Study participants
The study sample was from the Active Play Study. Data were collected between January 2015 and December 2016. To be included, participants must have been aged 10 to 13 years and lived and attended school in Kingston, Ontario, Canada. Children who did not speak English or French were excluded as were non-ambulatory individuals. Participants were recruited to ensure proportional representation of the city’s 10- to 13-year-old population by age, sex, and residence within its 12 electoral districts. An equal number of participants were studied in each of the four seasons. Recruitment strategies included word of mouth, social media, and advertisements posted and distributed in schools, stores, and community centers. Participants and a parent/guardian provided written informed consent prior to participation. The study was approved by the General Research Ethics Board at Queen’s University.
Twenty five (5.5%) of the 458 participants who participated in the Active Play Study were excluded from the present analyses because they did not have valid sleep and OAP measures for any of the 8 nights and 7 days (as explained below). Thus, the final sample consisted of 433 participants. Participants who were excluded from the analyses were similar in age (11.5 vs. 11.8 years) and ethnicity (90.5% vs. 88.0% white) to those who were included; however, there was a difference in gender distribution (68.0% vs. 49.2% male). Within these 433 participants, there were several days and nights with insufficient (invalid) data to determine sleep and/or OAP, and these nights and days were removed. The final number of observations were 2253 for the analyses where nighttime sleep characteristics predicted the following day’s OAP, and 2263 for the analyses where daytime OAP predicted the following night’s sleep characteristics.
Measurement of sleep
Four sleep characteristics were assessed to characterize the nocturnal sleep period: time in bed, sleep duration, sleep chronology, and sleep efficiency. They were measured over 8 consecutive nights. Time in bed (hours/day) was calculated as the difference between when participants turned off the lights to go to bed at night and when they got out of bed in morning, which they recorded on the log. These recorded times were verified, and corrected when necessary, by having a member of the research team visually inspecting the recorded log times against the Actical accelerometer data. In our laboratory, this process is highly reliable as 90% of the verified and corrected sleep times are within 10 min of each other based on repeated attempts completed by different researchers. Sleep chronology was determined using sleep midpoint, which was determined as the midpoint between the times participants turned off the lights to go to be at night and when they got out of bed in the morning.
Sleep duration (hours/day) was calculated as the total time spent sleeping during time in bed. To determine time spent sleeping, each of the 15-s long epochs (measurement intervals) from the Actical accelerometer was defined as either ‘sleeping’ or ‘awake’ based on a sleep likelihood score, and the time of the ‘sleeping’ epochs were summed. The sleep likelihood score was based on a weighted rolling average of the count value for the epoch in question and the 8 epochs that proceed and follow it. Sleep efficiency (%) was calculated as the ratio of sleep duration to time in bed [
28,
29]. The method of estimating sleep/awake status using a waist-worn Actical accelerometer was developed and cross-validated in a sub-study of 50 children from the present study. Results from the sub-study showed that there is no significant difference in sleep efficiency estimates between the waist-word Actical accelerometer and a wrist worn accelerometer developed specifically to assess sleep characteristics (89.0% vs 88.7%,
p = 0.665] [
30].
Measurement of OAP
The method for assessing OAP was developed by our laboratory and was one of the main objectives of the Active Play Study [Borghese MM, Janssen I. Development of a measurement approach to assess time children participate in outdoor active play, organized sport, active travel, and curriculum-based physical activity, submitted]. This method used data from several sources and involved a combination of automated steps, manual checks and corrections.
First, times recorded in the log (start and end times for sleep, organized sports, outdoor chores, and accelerometer non-wear periods) were checked by visually inspecting these times within the accelerometer data using Actical 3.10 software (Philips Respironics, Murrysville, PA). If necessary, corrections were made to the recorded times.
Second, Personal Activity and Location Measurement (PALMS) software (Center for Wellness and Population Health Systems, University of California, San Diego, CA) was used to merge the accelerometer and GPS data based on each 15-s accelerometer epoch. PALMS identified periods of time with missing GPS latitude and longitude coordinates, which occurred if the satellite signal was lost (e.g., when a participant entered a large building). When possible, missing geospatial coordinates were imputed by the research team by using Google Maps (Google, Mountain View, CA) and street view images to determine where the participant was immediately before the signal was lost and immediately after the signal returned. For instance, if the signal was lost for 30 min when a participant was in a building, the latitude and longitude coordinates for the center of that building were imputed for all 15-s epochs that occurred during that 30 min.
Third, PALMS software used a validated algorithm to identify all vehicle and non-vehicle (e.g., walking, bicycling) trips [
31], and all 15-s epochs that occurred during trips were flagged. Trip detection was based on distance traveled over time (i.e., travel ≥100 m over ≥180 s at a speed of ≥1 km/h). Trip modality was determined by the 90th percentile of travel speed (walking = 1 to 9.99 km/h, cycling = 10 to 24.99 km/h, vehicle = ≥ 25 km/h). Each of these trips identified by PALMS were checked by visually inspecting all of the GPS coordinates for that trip in Google Maps. During these visual inspections we identified and then deleted a number of false positive trips. An example of a false positive trip is a trip identified by PALMS that, upon visual inspection, was found to occur on school grounds during the recess period. This example reflected OAP and not active transportation. During our visual inspections of the GPS data we did not encounter false negative trips, and therefore we did not have a need to identify trips that were not captured by the PALMS algorithm.
The data from each participant was then exported from PALMS into ArcMap version 10.4 software (Esri, Redlands, CA). The longitude and latitude coordinates for each epoch were geocoded and a map layer of building footprints for the city of Kingston was used to determine whether each 15-s epoch was indoors (in a building) or outdoors (not in a building). The files for all participants were then merged together using SAS version 9.4 statistical software (SAS Inc., Carry, NC).
The merged file contained over 19 million rows, with > 40,000 rows per participant (i.e., one row of data for each 15-s long epoch that occurred over the 7 day/8 night measurement period). Additional “time” information was merged into the master file, including 1) the sleep, organized sport, and chore/work times in the logs; 2) the start and end times of the school day and school recess times; and 3) whether each day represented a school day or a non-school day (weekend or holiday).
A SAS program was then developed to determine the number of minutes spent in OAP on each measurement day. The SAS program started by identifying and deleting all 15-s epochs for the days in which there was insufficient (< 10 h) accelerometer and GPS watch wear time during waking hours [
32,
33]. The program then flagged all 15-s epochs that could not have occurred during OAP because it occurred during one or more of the following conditions: 1) time in bed, 2) indoors, 3) school curriculum time (but not recess time) on a school day, 4) a vehicle trip or non-vehicle trip, 6) while participating in an organized sport, or 7) while performing work or chores. All of the 15-s epochs that were not flagged in the proceeding step were then classified as either occurring during OAP or as sedentary time spent outdoors using a specifically designed algorithm that has a specificity of 85%, sensitivity of 85%, and positive predictive value of 99% for identifying OAP. This algorithm is not based on a single epoch intensity cut-off since OAP includes sedentary, light, and moderate-to-vigorous intensity movements. Rather, the algorithm is a function of the count value for the epoch in question, centered and forward rolling averages for the surrounding epoch count values, the duration of outdoor time session the epoch was contained within, and interactions of these variables. After the OAP epochs were identified, they were summed to determine the daily total. These daily OAP values were exported into a dataset with multiple observations for each participant (e.g., one row per day).
Confounders
Several factors that are associated with both sleep and physical activity and which may confound the bi-directional relationship between these behaviors were considered, including characteristics of the participant, their family, their neighborhood environment, and characteristics that reflected the calendar day when the sleep and physical activity measures were obtained. Participant factors consisted of biological sex (male or female); age (continuous); ethnicity (white or non-white including mixed race); the presence of a chronic medical condition (yes or no); body mass index (BMI) z-score based on the World Health Organization growth reference (continuous) [
34]; and biological maturity using the maturity offset method (continuous) [
35]. Family factors consisted of annual family income (≤ $50,000, $50,001–100,000, > $100,000, or not reported) and the number of parents in the household (single or dual parent). Neighborhood environment factors included the proportion of land area within a 1 km distance of the home devoted to green space (continuous) and a traffic volume index, which was calculated as: [(km of arterial roads * average daily vehicle counts on arterial roads in Kingston) + (km of collector roads * average daily vehicle counts on collector roads in Kingston) + (km of local roads * average daily vehicle counts on local roads in Kingston)] / total road distance in buffer in km (continuous). Finally, characteristics of the day the sleep and physical activity measures were obtained including the type of day (school day, a non-school day in which the participant was enrolled in an organized day camp program, or other non-school days such as weekends and holidays), minutes/day of daylight hours, minutes/day spent participating in the other domains of physical activity (i.e., organized sports, active transportation, physical activity during school curriculum); and minutes/day of accelerometer and GPS watch wear time during waking hours.
Statistical analysis
All statitical analyses were completed using SAS version 9.4 software. A p-value of < 0.05 was used to denote statistical significance. Standard descriptive statistics were used to describe the sample. Generalized estimating equation (GEE) models with a first order autoregressive matrix [AR(1)] were performed to assess the relationship of interest, which accounted for the repeated measures nested within participants and were adjusted for confounders. In analyses testing whether nighttime sleep was associated with OAP the following day, sleep characteristics represented the within-person independent variables and OAP the within-person dependent variable. Since ~ 20% of the days had zero values for OAP, a negative binomial distribution and a logit link were specified to adjust for the over-dispersion of non-zero values, and non-zero values were rounded to integer values to mimic count data. In analyses testing whether OAP during the day was associated with sleep characteristics the following night, OAP represented the within-person independent variable and sleep characteristics the within-person dependent variables. Three models were fit for each relationship: an unadjusted model, a partially adjusted model (adjusted for participant characteristics only), and a fully adjusted model (adjusted for participant characteristics, family characteristics, neighborhood environment, and characteristics of the measurement day).