Diet quality of children in the United States by body mass index and sociodemographic characteristics

Summary Objective The primary objective was to use the Healthy Eating Index‐2015 (HEI‐2015) to describe diet quality by categories of body mass index (BMI) and by sociodemographic characteristics within categories of BMI using a nationally representative sample of US children. Methods Dietary datasets from three cycles of the National Health and Nutrition Examination Survey (2009‐2014) were analysed for children 2 to 18 years of age (N = 8894). Using the population ratio method, mean and 95% confidence intervals for HEI‐2015 total and component scores were computed by BMI (underweight, normal weight, overweight, and obese) and by age (2‐5, 6‐11, and 12‐18 y), gender, race/ethnicity (non‐Hispanic black, non‐Hispanic white, Mexican American, other Hispanic, and other race), and family poverty to income ratio (below and at/above poverty threshold). Results HEI‐2015 mean total scores were 50.4, 55.2, 55.1, and 54.0 out of 100 points for children with underweight, normal weight, overweight, and obesity, respectively, and were not significantly different. Within BMI categories, significant differences in total and mean component scores were present for age and race/ethnicity groups. Conclusions Total and most components of diet quality did not significantly differ among child populations classified by BMI status. Within BMI categories, significant diet quality differences were found for age and race/ethnicity groups, although scores were low for all child groups. Researchers may need to address or target specific dietary components with low quality in various child populations to have the greatest effect on improving nutrition nationwide.

in childhood that follows into adulthood 3 , assessing the diet quality of children is important for addressing current and future health issues. Determining diet quality among child populations defined by weight status is crucial given the high prevalence of childhood obesity present in the United States. The relationship between diet quality and weight status in children has been studied by others, but results are inconclusive because both negative and positive associations have been reported 1,4 . Several sociodemographic factors have been associated with diet quality in children, including socio-economic status (positive association), age (negative association), sex (higher for girls), race/ethnicity (lower for black children, higher for Mexican American children), and geographic location (lower for rural) 1

| Study design and data source
The National Health and Nutrition Examination Survey (NHANES) is a programme of studies designed to assess the health and nutritional status of noninstitutionalized children and adults in the United States 11 . The NHANES interview includes demographic, socio-economic, dietary, and health-related questions. The 2009 to 2010, 2011 to 2012, and 2013 to 2014 cycles of NHANES data were used for this analysis because they were the most recent cycles with available dietary data at the time the analyses were conducted. To obtain reliable estimates for child populations, three cycles of data were used. The NHANES protocol was approved by the National Center for Health Statistics Research Ethics Review Board (protocol #2005-06 and #2011-17). Informed consent was obtained from participants 18 years of age and older. Written parental consent and child assent was obtained from participants 2 to 17 years of age.

| Dietary intake
Two days of dietary intake were collected using the US Department of Agriculture (USDA) Automated Multiple-Pass Method (AMPM), a fully computerized instrument for collecting 24-hour dietary recalls either in person or by telephone 12 . The first day's interview was conducted in person in the Mobile Examination Center. The second day's interview was collected by telephone 3 to 10 days after the first day.
Only the first day's data were used in these analyses. Proxies reported dietary intake for children 5 years of age and younger and assisted with the dietary interview for children 6 to 11 years of age. Children 12 years of age and older self-reported dietary intake. Only recalls that were coded as reliable and met the minimum criteria (first four steps of five-step AMPM were completed and food/beverages consumed for each reported eating occasion identified) were included in the analyses. Recalls were checked for consumption of human milk, which can be problematic because amounts are not quantified causing missing values for amounts of energy and nutrients from human milk.
No child reported consuming human milk on the first day of intake; hence, no recalls were excluded for this reason. Of the 9000 participants between 2 and 18 years of age with reliable dietary recalls, 106 were excluded due to missing BMI data.

| BMI and sociodemographic characteristics
Body measurements were conducted in the Mobile Examination Center by trained health technicians. BMI was calculated as weight (kg) divided by height (m 2 ), and categories were based on the Centers for Disease Control and Prevention's sex-specific 2000 BMI-for-age growth charts. The four categories are underweight (BMI < 5th percentile), normal weight (5th percentile ≤ BMI < 85th percentile), overweight (85th percentile ≤ BMI < 95th percentile), and obese (≥95th percentile). Age, gender, race/ethnicity, and family income were selfreported directly by participants 16 years of age or older and emancipated minors. For participants less than 16 years of age or those who could not answer the questions themselves, a proxy (eg, parent or guardian) provided this information. For analytic purposes, we categorized age into three groups-2 to 5 years, 6 to 11 years, and 12 to 18 years-based on mode of dietary recall (proxy, proxy assistance, and self-report) and education level (preschool, grade school, and middle/high school). Race/ethnicity included five categories-non-Hispanic black, non-Hispanic white, Mexican American, other Hispanic, and other race (including multiracial). Poverty to income ratio (PIR) was calculated by dividing family income by the Department of Health and Human Services poverty guidelines specific to the survey year. The PIR value was not computed if income was missing or insufficiently reported for computational purposes (ie, in dichotomous form as <$20 000 or ≥$20 000). PIR values at or above 5.00 were coded as ≥5.00 because of disclosure concerns (eg, potential for participant identification). For analytic purposes, we dichotomized PIR as <1.00 (below poverty threshold) or ≥ 1.00 (at or above poverty threshold).
For succinctness, children living in families with income below the poverty threshold were identified as PIR < 1 and children living in families with income at or above the poverty threshold were identified as PIR ≥ 1.

| Statistical analyses
Statistical analyses were performed using SAS software, version 9.4 (SAS Institute Inc, Cary, North Carolina). BMI and sociodemographic characteristics of the child dataset were summarized using procedures designed for complex sample surveys that can incorporate cluster, strata, and weight variables. Six-year weights were constructed for these analyses since three cycles of NHANES data were used. Based on SAS programs and examples provided by the National Cancer Institute 15 , the population ratio method was used to estimate intakes of dietary components for use in calculating HEI-2015 scores. For this method, intake of the relevant dietary constituents and energy were summed for all children in the population of interest to obtain estimates of the population's total intake, and then ratios of the constituent to energy were computed and scored 15 . Usual intake at the population level is estimated using the population ratio method. To aid in visualization of the multidimensional qualities of HEI-2015, the 13 component scores were graphed using radar charts. Component scores were converted to percentages of the maximum score such that the outer edge of the circle (or spoke) represented a score that is 100% of the maximum score for that component, while the centre of the circle represents a score of 0% for any component. Additionally, HEI-2015 mean total scores and 95% confidence intervals were graphed using interval plots. Because no direct test exists for making statistical comparisons between groups when using the population ratio method 15  between two populations, it was concluded that a significant difference did exist. Due to small sample sizes, estimates were unreliable for race/ethnicity within the underweight category, and therefore, these estimates were not reported.

| RESULTS
Anthropometric and sociodemographic characteristics of the children included in the analytic dataset (N = 8894) are presented in Table 1.
The majority (64%) of children were classified with normal weight while one-third (33%) were classified with overweight or obesity. Percentages of children increased with ascending age group. Percentages of boys and girls in the analytic sample were approximately equal. The majority of children were non-Hispanic white (55%) with other Hispanic representing the smallest racial/ethnic group (7%). Approximately three-fourths of the children lived in families that were at or above the poverty threshold.

| HEI-2015 total scores
HEI-2015 mean total scores for children by BMI category and by age, gender, race/ethnicity, and PIR within BMI category are presented in Tables 2-6  and obesity, respectively, and were not significantly different (Table 2 and Figure 1). Significant differences were present for age group within BMI category (Table 3 and Figure 2). Mean total scores were significantly higher for the youngest versus the two older age groups for children with normal weight (60.2 vs 54.0 and 52.0) and for children with obesity (59.7 vs 53.3 and 52.6). Mean total scores were significantly higher for the youngest versus the oldest age group for children with overweight (59.2 vs 51.5).
A significant difference was present for BMI category within gender for boys ( Table 4). The mean total score was significantly higher for boys with normal weight vs underweight (55.0 vs 48.2). Significant differences were present for race/ethnicity within BMI category (  The mean saturated fat scores were significantly higher for children with normal weight and obesity versus children with underweight.

| HEI-2015 component scores
Age group differences were somewhat similar among BMI categories (Table 3 and Figures 5-8). In all four BMI categories, the mean total and whole fruits scores were higher for the youngest versus one or both of the older age groups, while the total protein foods score was higher for the oldest versus one or both of the younger age groups. For children with normal weight, overweight, and obesity, the dairy, refined grains, and sodium scores were higher in the youngest versus one or both of the older age groups, while the fatty acid scores were higher in the oldest versus one of the younger age groups. For children with normal weight and overweight, the added sugars score was higher in the youngest versus the two older age groups. Additionally, for the oldest age group, the fatty acid and saturated fat mean scores were higher for children with normal, overweight, and obesity versus children with underweight.
For gender (Table 4), the mean fatty acids score was significantly higher for boys with obesity versus boys with underweight. Similarly, the mean saturated fats scores were significantly higher for boys with normal weight and obesity versus boys with underweight. The mean sodium score was significantly higher for girls with normal weight versus girls with obesity.
For race/ethnicity (Table 5, Figure 9 [normal weight], and Figure 10 [overweight]), the mean whole fruits score was significantly higher for non-Hispanic black children with normal weight versus those with overweight, while the mean added sugars score was significantly higher for other Hispanic children with obesity versus those with overweight. For children with normal weight, racial/ethnic differences in mean scores were found for all components except for total vegetables. For children with overweight, racial/ethnic differences in mean scores were found for seven components-total fruits, whole fruits, greens and beans, dairy, fatty acids, added sugars, and saturated fats. For children with obesity, racial/ethnic differences were found for two components-refined grains and added sugars.
For PIR ≥ 1 (Table 6), the mean saturated fats score was significantly higher for children with normal weight versus children with underweight.

| DISCUSSION
The study represents the most up-to-date estimates of diet quality for  between intake of fatty acids and task performance in children indicated that neither HEI-2010 scores nor n-6/n-3 fatty acid ratios were associated with BMI 16 . However, the studies were conducted in the Southeast and therefore are not nationally representative 16 . Likewise, a systematic review and meta-analysis found no relationship between reduced saturated fats intake and BMI in children 17 . Similar to the present study, a positive relationship between diet quality related to sodium intake and body weight in children has been reported in two other studies 18,19 .
Significant and similar age group differences in total diet quality and the majority of diet quality components were found in children with normal weight, overweight, and obesity. Age group differences also were similar in children with underweight, but significance was not reached likely due to smaller samples sizes. In agreement with the present study, others also have reported negative relationships between child age and intakes of fruit, whole and refined grains, and dairy (less healthful as age increases) [20][21][22] , positive relationships between age and intakes of protein foods and fatty acids (more healthful as age increases) 20 , and positive relationships between age and intakes of sodium and added sugars (less healthful as age increases) 18,20,23 . The present study's results also confirm the lack of significant relationships between age and vegetable intake reported by some researchers 21 .
Significant racial/ethnic differences in total diet quality and most components of diet quality were found, notably in children with normal weight. Others also have reported racial/ethnic differences in diet because parents report feeling more responsibility for feeding their younger children than their older children 27 . In black children, the potential for school and child care interventions to effectively improve nutrition and physical activity was found in three reviews 28 . In another systematic review of family-based and institutional nutrition interventions, the authors concluded that the most effective strategies for improving child eating habits included the use of rewards, cartoon characters promoting healthy foods, modelling by teachers, and the use of older peer educators 29 . Cultural tailoring of nutrition interventions may be particularly relevant for Hispanic child populations because food choices are affected by culturally mediated health beliefs and by popular food classifications (eg, hot-cold) 30 .
The use of nationally representative child datasets, the most recent HEI, and the population ratio method for estimating diet quality scores are strengths of the study. The population ratio method provides reasonably accurate population level estimates of usual intake means 31