Participants
This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines [20]. Participants were N = 2983 children in British Columbia (BC), Canada, who had participated in the Childhood Experiences Questionnaire (CHEQ; caregiver-report) Project and in the Early Development Instrument (EDI); teacher report) project in the 2019/20 school year. Primary caregivers’ responses on the CHEQ, collected in September 2019 [21] were linked to teachers’ responses on the EDI, collected in February 2020 [22]. The majority of primary caregivers were mothers (N = 2421, 81.2%) or fathers (N = 427, 14.3%); from here on they will therefore be referred to as parents. The CHEQ measures child and family demographics and children’s health behaviors at Kindergarten entry; the EDI measures children’s developmental health midway through the Kindergarten year. Detailed information on sampling for school districts for the CHEQ has been described elsewhere [23].
Children in the sample came from 144 schools within nine school districts in BC (there are a total of 60 school districts and 1038 public elementary schools in BC) [23]. The average participation rate for the CHEQ across school districts was 57%. The CHEQ-EDI linkage rate was 98.3%. The mean age of children was 5.2, SD = 0.3 years at the time of the CHEQ implementation and 5.6, SD = 0.3 at the time of the EDI implementation. Individual child-level data on the CHEQ were linked to their EDI survey by Population Data BC via unique identifiers (i.e. Personal Education Numbers) [24]. Data were accessed via a secure server at Population Data BC [25].
Measures
Information on all explanatory variables (i.e., screen time, sleep, physical activity ethnicity, sex, annual household income) was obtained from the CHEQ. Screen time was the main exposure variable of interest. Sleep, physical activity, family ethnicity, child sex, family annual household income, and rural versus urban community were included as covariates based on their role as potential confounders in predicting developmental health [17, 26–29]. Information on outcome variables (i.e., physical health/wellbeing, social competence, emotional maturity, language and cognitive development, communication skills) was obtained from the EDI. Descriptive statistics for all variables can be found in Table 1.
Table 1: Descriptive Statistics on Demographics and Health Behaviors Among Preschool-Aged Children in British Columbia, Canada
|
Variables
|
Total sample
N (%)
|
Explanatory Variables on the CHEQ
|
|
Screen time
|
> 1 hour
|
1183 (39.7%)
|
|
≤ 1 hour
|
1734 (58.1%)
|
|
Missing
|
66 (2.2%)
|
Sex
|
Male
|
1558 (52.2%)
|
|
Female
|
1425 (47.8%)
|
Ethnicity
|
European origins
|
1602 (53.7%)
|
|
Non-Euro. origins*
|
1048 (35.1%)
|
|
Missing
|
333 (11.2%)
|
Annual family income
|
< $75,000
|
826 (27.7%)
|
|
≥ $75,000
|
1640 (55.0%)
|
|
Missing
|
517 (17.3%)
|
Population Centre
|
Urban
|
1596 (53.5%)
|
|
Small to medium
|
1387 (46.5%)
|
Physical activity
|
Nonparticipant
|
526 (17.6%)
|
|
Participant
|
2416 (81.0%)
|
|
Missing
|
41 (1.4%)
|
Duration of sleep
|
< 10 hours
|
303 (10.2%)
|
|
≥ 10 hours
|
2616 (87.7%)
|
|
Missing
|
64 (2.1%)
|
Dependent Variables on the EDI
|
|
Physical health and wellbeing
|
Not vulnerable
|
2521 (84.5%)
|
|
Vulnerable
|
319 (10.7%)
|
|
Missing
|
143 (4.8%)
|
Social competence
|
Not vulnerable
|
2460 (82.5%)
|
|
Vulnerable
|
377 (12.6%)
|
|
Missing
|
146 (4.9%)
|
Emotional maturity
|
Not vulnerable
|
2408 (80.7%)
|
|
Vulnerable
|
427 (14.3%)
|
|
Missing
|
148 (5.0%)
|
Language and cognitive development
|
Not vulnerable
|
2580 (86.5%)
|
|
Vulnerable
|
244 (8.2%)
|
|
Missing
|
159 (5.3%)
|
Communication skills
|
Not vulnerable
|
2544 (85.3%)
|
|
Vulnerable
|
295 (9.9%)
|
|
Missing
|
144 (4.8%)
|
Note. N = 2983. * Non- European origins include families with Indigenous, East Asian, South East Asian, South Asian, Latin America, Middle East, African background
Screen time. Screen time was assessed by asking parents: “Think about the past 6 months, on average, how much time per day did your child use an electronic device like a tablet, smartphone, TV or computer (alone)?”. Response options were 1 = None; 2 = < 15 minutes; 3 = 15 minutes to 1 hour; 4 = 1-2 hours; 5 = > 2 hours. Based on the 24-hour movement guidelines [1] for the early years, screen time was categorized into 0 (≤ 1 hour) and 1 (>1 hour).
Physical activity. Physical activity was assessed by asking parents “In the last 6 months, about how many times per week did your child take part in energetic physical activity while participating in organized activities (for example, swimming lessons or gymnastics lessons)?” Response options were 1 = Never; 2 = Once a week or less; 3 = 2-3 times week; 4 = 4-5 times a week; 5 = 6-7 times a week. For subsequent analyses, responses were grouped into 1 = participant (participation at least once per week) and 0 = non-participant (never participated). Distinguishing between participants and nonparticipants is a commonly used indicator for preschool aged children’s involvement in organized physical activity in absence of objective record of number of minutes spent in a day [30].
Sleep. Sleep was assessed by asking parents “How many hours does your child usually sleep in a 24-hour period (Combining night time sleep and naps)?” In response parents indicated number of hours. Based on the 24-hour movement guidelines [1], responses were grouped into 1, indicating optimal sleep duration (i.e., ≥ 10 hours) and 0, indicating less sleep than recommended (i.e., <10 hours).
Demographics. Families indicated their ethnicity by endorsing one or more of nine ethnic categories they were presented with. The category indicating European origins was checked by 53% of families. A binary ethnicity variable was therefore created for subsequent analyses: 0 = European origins; 1 = Non-European origins. Income was assessed by asking parents to indicate their best estimate of their family’s overall household income in the previous year before taxes: 1 = under $20,000; 2 = $20,000 to $49,999; 3 = $50,000 to $74,999; 4 = $75,000 to $99,999; 5 = $100,000 to $149,999; 6 = $150,000 to $199,999; 7 = $200,000 or more. For further analyses, we created a binary variable indicating whether household income was less than $75,000 versus $75,000 or more. This cut-off was determined based on the 2019 living wages for families of two parent household in Metro Vancouver where both parents were full-time ($19.50/hour, resulting in approximately $75,000 before taxes) [31]. Statistics Canada's criteria were used to categorize school districts into urban (i.e. population of 100,000 and more) and rural (small and medium population centers i.e., <100,000) [32].
Developmental health. Children's developmental health was assessed in five domains: physical health and wellbeing (13 items); social competence (26 items); emotional maturity (30 items); language and cognitive development (26 items); and communication skills (8 items). Responses on these domains were assessed on 2- or 3- point Likert scales (i.e., yes, no, don't know; very true, sometimes or somewhat true, never or not true, don't know). In alignment with previous research with the EDI [33], all responses on binary items were coded 0 or 10; 3-point Likert-scale items were coded 0, 5, and 10; and 5-point Likert-scale items were coded 0, 2.5, 5, 7.5, and 10. All items contained an additional response option, I don't know (coded 99), which was not included in the statistical analyses. For every item, 10 designates the highest (i.e., most positive, most developmentally desirable) score. For every domain, the average score was calculated for each child, ranging from 0 to 10. Subsequently, children’s domain specific scores were converted into a dichotomous measure of vulnerability (0 = not vulnerable; 1 = vulnerable). As in previous research with the EDI, vulnerability for a given developmental domain was indicated by a child’s score being below a domain-specific cut-off score that marked the bottom 10th percentile in the normative distribution of kindergarten children, based on the first cycle of data collected in BC [29, 34, 35]. The scoring procedure for the EDI was developed by the instrument's authors in collaboration with educators and knowledge users [29, 34, 35]. Previous research has supported the use of vulnerability scores in research with the EDI [36, 37]. The EDI’s validity has been supported in previous research, and good interrater reliability has been established [38–40]. In the present study, the internal consistency was satisfactory for all domains (Cronbach’s alphas > .86), except for physical health (Cronbach’s alpha = .45).
Statistical analyses
Exploratory and bivariate analyses were conducted among all explanatory and dependent variables. To examine the differences in children exceeding screen time recommendations based on family income, Pearson’s Chi-square test was used [41]. Next, to examine the association between screen time and the five developmental health outcomes, five logistic regression models with generalizing estimating equations (GEE) were built (‘xtlogit’ command with ‘population average’ option in STATA). The GEE were used as estimators of variance, employing an exchangeable correlation structure to account for clustering in the data (i.e., children nested within schools) [42, 43]. School ID was used as a cluster variable. The school-level variability on developmental health outcomes ranged from ICC = 0.12 to ICC = 0.14.
Of 2,983 children in the linked sample, 853 had missing values on one or more variables; 165 had missing information on at least one developmental health domain (see Table 1). Based on bivariate analyses with missing indicator variable, information was presumed to be 'missing at random' (MAR) and multiple imputation (MI) using chained equation approach was used to impute missing values [44]. The process was repeated m=28 times [45, 46]. At the final stage, following Rubin's rules, [47] parameter estimates were pooled. Imputed values for outcome variables were excluded post imputation under 'multiple imputation then deletion' (MID) approach [48]. As a result, 2818 children with complete information on outcome variables were included in the GEE analyses.
In building the multivariable models, all the covariates were entered simultaneously into the model based on their role as potential confounder and bivariate association with the outcome [17, 26–29]. Unadjusted and adjusted odds ratios (ORs) and corresponding 95% confidence interval (CIs) were reported for each developmental health outcome. A two-way interaction between screen time and income was explored using analysis of variance (ANOVA). To support the stability of the findings, sensitivity analyses were conducted, using linear regression models predicting scores on the five developmental outcomes in their continuous, non-dichotomized form. A sensitivity analysis was also performed on the sub-sample of children with complete data before the imputation of missing values (N = 2130) to examine whether results were comparable. All analyses were completed using STATA version 16 [49].