Study design and participants
The study was conducted in local kindergartens, that is, schools that provide normal education to children aged three to six in the Dongcheng District, the eastern half of the downtown area of Beijing. The study was approved by the Ethics Committee of Dongcheng Center for Disease Control and Prevention (DCCDPCIRB-20180416-1).
As children in the third year of kindergarten are going to attend primary schools, eligible participants were children registering in the first year and the second year of the chosen kindergartens at the time of recruitment. One guardian of these children was also required to participate.
Sampling and survey procedures
Taking each class as a cluster, a stratified proportionate cluster sampling was used. The number of classes to recruit was determined by a sample size estimation formula:
Assuming the prevalence of overweight/obesity among children in the first year and the second year of kindergarten in the Dongcheng District (P) was 15% (estimated through a pilot study), a two-sided significance level of 5% ( = 1.96), a minimally detectable rate difference (δ) of 1.5%, and a 10% non-response rate and a 1.5 design effect of cluster sampling (k = 1.65), the number of child-guardian dyads needed was calculated to be 3,592. On the basis of 20 to 30 children per class, and 4 to 10 classes of the first year and the second year per kindergarten, approximately 150 classes of 20 kindergartens were required. A full list of the forty-four kindergartens in the Dongcheng District was extracted from the local education bureau, and stratified by financing sources and implementation of health promotion activities. Random numbers were generated to select kindergartens by stratification. Principals of the selected twenty kindergartens were contacted prior to the survey for their approval of participation, and fifteen agreed. With written informed consent from themselves and their legal guardians, all of the 4,237 children in classes of the first year and the second year from the fifteen kindergartens, along with one of their guardians, were invited to participate from April 2018 till the end of the Spring Semester of 2017/2018 Academic Year. Questionnaires with unique ID numbers were distributed by teachers in charge of classes who had received standardized training and were responsible for providing necessary instructions to guardians. Guardians then took questionnaires home and completed them anonymously on behalf of their children, considering preschool children were not capable of reading and writing. One week later, questionnaires were collected and preliminarily checked by teachers in charge of classes. Blank questionnaires were permitted if children or guardians refused to participate. A total of 3,585 child-guardian dyads participated in the survey, with a response rate of 84.61% (3585/4237), and incomplete information resulted in 212 dyads further eliminated from analyses. Figure 1 provides details on sampling procedures.
Weight and height
Children’s weight and height were measured annually (usually at the end of a semester, which is close to the survey date) by trained school nurses following standardized anthropometry measurement protocols developed by the Child Care Center of Bejing Health Bureau to the nearest 0.1 kg and 0.1 cm, with children wearing light clothes and no shoes. Children's weight and height were also attached to unique ID numbers, enabling matching with their questionnaire data. Children’s body mass index (BMI) was calculated, and weight status was defined by criteria developed by the World Health Organization (WHO) (22-24): the children aged three to four whose BMI z-scores greater than 2 standard deviations (SDs) and 3 SDs from medians of specific age and gender groups were classified as overweight and obesity, respectively, and for children aged five, the cut-points were 1 SD and 2 SDs.
Dietary consumption frequencies
In reference to The Dietary Guidelines for Chinese Preschool Children 2016 (25), a validated food frequency questionnaire (FFQ) (26) was adapted and developed to measure children’s usual consumption frequencies of twenty-five food and beverage groups frequently consumed by the Chinese with seven alternatives (never, less than once a week, once a week, twice to four times a week, five to six times a week, seven times a week, and more than seven times a week) scored 0 to 6, and several examples were annotated for some confusing groups to improve interpretability. Portion sizes were not collected, as better validity correlations were obtained in FFQs not assessing them (27). Additional file 1 is an English version of the FFQ. Food and beverage groups were deemed to be “consumed” when their consumption frequencies were between less than once a week and more than seven times a week. Test-retest reliability of the FFQ was pretested in a pilot study among a convenience sample of 175 pairs of preschool children and guardians, and the results are available in Additional file 2. Given the low consumption frequencies and poor reliability of coffee drinks and energy drinks or sports drinks, they were eliminated from the analyses of dietary patterns and their associations with overweight/obesity. In that condition, reliability of the adapted FFQ was moderate and comparable to the original one (26).
Age, gender, daily average time of moderate-to-vigorous physical activities (MVPA) on weekdays and weekends, and daily average sedentary time on weekdays and weekends of the children, as well as the highest level of educational attainment, occupation, weight, and height of their parents were also collected via single questiones. Parents’ SES scores (28) and BMI were calculated, and their weight status was defined by criteria developed by the Working Group on Obesity in China (WGOC) (29): BMI of 24 kg/m2 and 28 kg/m2 were taken as cut-points for overweight and obesity, respectively.
Normally-distributed continuous, skewedly-distributed continuous, and categorical variables were presented as means (SDs), medians (25th and 75th percentiles), and frequencies (percentages), respectively. Dietary patterns were identified through an exploratory factor analysis (EFA) (30) whose applicability was confirmed by the Bartlett’s test of sphericity and the Kaiser-Meyer-Olkin (KMO) value. The number of retained factors was determined by eigenvalues (greater than 1) and interpretability, and factors were rotated with an orthogonal (varimax) rotation to minimize mutual correlations and improve interpretability. Factor loadings represented correlations between each food and beverage group and each factor – loadings greater than 0.3 stood for positive correlations, while those less than -0.3 for negative. In accordance with descriptions of these correlations, each factor was named a specific dietary pattern. Through weighting standardized consumption frequencies of food and beverage groups by their factor loadings and adding up all these values, pattern-specific factor scores were calculated for each child, and the dietary pattern with the largest factor score was defined as the predominant dietary pattern. Differences across predominant dietary patterns in children’s age were tested by one-way analysis of variance, in children’s gender and consumption proportions of food and beverage groups by Pearson’s chi-square tests, and in children’s BMI and weight status, parents’ SES scores and weight status, and consumption frequencies of food and beverage groups by Kruskal-Wallis tests. Considering the multi-stage sampling procedure adopted where classes were taken as clusters, two-level random-intercept logistic models with cluster-robust standard errors, extensions of standard logistic models that treated intercepts as random variables to account for the clustering of one-level units (children) within two-level units (classes) (31), were conducted to estimate associations between predominant dietary patterns and overweight/obesity, a binary outcome. A null model without any independent variable was run firstly to verify the hierarchy of data, and subsequently, the first model (model 1) only including predominant dietary patterns, the second model (model 2) including adjustment for children’s age and gender, the third model (model 3) including further adjustment for children’s daily average time of MVPA and daily average sedentary time, on weekdays and on weekends respectively, and the fourth model (model 4) including further adjustment for parents’ SES scores and BMI were fitted. Correlations were expressed as odds ratios (ORs) with 95% confidence intervals (CIs). Please refer to Additional file 3 for variable definitions and Stata codes of these models. Lastly, a sensitivity analysis was carried out in the same way as model 4, but defining overweight/obesity by criteria specific to Chinese preschool children (32). Data imputation was applied in the EFA and the two-level models by using means, medians, and modes to replace missing data in normally-distributed continuous, skewedly-distributed continuous, and categorical variables, respectively. All the analyses were completed using Stata/SE 16.0 for Windows (StataCorp, College Station, Texas, USA). Statistical significance was considered when P ≤ 0.05 (two-sided).