 (−1 to 4) | .22 |
Net school night sleep duration (subtracting out wake-ups after sleep onset), h:mine | 6:40 (6:28 to 6:52) | 7:04 (6:54 to 7:14) | 6:59 (6:48 to 7:11) | 6:57 (6:47 to 7:07) | 6:45 (6:35 to 6:54) | 6:46 (6:33 to 6:58) | 37 (23 to 50) min | <.001 | 31 (16 to 46) min | <.001 |
Abbreviation: SFI, sleep fragmentation index.
Data are from the START study, spring 2016 through spring 2018. Baseline occurred in spring 2016 (9th grade); follow-up 1, spring 2017 (10th grade); and follow-up 2, spring 2018 (11th grade). Unless otherwise indicated, data are expressed as mean (95% CI).
Linear mixed-effects models adjusted for demographic factors and length of daylight, with school fixed effect in place of school random effect.
Calculated as the proportion of the sleep period characterized by movement.
Indicates number of minutes from “lights off” to sleep onset.
Linear mixed-effects models adjusted for biological sex, socioeconomic status (free or reduced lunch eligibility and parent educational attainment), race and ethnicity, and length of daylight at actigraphy, school- and student-level random effects.
Discussion
In schools that shifted their start time later, objective measurement shows that students slept longer on school nights at both 1 and 2 years after the change, relative to students in schools that maintained an early start time throughout the observation period. It was notable that the magnitude of initial sleep increase (>40 minutes) at follow-up 1 had not deteriorated at follow-up 2, and relative to comparison school students, sleep onset times had not become substantially delayed after the participants had experienced the later start time for nearly 2 academic years. Furthermore, concurrent with the increase in school night sleep duration, those attending delayed-start schools had a decrease in weekend night sleep duration, suggesting lesser accumulated sleep debt with the later start times. Sleep fragmentation, sleep efficiency, and sleep latency onset appeared to be minimally affected or unaffected by the shift to a later start time.
Despite the consistency of findings in the literature of high school start time delay and its association with longer sleep,12,13,14,17,18,26,27,28,29 no previous studies have been performed of sufficient quality to conclude that later start times cause students to get more sleep and that this effect can be sustained.21,30,31 Most of the research in this area has been cross-sectional, comparing high schools with different start times but at 1 point in time. The few longitudinal studies available used before-and-after schedule change designs without any contemporaneous comparison schools that continued to start early. Without this type of comparison group, which would offer a counterfactual to what would have happened if the schools that delayed start times had not delayed, it is difficult to infer what effect the policy change might have had, as opposed to other factors. Furthermore, few previous studies of high school start times14,22,28 used objective measures of sleep, and those that did had limited or no follow-up and/or small samples. Our use of wrist actigraphy not only allowed us to measure sleep duration and timing with greater validity, but it also enabled the measurement of sleep quality.
We were interested in whether there might have been a change in the relative timing of weekend sleep midpoint vs school night sleep midpoint for participants who attended delayed-start schools because weekends have fewer schedule constraints and are more likely to be days that adolescents can follow their natural circadian sleep timing. We hypothesized that students who attended later-start schools would have a lesser differential or less social jet lag, reflecting that their new weekday schedule had set them up to be in more alignment with their biologically set sleep timing on days without school. Social jet lag has been associated with circadian misalignment and increased risk for metabolic disease32,33 and dysfunction in regulation of reward processing,34,35 which can manifest as increased sensation seeking and diminished regulatory control.36,37 However, we did not see a sustained closing of the school night–weekend sleep midpoint difference gap.
Even in delayed-start schools, which compared favorably to the comparison schools as far as students’ sleep duration, the mean hours of total sleep time (8 hours and 5 minutes) barely meets the 8 to 10 hours per night recommended by the National Sleep Foundation for adolescents aged 14 to 17 years1 and certainly falls short of the roughly 9 hours per night that other authors38,39,40,41 have deemed to be an optimal amount of sleep for adolescents. Furthermore, given that this is a mean, a sizable group of students are below the 8-hour minimum total sleep time on school nights. This finding suggests that even later start times, perhaps in conjunction with other sleep-friendly policies and interventions, should be considered and tested in an effort to create a culture of sleep in which teens are able to meet their biological sleep needs. Investigation of additional structural changes schools could make to become even more sleep friendly is needed. Possibilities include policies limiting the hours that activities could end in the evening or start in the morning, guidelines for the time of day when electronically submitted assignments are due, and limitations on caffeinated beverage sales on campuses.
Limitations
This natural experiment study’s design has many protections against confounding or against the possibility that changes in sleep duration, timing, and/or duration attributed to delayed start times are actually due to other factors. However, there was no randomization to condition, so it remains possible that a factor correlated with both a school district’s decision to change start times and student sleep is actually responsible for the results seen. An additional limitation is START’s limited racial/ethnic diversity; it is possible that this limits its generalizability. However, previous research on delaying start times12,13,17,18,27 has demonstrated associations with outcomes for adolescents from various racial/ethnic, economic, and geographic backgrounds.
Conclusions
Complex pediatric public health challenges are posed by the various outcomes (driving safety,17,18 academic performance,19 mental health,12,13 school attendance,20 overweight/obesity42) that sleep duration appears to predicate. We believe that an important area for future research is to determine how school start times may relate to health and academic outcomes. The present study’s results suggest that later start times could be a durable and transformative strategy for dealing with the epidemic sleep insufficiency among adolescents.
References
- 1.Hirshkowitz M, Whiton K, Albert SM, et al. . National Sleep Foundation’s sleep time duration recommendations: methodology and results summary. Sleep Health. 2015;1(1):40-43. doi: 10.1016/j.sleh.2014.12.010 [DOI] [PubMed] [Google Scholar]
- 2.Keyes KM, Maslowsky J, Hamilton A, Schulenberg J. The great sleep recession: changes in sleep duration among US adolescents, 1991-2012. Pediatrics. 2015;135(3):460-468. doi: 10.1542/peds.2014-2707 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Institute of Medicine (US) Committee on Sleep Medicine and Research In: Colten HR, Altevogt BM, eds. Sleep Disorders and Sleep Deprivation: An Unmet Public Health Problem. National Academies Press; 2006. Accessed January 6, 2014. https://www.ncbi.nlm.nih.gov/books/NBK19960/ [PubMed] [Google Scholar]
- 4.Frey S, Balu S, Greusing S, Rothen N, Cajochen C. Consequences of the timing of menarche on female adolescent sleep phase preference. PLoS One. 2009;4(4):e5217. doi: 10.1371/journal.pone.0005217 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Carskadon MA, Acebo C, Jenni OG. Regulation of adolescent sleep: implications for behavior. Ann N Y Acad Sci. 2004;1021:276-291. doi: 10.1196/annals.1308.032 [DOI] [PubMed] [Google Scholar]
- 6.Taylor DJ, Jenni OG, Acebo C, Carskadon MA. Sleep tendency during extended wakefulness: insights into adolescent sleep regulation and behavior. J Sleep Res. 2005;14(3):239-244. doi: 10.1111/j.1365-2869.2005.00467.x [DOI] [PubMed] [Google Scholar]
- 7.Adolescent Sleep Working Group, Committee on Adolescence and Council on School Health. School start times for adolescents. Pediatrics. 2014;134(3):642-649. doi: 10.1542/peds.2014-1697 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Start School Later. Position statements and resolutions on sleep and school start times. Accessed March 21, 2020. https://www.startschoollater.net/position-statements.html
- 9.National Center for Education Statistics. Schools and Staffing Survey (SASS) Accessed March 21, 2020. https://nces.ed.gov/surveys/sass/tables/sass1112_201381_s1n.asp
- 10.Wheaton AG, Ferro GA, Croft JB. School start times for middle school and high school students—United States, 2011-12 school year. MMWR Morb Mortal Wkly Rep. 2015;64(30):809-813. doi: 10.15585/mmwr.mm6430a1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Paksarian D, Rudolph KE, He J-P, Merikangas KR. School start time and adolescent sleep patterns: results from the US National Comorbidity Survey–Adolescent Supplement. Am J Public Health. 2015;105(7):1351-1357. doi: 10.2105/AJPH.2015.302619 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Boergers J, Gable CJ, Owens JA. Later school start time is associated with improved sleep and daytime functioning in adolescents. J Dev Behav Pediatr. 2014;35(1):11-17. doi: 10.1097/DBP.0000000000000018 [DOI] [PubMed] [Google Scholar]
- 13.Owens JA, Belon K, Moss P. Impact of delaying school start time on adolescent sleep, mood, and behavior. Arch Pediatr Adolesc Med. 2010;164(7):608-614. doi: 10.1001/archpediatrics.2010.96 [DOI] [PubMed] [Google Scholar]
- 14.Nahmod NG, Lee S, Buxton OM, Chang A-M, Hale L. High school start times after 8:30 am are associated with later wake times and longer time in bed among teens in a national urban cohort study. Sleep Health. 2017;3(6):444-450. doi: 10.1016/j.sleh.2017.09.004 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Bowers JM, Moyer A. Effects of school start time on students’ sleep duration, daytime sleepiness, and attendance: a meta-analysis. Sleep Health. 2017;3(6):423-431. doi: 10.1016/j.sleh.2017.08.004 [DOI] [PubMed] [Google Scholar]
- 16.Owens JA, Dearth-Wesley T, Herman AN, Oakes JM, Whitaker RC. A quasi-experimental study of the impact of school start time changes on adolescent sleep. Sleep Health. 2017;3(6):437-443. doi: 10.1016/j.sleh.2017.09.001 [DOI] [PubMed] [Google Scholar]
- 17.Danner F, Phillips B. Adolescent sleep, school start times, and teen motor vehicle crashes. J Clin Sleep Med. 2008;4(6):533-535. doi: 10.5664/jcsm.27345 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Wahlstrom K, Dretzke B, Gordon M, Peterson K, Edwards K, Gdula J Examining the impact of later high school start times on the health and academic performance of high school students: a multi-site study. Issued February 2014. Accessed March 16, 2015. https://conservancy.umn.edu/handle/11299/162769
- 19.Carrell SE, Maghakian T, West JE. A’s from zzzz’s? the causal effect of school start time on the academic achievement of adolescents. Am Econ J Econ Policy. 2011;3(3):62-81. doi: 10.1257/pol.3.3.62 [DOI] [Google Scholar]
- 20.Wahistrom K. Changing times: findings from the first longitudinal study of later high school start times. NASSP Bull. 2002;86(633):3-21. doi: 10.1177/019263650208663302 [DOI] [Google Scholar]
- 21.Marx R, Tanner-Smith EE, Davison CM, et al. . Later school start times for supporting the education, health, and well-being of high school students. Cochrane Database Syst Rev. 2017;7:CD009467. doi: 10.1002/14651858.CD009467.pub2 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Lufi D, Tzischinsky O, Hadar S. Delaying school starting time by one hour: some effects on attention levels in adolescents. J Clin Sleep Med. 2011;7(2):137-143. doi: 10.5664/jcsm.28100 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Ancoli-Israel S, Cole R, Alessi C, Chambers M, Moorcroft W, Pollak CP. The role of actigraphy in the study of sleep and circadian rhythms. Sleep. 2003;26(3):342-392. doi: 10.1093/sleep/26.3.342 [DOI] [PubMed] [Google Scholar]
- 24.Cole RJ, Kripke DF, Gruen W, Mullaney DJ, Gillin JC. Automatic sleep/wake identification from wrist activity. Sleep. 1992;15(5):461-469. doi: 10.1093/sleep/15.5.461 [DOI] [PubMed] [Google Scholar]
- 25.Allison PD. Paper 312-2012: Handling missing data by maximum likelihood. 2012. Accessed March 21, 2020. https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&cad=rja&uact=8&ved=2ahUKEwj5lYLg5qvoAhWCZM0KHRg6DkoQFjAAegQIAxAC&url=http%3A%2F%2Fwww.statisticalhorizons.com%2Fwp-content%2Fuploads%2FMissingDataByML.pdf&usg=AOvVaw3Kp48dSCXz64zGtDWvbXpD [Google Scholar]
- 26.O’Malley EB, O’Malley MB. School start time and its impact on learning and behavior In: Ivanenko A, ed. Sleep and Psychiatric Disorders in Children and Adolescents. Informa Healthcare; 2008:79-94. [Google Scholar]
- 27.Wahlstrom KL. Accommodating the sleep patterns of adolescents within current educational structures: an uncharted path In: Carskadon MA, ed. Adolescent Sleep Patterns. Cambridge University Press; 2002. doi: 10.1017/CBO9780511499999.014. [DOI] [Google Scholar]
- 28.Dunster GP, de la Iglesia L, Ben-Hamo M, et al. . Sleepmore in Seattle: later school start times are associated with more sleep and better performance in high school students. Sci Adv. 2018;4(12):eaau6200. doi: 10.1126/sciadv.aau6200 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Patte KA, Qian W, Cole AG, et al. . School start time changes in the COMPASS study: associations with youth sleep duration, physical activity, and screen time. Sleep Med. 2019;56:16-22. doi: 10.1016/j.sleep.2018.09.020 [DOI] [PubMed] [Google Scholar]
- 30.Wheaton AG, Chapman DP, Croft JB. School start times, sleep, behavioral, health, and academic outcomes: a review of the literature. J Sch Health. 2016;86(5):363-381. doi: 10.1111/josh.12388 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Minges KE, Redeker NS. Delayed school start times and adolescent sleep: a systematic review of the experimental evidence. Sleep Med Rev. 2016;28:86-95. doi: 10.1016/j.smrv.2015.06.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Kantermann T, Juda M, Merrow M, Roenneberg T. The human circadian clock’s seasonal adjustment is disrupted by daylight saving time. Curr Biol. 2007;17(22):1996-2000. doi: 10.1016/j.cub.2007.10.025 [DOI] [PubMed] [Google Scholar]
- 33.Malone SK, Zemel B, Compher C, et al. . Social jet lag, chronotype and body mass index in 14-17-year-old adolescents. Chronobiol Int. 2016;33(9):1255-1266. doi: 10.1080/07420528.2016.1196697 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Hasler BP, Smith LJ, Cousins JC, Bootzin RR. Circadian rhythms, sleep, and substance abuse. Sleep Med Rev. 2012;16(1):67-81. doi: 10.1016/j.smrv.2011.03.004 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Hasler BP, Clark DB. Circadian misalignment, reward-related brain function, and adolescent alcohol involvement. Alcohol Clin Exp Res. 2013;37(4):558-565. doi: 10.1111/acer.12003 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Edwards S, Reeves GM, Fishbein D. Integrative model of the relationship between sleep problems and risk for youth substance use. Curr Addict Rep. 2015;2(2):130-140. doi: 10.1007/s40429-015-0052-0 [DOI] [Google Scholar]
- 37.Hasler BP, Soehner AM, Clark DB. Sleep and circadian contributions to adolescent alcohol use disorder. Alcohol. 2015;49(4):377-387. doi: 10.1016/j.alcohol.2014.06.010 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Eaton DK, McKnight-Eily LR, Lowry R, Perry GS, Presley-Cantrell L, Croft JB. Prevalence of insufficient, borderline, and optimal hours of sleep among high school students—United States, 2007. J Adolesc Health. 2010;46(4):399-401. doi: 10.1016/j.jadohealth.2009.10.011 [DOI] [PubMed] [Google Scholar]
- 39.Perlus JG, O’Brien F, Haynie DL, Simons-Morton BG. Adolescent sleep insufficiency one year after high school. J Adolesc. 2018;68:165-170. doi: 10.1016/j.adolescence.2018.07.016 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Short MA, Weber N, Reynolds C, Coussens S, Carskadon MA. Estimating adolescent sleep need using dose-response modeling. Sleep. 2018;41(4). doi: 10.1093/sleep/zsy011 [DOI] [PubMed] [Google Scholar]
- 41.National Heart, Lung, and Blood Insititute Your Guide to Healthy Sleep. US Department of Health and Human Services, National Cancer Institute, National Institutes of Health; 2011. Accessed September 16, 2019. https://pediatrics.aappublications.org/content/pediatrics/129/3/548.full.pdf
- 42.Gupta NK, Mueller WH, Chan W, Meininger JC. Is obesity associated with poor sleep quality in adolescents? Am J Hum Biol. 2002;14(6):762-768. doi: 10.1002/ajhb.10093 [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.