Socioeconomic Status and Other Factors Associated with Childhood Obesity

Introduction:

Childhood obesity is a critical public health issue, with one-third of all children and adolescents in the U.S. being either overweight or obese.¹ Children who are overweight in kindergarten are four times more likely than healthy-weight children to be obese at age fourteen.² Despite advances in obesity research, insufficient evidence exists about how children develop obesogenic behaviors such as inactivity and poor nutritional preferences, especially in families of low socioeconomic status (SES). Recent studies suggest that children begin to develop health behaviors and attitudes as young as 5 years of age.³ Children as young as 4 or 5 may begin internalizing their parents’ physical activity and dietary habits. A previous study shows that 71 percent of childhood obesity is explained by the influence of family factors on young children.⁴ This suggests that early childhood may provide the best window of opportunity for modifying environmental risk factors for childhood obesity.

The etiology of childhood obesity appears to be multifactorial. Child behavioral risk factors known to increase obesity risk include decreased physical activity rates,⁵ increased time playing video games and watching television,⁶ and being put to bed with a bottle.⁷ Some studies also show a strong association between birth weight and childhood obesity.⁸ Multiple studies document parental risk factors for their children’s overweight status, including maternal obesity,^8,9 lower educational attainment,¹⁰ African American race,¹¹ lower physical activity rates,¹² poor nutrition knowledge,¹² food insecurity,¹³ smoking,⁸ rules about food consumption and eating at regular times,¹⁴ and perceived neighborhood safety.¹⁴ The increased childhood obesity associated with these risk factors results in poor health outcomes in these children that persist as they become adults through the development of chronic diseases such as diabetes, hypertension, and coronary heart disease.¹⁵

Low family SES is associated with increased childhood obesity rates. Despite recent modest improvements in obesity rates among U.S. low-income, preschool-aged children,¹⁶ obesity rates continue to be higher among low-income children.¹⁷ However, this trend is not consistent in all races and at all SES levels.¹⁸ Some attribute the increased rate of obesity in minorities to their greater poverty rates.¹⁹ However, studies show that black children with higher SES do not exhibit the trend of lower obesity prevalence as do higher SES white children.^20–22 In this study, we use a large national database to evaluate whether SES modifies risk factors for childhood obesity, including race. With this knowledge at hand, health care providers who care for children who are racial minorities can better address their pediatric patients’ obesity by understanding the relationship between race, modifiable risk factors, SES and childhood obesity.

Many modifiable risk factors for childhood obesity are related to SES, including neighborhood safety, smoking, drinking soda, and watching television.⁸ Demographic risk factors associated with increased childhood weight, such as race and birthweight, also vary with SES. Indeed, some have assumed these risk factors are SES indicators rather than risk factors for childhood overweight status that are independent of SES.²³ SES is importantly associated with health, including childhood obesity, but it is difficult to alter. Preventive health researchers and providers would benefit from a model that evaluates whether SES modifies the influence of these risk factors on early childhood weight. With such a model, they could design population- and individual-level interventions that target modifiable risk factors for childhood obesity. Therefore, this study sought to test SES as a modifying risk factor for other demographic and behavioral obesity risk factors associated with childhood obesity.

Methods:

Sample:

The Early Childhood Longitudinal Birth Cohort (ECLS-B) is a nationally representative sample of 14,000 children born in the U.S. in 2001. The subjects, from diverse socioeconomic and racial backgrounds, were followed during the formative years of birth through kindergarten entry. Parents were surveyed at their child’s birth, at 9 months, 24 months, and 4–5 years of age. Surveys focused on child health, development, care, and education. In this study, we analyzed data from kindergarten entry (ages 4–5 years), which included 7,022 children. The ECLS-B employed a multistage, stratified and clustered design for data collection.²⁴ Data were collected at every round by trained assessors who visited the child and parents in their homes. At the visits, assessors measured height and weight, which were used to calculate body mass index (BMI). Children’s BMI is classified by standard percentiles; those with a BMI greater than 85% were classified as overweight or obese, according to CDC recommendations.²⁵ The database also includes birthweight from U.S. birth certificate data.

The ECLS-B oversamples children who are twins, Chinese and other Asian and Pacific Islanders, American Indians and Alaska Natives, and those born with low or very low birth weight.²⁴ Researchers utilized weighted analysis to offset oversampling. The University of Missouri Health Sciences Institutional Review Board approved this project. There were no conflicts of interest.

Measures:

Our study’s major outcome measure is the child’s weight status (overweight or obese, ≥ 85th percentile, vs. healthy weight, < 85th percentile, based on age and gender) as measured during the ECLS-B kindergarten-entry visit. Assessors determined food security based on the survey at the same visit. The survey options were food secure, food insecure without hunger, and food insecure with hunger. Assessors also surveyed parents regarding parental household status at kindergarten entry, and we consolidated the categories into (1) two-parent household or (2) other household status, including single-parent household. Assessors recorded the child’s sex at birth as well as birthweight (in grams). We evaluated six categories of race ascertained at birth: White, Black, Hispanic, Asian, Native American, and other. At nine months, assessors asked parents if the child was routinely put to bed with bottle (yes or no). Assessors determined the remainder of the behavioral variables at kindergarten entry. The parent survey included whether they had rules about which foods the child could eat (yes or no); if they had rules about what the child could watch on television (yes or no); how many hours of television the child watched each weekday (continuous variable up to 24 hours); how much soda their child consumed in the past 7 days (0/week, 1–3/week, 4–6/week, 1/day, 2/day, 3/day, 4/day); how many days per week the family eats at a regular time (0–7); how safe the parent felt the neighborhood was (very safe, fairly safe, fairly unsafe, very unsafe); and how many days in the past week the parent had exercised for 30 minutes or more (0–7). We considered but ultimately excluded the following variables from the model due to too much missing data: participation in free or reduced lunch at school, breastfeeding vs. bottle-feeding, child’s habit of snacking after school, and amount of the child’s aerobic exercise.

Household socioeconomic status (SES) is a measure of a family’s relative social position. The SES measure is based on five equally weighted, standardized components: family income, father’s/guardian’s educational attainment, mother’s/guardians educational attainment, father’s/guardian’s prestige of occupation, and mother’s/guardians prestige of occupation. This is a composite variable included in the ECLS database, with the fifth quintile being the highest. Parental education was coded nine categories (grades 0–11; grade 12 without diploma or equivalent; high school diploma or equivalent; vocational/technical after high school; some college; Bachelor’s degree; graduate or professional school but no degree; Master’s degree; doctorate or professional degree after bachelor’s degree) with the highest level of education in the household recorded and included birth, adoptive, step, and foster parents. Occupation was classified according to the Standard Occupation Classification Manual²⁶ and collapsed into 23 codes, and the alternative codes of retired/unemployed and “uncodable”. Each parent’s occupation was scored using the average of the 1989 General Social Survey (GSS) prestige scores for the 1980 census occupational category codes that correspond to the ECLS-K occupation code. These prestige scores ranged from 77.5 for physicians/dentists/veterinarians to 29.6 for handlers/equipment cleaners/helpers/laborers. Household income was not relative to household size but rather was is comprised of 18 categories that increased by intervals of $5,000 starting at less than $5,000 and increasing to greater than $200,000. Rather than use income markers such as poverty level, we choose SES because it includes parental education and occupation in addition to household income. Thus, it is a more robust variable that may more fully describe children’s home environment. Other researchers have employed the SES composite variable to evaluate children’s weight.²⁷

Statistical Analysis:

Simple logistic regression was used to assess the association between overweight status and each individual risk factor. Risk factors with p < 0.2 in the simple analyses were candidates for inclusion in a multivariable model. Logistic regression with stepwise variable selection was used to identify our main-effects risk model. A significance level of 0.05 was used as the variable entry and retention criteria in the stepwise process. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated. The final phase of the regression analysis was to include two-way interaction terms involving the selected main effects and SES. Only significant variables were included in the main effects model to test SES interaction term. Statistically significant interactions would suggest that the actual effects of a risk factor are influenced by SES. SAS 9.4 was used for all analyses (Cary, NC, SAS institute Inc.). SAS SURVEY procedures were used to accomodate the complex survey sample design. We conducted the analysis in 2015 and reviewed it in 2016 and 2017.

Results:

Of the 7,022 children in our analysis, 49.06% were female. The majority of children were White (53.80%), followed by Hispanic (25.12%), Black (13.88%), other races (3.94%), Asian (2.62%), and Native American (0.65 %). Most of the children (68.96%) resided in two-parent households with both their biological mother and father. Overall, 36.31% of children in the sample were overweight or obese (Table 1).

Table 1.

Univariable analyses of factors associated with child obesity in ECLS-B data.

Risk factors	n	Population %	% overweight or obese (95% CI)	Odds ratio ( 95% CI)
Categorical variables
Socioeconomic status (P<0.0001)
Quintile1 (lowest)	1263	19.91	44.94 (40.47–49.41)	REF
Quintile2	1265	19.88	37.24 (33.57–40.90)	0.73 (0.57–0.92)
Quintile3	1288	19.96	36.43 (32.23–40.63)	0.70 (0.55–0.90)
Quintile4	1332	20.20	35.41 (31.92–38.91)	0.67 (0.53–0.86)
Quintile5 (highest)	1622	20.06	27.09 (24.26–29.91)	0.46 (0.37–0.56)
Food security (P=0.043)
Food secure	6034	89.71	35.45 (33.82–37.07)	0.61 (0.38–0.98)
Food insecure without hunger	547	8.03	40.98 (34.96–47.00)	0.77 (0.68–1.27)
Food insecure with hunger	171	2.26	47.39 (35.47–59.30)	REF
Parents who reside in household (P=0.0078)
Biological mother and biological father	4705	68.96	34.61 (32.73–36.49)	0.80 (0.68–0.94)
Other household composition	2065	31.04	39.75 (36.53–42.97)	REF
Race/ethnicity (P<0.0001)
White	2767	53.80	31.35 (29.03–33.66)	REF
Black	1056	13.88	40.32 (35.62–45.02)	1.48 (1.18–1.86)
Hispanic	1359	25.12	43.47 (39.76–47.18)	1.68 (1.39–2.04)
Asian	765	2.62	28.69 (22.82–34.57)	0.88 (0.67–1.16)
Native Hawaiian, Pacific Islander, Alaska, or American Indian	251	0.65	48.44 (36.96–59.92)	2.06 (1.28–3.30)
Other race (includes more than one race)	560	3.94	43.39 (36.77–50.00)	1.70 (1.26–2.24)
Child’s sex (P=0.28)
Female	3348	49.06	37.12 (35.13–39.10)	1.08 (0.94–1.25)
Male	3422	50.94	35.26 (32.63–37.89)	REF
Child was put to bed with bottle (P=0.014)
Yes	2089	29.69	39.48 (36.12–42.84)	1.22 (1.04–1.43)
No	4675	70.31	34.80 (33.03–36.57)	REF
Parental rule about which kinds of food child can eat (P=0.0039)
Yes	5142	76.77	34.74 (32.98–36.51)	0.77 (0.64–0.92)
No	1627	23.23	40.94(37.12–44.76)	REF
Current smoking (P=0.0033)
Yes	1303	19.56	41.43 (37.36–45.49)	1.32 (1.10–1.60)
No	5462	80.44	34.89 (33.14–36.64)	REF
Amount of soda consumed by child (P=0.034)
More than 7 per week	1839	26.99	39.23 (36.21–42.26)	1.03 (0.87–1.20)
Less than 6 time per week	3019	44.08	35.12 (32.97–37.28)	1.22 (1.03–1.45)
Never	1903	28.93	34.56 (31.54–37.58)	REF
Mother’s hours work/week (p=0.48)				0.4827
35 hours or more per week	2971	42.93	37.55 (35.20–39.90)	1.13 (0.95–1.35)
Less than 35 hours per week	1257	19.13	34.99 (31.02–38.96)	1.02 (0.83–1.24)
Looking for work	358	5.68	35.72 (28.24–43.21)	1.05 (0.75–1.64)
Not in the labor force	2071	32.27	34.66 (31.89–37.44)	REF
Parental rule about television watching (P=0.041)
Yes	6274	93.82	35.77 (34.09–37.45)	0.73 (0.55–0.99)
No	466	6.18	43.16 (36.18–50.14)	REF
Neighborhood safety (P=0.0005)
Very safe	4012	59.40	33.71 (31.77–35.66)	0.55 (0.33–0.91)
Fairly safe	2363	34.60	38.83 (35.83–41.82)	0.68 (0.41–1.15)
Very unsafe	115	4.30	45.32 (37.07–53.57)	0.89 (0.48–1.65)
Refuse	93	1.70	48.16 (35.15–61.16)	REF
Overall population			36.31 (34.62–37.79)
Risk factors	Overweight/obese (95% CI)	Non overweight/obese (95% CI)		P –value
	N=2246	N=4572
Continuous variables
Birth weight (grams)	3436.0 (3405.9–3466.1)	3248.4 (3224.0–3272.9)		<0.0001
Hours TV watched per weekday	2.01 (1.92–2.10)	1.86 (1.79–1.93)		0.0045
Days per week eat as a family	5.49 (5.38–5.59)	5.66 (5.58–5.74)		0.046
Parent days per week exercises for 30 min	1.75 (1.61–1.90)	1.94 (1.83–2.06)		0.047

Boldface indicates statistical significance (p<0.05)

ECLS- Early Childhood Longitudinal Study

REF- Reference category for odd ratios

In univariable logistic models, all of the variables we tested were significantly and independently associated with children being overweight or obese at kindergarten except child’s sex and number of hours worked by the mother (Insert Table 1). In the multivariable logistic regression analysis, SES, race, birthweight, smoking status, and eating dinner as a family remained statistically significantly associated with children being overweight or obese (Insert Table 2). Interactions between these variables and SES were not statistically significant, indicating that that SES did not modify the relationship between these variables and childhood obesity (Table 2).

Table 2.

Multiple logistic regression analysis of a child being obese or overweight at kindergarten entry

Variables	Parameter estimate (SE)	Odds ratio for overweight/obesity (95% CI)	P-value
Socioeconomic status (all compared to ref=5th quintile, highest)			0.0018
1st (lowest)	0.5384 (0.1360)	1.71 (1.31, 2.23)
2nd	0.3049 (0.1080)	1.36 (1.10, 1.68)
3rd	0.3222 (0.1321)	1.38 (1.07, 1.79)
4th	0.3104 (0.1286)	1.36 (1.06, 1.76)
Race (reference category = White)			<0.0001
Black	0.4534 (0.1337)	1.57 (1.21, 2.05)
Hispanic	0.4999 (0.1183)	1.65 (1.31, 2.08)
Asian	0.1199 (0.1559)	1.13 (0.83, 1.53)
Native American	0.6823 (0.2667)	1.98 (1.17, 3.34)
Other (includes more than 1 race)	0.5558 (0.1542)	1.74 (1.29, 2.36)
Birthweight (per 100 grams)	0.000684 (0.000059)	1.07 (1.06, 1.08)	<0.0001
Current smoker (yes vs. no)	0.3361 (0.0905)	1.40 (1.17, 1.67)	0.0002
Eat dinner as a family (days per week)	−0.038 (0.0189)	0.96 (0.93, 0.99)	0.0446
Interaction Effects1
Race*SES			0.4263
Birthweight*SES			0.3946
Parental Smoking*SES			0.5884
Eating dinner as a family*SES			0.8809

SE=standard error, CI=confidence interval, SES = socioeconomic status

¹Main effect model included Race, Birthweight, Parental Smoking, and Eating dinner as a family and evaluated SES as interaction term

We found a significant association of SES for children who were overweight or obese, compared to healthy-weight children. Children in the lowest quintile of SES were 70% more likely to be overweight or obese than children in the highest quintile (OR 1.7, 95% CI 1.3–2.2) Children who were Black, Hispanic, Native American or other races had a statistically significantly increased risk of being overweight or obese compared to White children. Only Asian children had a nonsignificant odds of being overweight or obese compared to White children. Black or Hispanic children had a 60% increased odds of being overweight or obese when compared to their White counterparts. Native American children had an almost doubled odds of being overweight or obese compared to White children (Table 2). Elevated birthweight was significantly associated with overweight status in children (OR 1.07, 95% CI 1.06–1.08). Every 100g increase in birthweight was associated with a 7% increased risk of overweight or obesity (Table 2). Parental smoking was associated with 40% higher odds of child overweight or obesity, while eating dinner as a family was associated with 4% lower odds.

Discussion:

We found that race, birthweight, parental smoking, and not having family meals were associated with obesity, but there were no statistically significant interaction terms between SES and any of these variables. This suggests the relationship between these variables and childhood obesity were not modified by SES. This is important because the significant risk factors that are modifiable: birthweight, parental smoking, and not eating meals as a family, could be areas that health care providers could focus anticipatory guidance. Changing the home environment with these modifiable risk factors could indirectly reduce the child’s risk for obesity, although it is important to clarify that our study design supports association, not causation.

Our results confirm previous findings that children’s weight status in kindergarten varies with SES.¹¹ The largest odds ratio of being overweight was between the first and the fifth quintiles for SES. Although many believe that food security contributes to increased obesity rates in low-SES families,²⁸ our model did not find that self-reported food security was a significant predictor. Other studies of food security in children similarily show an inconsistent correlation with weight status.^29–31 This may result from the difficulty of defining and measuring food security. Other factors felt to contribute to obesity in low-SES families include different feeding behaviors^32,33 and disproportionately more psychosocial stressors.³⁴

By using the SES composite variable, which includes household income as well as parents’ education and occupation, our study demonstrated that a combination of socioeconomic factors contribute to children’s early obesogenic environment. Thus, a more comprehensive approach to childhood obesity could include interventions targeting parents not only of different income levels, but also of different education levels and occupations.¹⁰ We found higher rates of overweight and obesity in Black, Native American, and other children in the fourth and fifth quintiles, compared to White children. Because SES did not significantly modify the relationship between race and childhood obesity, other differences should also be considered when addressing childhood obesity. This study supports an increasing body of evidence that racial differences in childhood obesity are not modified by SES.¹⁸ The higher prevalence of childhood obesity among racial minorities is likely the result of complex interactions between a multitude of factors.

Racial neighborhood differences are believed to be one contributing factor. Specifically, living in neighborhoods with higher poverty levels, lower educational levels and a high proportion of Black residents is associated with increased risk of childhood obesity.³⁵ Although our univariate analysis found that neighborhood safety was a risk factor for child overweight and obesity, it was not significantly associated with child weight in the multivariable logistic regression model. Other racial neighborhood differences beyond perceived safety, such as inclusivity of the residential community and availability of outside play equipment and recreation resources, also may contribute.

Limitations:

This study was limited by its use of previously collected data. Although our model included many factors that contribute to childhood obesity, objective data about the child’s activity level, calories consumed, or maternal weight, were not available. However, our model was consistent with many prior studies on childhood obesity. Additionally, we only examined weight at kindergarten entry. Further longitudinal analysis may help determine whether the associations of these factors with increased child weight persist into older childhood.

A particular strength of this study is its large, nationally representative population, which reduces the risk of sampling bias. The large sample also allowed us to detect smaller differences in child overweight due to increased power, increasing the model’s sensitivity.

Conclusion:

In this ECLS-B secondary data analysis study, we created a regression model from known parental and child risk factors for childhood obesity. Socioeconomic status, race, birthweight parental smoking, and not eating dinner as a family were associated with overweight and obesity in kindergarten-age children. SES did not modify the relationship between these variables and childhood obesity. Child healthcare providers should consider discussing these behaviors with families when addressing childhood obesity. Public health programs that influence risk factors, such as promoting healthy birthweight, reducing parental smoking, and eating meals together as a family, may help improve family health behaviors and environment and thus reduce risk of childhood obesity.