Abstract
Younger adults have a biological disposition to sleep and wake at later times that conflict with early morning obligations like work and school; this conflict leads to inadequate sleep duration and a difference in sleep timing between school days and weekends. The COVID-19 pandemic forced universities and workplaces to shut down in person attendance and implement remote learning and meetings that decreased/removed commute times and gave students more flexibility with their sleep timing. To determine the impact of remote learning on the daily sleep–wake cycle we conducted a natural experiment using wrist actimetry monitors to compare activity patterns and light exposure in three cohorts of students: pre-shutdown in-person learning (2019), during-shutdown remote learning (2020), and post-shutdown in-person learning (2021). Our results show that during-shutdown the difference between school day and weekend sleep onset, duration, and midsleep timing was diminished. For instance, midsleep during school days pre-shutdown occurred 50 min later on weekends (5:14 ± 12 min) than school days (4:24 ± 14 min) but it did not differ under COVID restrictions. Additionally, we found that while the interindividual variance in sleep parameters increased under COVID restrictions the intraindividual variance did not change, indicating that the schedule flexibility did not cause more irregular sleep patterns. In line with our sleep timing results, school day vs. weekend differences in the timing of light exposure present pre- and post-shutdown were absent under COVID restrictions. Our results provide further evidence that increased freedom in class scheduling allows university students to better and consistently align sleep behavior between school days and weekends.
Keywords: sleep, COVID shutdown, university students, circadian rhythms, social jetlag
Graphical abstract
Graphical Abstract.
Statement of Significance.
We show that under remote learning conditions that resulted from the COVID-19 pandemic, university students had better alignment of sleep timing between school days and weekend as well as longer duration of sleep during school days.
Introduction
Adolescents have a natural inclination to go to bed later and wake up later in the day than adults and children [1, 2]. There are several proposed mechanisms to explain the late sleep timing in adolescents, with the most prevalent being modifications to the two-process model of sleep regulation [3]. This model presents the interaction between the homeostatic and circadian regulation processes and predicts sleep–wake timing and duration. Puberty leads to developmental changes in both processes, which presumably persist throughout adolescence and young adulthood. First, adolescents can have a longer circadian period [4, 5], which combined with their lower sensitivity to morning light and higher sensitivity to evening light [6], would result on an overall later circadian timing. Second, adolescents have altered homeostatic regulation with decreased accumulation of sleep pressure during wake periods, so they are able to stay awake longer [7, 8]. Third, adolescents and young adults have more autonomy in self-selecting their bedtimes and that combined with evening social and school-work activities contributes to later bedtimes. Although the exact mechanisms behind the delayed adolescent chronotype are still a matter of debate [9], it is clear that adolescents, extending to college students, display a delayed circadian phase and late timing of sleep [10–14].
The late chronotype—or preference to sleep and wake up at a certain time—of college students leads to later bedtimes while their social responsibilities, like school and work start times, force earlier wake-up times than necessary for an adequate sleep duration. Furthermore, the misalignment between their biological and social timing leads to social jetlag (SJL)—the difference in sleep timing between school days and weekends [15, 16]. SJL is prevalent in university students, whose sleep onset is delayed and sleep duration is lengthened on free vs. school days [10, 11, 13, 14, 17–19]. This timing difference has been confirmed in adults by dim light melatonin onset (DLMO), which showed late and moderate chronotypes have a delayed circadian phase on free days compared to workdays [20, 21]. SJL is also present in middle and high school students, and is clearly a consequence of early school times, as it is reduced or abolished by later school start times [22, 23]. Increased SJL is associated with mental and physical health problems as well as reduced academic performance [10, 24, 25], supporting the belief that aligning social responsibilities with the adolescent chronotype should represent a goal to improve health and performance of university students.
During 2020–2021, the COVID-19 pandemic led to many changes in society including the implementation of remote learning. Remote learning with both synchronous and asynchronously held classes removed the school commute time and allowed students to have more flexible schedules, changing sleep timing [17]. However, studies on the pandemic’s effect on sleep have generated a heterogenous collection of results. While many studies showed improved sleep in the form of reduced SJL, increased sleep duration, and decreased sleep timing variability [26–31], others revealed increased sleep disturbance [32–35]. We analyzed sleep in US university students during the first months of the pandemic and found that the COVID-19 shutdown resulted in later sleep timing on school days and weekends, increased sleep duration in most participants, and reduced SJL [17].
However, our previous study, as well as others addressing sleep changes associated with the COVID-19 shutdown, are limited as they are based on sleep diaries or retrospective questionnaires, most of which are less accurate in assessing the times of sleep onset and offset than wrist actimetry. Although Ong et al. used ring-measured actimetry, this study was not conducted specifically with adolescents and young adults [28]. Furthermore, wrist actimeters can also record light exposure, which could change under social isolation and affect sleep timing. To determine changes more objectively in sleep timing in university students, we measured actimetry-based sleep in a cohort of students taking a class during the fall of 2020 (during-shutdown remote learning), under strict COVID social isolation, and compared it to two other cohorts of students taking the same class and within the same season in 2019 (pre-shutdown), before the pandemic, and in 2021 (post-shutdown) when the class and remaining courses had returned to a predominantly in-person format.
Methods
Participants
We recorded data from 60 college students at the University of Washington (UW, Seattle, WA) in three separate years. The students were enrolled in the same fall quarter course (late September-early December) led by the same instructor. Activity and light exposure were recorded from 20 students in the pre-shutdown cohort (age range 21 to 26 [mean age = 23.5; 92% self-identified women 38% Caucasian, 15% Asian, 15% Hispanic, 46% other/no ethnicity identified]), 20 students in the during-shutdown cohort (age range 20–41 [mean age = 23.4; 95% self-identified women; 40% Caucasian, 25% Asian, 15% Hispanic, 5% Black, and 15% other/no ethnicity identified]), and 20 students in the post-shutdown cohort (age range 20–35 [mean age = 22.9; 90% self-identified women; 60% Caucasian, 10% Asian, 5% Hispanic, 5% Black, and 20% other/no ethnicity identified]). Pre-shutdown (Fall 2019), 100% of students were attending campus every day and taking all their UW classes in person. During-shutdown (Fall 2020), 10% of students were attending campus every day, 65% were going to campus some days but not every day, and 25% were not going to campus at all (the course from which students in this study were enrolled was fully remote, but required students to come to campus once a week for actiwatch data downloads). Post-shutdown (Fall 2021) 40% of students were going to campus every day, 55% were going to campus some days but not every day, and 5% were not going to campus at all (the course from which students in this study were enrolled was hybrid, with an in-person lab and remote option for lectures). A chi-squared analysis of the proportion of students in each category showed a statistically significant difference between years (X2 = 36.05; p = 28.25 × 10−8). Sleep recordings were done as part of a laboratory practice that included other self-collected biological data, which students received participation credit for recording and analyzing. At the end of the class, students signed an informed consent form for their anonymized data to be used in this or other research studies.
All studies were approved by the Human Subjects Division at UW (IRB Study number 00010431).
Data collection
We recorded each participant’s wrist activity and light exposure for 6 weeks pre-shutdown, 8 weeks during-shutdown, and 4 weeks post-shutdown using Actiwatch Spectrum Plus (Phillips, Respironics, Bend, OR), with loggers programmed to collect data in 1-min epochs. Students also submitted daily sleep logs through online forms asking their bedtime, wake time, nap time (if taken), meal (lunch and dinner) times, work/school times (if attended).
We obtained sunset and sunrise data from NASA’s Jet Propulsion Laboratory HORIZONS Web-Interface (https://ssd.jpl.nasa.gov/horizons.cgi) with a 1-min precision. Plotted sunset and sunrise times represent the average of sunset and sunrise times for all days of recordings.
All preparation, analysis (including statistical analysis), and plotting of data were performed using R studio version 2022.12.0 + 353, unless otherwise indicated. A p-value of <0.05 was considered statistically significant.
Inclusion criteria
Individual actograms were inspected with Philips Actiware (version 6.0.9) software to check for watch malfunction and that watches were worn an appropriate number of days during recording. Data were separated into school (Sunday 1200—Friday 1200) and weekend (Friday 1200—Sunday 1200 and the night before Veteran’s Day) nights. During the recording period, students went through the transition of daylight savings time (DST) to standard time (ST); because this transition day was recorded in all three years it was not excluded from the analysis.
In final statistical analyses, students pre- and during-shutdown who missed recordings from more than 14 school days or 7 weekend days were not included. Students post-shutdown missing more than 7 school days, or 3 weekend days were not included in the analysis. Students who were night-shift workers or not residing in Seattle at the time of data collection were also excluded from the analysis. This led to 1 student excluded pre-COVID for missing too many days of recording, 4 students excluded during-COVID (2 students were not in Seattle and 2 students missed too many days of recording), and 6 students excluded post-COVID for missing too many days of recording.
Sleep parameters
Actiwatch recordings were downloaded and exported using the Philips Actiware software. The software estimates sleep parameters including onset, offset, and duration from activity measurements.
Before data analysis, we used sleep duration measurements to detect and disregard outliers. We used the median absolute deviation (MAD) method setting a threshold of 3 MADs, using the Routliers package for R [36, 37]. Twenty-two values determined as outliers were removed from the analysis for any sleep parameter.
Statistical analysis.
Each participant’s daily sleep events (sleep onset, sleep offset, duration, midsleep) were averaged for school days and weekends separately, leading to two values per parameter per participant. Data were analyzed using a linear mixed effects model (LMEM) using the lme4 package [38]. QQ plots for every LMEM fit were first analyzed to check normality. The model was then run on each sleep parameter separately considering condition (pre-, during-, or post-shutdown) and day type (school day or weekend) as factors and each participant as a random factor. Tukey post hoc comparisons were calculated to determine differences in the sleep parameters between years and day types. Gender was not used as a factor in any of our analysis as overwhelming majorities (≥90%) of participants each year self-identified as women.
To assess differences in interindividual variance of sleep parameters between COVID restrictions we used a Bartlett’s test to conduct pairwise comparisons between years for school days and weekends separately, with an α of 0.025 to account for multiple comparisons.
To assess differences in intraindividual variance of sleep parameters between years we calculated the standard deviation of each sleep parameter across all days recorded for each participant irrespective of whether they corresponded to school days or weekends. Then, a one-way ANOVA or Kruskal–Wallis test was used to compare intraindividual variances between years for groups that respectively had homogenous variances or not.
Light exposure
All light exposure data was analyzed using the Actiwatch white-light lux reading. The first and last daily exposure times to a specific light intensity were determined using the 1-minute epoch raw light data for each individual. We analyzed the mean time of the first and last exposure to a 50-lux light intensity for each student separately for school days and weekends. Importantly, 50 lux is not necessarily a circadian threshold for non-visual responses [39, 40], but instead it was chosen as an arbitrary value that would likely be above the threshold for circadian photic stimulation for most of the young adult participants (similar results were seen if 5 and 100 lux were set as the threshold value).
Statistical analysis.
Differences in light exposure times between years and type of day were analyzed using a similar LMEM that was used for sleep parameters. QQ plots were inspected, then the model with COVID condition and day type as factors, and participant as a random factor, was run on first and last exposure separately.
Waveform analysis
To determine differences in overall activity and light exposure, raw 1-minute activity and light data for each student was binned into means of 10-min intervals. After 10-min light data were log-transformed, both activity and light were smoothed by a one-hour running average. The resulting data were used to generate individual 24-hour activity and light waveforms for each student and day type. Individual waveforms were in turn used to generate mean waveforms, leading to six light and six activity waveforms (2 school day types × 3 years).
Statistical analysis.
To test whether the 24-h patterns of activity and light exposure differed between school days and weekends, we used, for each condition (pre-, during-, or post-shutdown) a two-way repeated measures ANOVA with day type and clock time as within-participant factors. In cases in which day type (school day vs. weekend) or the interaction was statistically significant, and after visual inspection of the waveforms, we compared the school day vs. weekend activity (or light exposure) during the morning (5:00–11:00) or evening (18:00–2:00). For this comparison, we added the cumulative activity counts (or lux intensity) for each student within each interval and used paired Student t-test to compare school vs. weekend to test the prediction that activity and light exposure would be delayed on weekends relative to school days.
Time windows for morning and evening analysis were chosen based on visual inspection and the slope of the waveforms. The evening time window was longer than the morning due to the use of electricity in the evening, which decreases the slope of descending activity and light exposure.
Analysis of school day or weekend waveforms between COVID conditions was done using a two-way mixed ANOVA with year as a between-participants factor and time as a within-participants factor. Only the interaction for the light analysis yielded a significant effect, and the morning and evening light exposures within the same intervals defined above were compared between COVID conditions with a one-way ANOVA.
Results
COVID stay-at-home restrictions and sleep timing between school days and weekends
Our analysis of actigraphic sleep indicated that during the COVID stay-at-home restrictions students displayed a better alignment between their school day and weekend sleep timing than during pre- and post-shutdown conditions when classes were fully or mostly in-person. Figure 1 displays the timing of sleep parameters under different COVID restrictions. We used an LMEM to find associations between changes in sleep and COVID restrictions and Tukey comparisons for post hoc comparisons. Table 1 displays results from the type III ANOVA analysis.
Figure 1.
Sleep patterns pre-, during- and post-shutdown restrictions. Bar plots of the average sleep duration, onset, and offset. Diamond shapes within the bar plots represent the midpoint of sleep. Sleep plots per year (pre-shutdown [2019], during-shutdown [2020], and post-shutdown [2021]) are displayed separately for school days and weekends. Error bars represent the standard deviation. n = 19 (pre-), n = 16 (during-), n = 14 (post-). Symbols represent the result of post hoc Tukey comparisons between school days and weekend: +++P < 0.001, difference in midsleep; ###P < 0.001, difference in sleep duration; *p < 0.05; ***p < 0.001, differences in sleep onset or offset. Note that the lengths of the bars do not necessarily reflect the duration indicated in hours and minutes because the average duration is not calculated as the difference between average onset and average offset. See Table 1 for linear model statistical results.
Table 1.
Results for type III analysis of variance for each sleep parameters
| Fixed effect | F-Value | P-value | |
|---|---|---|---|
| Sleep onset | COVID condition | F (2,46) = 0.16 | 0.8553 |
| Day | F (1,46) = 29.37 | 2.1 × 10 −6 | |
| Interaction | F (2,46) = 1.41 | 0.2555 | |
| Sleep offset | COVID condition | F (2,46) = 0.63 | 0.5371 |
| Day | F (1,46) = 42.00 | 5.5 × 10 −8 | |
| Interaction | F (2,46) = 6.36 | 3.6 × 10 −3 | |
| Midsleep | COVID condition | F (2,46) = 0.55 | 0.5800 |
| Day | F (1,46) = 22.04 | 2.4 × 10 −5 | |
| Interaction | F (2,46) = 1.65 | 0.2032 | |
| Sleep duration | COVID condition | F (2,46) = 0.68 | 0.5126 |
| Day | F (1,46) = 14.89 | 3.5 × 10 −4 | |
| Interaction | F (2,46) = 3.62 | 3.46 × 10 −2 |
Sleep onset showed an effect of the type of day (Table 1), but no effect of COVID restriction or the interaction. Pre-shutdown students fell asleep 27 min later on weekends than school days (00:45 ± 53 min and 1:12 ± 58 min, p = 0.049). During-shutdown students fell asleep 49 min later on weekends than school days (00:20 ± 1 h 25 min and 1:09 ± 2 h 2 min, p ≤ 1 × 10−4), but post-shutdown this difference was not significant (00:44 ± 1 h 26 min and 1:10 ± 1 h 29 min).
Sleep offset showed an effect of the type of day and of the interaction, but no effect of COVID restriction. Students woke up 74 min later on weekends than school days pre-shutdown (8:04 ± 1 h 11 min & 9:18 ± 1 h 8 min, p = <1 × 10−4) and the weekend delay in wake time was of 48 min post-shutdown (8:14 ± 1 h 24 min and 9:02 ± 1 h 18 min, p = 2.09 × 10−3). In contrast, during-shutdown students woke up at similar times on school days and weekends (9:05 ± 2 h 26 min and 9:20 ± 2 h 7 min, p = 0.7663).
Midsleep was delayed on weekends compared to school days, but showed no difference between COVID restrictions or an interaction. Pre-shutdown weekend midsleep was 50 min later than school days (4:24 ± 59 min and 5:14 ± 54 min, p ≤ 1 × 10−4). In contrast, there was no discrepancy between midsleep times either during-shutdown (5:10 ± 2 h 40 min and 5:30 ± 2 h 15 min, p = 0.451) or post-shutdown (4:35 ± 1 h 18 min and 5:03 ± 1 h 23 min, p = 0.170).
Students’ sleep duration showed a significant effect of type of day and of the interaction but not of COVID restrictions. Pre-shutdown, students slept 50 minutes longer on weekends (7 h 19 min ± 42 min and 8 h 08 min ± 1 h 9 min, p ≤ 1 × 10−4). In contrast, during-shutdown (8 h 01 min ± 1 h 17 min and 8 h 06 min ± 1 h 25 min, p = 0.998) and post-shutdown (7 h 30 min ± 33 min and 7 h 57 min ± 35 min, p = 0.251) students displayed no difference in their sleep duration. Visual inspection of the data (Fig. 1) suggested longer school day sleep duration during- than pre-shutdown; a student t-test between these two groups yielded a significant difference pointing a bout of sleep that was on average 42 min longer during-shutdown (t = 2.035; p = 0.049).
SJL did not differ under different COVID restrictions (one-way Kruskal–Wallis ANOVA p = 0.1649) although it showed the same trend revealed by our LMEM analysis of midsleep (50 min ± 34 min pre-shutdown; 20 min ± 1 h 11 min during-shutdown; 29 min ± 33 min post-shutdown, Fig. S1).
Our waveform analysis further supported the general conclusion of increased alignment in sleep parameters between school days and weekends during peak COVID (2020) restrictions. There was an effect of time of day on activity irrespective of COVID condition (Fig. 2A, Table S1). There was a statistically significant interaction between time of day and type of day for activity pre-, during-, and post-shutdown. Morning and evening comparisons of activity revealed that pre- and post-shutdown, students were active later during mornings and evenings of weekends than those of school days (Fig. 2A, Table S2). In contrast, during COVID the timing of activity was only later in the evening but not in the morning on weekends (Fig. 2A). The analysis of variance of activity waveforms between COVID restrictions only revealed an effect of time of day both for school days and weekends (Fig. 2B, Table S3).
Figure 2.
Activity waveforms compared between (A) days and (B) COVID restrictions. Dots represent average activity level of students for each time point, split either between days or COVID restrictions, and SEM for error bars on points. Grey shaded areas represent morning and evening time windows in which cumulative activity was compared between groups. *p < 0.05, **p < 0.01, paired t-test analysis.
COVID stay-at-home restrictions and interindividual and intraindividual variability in sleep timing and duration
To determine the degree to which COVID restrictions affected sleep timing choices, we analyzed interindividual variability (Fig. 3) in sleep timing between COVID restrictions through a Bartlett’s test. On school days (Table 2), sleep timing was more variable between participants during-shutdown than pre-shutdown according to all sleep parameters except sleep onset. The increased variance in sleep duration and midsleep during COVID restrictions decreased post-shutdown.
Figure 3.
Violin plots of sleep onset, sleep offset, midsleep, and sleep duration pre, during, and post-shutdown restrictions. Large points in the middle of the plots represent the overall average for each sleep parameter. Smaller points in the middle of the plots represent individual student values, each color corresponding to a different student. Asterisks represent the results of Bartletts tests: *p < 0.025. See Tables 2 and 3 for Bartletts test statistical results.
Table 2.
Interindividual variability in school day sleep parameters under different restrictions
| Sleep Onset | Pre-shutdown | During-shutdown | Post-shutdown | Sleep Offset | Pre-shutdown | During-shutdown | Post-shutdown |
|---|---|---|---|---|---|---|---|
| Pre-shutdown | Pre-shutdown | ||||||
| During-shutdown | −0.532 | During-shutdown | −1.253 | ||||
| (0.0635) | (4.69 × 10 −3 ) | ||||||
| Post-shutdown | −0.536 | −0.004 | Post-shutdown | 0.21 | 1.042 | ||
| (0.0703) | (0.9922) | (0.5277) | (0.0479) | ||||
| Midsleep | Pre-shutdown | During-shutdown | Post-shutdown | Sleep Duration | Pre-shutdown | During-shutdown | Post-shutdown |
| Pre-shutdown | Pre-shutdown | ||||||
| During-shutdown | −1.687 | During-shutdown | −0.577 | ||||
| (1.42 × 10 −4 ) | (1.94 × 10 −2 ) | ||||||
| Post-shutdown | −0.326 | 1.361 | Post-shutdown | 0.159 | 0.736 | ||
| (0.2695) | (1.33 × 10 −2 ) | (0.3419) | (3.81 × 10 −3 ) |
Top values represent the effect size for school day sleep parameters measured as the standard deviation difference between compared years. Negative values indicate that the variability (standard deviation) was lower on the column header than the row header and vice versa. Values in brackets represent Bartlett’s test p-values (boldface indicates statistical significance). For example, school day sleep offset pre-shutdown was less variable than during-shutdown.
During weekends (Table 3), variability between students increased during-shutdown compared to pre-shutdown for all parameters except sleep duration. Sleep duration variability was lower post-shutdown compared to both pre- and during-shutdown conditions.
Table 3.
Interindividual variability in weekend sleep parameters under different restrictions
| Sleep Onset | Pre-shutdown | During-shutdown | Post-shutdown | Sleep Offset | Pre-shutdown | During-shutdown | Post-shutdown |
|---|---|---|---|---|---|---|---|
| Pre-shutdown | Pre-shutdown | ||||||
| During-shutdown | −1.075 | During-shutdown | −0.987 | ||||
| (3.52 × 10 −3 ) | 1.38 × 10 −2 | ||||||
| Post-shutdown | −0.523 | 0.553 | Post-shutdown | −0.166 | 0.821 | ||
| (0.0955) | (0.2543) | (0.5999) | (0.08152) | ||||
| Midsleep | Pre-shutdown | During-shutdown | Post-shutdown | Sleep Duration | Pre-shutdown | During-shutdown | Post-shutdown |
| Pre-shutdown | Pre-shutdown | ||||||
| During-shutdown | −1.357 | During-shutdown | −0.266 | ||||
| (4.13 × 10 −4 ) | (0.4062) | ||||||
| Post-shutdown | −0.483 | 0.874 | Post-shutdown | 0.566 | 0.832 | ||
| (0.0977) | −0.0813 | (1.56 × 10 −2 ) | (2.62 × 10 −3 ) |
Effect sizes and p-values are presented as in Table 2.
As an estimate of the intraindividual variability in sleep timing, we calculated the standard deviation of the mean of each sleep parameter for each individual throughout all recorded days. Comparison of the intraindividual variance did not show differences between COVID restrictions (Fig. 4, Table 4). This result indicates that although students differed more from each other in their sleep timing under the COVID stay-at-home mandate the social isolation did not make the sleep of each individual more variable from day to day. Of note, visual inspection of Fig. 3 reveals there were two students with highly variable sleep patterns during-shutdown.
Figure 4.
Interindividual variation of sleep onset, sleep offset, midsleep, and duration. Each point represents individual student standard deviation (school day and weekend values combined) for each sleep parameter. See Table 4 for Kruskal–Wallis and ANOVA statistical results.
Table 4.
Statistical analyses (Kruskal–Wallis or One-Way ANOVA) for intraindividual variability of sleep parameters
| Fixed effect | Df | F-value | P-value | |
|---|---|---|---|---|
| Sleep Onset (Kruskal–Wallis) |
Year | 2 | 1.87 | 0.166 |
| Sleep Offset (Kruskal–Wallis) |
Year | 2 | 0.972 | 0.386 |
| Midsleep (Kruskal–Wallis) |
Year | 2 | 2.034 | 0.142 |
| Duration (ANOVA) |
Year | 2 | 1.15 | 0.326 |
COVID stay-at-home restrictions and differences in light exposure between school days and weekends
To determine the time course of light exposure during the day we measured students’ first time and last time of exposure to light above 50-lux (Fig. 5). The 50-lux threshold was chosen as an arbitrary threshold for circadian responses, but similar results were obtained with higher (500 lux) or lower (5 lux) thresholds (data not shown). The LMEM for first exposure to 50-lux light showed an effect of type of day (F = 18.76, p = 7.96 × 10−5) and the interaction between type of day and COVID restrictions (F = 5.07, p = 1.03 × 10−2). Pre-shutdown, students were exposed to a 50-lux light intensity one hour and 5 min earlier on school days (9:24 ± 15 min) than weekends (10:29 ± 12 min; p ≤ 1 × 10−4). The time of the first exposure to 50-lux light did not differ between school days (9:22 ± 23 min) and weekends (9:23 ± 24 min; p = 1.0) during-shutdown restrictions, but a 50-min difference reappeared post-shutdown (9:49 ± 1 h 27 min on school days, 10:39 ± 1 h 56 min on weekends; p = 2.05 × 10−2).
Figure 5.
Light exposure before and during COVID restrictions. Bar plots of the average time of first exposure to 50 lux, last exposure to 50-lux, and light exposure duration. Years are split (top = pre-shutdown [2019], middle = during-shutdown [2020], bottom = post-shutdown [2021]) with school days and weekends split within each year. n = 19 (pre), n = 16 (during), n = 14 (post). Error bars represent the standard deviation.
The last exposure to 50-lux light showed no significance effect for any factors or the interaction. On the other hand, the daily duration of exposure to at least 50-lux light showed a significant effect of type of day and of the interaction. Pre-shutdown, students were exposed to 50 lux light intensity for about one hour longer on school days than weekends (12 h 59 min ± 1 h 51 min and 11 h 54 min ± 1 h 50 min, p = 4.09 × 10−3). In contrast, exposure duration during-shutdown (12 h 23 min ± 42 min on school days, 12 h 38 min ± 48 min on weekends; p = 0.9620) and post-shutdown (11 h 24 min ± 26 min on school days and 10 h 45 min ± 37 min on weekends; p = 0.3563) did not differ between school days and weekends.
Our waveform analysis of light exposure further confirmed more consistent timing of light exposure between school days and weekends under COVID restrictions (Fig. 6A). The day of the week did not influence light exposure except under pre-shutdown conditions when students were exposed to lower levels of light during the weekend than during school days (Fig. 6A, Table S1). Interestingly, the comparison between school day and weekend waveforms revealed a difference pre- and post-shutdown conditions but not during-shutdown. Pre- and post-shutdown, students were exposed to brighter light during school days than during weekends, both during the morning and evening (Fig. 6A, Table S2). These results indicate that the 24-h profile of light exposure was similar between school days and weekends only for students under strict COVID restrictions.
Figure 6.
Light waveforms compared between (A) days and (B) COVID restrictions. Dots represent average illuminance level of students for each time point, split either between days or COVID restrictions, and SEM for error bars on points. Grey shaded areas represent morning and evening time windows in which cumulative light exposure was compared between groups. *p < 0.05, ***p < 0.001, paired t-test analysis. +p < 0.05 Tukey between pre- and during-shutdown conditions.
The waveform analysis comparing different COVID restrictions revealed the expected effect of time of day, and no effect of COVID restrictions except for a trend for exposure to brighter light intensity during school days pre-shutdown compared to the other two conditions (Fig. 6B, Table S3). There was an effect of the interaction between time of day and COVID restrictions both during school days and weekends. Further analysis showed that during the weekend mornings, students were exposed to higher light intensities pre- than during-shutdown (F(2,46) = 4.39, p = 0.0179; p = 0.0192 post hoc Tukey comparison).
Discussion
In this natural experiment, we observed increased consistency between school day and weekend sleep timing among university students in Seattle, WA learning under remote and hybrid conditions during the 2020 and 2021 fall quarters of the COVID-19 pandemic, as compared to students observed in the fall of 2019. Students pre-shutdown showed differences in timing of all sleep parameters between school days and weekends, while only sleep onset and offset differed during- and post-shutdown, respectively (Fig. 1). These findings are consistent with the notion that school closures during the COVID-19 shutdown ended the need to commute to campus and allowed university students more freedom to choose their bed and wakeup times. Although some studies reported poorer sleep quality in students during the COVID shutdown [32–35], our actimetry results are consistent with our previously published results, which showed decreased SJL and increased sleep duration in university students during the shutdown [17] through self-reported surveys. This latter result was supported in the current study when the school day sleep duration during the COVID shutdown was compared to pre-shutdown conditions.
Increased SJL is associated with physical and mental health problems [24, 25]. In students, this difference in sleep timing between school days and weekends likely reflects the fact that during school days they are pressed to sleep out of phase with their circadian clock to meet their class schedule, a mismatch that is magnified by the delaying effect that the weekend has on circadian phase [21]. Our finding that the interindividual variability in sleep times increased under the COVID-19 shutdown supports this idea, suggesting that more sleep timing freedom for students led them to consistently sleep at a preferred time, which is likely determined by circadian phase and differs between students. Importantly, the intraindividual variance in sleep timing did not increase during the shutdown, indicating that the increased freedom for bedtime choices did not necessarily result in more erratic sleep patterns, and supporting previous findings indicating that young adults present a stable circadian phase [12, 41]. Together these findings suggest that sleep among teenager and young adult university students may benefit from later school start times or flexible schedules that allow them to choose specific times of instruction and reduce commute time. Alternatively, if earlier bedtimes and waketimes are desired, then adopting and maintaining a consistent wake-sleep/light-dark schedule with exposure to bright light in the morning and dim light in the evening in the home will facilitate earlier sleep times.
Observations of students’ light exposures during school days vs. weekends agreed with those of sleep timing results. Pre- and post-shutdown, students showed delayed weekend light exposure in the morning compared to school days, while this wasn’t the case during-shutdown (Fig. 5). This further suggests the COVID-19 shutdown minimized early wakings for classes on school days, so students were able to expose themselves to light with the same timing as on weekends. Because light exposure timing affects sleep–wake timing [42] and a single weekend of changed light exposure is sufficient to change circadian phase [21, 43] the delayed light exposure on weekends pre- and post-shutdown likely exacerbated students’ SJL by delaying their sleep timing further. The impact of reduced morning light exposure during the weekend mornings may be particularly detrimental during the short photoperiod and overcast skies of Seattle during the fall quarter [19].
Some studies have found that during the COVID-19 school closures, students reported living more sedentary lifestyles with less interactions [44, 45]. When comparing the daily activity levels of pre-, during, and post-shutdown (Fig. 2B), we surprisingly found no difference on school days or weekends. Similarly, daily light exposure levels between the COVID conditions did not differ, except for a trend to brighter light exposure on school days pre-shutdown, which can be explained by the fact that students were forced to attend all classes in person.
Some limitations in our study should be noted. First, each condition was measured in different cohorts of students across each year of observation, though students were observed during the same fall course. Our pre-shutdown recordings were concluded before the COVID-19 pandemic and the pandemic could not have been predicted at the time. Second, we did not include gender as a factor in our LMEM, as a majority of participants in all cohorts identified as women, and we had no statistical power to detect gender differences. Third, although SJL has been correlated with reduced academic performance [10] this was not analyzed in our study because expectations and evaluations of course assessments were handled differently during remote learning. Finally, our post-shutdown condition did not represent a return to the pre-shutdown conditions, which may explain why some of the sleep parameters did not return to the values recorded in 2019.
In summary, we found that the COVID-19 shutdown led to better consistency in the phase of sleep and light exposure between school days and weekends, and a longer duration of sleep during school days. As these changes are associated with better health and behavioral outcomes such as decreased obesity, better academic performance, and less depression, our findings support the importance of alignment between circadian/sleep and social timing in adolescents and young adults.
Supplementary Material
Acknowledgments
We thank the University of Washington students who have graciously consented for their data to be used for research.
Contributor Information
Alicia Rice, Department of Biology, University of Washington, Seattle, WA, USA.
Olivia Sather, Department of Biology, University of Washington, Seattle, WA, USA.
Kenneth P Wright, Jr., Sleep and Chronobiology Laboratory, Department of Integrative Physiology, University of Colorado Boulder, Boulder, CO, USA
Céline Vetter, Circadian and Sleep Epidemiology Laboratory, Department of Integrative Physiology, University of Colorado Boulder, Boulder, CO, USA.
Melanie A Martin, Department of Anthropology, University of Washington, Seattle, WA, USA.
Horacio O de la Iglesia, Department of Biology, University of Washington, Seattle, WA, USA.
Funding
This work was supported by US National Institutes of Health Research Grant R01 HL16231 to H.O.D., the National Science Foundation Grant # 1743364 to H.O.D., the University of Washington Center for Studies in Demography and Ecology (NICHD P2C HD042828) to M.A.M., and the University of Washington Department of Biology. A.R. was supported by the Riddiford-Truman Award.
Disclosure Statement
K.P.W. reports research support/donated materials: DuPont Nutrition & Biosciences, Grain Processing Corporation, and Friesland Campina Innovation Centre and being a consultant to and/or receiving personal fees from Circadian Therapeutics, Inc., Circadian Biotherapies, Inc., Philips, Inc, and U.S. Army Medical Research and Materiel Command - Walter Reed Army Institute of Research, outside the submitted work.
Data Availability Statement
Research data will be available upon request to the authors.
References
- 1. Roenneberg T, et al. A marker for the end of adolescence. Curr Biol. 2004;14(24):R1038–R1039. doi: 10.1016/j.cub.2004.11.039. [DOI] [PubMed] [Google Scholar]
- 2. Crowley SJ, et al. An update on adolescent sleep: new evidence informing the perfect storm model. J Adolesc. 2018;67:55–65. doi: 10.1016/j.adolescence.2018.06.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Borbely AA, et al. The two-process model of sleep regulation: a reappraisal. J Sleep Res. 2016;25(2):131–143. doi: 10.1111/jsr.12371. [DOI] [PubMed] [Google Scholar]
- 4. Carskadon MA, et al. 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]
- 5. Carskadon MA, et al. Intrinsic circadian period of adolescent humans measured in conditions of forced desynchrony. Neurosci Lett. 1999;260(2):129–132. doi: 10.1016/s0304-3940(98)00971-9. [DOI] [PubMed] [Google Scholar]
- 6. Crowley SJ, et al. Increased sensitivity of the circadian system to light in early/mid-puberty. J Clin Endocrinol Metab. 2015;100(11):4067–4073. doi: 10.1210/jc.2015-2775. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Jenni OG, et al. Homeostatic sleep regulation in adolescents. Sleep. 2005;28(11):1446–1454. doi: 10.1093/sleep/28.11.1446. [DOI] [PubMed] [Google Scholar]
- 8. Taylor DJ, et al. 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]
- 9. Crowley SJ, et al. Free-running circadian period in adolescents and adults. J Sleep Res. 2018;27(5):e12678. doi: 10.1111/jsr.12678. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Smarr BL, et al. 3.4 million real-world learning management system logins reveal the majority of students experience social jet lag correlated with decreased performance. Sci Rep. 2018;8(1):4793. doi: 10.1038/s41598-018-23044-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Smarr BL. Digital sleep logs reveal potential impacts of modern temporal structure on class performance in different chronotypes. J Biol Rhythms. 2015;30(1):61–67. doi: 10.1177/0748730414565665. [DOI] [PubMed] [Google Scholar]
- 12. McHill AW, et al. Robust stability of melatonin circadian phase, sleep metrics, and chronotype across months in young adults living in real-world settings. J Pineal Res. 2021;70(3):e12720. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Friborg O, et al. Associations between seasonal variations in day length (photoperiod), sleep timing, sleep quality and mood: a comparison between Ghana (5 degrees) and Norway (69 degrees). J Sleep Res. 2012;21(2):176–184. doi: 10.1111/j.1365-2869.2011.00982.x. [DOI] [PubMed] [Google Scholar]
- 14. Shochat T, et al. Functional consequences of inadequate sleep in adolescents: a systematic review. Sleep Med Rev. 2014;18(1):75–87. doi: 10.1016/j.smrv.2013.03.005. [DOI] [PubMed] [Google Scholar]
- 15. Wittmann M, et al. Social jetlag: misalignment of biological and social time. Chronobiol Int. 2006;23(1-2):497–509. doi: 10.1080/07420520500545979. [DOI] [PubMed] [Google Scholar]
- 16. Roenneberg T, et al. Life between clocks: daily temporal patterns of human chronotypes. J Biol Rhythms. 2003;18(1):80–90. doi: 10.1177/0748730402239679. [DOI] [PubMed] [Google Scholar]
- 17. Wright KP, Jr, et al. Sleep in university students prior to and during COVID-19 Stay-at-Home orders. Curr Biol. 2020;30(14):R797–R798. doi: 10.1016/j.cub.2020.06.022. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. McHill AW, et al. Stability of the timing of food intake at daily and monthly timescales in young adults. Sci Rep. 2020;10(1):20849. doi: 10.1038/s41598-020-77851-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Dunster GP, et al. Daytime light exposure is a strong predictor of seasonal variation in sleep and circadian timing of university students. J Pineal Res. 2022;74(2). [DOI] [PubMed] [Google Scholar]
- 20. Zerbini G, et al. Weekly, seasonal, and chronotype-dependent variation of dim-light melatonin onset. J Pineal Res. 2021;70(3):e12723. doi: 10.1111/jpi.12723. [DOI] [PubMed] [Google Scholar]
- 21. Stothard ER, et al. Circadian entrainment to the natural light-dark cycle across seasons and the weekend. Curr Biol. 2017;27(4):508–513. doi: 10.1016/j.cub.2016.12.041. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Carvalho-Mendes RP, et al. Afternoon school start times are associated with a lack of both social jetlag and sleep deprivation in adolescents. J Biol Rhythms. 2020;35(4):377–390. doi: 10.1177/0748730420927603. [DOI] [PubMed] [Google Scholar]
- 23. Dunster GP, 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] [PMC free article] [PubMed] [Google Scholar]
- 24. Wong PM, et al. Social jetlag, chronotype, and cardiometabolic risk. J Clin Endocrinol Metab. 2015;100(12):4612–4620. doi: 10.1210/jc.2015-2923. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Roenneberg T, et al. Social jetlag and obesity. Curr Biol. 2012;22(10):939–943. doi: 10.1016/j.cub.2012.03.038. [DOI] [PubMed] [Google Scholar]
- 26. Blume C, et al. Effects of the COVID-19 lockdown on human sleep and rest-activity rhythms. Curr Biol. 2020;30(14):R795–R797. doi: 10.1016/j.cub.2020.06.021. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Facer-Childs ER, et al. Sleep and mental health in athletes during COVID-19 lockdown. Sleep. 2021;44(5):zsaa261. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Ong JL, et al. A longitudinal analysis of COVID-19 lockdown stringency on sleep and resting heart rate measures across 20 countries. Sci Rep. 2021;11(1):14413. doi: 10.1038/s41598-021-93924-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Leone MJ, et al. Effects of lockdown on human sleep and chronotype during the COVID-19 pandemic. Curr Biol. 2020;30(16):R930–R931. doi: 10.1016/j.cub.2020.07.015. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Massar SAA, et al. Reopening after lockdown: the influence of working-from-home and digital device use on sleep, physical activity, and wellbeing following COVID-19 lockdown and reopening. Sleep. 2021. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Lee A, et al. Impact of COVID-19 on life of students: case study in Hong Kong. Int J Environ Res Public Health. 2021;18(19):10483. doi: 10.3390/ijerph181910483. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Beck F, et al. Covid-19 health crisis and lockdown associated with high level of sleep complaints and hypnotic uptake at the population level. J Sleep Res. 2021;30(1):e13119. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Romero-Blanco C, et al. Sleep Pattern Changes in Nursing Students during the COVID-19 Lockdown. Int J Environ Res Public Health. 2020;17(14):5222. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Marelli S, et al. Impact of COVID-19 lockdown on sleep quality in university students and administration staff. J Neurol. 2021;268(1):8–15. doi: 10.1007/s00415-020-10056-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Simonelli G, et al. Sleep in times of crises: a scoping review in the early days of the COVID-19 crisis. Sleep Med Rev. 2021;60:101545. doi: 10.1016/j.smrv.2021.101545. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36. Delacre M, Klein O.. Routliers: Robust outliers detection. R package version 0003. 2019; https://CRAN.R-project.org/package=Routliers. [Google Scholar]
- 37. Leys C, et al. Detecting outliers: do not use standard deviation around the mean, use absolute deviation around the median. J Exp Soc Psychol. 2013;49(4):764–766. [Google Scholar]
- 38. Bates D, et al. Fitting linear mixed-effects models using lme4. J Stat Softw. 2015;67(1):1–48. [Google Scholar]
- 39. Phillips AJK, et al. High sensitivity and interindividual variability in the response of the human circadian system to evening light. Proc Natl Acad Sci U S A. 2019;116(24):12019–12024. doi: 10.1073/pnas.1901824116. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40. Wright KP, Jr, et al. Intrinsic near-24-h pacemaker period determines limits of circadian entrainment to a weak synchronizer in humans. Proc Natl Acad Sci U S A. 2001;98(24):14027–14032. doi: 10.1073/pnas.201530198. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41. Kantermann T, et al. Circadian phase, circadian period and chronotype are reproducible over months. Chronobiol Int. 2018;35(2):280–288. doi: 10.1080/07420528.2017.1400979. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42. Duffy JF, et al. Effect of Light on Human Circadian Physiology. Sleep Med Clin. 2009;4(2):165–177. doi: 10.1016/j.jsmc.2009.01.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43. Crowley SJ, et al. Modifications to weekend recovery sleep delay circadian phase in older adolescents. Chronobiol Int. 2010;27(7):1469–1492. doi: 10.3109/07420528.2010.503293. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44. Elmer T, et al. Students under lockdown: comparisons of students’ social networks and mental health before and during the COVID-19 crisis in Switzerland. PLoS One. 2020;15(7):e0236337. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45. Huckins JF, et al. Mental health and behavior of college students during the early phases of the COVID-19 pandemic: Longitudinal Smartphone and Ecological Momentary Assessment Study. J Med Internet Res. 2020;22(6):e20185. doi: 10.2196/20185. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
Research data will be available upon request to the authors.