Statistical analysis

Descriptive analyses of the primary exposure, outcomes and covariates were examined for all subjects at their first recorded visit (Table ​(Table1).1). For all analyses, repeated measures of exposures and outcomes measured concurrently over time were used to investigate the association between total mealtime electronic media use and outcome variables. Linear regression modelling using generalised estimating equations was used to account for within-subject correlation. Such models can accommodate subjects with available data at both single and multiple time points, and they do not require that all subjects have repeated measures. All subjects with at least one measure of total mealtime media use and non-HDL-cholesterol were included. A first-order autoregressive covariance matrix, AR(1), with observations ordered by subjects’ age in months, was used to account for correlations among repeated measures in all analyses⁽⁴⁶⁾. The AR(1) covariance matrix accounts for correlation over time for a subject with repeated measures. Under this correlation structure, correlations decline with increasing time between visits.

Baseline subject characteristics

In the primary analysis, the association between total mealtime electronic media use and CMR outcomes was investigated using both unadjusted and fully adjusted models. The fully adjusted model accounted for covariates that have been identified in the literature as potential confounders between mealtime media use and CMR outcomes or that have been directly associated with the outcome variables. Secondary analysis between mealtime media exposure during specific meals (breakfast, lunch, dinner and snack) and CMR outcomes was conducted using the same fully adjusted models. All analyses were stratified by a priori age groups (1–4 and 5–13 years). These age groups were selected since recent position statements on media use from both the American Academy of Pediatrics and the Canadian Paediatric Society specifically pertain to children ≥5 years^(47,48). Additionally, the majority of studies investigating mealtime television use and risk of overweight/obesity have been conducted in children aged 5 years and older⁽⁷⁾, and associations between mealtime media use and CMR markers may differ between these age groups. Interactions between mealtime media use and age group (1–4 years and 5–13 years) on each of the outcome variables were examined in models without stratification, but P-values and estimates of main effects are reported from models stratified by age groups, as this was specified a priori. For identified associations in either age group with P < 0·05, additional adjustment for zBMI was done as an exploratory analysis to determine whether the identified associations were likely independent from changes in zBMI.

In order to facilitate the inclusion of subjects with missing covariate data, multiple imputation (m = 20) was performed using the MICE package⁽⁴⁹⁾ for all adjusted models. All covariates had missing observations <15 %, with the exception of maternal BMI (17 %). Participants with zBMI values > +5 or < −5 were excluded (n 2) in accordance with WHO guidelines⁽⁵⁰⁾. The distributions of outcome variables were assessed, and both TAG and insulin were log-transformed in order to achieve normality. For all analyses, the α-error was set at 0·05 and statistical tests were two-sided. Statistical analysis was conducted using R version 3.5.1⁽⁵¹⁾.