Participants
Participants were from the Parents’ Role in Establishing healthy Physical activity and Sedentary behaviour habits (PREPS) project. From October, 2014 to December, 2015, parents and their toddlers were recruited during routine 18-month immunization appointments at one of four public health centers in Edmonton, Canada. These health centers where chosen because they were large and busy centers that served diverse communities. To be eligible for the study, children needed to be ambulatory, and one parent needed to be comfortable speaking and reading in English. Of the 491 eligible families approached, 52% or 257 families agreed to participate in the baseline portion of the study. The reasons why families did not agree to participate are published elsewhere [19].
At baseline (time 1, 2014–2015) of the PREPS project, the majority of participating parents (n = 242 or 94%) indicated that they would be interested in being contacted for future research. In 2015, funding was secured to add 1-year (time 2, 2015–2016) and 2-year (time 3, 2016–2017) follow-ups to the PREPS project. Of the 242 eligible participants, 32 were not contacted at times 2 or 3 due to challenges associated with accelerometer return at time 1. Of the 210 remaining participants at time 2, 99 agreed to participate, 21 declined (17 agreed to stay on contact list for time 3 and four asked to be removed), 66 were not reachable (e.g., telephone or e-mail no longer working, no response after multiple attempts), and 24 were missed by error. Data were collected between October, 2015 and September, 2016. Of the 206 remaining participants at time 3, 92 agreed to participate, 30 declined, and 84 were not reachable. Data were collected between October, 2016 and December, 2017. Ethics approval was obtained from the University of Alberta Human Research Ethics Board and all participating parents provided written informed consent at all three time-points.
Procedures
At time 1, eligible participating parents completed the PREPS questionnaire and were given an accelerometer and pre-paid courier return envelope during the 15-min wait period that is required after all immunizations. At times 2 and 3, after parents agreed to participate via phone or e-mail, study materials, including a follow-up PREPS questionnaire, accelerometer, and pre-paid courier return envelope were mailed to participants. At all three time-points, questionnaires were checked for missing data and parents were followed up when necessary. Additionally, mid-week reminders on the accelerometer procedures were provided to parents via e-mail. Finally, after the accelerometer was returned, a $25 gift certificate was mailed to families.
Physical activity and sedentary behavior
At all three time-points, sedentary time and physical activity were measured with Actigraph wGT3X-BT accelerometers (ActiGraph Corp, Pensacola, FL, USA). Parents were informed that for the entire week their child should wear the accelerometer over their right hip, except during overnight sleep and water-based activities (e.g., swimming, bathing). Data were collected in 15-s epochs. For participants’ data to be included at each time-point, they were required to have ≥4 days with ≥1440 total 15 s intervals (equivalent to ≥6 h) of wear time. Previous studies have shown that these wear time parameters provide reliable estimates (Intra class correlation [ICC] = 0.70–0.80) of physical activity in toddlers and preschoolers [20, 21]. Non-wear time was defined as ≥80 consecutive 15-s intervals of zero counts (equivalent to ≥20 min of consecutive zeros counts) [22]. Daytime naps were assumed to be removed with non-wear time. For wear time data, sedentary time was defined as 0–24 counts per 15 s, light-intensity physical activity (LPA) as 25–420 counts per 15 s, and moderate- to vigorous-intensity physical activity (MVPA) as > 420 counts per 15 s. When compared to direct observation, these cut-points have shown fair to excellent validity in toddlers and preschoolers (Receiver operating characteristics – Area under the curve [ROC-AUC] = 0.72–0.90) [23, 24]. Minutes per day of sedentary time, LPA, and MVPA were calculated by dividing the number of 15-s intervals for sedentary time, LPA, and MVPA by 4 and then dividing by the total number of valid days. To adjust for wear time, standardized sedentary time, LPA, and MVPA variables were calculated at each time point [25].
At all three time-points, parents reported their toddler’s average screen time through four items in the PREPS questionnaire. Specifically, parents reported the average hours and minutes per weekday and weekend day that their toddler: 1) watches television, videos, or DVDs on a television, computer, or portable device; 2) plays video/computer games on devices such as a learning laptop, leapfrog leapster, computer, laptop, tablet, cell phone, the internet, Playstation, or XBOX. Weighted averages ([weekday*5 + weekend*2]/7) were calculated for television/videos and video/computer games variables, and minutes per day of toddlers’ screen time was calculated by summing the weighted averages. These screen time items were modified from a national survey in Canada [26], and had good 1-week test re-test reliability (Intra-class correlation [ICC] = 0.82) in a sub-sample of PREPS participants [27].
Social skills
At times 2 and 3, parents reported their toddler’s social skills using the Adaptive Social Behavior Inventory (ASBI) [28, 29] as part of the PREPS questionnaire. Parents did not report on the ASBI at time 1 because the tool is considered developmentally appropriate for children aged approximately 2.5 to 5 years [29]. The ASBI includes 30 items, each with three response options (“rarely/never,” “sometimes,” or “always”), which form three sub-scales: express (e.g., understands others’ feelings; will join a group of children playing; 13 items), comply (e.g., helpful to other children; is calm and easy going; 10 items), and disrupt (e.g., teases other children; is bossy; 7 items) [28, 29]. Items were summed to create each sub-scale score. In this sample, internal consistency reliability was α = 0.79, 0.81, 0.65 at time 2 and α = 0.78, 0.76, 0.53 at time 3 for express, comply, and disrupt sub-scales, respectively. The lower alpha for the disrupt score is consistent with the study that developed the ASBI [28]. Specifically, due to the low number of items in the disrupt score with factor loadings of ≥0.40, two items with factor loadings of ≥0.35 were included [28].
Covariates
Based on previous research [6,7,8], child age at all three time-points, child sex at time 1, and parental education at time 1 were included as covariates in the analyses. Questionnaire completion date at times 2 and 3, child birthdate, child sex, and parental education were reported by parents as part of the PREPS questionnaire. Child age (years) was calculated by subtracting the exam date (time 1) or questionnaire completion dates (times 2 and 3) from the birthdate. Parent education was the highest grade or level of education of the parent completing the questionnaire. There were six response options ranging from “no schooling” to “post-graduate”.
Statistical analyses
Statistical analyses were completed using SAS version 9.4 (SAS Institute Inc., Cary, NC). Descriptive statistics were calculated, and differences on time 1 demographic, physical activity, and sedentary behavior variables between participants included and excluded at time 2 and 3 were examined using t-tests or Wilcoxon Rank Sum tests and chi-squared statistics. To examine the longitudinal associations of each physical activity and sedentary behavior variable with social skills, generalized estimating equations (GEE) were conducted that adjusted for child age, child sex (time 1), and parental education (time 1). GEE uses all available data, to produce a single regression coefficient, which represents pooled cross-sectional (between-subjects) and longitudinal (within-subject) associations [30]. To correct for the non-independent repeated measures data, GEE uses an a priori determined correlation structure [30]. An exchangeable correlation structure was used for all models addressing objective one [30]. Sedentary behavior and physical activity variables were expressed as 10 min/day within these models to make the interpretation of the regression coefficients more meaningful. The disrupt time 2 and 3 variables were log transformed to meet the assumption of normality. When observations were identified as potential influential cases by examining Cook’s distance values and data entry errors were ruled out, models were run with and without those observations to determine if findings differed.
To examine how physical activity and sedentary behavior track over three time-points, longitudinal tracking coefficients (β1) were calculated using GEE [30]. Z-scores were first calculated for sedentary behavior and physical activity variables so a standardized longitudinal tracking coefficient could be obtained. Then the time 1 sedentary behavior or physical activity variable was regressed on the corresponding longitudinal sedentary behavior or physical activity variables from time 2 to time 3 [30]. An unstructured correlation structure was used for all models addressing objective two [30]. Tracking coefficients were defined as low (β1 < 0.30), moderate (β1 = 0.30–0.59) moderate-high (β1 = 0.60–0.90), and high (β1 > 0.9) [31]. Statistical significance was defined as p < 0.05.