Setting and Participants. This study is part of a larger natural experiment that aims to examine changes in BMI and fitness during the traditional summer vacation and during the school year for children attending a year-round school and two match paired traditional schools (28). Three schools in one urban school district in the southeastern United States participated in the current study. One school (i.e., school A) converted to a year-round schedule in the fall of 2016. The year-round calendar called for children to attend school for 9 weeks and then to take a 3-week break from school. During June and July, the traditional summer vacation, the year-round school took an extended 5-week break. This school is the only school in the school district following a year-round calendar. With the exception of the year-round calendar, the school follows all district policies and procedures, including school zoning practices (i.e., how children are assigned to attend specific schools within the district). Specifically, the decision for children to attend the year-round school is made by the district, not families, and based on home address. Two match-paired schools (i.e., schools B and C) were selected to participate because of similar school day structure, daily start and end times, student race/ethnicity, gender, number of students enrolled, age/grade levels served, percentage of students receiving free and reduced lunch, and academic test scores. Table 1 presents the demographics of the participating schools and individual participants. Table 2 displays the flow of participants throughout the study.
Power Analysis. A power analysis for the smallest detectable effect was performed using G*Power (v.3.1.9.2), and was based on the difference in change of behaviors between groups. With a total of 240 children (using a variance inflation correction factor of 1.74 to account for clustering within grades per school)(29), and according to the statistical software G*power 3.1.9.7, the study is sufficiently powered to detect a difference between intervention groups of d=0.23 with a power=80% and α=0.05. This was determined to be sufficient power as previous studies that have examined changes in obesogenic behaviors during the summer have found Cohen’s d effects of 0.21 (i.e., diet) to 0.78 (i.e., physical activity) (15-22, 30).
Procedures. Behavioral data were collected on a subset of children participating in the larger study from Spring 2018-Fall 2019. This study presents the behavioral data in addition to changes in BMI during the corresponding school years (2017-2018, 2018-2019) and summers (2018, 2019). All kindergarten through third grade (i.e., 5-8 years) students participating in the larger study were invited to participate in the behavioral data collection in the Spring of 2018. A consent letter was sent home to the parents describing the study procedures. Parents who consented to their child’s participation were asked to sign and return the letter to the school where it was retrieved by research staff. From the 254 children whose parents consented a total of 240 were randomly selected to participate in the study. Measurements commenced in the spring academic semester of 2018 (i.e., March) and were completed in the fall academic semester of 2019 (i.e., August). In the fall of 2018 a refreshment sample was recruited to replace children who dropped out of the study (e.g., did not complete behavioral measures, family moved). The refreshment sample was recruited using the same procedures described for the original sample at the participating schools and were matched on age, sex, and race/ethnicity of the children they replaced. Table 2 presents data on the number of children recruited to participate in the original sample and for the refreshment sample. Data were collected during three distinct conditions: Condition 1 when both traditional and year-round school children were attending school, Condition 2 when traditional children were on summer vacation, but year-round children were attending school, and Condition 3 when both traditional and year-round school children were on summer vacation. Figure 1 depicts the schedule of measurements. Condition 1 was collected during March of 2018 and 2019. Each measurement period for Condition 1 lasted approximately 1 month. Condition 2 and 3 were collected during late May, June, July, and early August of 2018 and 2019. Data during these conditions were collected during one extended measurement period lasting approximately 3 months (Condition 2 lasts 6 weeks, Condition 3 lasts 5 weeks). All protocols were approved by the lead author’s University Institutional Review Board.
Measures
Physical Activity and Sleep. As described previously (31), physical activity and sleep were measured using a Fitbit Charge 2TM (Fitbit Inc., San Francisco, California, USA). Fitbits were chosen because Fitbit Charge devices have be shown to provide sleep and heartrate estimates in elementary aged school children and adolescents that demonstrate good agreement with polysomnography and electrocardiography, respectively (32, 33), and wrist-worn scientific grade devices used to assess free-living sleep (34). Further, because participants can charge Fitbit devices at home and data is stored in the cloud Fitbits allowed for data collection over extended periods of time (e.g., 3-month summer vacation) without the need to replace devices due to battery or data storage limitations. Data processing was informed by the ISCOLE data processing protocols (35).
Fitbit sleep data were exported to identify child sleep episodes. For this study sleep duration, timing and quality were considered as they have all be linked to risk for overweight or obesity (11, 14). For sleep timing, sleep onset was defined as the first minute that a sleep episode began. Sleep offset was defined as the last minute that a sleep episode was recorded. Sleep midpoint was calculated by identifying the time halfway between sleep onset and sleep offset. Sleep midpoint is a common indicator of shifts in sleep timing in sleep research as it takes into account both sleep onset and offset (36, 37). For duration, total sleep time was identified as the number of minutes that the Fitbit device classified a child as asleep during a sleep episode. For quality, sleep efficiency was calculated by dividing the total sleep time by time in bed. For this paper, only nocturnal sleep was considered. Nocturnal sleep was defined as sleep onset times that occurred between 5pm and 6am and lasted for greater than 240 minutes (38). If sleep segments were separated by less than 20 minutes they were considered one continuous sleep segment (35).
To distill the heartrate data into activity intensity levels, each child’s resting heartrate was identified as the lowest mean beats-per-minute for 10 consecutive minutes each day (39-42). Resting heartrates were calculated for each child each wear day. Heartrate has been widely used to determine activity intensity in children (43). Heartrates were distilled into activity intensity levels based on percent heart rate reserve (HRR). That is, 0.0-19.9% of HRR equaled sedentary, 20.0-49.9% of HRR equaled light physical activity, and ≥50.0% equaled MVPA (44, 45). Sleep episode data were mapped onto a child’s physical activity data to determine sleep and wake times. A day with at least 10 hours of waking wear was considered as a valid day of wear (35). Valid days were distilled into total waking time children spent sedentary and in MVPA on each day.
Healthy Foods, Unhealthy Foods, and Screen time. Children’s consumption of healthy and unhealthy foods and screen time were assessed via parent proxy-report. Parents received a daily diary via text message which asked them to report their child’s screen time and foods consumed twice per week. Parents were asked to report on their child’s screen time and foods on at least 4 days during each measurement condition (i.e. Condition 1-3). Parents were encouraged to complete the diaries along with their child to enhance the accuracy of the estimates. Parents/children estimated the total amount of time (hours and minutes) spent in front of a screen that day (e.g., TV, computer, video game, smartphone, and tablet) (46, 47). Similar to past studies (22, 48), healthy and unhealthy foods were assessed using the Beverage and Snack Questionnaire (49). Items were scored by four possible response categories: 0 (‘child did not consume’), 1(‘child consumed a little’), 2(‘child consumed some’), and 3 (‘child consumed a lot’) with those individual items. For this study, individual food items were grouped in accordance with the Healthy Meal Index (50). Food categories included: fruits, vegetables, dairy (non-sugar sweetened), convenience foods, sweets and desserts, and sugar sweetened beverages (including dairy), water. Two groups were created for analysis of foods consumed: healthy foods/drinks (fruits, vegetables, and unsweetened dairy, water), and unhealthy foods/drinks (convenience foods, and sweets/desserts, sugar sweetened beverages). Consumption was dichotomized (i.e., ‘did’ vs. ‘did not’ consume) and reported as mean days per week that a healthy or unhealthy food/drink was consumed (49).
Body Mass Index. Changes in children’s heights and weights were measured for the 2017/18 school year (August 2017-May 2018), the 2018 summer (May-August 2018), the 2018/19 school year (August 2018-May 2019), and the 2019 summer (May-August 2019). All measures in both the year-round and traditional schools were based upon the traditional school calendar and occurred during the same two-week period. Measures were completed during the last (end of school year) or first (beginning of school year) two weeks of the traditional school year. All measures were obtained during regularly scheduled physical education class. Using a portable stadiometer (Model S100, Ayrton Corp., Prior Lake, Minn.) and digital scale (Healthometer model 500KL, Health o meter, McCook, Ill.), children’s heights (nearest 0.1 cm) and weights (nearest 0.01 lbs.), without shoes, were collected by research assistants. BMI was calculated (BMI= kg/m2) and transformed into age and sex specific z-scores (zBMI).(51)
Statistical Analyses.
All analyses were completed in Stata (v14.2, College Station, TX) during April of 2019. Prior to completing the primary analyses descriptive means and standard deviations of school and child characteristics were examined. To examine the differences in obesogenic behaviors during the school year and summer break from school, multi-level mixed effect linear regressions, with days nested within children, were estimated. Separate models were estimated for each variable related to the four measured obesogenic behaviors including (1) sedentary time, (2) light physical activity, (3) MVPA, (4) screen time, (5) unhealthy foods index, (6) healthy foods index, (7) total sleep time, (8) sleep midpoint, and (9) sleep efficiency. A three-level condition (0=Condition 1, 1=Condition 2, 3=Condition 3) and two-level school calendar (i.e., 0=traditional or 1=year-round) variable were considered the independent variables. A condition by school calendar interaction was also included to test our hypotheses that behaviors would change differently between groups from one condition to the next. In order to ensure that comparisons were between school days and summer break days, weekend days and school break days (i.e., spring break, teacher workdays) were excluded from all models.
Monthly zBMI change was also examined among the participating children for the corresponding school and summers using multi-level mixed effect linear regressions, with repeated measures nested within children. Monthly zBMI change was the dependent variable with a two-level condition (school vs. summer), two-level school calendar (traditional vs. year-round) variable, and the condition-by-school calendar interaction as the independent variables. Monthly zBMI change was used to standardize change during the summer (i.e., 3 months) and school year (i.e., 9 months).
All statistical models included sex, race/ethnicity, grade, academic year, and refreshment status (original vs. refresh participants) as covariates. Physical activity models included wear time as an additional covariate. The Benjamini-Hochenberg procedure with a false discovery rate of 10% was utilized to account for multiple comparisons (52).