Participants
The present study was conducted in Xi'an, Hanzhong, and Yanan, China, between March 2017 and October 2018. To attain a representative population of school-aged children with difference in body weight, a random sampling technique was used to select schools located at different areas in these cities. The detailed study design is presented in Figure 1.
Caregiver was defined as someone who takes care of a child on a daily basis. The current study used the following criteria to include participants: 1) the child was between 6 and 12 years old, 2) the child did not suffer from any disease that might influence the child’s appetite or eating behaviors in the past month, and 3) the caregiver of the child provided informed consent to take part in the survey. We excluded caregivers who were uneducated or not willing to participate.
The Ethics Committee of the Fourth Military Medical University approved the current study. Written informed consent was provided by all recruited caregivers prior to the study, and the study complied with related regulations and guidelines. Data was collected anonymously.
Information on the age, sex, height, and weight of a child, and the caregiver’s levels of education, family structure, place of residence, and (family per capita) monthly income was collected using structured questionnaires. We also evaluated eating behaviors of children using the scale developed in the current study, namely the Chinese School-aged Children’s Eating Behavior Scale (CSCEBS). Prior to data collection, we explained the aims, procedures, methods, each item’s meaning, and the significance of the study, and we also gave instructions for filling out questionnaires.
Development of the Conceptual Model and the Draft Scale
We reviewed literature in English and Chinese published in the past 30 years, conducted two qualitative interviews, and developed a ten-factor conceptual model that systematically summarizes the eating behaviors of school-aged children (6 - 12 years old). During the first in-depth interview with 20 caregivers, we collected information on school-aged children's eating behaviors to the greatest extent possible using qualitative interviews. We then summarized the eating behaviors from the first interview and used the information in the second interview. The second interview was conducted with 30 caregivers and was structured as a focus group discussion examining the generalizability of items in the outline. Results from the second interview were used to build the conceptual model that included 10 dimensions: food fussiness, responsiveness to food, responsiveness to satiety, bad eating habits, external eating, emotional eating, independent and initiative eating, enjoyment of food, restrained eating, and junk food addiction.
Subsequently, 108 items were identified from the conceptual model and previous questionnaires (i.e., the DEBQ-C [14, 15], CEBQ [13], the Children Eating Behavior Inventory [16], the Oregon Research Institute Child Eating Behavior Inventory [17], Children’s Binge Eating Disorder Scale [25], Mealtime Behavior Questionnaire [26]). In addition, a third interview was conducted among six caregivers and three nutrition experts to combine the characteristics of Chinese eating culture, and 60 additional items were identified capturing the ten-factor conceptual model. We used a forward-backward procedure to translate items from existing instruments. A nutritionist and two bilingual professional translators conducted the forward (English to Chinese) and backward (Chinese to English) translation. One psychologist and four nutrition experts further evaluated the content validity (relevance, clarity, and ambiguity of items) of the Chinese scale version. As a result, we formed an item pool of 168 items.
Fifty caregivers of school-aged children and 10 experienced pediatricians reviewed the item pool and critically evaluated each item, including the importance of each item (correlations with eating behaviors) and the frequency. Each item ranged from 1 (not very frequent or important) to 5 (very frequent or important). Therefore, 60 scores for frequency and 60 scores for importance were scored for each item. A higher mean score of an item indicates increased importance or frequency. We further deleted items that had low frequency or importance (< 50th percentile). After the review, 55 items were removed due to low frequency or importance. Ultimately, a draft of the Chinese School-aged Children’ Eating Behavior Scale (CSCEBS) was created, which included 113 items.
Methods for Scoring
Each item of the scale measures the frequency of the corresponding eating behavior over the past two months. Five options were given for each item (“never”, “rarely”, “sometimes”, “often”, and “always”), and a corresponding number ranging from 1 to 5 was assigned respectively. Negative scores were given for reverse items. The mean score was calculated by dividing the total of all items by the number of items answered in each dimension. The score of the scale was the total of the scores in each dimension. A greater score for each dimension suggested a greater likelihood of children with this eating behavior.
Investigation Methods
The CSCEBS was administered to the main caregiver who had been feeding the child for more than 1 year and was very familiar with the child's daily diet. Five trained investigators were responsible for administering the questionnaire. The investigators first clarified the aim and process of the assessment, as well as the meaning of the questionnaire, to caregivers. Second, the height and weight of the child and the caregiver were measured using calibrated equipment (JT-918) by the investigators. Subsequently, caregivers completed the questionnaires to report their children’s eating behaviors over the past 2 months, and returned the filled questionnaires to the investigators.
First Investigation: Establishing the Trial Scale
The first investigation included 140 caregivers from two kindergartens in urban and suburban Xi'an using the stratified sampling methods. The first draft of the CSCEBS was completed by the caregivers independently, and was used for analyzing items of the draft scale.
Second Investigation: Constructing the Final Scale
The second investigation included 400 caregivers and utilized the same method and criteria from one urban and one suburban kindergarten in Xi’an, Hanzhong, and Yanan, respectively. The caregivers finished the trial scale of the CSCEBS independently for constructing the final CSCEBS.
Third Investigation: Assessing the Final Scale
The third investigation included 700 caregivers from two suburban and three urban kindergartens in Xi’an, Hanzhong, and Yanan. The caregivers completed the final scale independently, and the dimensions of the scale and the reliability and validity were assessed. To test the test-retest reliability, 120 caregivers were randomly selected to complete the scale again after two weeks.
Methods for Quality Control
The investigators carefully checked all questionnaires and conducted telephone interviews when spotting any missing information. Valid data from all completed questions were entered by using EpiData software. Double-entry and random check were used to ensure the data accuracy. SPSS was used to perform data analysis.
Body Mass Index (BMI) Classification
BMI (kg/m2) was calculated by measuring weight and height, and was classified separately for children and caregivers. For children, overweight and obesity were defined using the Chinese guideline for children aged younger than 18 years of age [27]. Three groups were created: thinner weight (age- and sex-specified BMI < 10th percentile), normal weight (BMI ≥ 10th percentile to < 85th percentile), overweight/obesity (BMI ≥ 85th percentile).
Data Analysis
Item Analysis
First, items that received the highest or lowest scores from over 15% of the caregivers indicated ceiling or floor effects, respectively [28], and were thus discarded. Second, we converted the reverse scoring items accordingly (5 = 1, 2 = 4, 4 = 2, and 1 = 5). Subsequently, we used five methods to select items [29]: the critical ratio analysis method, the discrete trend method, the correlation coefficient method [30], the exploratory factor analysis method [31], and the Cronbach’s α coefficient method [32]. The details of the five methods have been described previously [29]. Based on these methods, an item was deleted when it met exclusion criteria of three or more methods; if an item met the exclusion criteria of two methods, it was discussed with experts to decide whether it should be deleted or merged.
Reliability Analysis
The Cronbach’s α coefficient, test-retest reliability coefficient, and split-half reliability coefficient were used for reliability testing. If coefficients were ≥ 0.70 and 0.60 [32] for total scale and dimensions, respectively, then the reliability was considered satisfactory.
Validity Analysis
We split the samples into half and used exploratory factor analysis and confirmatory factor analysis to explore and validate the structure [33]. A few fit indices were used to assess how well the model fit the data [31, 34]. The details have been described previously [31, 33-35]. In brief, the standardized root mean squared residual (SRMR) > 0.08, the adjusted goodness-of-fit index (AGFI) and the goodness-of-fit index (GFI) > 0.90, the comparative fit index (CFI) and the non-normed fit index (NNFI) > 0.95, the χ2/df < 5, and the root mean square error of approximation (RMSEA) < 0.05 indicated good model fit [31].
Discrimination Analysis
We used the Student’s t-test to compare scores of various dimensions between sex and place of residence. We used one-way analysis of variance to compare scores by age, weight, education level of caregiver, monthly income, and family structure. We used SPSS to perform all statistical analyses. Continuous variables are presented as the mean ± standard deviation (), and categorical variables are expressed as frequencies and percentages. Two-sided P values < 0.05 indicate statistical significance.