Day-to-day and longer-term longitudinal associations between physical activity, sedentary behavior, and sleep in children

Abstract

Study Objectives

To determine the day-to-day and longer-term longitudinal associations between daytime physical activity and night-time sleep.

Methods

We used data from a 2-year longitudinal study which included three time points (i.e. baseline, year 1, and year 2). Participants were recruited from primary schools and included 1059 children (50% girls) with a mean age of 8.81-years-old (SD = 0.72) at baseline. Sleep variables included sleep duration, sleep efficiency, time in bed, sleep onset, and wake time. Physical activity variables included light, moderate, moderate-to-vigorous, and vigorous physical activity as well as sedentary time. We objectively assessed physical activity and sleep behaviors using the GENEActiv wrist-worn accelerometer over an 8-day period at each timepoint for a potential 21 190 observed days.

Results

We used fixed-effects multilevel models and parallel latent growth curve modeling to examine day-to-day and longer-term associations, respectively. Day-to-day, physical activity, and sleep variables were significantly, positively, and bidirectionally associated, except for sleep efficiency, which showed little association with physical activity. Longer-term, we found little association between physical activity and sleep variables.

Conclusions

Overall, our findings indicate that there is a day-to-day association between the amount of time spent being physically active and improved sleep. The lack of a longer-term association indicates that a focus on children’s daily behavior may be most appropriate to help children improve sleep and increase physical activity.

sleep, physical activity, child, pediatric, longitudinal, actigraphy

Statement of Significance

Insufficient sleep is common in children and sleep duration has declined over the years. Intervention is needed to improve the sleep behavior of children. Physical activity is often recommended as a strategy to improve sleep; however, previous research has often shown neutral or even negative results. In this longitudinal study, we found positive, significant, and bidirectional associations between daily daytime physical activity and night-time sleep behavior. However, over longer periods of time, we found little longer-term association. These findings (1) support the recommendation to engage in increased physical activity to improve sleep in children, (2) suggest that the benefits may only be effective short-term, and (3) indicate that for longer-term benefits, additional intervention beyond increased physical activity may be needed.

Introduction

Inadequate physical activity and insufficient sleep are common in children [1, 2]. Moreover, the amount of physical activity and sleep that children obtain has substantially declined over the years [3, 4]. These trends are concerning for children’s physical and mental well-being and have been linked with obesity, diabetes, cardiovascular disease, depression, poorer cognitive ability, and poorer academic performance when compared to children who are more active and sleep better [5–11]. Twenty-four-hour movement behavior guidelines recommend that children should obtain at least 60 minutes of moderate to vigorous physical activity, break up sedentary time as much as possible, and sleep between 9 and 11 hours per night [12]. Given that many children do not meet these guidelines, investigating ways to improve adherence to the recommendations is warranted.

Recent research has investigated whether one day’s participation in one movement behavior is bidirectionally associated with participation in other movement behaviors (e.g. daytime physical activity and the following night of sleep) [13–21]. In other words, are children more likely to sleep better if they were more active during the day or are children more likely to be more active if they slept better the night before? Of the studies that have examined day-to-day movement behaviors in children across multiple days to 1 week, most have reported little [15, 16, 18, 21] or negative [13, 14, 17, 20] associations between objectively measured physical activity and sleep. Only one of these studies has reported positive and bidirectional associations between sleep duration and moderate-to-vigorous physical activity [19]. The discrepancies in these findings are consistent with a recent meta-analysis which reported that there is little association between physical activity and sleep in children, and that the relationship is largely inconsistent [22]. The meta-analysis [22] also reported only one longer-term longitudinal study had been conducted among children. The one longitudinal study reported a positive association between physical activity and subjectively reported sleep disturbances across a school year (i.e. 9 months with 3-month lags) in 13-year-old girls [23]. A positive longer-term association may suggest that there are habitual adaptations from movement behaviors that are not captured in day-to-day research studies. This discrepancy in findings among day-to-day and longer-term studies may be due to the study designs and measures or the different samples of participants. Thus, further research is needed to clarify the possible bidirectional relationship between physical activity and sleep.

The purpose of this study was to objectively examine the relationship between both day-to-day and longer-term (i.e. habitual) physical activity and sleep behavior in children. Specifically, for day-to-day associations, does daytime physical activity predict improved sleep the following night and does sleep at night predict increased physical activity the following day? For longer-term associations, is there an association between habitual daytime physical activity patterns and habitual night-time sleep behavior over time?

Methods

This article is reported following the STROBE statement [24].

Sample

We used data from the “Internet-based Professional Learning to help teachers support Activity in Youth” (iPLAY) cluster randomized controlled trial [25]. Three cohorts of Grade 3 and 4 children (aged 8–9 years) participated in iPLAY over a 3.5-year period. Within each cohort, we collected children’s physical activity, sedentary time, and sleep data at three time points over 2 years (i.e. baseline, 12-months follow-up, and 24-months follow-up). To control for possible seasonal and weather effects and ensure that time points were as similar as possible, we scheduled data collection for each cohort as close to the same date and weather conditions as possible. We collected data between June 2016 and December 2019.

Measures

We measured daily physical activity, sedentary time, and sleep using the wrist worn GENEActiv triaxial accelerometer (Activinsights, Cambridge, United Kingdom). We asked participants to wear the accelerometer all day and night on their non-dominant wrist for a period of 8 days with the only exception being during organized contact sports (e.g. rugby) when the device could pose a risk of injury. We sampled the data at a frequency of 87.5 Hz. The GENEActiv accelerometer has been validated for physical activity [26], sedentary time [27], and sleep [28, 29].

We processed the accelerometer data into 5-second epochs using the R-package GGIR (ver. 1.10–7) [30] in the R environment (ver. 3.6.1) [31]. GGIR detects non-wear time and calculates physical activity intensities by converting the raw acceleration values into one omnidirectional value of acceleration. We used the Euclidian norm minus one with negative values set to zero (ENMONZ) metric of acceleration. For valid days, we calculated physical activity intensity variables based on ENMONZ value cut-points [26, 27]. sedentary activity (0–56.3 mg), light-intensity physical activity (56.3–191.6 mg), moderate-intensity physical activity (191.6–695.8 mg), vigorous-intensity physical activity (greater than 695.8 mg), and moderate-to-vigorous physical activity (greater than 191.6 mg). For sleep detection, GGIR uses estimated change in arm angle relative to the horizontal plane. In this study, a change in arm angle of less than five degrees over a 5-minute period was characterized as a possible sleep period [28]. We calculated the following sleep variables: sleep duration (minutes/night), time in bed (minutes/night), sleep efficiency (TST/TIB*100), sleep onset (number of hours from previous midnight), and wake time (number of hours from previous day’s midnight). We included a day of participant data if the day had greater than 16 hours of valid wear time [29]. For the longer-term analyses, we used additional exclusion criteria to include only participants with at least 4 days, including one weekend day, of valid data (i.e. 16 hours valid wear time) [32, 33]. We included all participants with valid data from at least one timepoint.

Statistical analysis

Day-to-day analyses

We used multilevel models with lagged effects using the Linear Models for Panel Data (plm) R-package [34] to test the day-to-day associations between physical activity, sedentary time, and sleep variables. We used fixed-effects models to focus on the within-person variation. That is, fixed-effects models account for all between-person time-invariant variation and allows us to treat each individual as their own control. Therefore, using fixed effects allows us to examine how an individual’s behavior predicts changes in future behavior.

The two-level models incorporated data from multiple time points (level 1) nested within individuals (level 2). For each physical activity outcome variable (i.e. light physical activity, moderate physical activity, vigorous physical activity, moderate-to-vigorous physical activity, and sedentary time), either sleep duration, sleep efficiency, time in bed, sleep onset, or wake time was the predictor variable and vice versa. There were, therefore, a total of 50 separate models (i.e. 25 predicting physical activity outcomes and 25 predicting sleep outcomes). To account for the relative and constrained nature of these data in our analyses, we converted physical activity variables from minutes per day to be analyzed as proportions of daily wake time spent being active. We used a Bonferroni correction to account for multiple comparisons in our day-to-day analyses and considered a p value of <.001 to be statistically significant. We controlled for wear time in all models. We also controlled for sleep duration in sleep efficiency predicting next-day physical activity models. Lastly, we investigated differences across sex by conducting separate analyses for boys and girls.

Longer-term analyses

We used parallel process latent growth curve models to analyze longer-term longitudinal associations between physical activity and sleep [35]. Using this analytical method, we were able to compare how changes in one variable over time (i.e. physical activity or sleep) are related to changes in another variable (i.e. sleep or physical activity). In other words, we modeled the trajectory of change in physical activity (i.e. growth process) in parallel with the trajectory of change in sleep. This process allows us to examine how both the intercept and growth of physical activity are related to both the intercept and growth of sleep (Figure 1). Latent growth curve modeling accounts for both between-person and within-person variability to model change. We aggregated data at each time point within individuals by calculating a weighted mean of weekday and weekend data for each variable. All analyses used all available information and maximum likelihood estimation with robust standard errors to control for missing data. We determined model fit using the comparative fit index, root-mean-square error of approximation, and the Tucker-Lewis index [36]. For the comparative fit index, values greater than 0.90 and 0.95 demonstrated good and excellent fit, respectively. For the root-mean-square error of approximation, values less than 0.10 and 0.05 demonstrated good fit and excellent fit, respectively. Finally, for the Tucker-Lewis index, a value greater than 0.95 indicated good model fit. We used the Latent Variable Analysis package (lavaan) [37] within the R environment (ver. 3.6.2) [31] to conduct the analyses. We also tested for sex differences by conducting separate analyses for boys and girls.

Figure 1.

Parallel process growth model.

Results

Preliminary analyses

From 1138 participants, there were 2876 total participant observations (i.e. 8-day periods), covering a potential 23 008 individual days, across the three time points. First, we identified and removed days that did not meet wear time criteria (i.e. less than 16 valid wear time hours). We then calculated the interquartile range for physical activity, sleep variables, and ENMONZ values. We identified extreme outliers (i.e. 3*interquartile range ± upper/lower quartile) in the data, suggesting accelerometer measurement error. Finally, we removed data for observations with accelerometer calibration errors, including those indicated by extreme values. The final study sample consisted of 2745 observations (i.e. 8-day periods) over three time points (i.e., baseline, 12-month follow-up, and 24-month follow-up) or a potential 21 960 individual days for the day-to-day analyses. We applied additional criteria (i.e. at least 4 valid days with at least one weekend day) to the data for the habitual analyses. Therefore, the final sample for the habitual analyses consisted of 2119 observations.

Participant characteristics

We present the participant characteristics descriptive data of each physical activity and sleep variable for each time point and for the overall sample in Table 1. We report the distributions of each variable in Supplementary Figure S1. The number of participants declined over time as students moved to different schools, were absent from school during data collection, failed to return the accelerometer, or the accelerometer data was unusable (i.e. failed to record, accelerometer was returned broken, data could not be extracted, etc.).

Table 1.

Participant characteristics and descriptive data for physical activity and sleep variables (n = number of observed days)

	All (n = 21 960 days)	Baseline (n = 8472 days)	Year 1 (n = 7248 days)	Year 2 (n = 6240 days)
Number of participant observations (% male)	2,745 (49%)	1,059 (50%)	906 (49%)	780 (48%)
Age (years)	9.72 ± 1.08	8.81 ± 0.72	9.82 ± 0.71	10.83 ± 0.70
Light physical activity (minutes)	214.00 ± 53.51	220.89 ± 52.51	213.09 ± 53.00	204.35 ± 54.17
Moderate physical activity (minutes)	69.55 ± 31.12	73.72 ± 30.80	69.60 ± 31.58	62.94 ± 29.87
Vigorous physical activity (minutes)	13.34 ± 10.74	14.90 ± 11.11	13.46 ± 10.79	10.73 ± 9.51
Moderate-to-vigorous physical activity (minutes)	83.14 ± 40.01	88.92 ± 39.79	83.32 ± 40.75	73.82 ± 37.59
Sedentary time (minutes)	691.72 ± 126.13	666.54 ± 113.84	698.51 ± 127.29	724.81 ± 135.07
Sleep duration (minutes)	455.88 ± 81.83	468.05 ± 72.85	449.27 ± 83.24	444.01 ± 91.14
Sleep efficiency (%)	86.82 ± 5.67	86.53 ± 5.62	86.81 ± 5.69	87.33 ± 5.69
Time in bed (minutes)	529.60 ± 86.71	545.00 ± 74.99	520.72 ± 90.84	515.46 ± 95.07
Sleep onset (hours)*	22.00 ± 1.33	21.89 ± 1.25	22.01 ± 1.34	22.18 ± 1.43
Wake time (hours)*	31.11 ± 1.07	31.06 ± 1.02	31.11 ± 1.06	31.20 ± 1.15
Meeting physical activity guideline†	69.66%	75.67%	69.49%	60.44%
Meeting sleep duration guideline‡	10.53%	11.82%	9.31%	9.94%

Data are presented as mean ± (SD).

*Sleep onset and wake time in hours from the previous day’s midnight.

^†Physical activity guideline of at least 60 minutes of moderate to vigorous physical activity per day.

^‡Sleep duration guideline of 9–11 hours per night.

Day-to-day analyses

We show the results of the fixed effects panel models in Table 2 for sleep predicting next-day physical activity and Table 3 for physical activity predicting the following night’s sleep.

Table 2.

Fixed effects models with sleep predicting next day physical activity outcomes

Predictor	Outcome	Std. Beta	SE	p Value	Average difference in outcome with 30-minute increase in predictor‡ (minutes)
Sleep duration (minutes)	Light physical activity (% of wake time)	0.032	0.037	<.001	03:39
Sleep duration (minutes)	Moderate physical activity (% of wake time)	0.022	0.020	.014	00:48
Sleep duration (minutes)	Moderate to vigorous physical activity (% of wake time)	0.028	0.026	<.001	01:43
Sleep duration (minutes)	Vigorous physical activity (% of wake time)	0.045	0.007	<.001	00:12
Sleep duration (minutes)	Sedentary time (% of wake time)	−0.034	0.057	<.001	−09:07
Sleep efficiency (%)†	Light physical activity (% of wake time)	−0.007	0.010	.480	−00:24
Sleep efficiency (%)†	Moderate physical activity (% of wake time)	−0.180	0.006	.061	−03:11
Sleep efficiency (%)†	Moderate to vigorous physical activity (% of wake time)	−0.019	0.007	.060	−00:34
Sleep efficiency (%)†	Vigorous physical activity (% of wake time)	−0.012	0.002	.220	−00:02
Sleep efficiency (%)†	Sedentary time (% of wake time)	0.014	0.016	.187	01:47
Time in bed (minutes)	Light physical activity (% of wake time)	0.039	0.034	<.001	04:16
Time in bed (minutes)	Moderate physical activity (% of wake time)	0.037	0.019	<.001	01:16
Time in bed (minutes)	Moderate to vigorous physical activity (% of wake time)	0.042	0.024	<.001	02:25
Time in bed (minutes)	Vigorous physical activity (% of wake time)	0.049	0.006	<.001	00:12
Time in bed (minutes)	Sedentary time (% of wake time)	−0.046	0.052	<.001	−11:32
Sleep onset (hours)	Light physical activity (% of wake time)	−0.101	0.044	<.001	−11:56
Sleep onset (hours)	Moderate physical activity (% of wake time)	−0.094	0.025	<.001	−03:33
Sleep onset (hours)	Moderate to vigorous physical activity (% of wake time)	−0.098	0.032	<.001	−06:09
Sleep onset (hours)	Vigorous physical activity (% of wake time)	−0.087	0.008	<.001	−00:24
Sleep onset (hours)	Sedentary time (% of wake time)	0.114	0.069	<.001	30:57
Wake time (hours)	Light physical activity (% of wake time)	−0.145	0.051	<.001	−21:26
Wake time (hours)	Moderate physical activity (% of wake time)	−0.097	0.029	<.001	−04:34
Wake time (hours)	Moderate to vigorous physical activity (% of wake time)	−0.089	0.037	<.001	−06:55
Wake time (hours)	Vigorous physical activity (% of wake time)	−0.051	0.010	<.001	−00:17
Wake time (hours)	Sedentary time (% of wake time)	0.138	0.079	<.001	46:57

All models controlled for accelerometer wear time.

^†Model adjusted for wear time and sleep duration.

^‡For sleep efficiency outcomes, Average change in outcome with a 1% increase in efficiency. Statistically significance was set at p < .001. Effect sizes represent changes in SD units, for example, one SD unit change in sleep duration is associated with a 0.032 SD unit increase in light physical activity.

Table 3.

Fixed effects models with physical activity predicting sleep outcomes the following night

Predictor	Outcome	Std. Beta	SE	p Value	Average difference in outcome with 30-minute increase in predictor
Light physical activity (% of wake time)	Sleep duration (minutes)	0.327	0.002	<.001	14:23 minutes
Light physical activity (% of wake time)	Sleep efficiency (%)	0.024	0.008	.003	0.07%
Light physical activity (% of wake time)	Time in bed (minutes)	0.271	0.002	<.001	12:38 minutes
Light physical activity (% of wake time)	Sleep onset (hours)	−0.163	0.002	<.001	−07:00 minutes
Light physical activity (% of wake time)	Wake time (hours)	0.000	0.002	.984	00:00 minutes
Moderate physical activity (% of wake time)	Sleep duration (minutes)	0.203	0.004	<.001	15:50 minutes
Moderate physical activity (% of wake time)	Sleep efficiency (%)	0.014	0.015	.087	0.08%
Moderate physical activity (% of wake time)	Time in bed (minutes)	0.182	0.004	<.001	15:00 minutes
Moderate physical activity (% of wake time)	Sleep onset (hours)	−0.129	0.004	<.001	−09:46 minutes
Moderate physical activity (% of wake time)	Wake time (hours)	−0.017	0.003	.075	−01:01 minutes
Moderate to vigorous physical activity (% of wake time)	Sleep duration (minutes)	0.195	0.003	<.001	11:46 minutes
Moderate to vigorous physical activity (% of wake time)	Sleep efficiency (%)	0.013	0.012	.135	0.05%
Moderate to vigorous physical activity (% of wake time)	Time in bed (minutes)	0.179	0.003	<.001	11:27 minutes
Moderate to vigorous physical activity (% of wake time)	Sleep onset (hours)	−0.137	0.003	<.001	−08:03 minutes
Moderate to vigorous physical activity (% of wake time)	Wake time (hours)	−0.026	0.002	.007	−01:12 minutes
Vigorous physical activity (% of wake time)	Sleep duration (minutes)	0.134	0.012	<.001	29:59 minutes
Vigorous physical activity (% of wake time)	Sleep efficiency (%)	0.005	0.044	.580	0.07%
Vigorous physical activity (% of wake time)	Time in bed (minutes)	0.138	0.013	<.001	32:53 minutes
Vigorous physical activity (% of wake time)	Sleep onset (hours)	−0.134	0.011	<.001	−29:13 minutes
Vigorous physical activity (% of wake time)	Wake time (hours)	−0.042	0.009	<.001	−07:17 minutes
Sedentary time (% of wake time)	Sleep duration (minutes)	−0.294	0.001	<.001	−08:31 minutes
Sedentary time (% of wake time)	Sleep efficiency (%)	−0.021	0.005	.009	−0.04%
Sedentary time (% of wake time)	Time in bed (minutes)	−0.251	0.002	<.001	−07:43 minutes
Sedentary time (% of wake time)	Sleep onset (hours)	0.164	0.001	<.001	04:38 minutes
Sedentary time (% of wake time)	Wake time (hours)	0.012	0.001	.204	00:16 minutes

All models controlled for accelerometer wear time. Statistically significance was set at p < .001. Effect sizes represent changes in SD units, for example, one SD unit change in light physical activity is associated with a 0.327 SD unit increase in sleep duration.

Sleep predicting next-day physical activity outcomes

We found statistically significant associations for all the models, except for the sleep efficiency models and the association between sleep duration and moderate physical activity. Increased sleep duration and time in bed both predicted an increase in all physical activity outcomes and a decrease in proportion of time spent sedentary the next day. Similarly, earlier sleep onset and earlier wake time predicted a greater proportion of time spent being active at all intensity levels and a decrease in time spent sedentary the following day. We examined the differences across boys and girls in separate models and found the overall results were similar for boys (Supplementary Table S1). For girls, however, sleep duration and time in bed showed little association with next day physical activity and only sleep onset and wake time were significant predictors (Supplementary Table S2).

Physical activity predicting the following night’s sleep outcomes.

All physical activity variables were statistically significantly associated with the following night’s sleep duration, time in bed, and sleep onset. That is, an increase in the proportion of time spent active at any intensity and a decrease in the time spent sedentary were associated with longer sleep duration, longer time in bed, and earlier sleep onset. Physical activity was not associated with sleep efficiency the following night. Physical activity was also generally not associated with the next day’s waketime, as only increases in the proportion of time spent in vigorous physical activity was associated with earlier wake times. Separate analyses for boys and girls showed similar results (Supplementary Tables S3 and S4).

Longer-term analyses

The results of the parallel process growth models (Figure 1) used in our analyses are shown in Table 4. Due to errors with convergence among the light physical activity and sedentary time variables, we did not include these models. We report plots for the trajectories of change of all variables in Supplementary Figure S2.

Table 4.

Parallel process latent growth model results

	Moderate physical activity (minutes)			Moderate-to-vigorous physical activity (minutes)			Vigorous physical activity (minutes)
	Std. Est. (r)	Std. Error	p value	Std. Est. (r)	Std. Error	p value	Std. Est. (r)	Std. Error	p value
Sleep duration (minutes)
s2 on i1	0.027	0.037	.834	0.014	0.043	.908	−0.009	0.009	.950
s1 on i2	0.089	0.114	.783	−0.063	0.087	.847	−0.381	0.355	.258
s1 with s2	−0.333	0.004	.397	−0.246	0.005	.516	−0.206	0.001	.706
i1 with i2	−0.118	0.013	.253	−0.087	0.016	.387	0.024	0.004	.814
s1 on i1	0.372	0.174	.557	0.343	0.172	.589	0.252	0.166	.673
s2 on i2	−0.351	0.048	.007*	−0.377	0.045	.003*	−0.516	0.051	.011*
Sleep efficiency (%)
s2 on i1	−0.020	0.132	.783	−0.027	0.165	.706	−0.015	0.051	.901
s1 on i2	−0.036	0.005	.722	−0.034	0.004	.739	−0.046	0.016	.686
s1 with s2	−0.021	0.690	.865	−0.017	0.833	.887	−0.021	0.199	.902
i1 with i2	−0.129	4.824	.072	−0.159	6.009	.021*	−0.233	1.587	.001*
s1 on i1	−0.341	0.052	.092	−0.340	0.053	.096	−0.342	0.054	.112
s2 on i2	−0.346	0.049	.011*	−0.375	0.047	.004*	−0.524	0.054	.019*
Time in bed (minutes)
s2 on i1	0.007	0.028	.947	−0.009	0.044	.940	−0.076	0.584	.590
s1 on i2	0.048	0.103	.613	−0.156	0.089	.650	−0.519	0.007	.157
s1 with s2	−0.107	0.004	.367	−0.169	0.005	.636	0.080	0.077	.885
i1 with i2	0.024	0.012	.810	0.103	0.016	.339	0.313	0.248	.001*
s1 on i1	−0.302	0.118	.041*	−0.163	0.263	.861	−0.221	0.246	.770
s2 on i2	−0.343	0.049	.013*	−0.368	0.047	.006*	−0.485	0.057	.028*
Sleep onset (hours)
s2 on i1	−0.119	0.839	.276	−0.115	0.017	.249	−0.153	0.284	.297
s1 on i2	−0.172	0.001	.499	−0.058	0.063	.815	0.051	0.005	.855
s1 with s2	−0.018	0.186	.958	−0.103	0.004	.737	0.087	0.058	.852
i1 with i2	−0.093	1.175	.279	−0.147	0.025	.075	−0.282	0.383	.001*
s1 on i1	−0.190	0.040	.507	−0.188	0.040	.520	−0.185	0.041	.536
s2 on i2	−0.281	0.063	.166	−0.346	0.059	.056	−0.536	0.067	.066
Wake time (hours)
s2 on i1	−0.069	0.017	.535	−0.079	0.020	.426	−0.157	0.005	.160
s1 on i2	0.011	0.072	.921	−0.030	0.053	.782	−0.148	0.213	.181
s1 with s2	−0.320	9.656	.039*	−0.238	11.150	.071	0.016	2.435	.921
i1 with i2	−0.091	65.374	.310	−0.104	82.003	.228	−0.109	21.375	.198
s1 on i1	−0.321	0.057	.095	−0.330	0.057	.081	−0.353	0.057	.056
s2 on i2	−0.290	0.063	.137	−0.352	0.057	.040*	−0.520	0.062	.033*

*p < .05.

i1 = intercept for sleep variable. i2 = intercept for physical activity variable. s1 = slope for sleep variable. s2 = slope for physical activity variable. with = covariance path. on = regression path.

For the covariances between change in physical activity and change in sleep over time (s1 with s2), only one significant association was found between moderate physical activity and wake time. Other covariances were not significant, indicating that sleep and physical activity generally did not change together over time.

The covariances between initial physical activity and initial sleep (i1 with i2) showed several significant associations. In contrast to our day-to-day analyses, initial sleep efficiency was negatively associated with initial moderate-to-vigorous physical activity and vigorous physical activity. These results indicated that those with higher sleep efficiency were associated with significantly less moderate-to-vigorous and vigorous physical activity. Initial vigorous physical activity was also associated with significantly longer initial time in bed and earlier initial sleep onset. No other significant associations were found between physical activity variables and sleep variables. Separate models for boys and girls showed largely the same results (Supplementary Tables S5 and S6).

Discussion

Our study aimed to examine the nature of the relationship between physical activity and sleep in children. We examined both day-to-day and longer-term physical activity and sleep behaviors of children aged 7 to 12 years. Our study is the first to investigate both day-to-day and longer-term longitudinal associations within the same sample of children.

Our day-to-day findings indicated that physical activity and sleep were significantly and positively related. Increased physical activity, at all intensity levels, was associated with longer sleep duration, longer time in bed, earlier bedtime, and earlier wake time the following night. These associations may also be practically meaningful. For example, a 1% increase (i.e. about 11 minutes) in wake time spent in vigorous physical activity was associated with an increase of about 11 minutes in sleep duration (β = .134), 12 minutes of time in bed (β = .138), 11 minutes earlier sleep onset (β = −.134), and 3 minutes earlier waketime the next day (β = −.042). These same sleep variables were also associated with an increase in time spent being physically active the following day. However, sleep efficiency, as a predictor and as an outcome, showed little association with changes in physical activity or sedentary time.

Our positive associations are consistent with Lin et al. [19] but are inconsistent with many other similar studies in children. Multiple studies reported little to no day-to-day associations [15, 16, 18, 21], while others have shown negative associations for some physical activity and sleep variables [13, 14, 17, 20]. While these studies use similar participants (i.e. 6- to 12-year-old children), a single device to objectively measure both physical activity and sleep, and a similar protocol length (i.e. 7–8 days), there may be some important differences in methods that impact their findings. For example, each study used different wear time criteria for number of days (e.g. 2–7 days), number of hours/day (e.g. 10–20 hours/day or no criteria), and whether a weekend day was required. Additionally, only two studies controlled for wear time in their analyses [15, 19]. We used 16 hours/day [29] but included all days meeting that criteria. This protocol allowed our analyses to use all available valid data while controlling for wear time. More research, however, is needed to clarify the optimal wear time criteria for this research or if it varies by device and sleep detection algorithm used [38, 39]. Another possibility relates to the activity levels of our sample. We found that nearly 70% of our sample adhered to the physical activity guideline to obtain at least 60 minutes of moderate-to-vigorous physical activity each day. This is unusually high for children both in Australia [2] and globally [40]. It may be that positive associations between physical activity and sleep are more likely among these more active children.

We also found some sex differences in our sample. While there was little day-to-day bi-directional association for sleep efficiency for both sexes, boys showed a positive bi-directional association between most other sleep and physical activity variables. In contrast, girls only showed a positive bi-directional association between sleep onset and wake time and physical activity. For other sleep variables, physical activity promoted sleep duration and time in bed the following night, but sleep duration and time in bed did not influence next-day physical activity. In terms of sleep duration being associated with next day light or moderate-to-vigorous physical activity, other studies have shown either little difference between sexes [14, 17, 21] or a stronger positive association for girls than boys [19]. Lin and colleagues [19], suggest that floor and ceiling effects may be the reason for these differences between sexes due to the shorter average sleep duration of girls (i.e. 1.4 hours shorter than boys) in their sample. In our study, however, average sleep durations were similar for boys (7.5 hours) and girls (7.7 hours), so we are unable to determine whether this effect influenced our results.

Our longer-term findings are also inconsistent with previous research. Our study, which is the first longer-term longitudinal investigation of physical activity and sleep in children using device-measured outcomes, found that there is little association between habitual physical activity and sleep behaviors over a 2-year period. Raudsepp [23], however, used a similar analysis but employed subjective measures of physical activity and sleep disturbances, and found a positive association. That is, increases in reported physical activity was associated with decreases in reported sleep disturbances over 9 months. A possible reason for our different findings is the use of objective rather than subjective measures of physical activity and sleep, which are susceptible to recall and reporting bias [41]. Alternatively, the difference may be related to the increasing age and maturation of children in our sample into adolescence, which has been associated with poorer sleep and decreased physical activity [2, 42, 43]. These long-term developmental factors may have affected the 2-year trajectories of our sleep and physical activity growth curves differently compared to Raudsepp’s [23] older sample of teenage girls during one school year. In other words, the developmental processes in children may have minimized the capacity for sleep and physical activity to influence each other.

Our day-to-day and longer-term findings seem to contradict each other. One possible explanation is that day-to-day variation of behavior is greater than average weekly variation of behavior across time points. To examine this, we compared the average individual standard deviation of physical activity and sleep variables within weeks and within time points (Supplementary Table S7). There was less variance within time points than within weeks, meaning there was less variance to predict in the longer-term models. More frequent observations may be needed to capture long term variation in physical activity and sleep.

Nevertheless, our findings suggest that these behaviors may need to be emphasized daily to achieve the best health outcomes and supports the clinical recommendation that physical activity may improve sleep behavior in children [44]. Our findings also suggest that physical activity and sleep may be reciprocally associated in children [45], and that promoting improvements in one behavior may have positive implications for the other. Interventions for meeting movement behavior guidelines, including physical activity and sleep [46], should consider strategies that promote earlier sleep onset, improve consistency of sleep timing, and increase sleep duration and physical activity to improve both physical activity and sleep bidirectionally.

The strengths of our study include the large sample size and the use of both day-to-day data in fixed-effects multilevel models and 2-year longitudinal data in parallel process growth models. Our fixed-effects models allowed us to specifically examine individual-level changes from one day to the next, while parallel process growth modeling allowed us to account for both between- and within-person variation and examine the bidirectional association of physical activity and sleep over a 2-year period. Taken together, we get a more complete picture of the relationship between physical activity and sleep in children. We also examined a variety of sleep variables including, duration, efficiency, and timing. Previous research has discussed the need to investigate a variety of sleep dimensions and not just sleep duration [15, 47].

Our study also has some limitations. It is unknown how long the appropriate amount of lag between measures should be for long-term longitudinal analyses between physical activity and sleep. Moreover, we have assumed that 1-week of accelerometer-derived movement behavior data is enough to represent habitual movement behavior for a child at that time. While studies have reported that 1-week of data is typically enough to establish reliable estimates of physical activity and sleep [48, 49], it is unknown how stable these estimates are over time. Additionally, the method we used for sleep detection with the GENEActiv accelerometer has only been validated in adults [28, 29]. Our findings should be interpreted with a degree of caution until sleep validation studies in children are conducted. Furthermore, while accelerometers are considered valid measures of movement behaviors [29, 50, 51] and wrist-worn devices tend to achieve greater wear time compliance [52], some devices malfunctioned during data collection, were lost or broken, or did not meet wear time criteria, resulting in data loss. Therefore, we do not have complete data for all participants at all time points, which may influence our findings. Finally, while our use of fixed effects and within-subjects analyses allowed us to control for many time-invariant and between-subject confounding variables, we cannot rule out all residual confounding variables and time-variant covariates (e.g. screen use, weather conditions, and secular events) which may influence our results.

Conclusions

This study contributes to a growing body of knowledge of the day-to-day and longer-term relationship between physical activity and sleep in children and provides the first objectively measured longer-term longitudinal study in children. Generally, we found positive associations between day-to-day physical activity and sleep, but little association over time. Our findings were largely inconsistent with previous research in this area. Further research is needed to confirm our findings, especially longer-term longitudinal studies. Overall, our findings suggest that day-to-day physical activity and sleep are bidirectionally associated in children and regular, short-term intervention may be needed to help children sustainably improve sleep and increase physical activity levels.

Supplementary material

Supplementary material is available at SLEEP online.

Supplement Figure S1 Distribution of Physical Activity and Sleep Variables

Supplement Figure S2 Trajectories of Change in Physical Activity and Sleep Variables at Baseline, Year 1, and Year 2

Supplement Table S1 Boys - Fixed Effects Models with Sleep Predicting Next-Day Physical Activity Outcomes

Supplement Table S2 Girls - Fixed Effects Models with Sleep Predicting Next-Day Physical Activity Outcomes

Supplement Table S3 Boys - Fixed Effects Models with Physical Activity Predicting Sleep Outcomes the Following Night

Supplement Table S4 Girls - Fixed Effects Models with Physical Activity Predicting Sleep Outcomes the Following Night

Supplement Table S5 Parallel Process Latent Growth Model Results - Boys

Supplement Table S6 Parallel Process Latent Growth Model Results - Girls

Supplement Table S7 Average individual SDs within weeks (i.e., daily) vs. within timepoints (i.e., yearly)

Funding

This research was supported by the National Health and Medical Research Council (Australia) Partnership Project Grant (Ref: GNT1114281) and the New South Wales Department of Education.

Conflict of interest statement. None declared.

Author contributions

D.A. was the lead researcher on the study. D.A. drafted the manuscript, prepared the data, and conducted all the analyses. C.L. and T.S. contributed to the design of the study and commented on all manuscript drafts. B.dP.C. and P.P. provided advice on analyses, provided interpretation of results, and commented on all manuscripts drafts. All authors have approved the final text.