Associations between eating behaviors and cardiometabolic risk among adolescents in the HOME Study

Abstract

Background:

Eating behaviors are associated with childhood obesity, but their associations with cardiometabolic risk are less clear.

Objectives:

We evaluated cross-sectional associations between eating behaviors and cardiometabolic risk among 185 adolescents (age 12.4 ± 0.7 years; 53% female; BMI-z 0.72 ± 1.37) from Cincinnati, Ohio (HOME Study; enrolled 2003-2006).

Methods:

Caregivers assessed adolescents’ eating behaviors with the Child Eating Behavior Questionnaire. We computed adolescents’ cardiometabolic risk scores based on HOMA-IR, triglycerides to high-density lipoprotein cholesterol ratio, adiponectin to leptin ratio, systolic blood pressure, and cross-sectional area of fat inside the abdominal cavity. Using multivariable linear regression models, we estimated associations of eating behavior subscales with cardiometabolic risk scores or individual risk components.

Results:

Emotional overeating (ß = 1.34, 95% CI: 0.67, 2.01), food responsiveness (ß = 0.99, 95% CI: 0.41, 1.57), and emotional undereating (ß = 0.64, 95% CI: 0.08, 1.21) were associated with higher cardiometabolic risk scores. Satiety responsiveness (ß = −0.79, 95% CI: −1.59, 0.00) was associated with lower cardiometabolic risk scores. Adjusting for adolescent BMIz at age 12 attenuated these associations, suggesting that adiposity may mediate these associations.

Conclusions:

Hedonistic eating behaviors were associated with higher cardiometabolic risk in these adolescents.

Introduction

Cardiometabolic diseases, like diabetes mellitus and cardiovascular diseases, are the leading causes of death in the US.¹ Heart disease alone accounted for one in four deaths in the US in 2021.² Cardiometabolic risk tracks across the life course.³ Without effective intervention, youth with poor cardiometabolic health will likely suffer from cardiometabolic diseases in adulthood.

While cardiometabolic disease and risk factors are generally associated with lifestyle behaviors, such as diet and physical activity,⁴ eating behavior is a modifiable factor that may affect cardiometabolic risk in children.⁵ Prior studies showed that children’s eating behaviors were associated with greater adiposity.⁶ In particular, fast eating, eating in the absence of hunger, and loss of control of eating may increase the risk for childhood obesity.^7,8 Eating behaviors thus can serve as novel targets for intervention and prevention strategies to improve cardiometabolic health among youth.⁹

Prior research has described the relation of child eating behavior with weight,⁶ but few examined its relation with other key components of cardiometabolic health, such as blood pressure, glucose-insulin homeostasis, and dyslipidemia.¹⁰ The study by Warkentin et al. has assessed associations between eating behaviors and overall cardiometabolic risk in school-age children.⁵ They found that food approach behaviors, measured by the Child Eating Behavior Questionnaire (e.g., Food Responsiveness, Enjoyment of Food, and Emotional Overeating) were positively associated with cardiometabolic risk, while food avoidance behaviors (e.g., Satiety Responsiveness and Slowness in Eating) were negatively associated with cardiometabolic risk. However, Warkentin’s study focused on school-age children, and it remains unclear whether eating behaviors may influence cardiometabolic risk in adolescence.

To better understand the potential impact of eating behaviors on adolescent cardiometabolic risk, we estimated cross-sectional associations between caregiver-reported eating behaviors and cardiometabolic risk scores among adolescents in the HOME Study. We hypothesized that (1) food avoidance behaviors, including Satiety Responsiveness, Slowness in Eating, Emotional Undereating, and Food Fussiness, would be related to lower cardiometabolic risk; (2) food approach behaviors, including Food Responsiveness, Enjoyment of Food, Emotional Overeating, and Desire to Drink, would be related to higher cardiometabolic risk.

Methods

Study Participants

The data for this analysis comes from a longitudinal pregnancy and birth cohort study, the Health Outcomes and Measures of the Environment (HOME) Study.^11,12 The HOME Study recruited pregnant women (age range: 18 - 45 years) from the Cincinnati, Ohio region between 2003 and 2006. Women aged 18 years old and above, 16 ± 3 weeks of gestation, and living in the Cincinnati, Ohio region in a home built before 1978 were eligible to participate. Those with a diagnosis of diabetes, schizophrenia, bipolar disorder, or cancer, a history of HIV infection, or taking medications for thyroid disorders or seizures were excluded. The HOME Study conducted follow-up visits with their offspring (n = 407) at 4 weeks, and 1, 2, 3, 4, 5, 8, and 12 years of age.

Two hundred and fifty-six adolescents completed the 12-year study visit.¹² Our analysis excluded twins and individuals with congenital anomalies, which may lead to low birth weight and affect cardiometabolic characteristics during childhood.^13,14 A total of 185 adolescents had complete data for cardiometabolic risk markers, caregivers’ responses to the Child Eating Behavior Questionnaire (CEBQ), and relevant covariates and thus were included in the analysis (Figure 1).

Figure 1.

Participant flow diagram.

The HOME Study protocols received IRB approval from Cincinnati Children’s Hospital Medical Center and participating delivery hospitals. Adolescents’ mothers or primary caregivers gave written informed consent for themselves and their offspring at all visits. Adolescents provided written informed assent at the 12-year study visit.

Eating Behavior Assessment

At the 12-year study visit, caregivers completed the CEBQ, a parent-report questionnaire evaluating eight distinct eating behaviors hypothesized to affect childhood obesity.¹⁵ This questionnaire has 35 questions that assess eating behaviors on a five-point Likert scale. A higher score indicates a greater parent-perceived presence of the eating behavior.

The CEBQ contains four subscales that measure food approach behaviors: Food Responsiveness evaluates how reactive a child is to external cues for eating; Enjoyment of Food indicates a child’s interest in food; Emotional Overeating assesses the behavior of eating more in response to negative emotions; Desire to Drink reveals the need for a frequent drink or beverage intake. The other four CEBQ subscales measure food avoidance behaviors: Satiety Responsiveness assesses a child’s ability to stop eating when feeling full; Slowness in Eating evaluates the speed of eating; Emotional Undereating measures the behavior of eating less when encountering negative emotions; Food Fussiness indicates selectiveness about food.

The CEBQ showed good internal consistency in prior research (Cronbach α’s 0.79 - 0.91)¹⁵ and in our sample (Cronbach α’s 0.70 - 0.90) (Table S1). Furthermore, the subscales Satiety Responsiveness, Enjoyment of Food, and Food Responsiveness have been validated against laboratory-based measures of eating behaviors (eating rate, eating without hunger, caloric compensation, and energy intake) in 4- to 5-year-old children.¹⁶

Cardiometabolic Risk Assessment

Cardiometabolic risk measures were assessed at the 12-year visit. We collected fasting serum samples from the adolescents and quantified glucose, insulin, triglycerides (TG), high-density lipoprotein cholesterol (HDL), leptin and adiponectin concentrations. We then computed TG to HDL ratio (a biomarker positively associated with cardiometabolic risk)¹⁷ and adiponectin to leptin ratio (a biomarker of adipose tissue dysfunction inversely associated with atherosclerosis risk).¹⁸ Next, we calculated the homeostatic model assessment for insulin resistance (HOMA-IR) using the formula: fasting insulin (mIU/L) × fasting glucose (mg/dL)/405. Higher values of HOMA-IR suggest more severe insulin resistance and greater cardiometabolic risk.¹⁹ Also, three sitting blood pressures were measured for the adolescents, each 1 minute apart, with a Dinamap Pro100 automated monitor.²⁰ Our analysis used the average of the second and third measures.²¹ Lastly, cross-sectional area of fat inside the abdominal cavity was assessed using a whole-body dual-energy X-ray absorptiometry (DXA, Hologic Horizon densitometer) scan.^22,23

Z-scores were created for each cardiometabolic risk measure. Systolic blood pressure (SBP) was standardized by sex, age, and height using references from the National High Blood Pressure Education Program.²⁴ Our analysis used SBP z-scores for comparability with the other risk score components. For components without US references, age- and sex-standardized z-scores were computed using residuals from linear regression models with individual cardiometabolic risk components as the dependent variable and age and sex as the independent variables.²⁵ HOMA-IR, TG to HDL ratio, adiponectin to leptin ratio, and cross-sectional area of fat inside the abdominal cavity did not approximate a normal distribution and were log₂-transformed before standardization.

We computed a continuous cardiometabolic risk score by summing the standardized z-scores of the cardiometabolic risk components (adiponectin to leptin ratio was multiplied by −1 as it inversely correlates with cardiometabolic risk).²⁶ Higher scores indicate greater cardiometabolic risk.

Covariate Assessment

Mothers’ demographic information, including maternal age, income, and education, was collected at baseline. Mothers reported their offspring’s race during the postpartum visit. Information on adolescents’ sex was collected from hospital medical charts. Gestational tobacco smoke exposure was assessed with the mean of log₁₀-transformed serum cotinine concentrations measured at 16 and 26 weeks of gestation. We created a three-category variable in Table 1 to describe gestational tobacco smoke exposure status: unexposed (gestational serum cotinine concentrations <0.015 ng/ml), exposed to secondhand tobacco smoke (0.015 - 3 ng/ml), and active smoker (>3 ng/ml).²⁶ Mothers reported the duration of any breastfeeding during the first three years of life.

Table 1.

Food approach behavior, food avoidance behavior, and cardiometabolic risk scores at age 12 years according to covariates in the Health Outcomes and Measures of the Environment (HOME) Study adolescents.

		Food Approach Behavior	Food Avoidance Behavior	Cardiometabolic Risk Score
	N	Mean ± SD	Mean ± SD	Mean ± SD
Overall	185	10.9 ± 2.5	9.9 ± 2.0	0.1 ± 3.4
Adolescent Sex
Female	98	11.3 ± 2.6	10.0 ± 2.0	0.1 ± 3.7
Male	87	10.5 ± 2.5	9.7 ± 1.9	0.1 ± 3.1
Adolescent Race
Non-Hispanic White	102	10.6 ± 2.3	10.1 ± 2.0	−0.5 ± 3.4
Non-Hispanic Black	71	11.7 ± 2.7	9.7 ± 1.9	1.1 ± 3.4
Other	12	9.5 ± 2.3	9.6 ± 2.3	−1.2 ± 2.4
Adolescent Pubic Hair Stage
Stage 1	17	9.8 ± 2.0	9.8 ± 2.1	−0.7 ± 3.5
Stage 2	47	10.4 ± 2.3	10.5 ± 2.0	−0.6 ± 3.7
Stage 3	54	11.0 ± 2.6	9.7 ± 1.8	0.1 ± 3.4
Stage 4	38	11.1 ± 2.4	9.9 ± 1.8	0.9 ± 3.4
Stage 5	29	12.3 ± 2.8	9.3 ± 2.0	0.5 ± 2.9
Maternal Age at Delivery
18-25 Years	46	11.2 ± 2.8	9.7 ± 2.0	0.4 ± 3.1
>25-35 Years	113	10.8 ± 2.5	9.8 ± 2.0	−0.1 ± 3.5
>35 Years	26	11.2 ± 2.2	10.5 ± 1.8	0.3 ± 3.6
Household Income During Pregnancy ($)
<45, 000	73	11.5 ± 2.9	9.7 ± 1.9	0.7 ± 3.6
45,000-75,000	67	10.4 ± 2.1	9.9 ± 2.0	0.0 ± 3.3
>75,000	45	10.8 ± 2.5	10.2 ± 2.0	−0.8 ± 3.2
Child Depression Inventory Scores
<65	171	10.8 ± 2.5	9.9 ± 1.9	−0.1 ± 3.4
≥65 (At risk of depression)	14	12.3 ± 3.0	9.8 ± 2.1	1.4 ± 3.7
Gestational Serum Cotinine (ng/ml)
<0.015 (Unexposed)	52	10.1 ± 2.2	10.0 ± 2.0	−0.3 ± 3.1
0.015-3 (Secondhand)	113	11.2 ± 2.6	9.9 ± 1.9	0.0 ± 3.6
>3 (Active Smoker)	20	11.7 ± 2.6	9.7 ± 2.1	1.6 ± 3.3
Child Physical Activity Scores
0-2.5	91	11.4 ± 2.6	9.9 ± 1.8	0.6 ± 3.5
>2.5	94	10.5 ± 2.5	9.9 ± 2.1	−0.5 ± 3.3

Note: Food approach summary score was the sum of scores on food responsiveness, emotional overeating, enjoyment of food, and desire to drink subscales from the Child Eating Behavior Questionnaire (CEBQ). Food avoidance summary score was the sum of scores on satiety responsiveness, slowness in eating, emotional undereating, and food fussiness CEBQ subscales. Cardiometabolic risk score was the sum of standardized z-scores of HOMA-IR, triglyceride to high-density lipoprotein cholesterol ratio, adiponectin to leptin ratio (multiplied by −1, as this ratio is inversely related to cardiometabolic risk), systolic blood pressure, and cross-sectional area of fat inside the abdominal cavity. SD: standard deviation.

At the 12-year visit, adolescents followed standardized instructions and self-assessed their pubertal stage (stages I-V) based on pubic hair (males and females) and breast development (females only).²⁷ In addition, they completed the Child Depression Inventory-II (CDI-II)²⁸ and the Physical Activity Questionnaire for Older Children (PAQ-C).²⁹ Trained research assistants administered three 24-hour dietary recalls (2 weekdays and 1 weekend day) to the adolescents. We then calculated adolescents’ total daily energy, macronutrient, and micronutrient intake with the Nutrition Data System for Research software³⁰ and assessed their diet quality with the Healthy Eating Index (HEI-2010).³¹

Using prior literature, we created directed acyclic graphs to identify potential confounders that may impact both adolescent eating behaviors and cardiometabolic risk while ensuring that we did not adjust for mediators and colliders (Figure S1).

Statistical Analysis

In Table 1, for descriptive purposes, we calculated food approach summary scores by taking the sum of four subscales that reflect food approach behaviors. Likewise, we computed food avoidance summary scores from the remaining four subscales. Next, we calculated univariate statistics of the food approach summary score, food avoidance summary score, and cardiometabolic risk score across strata of potential confounders. We computed univariate statistics of the CEBQ subscale scores and cardiometabolic characteristics, stratified by sex. Furthermore, we calculated Pearson correlation coefficients and Cronbach’s alphas of the CEBQ subscales in our sample.

With multivariable linear regression models, we estimated the adjusted difference in cardiometabolic risk scores per unit increase in the CEBQ subscale scores. Additionally, we assessed the associations between each eating behavior subscale and individual cardiometabolic risk components with multivariable linear regression models. Next, we examined effect measure modification by sex by including cross-product terms for each eating behavior subscale and adolescent sex in the models. HOMA-IR, triglyceride to HDL ratio, leptin to adiponectin ratio, and cross-sectional area of fat inside the abdominal cavity had a right-skewed distribution and were log₂-transformed in the regression models.

In the main analysis, we adjusted for the following covariates: adolescent sex (categorical), race (categorical), pubic hair stage (ordinal), depression (CDI-II, continuous), physical activity (PAQ-C, continuous), maternal age (continuous), maternal income (continuous), and gestational serum cotinine concentrations (continuous).

In sensitivity analyses, we further adjusted for adolescent body mass index (BMI age- and sex-standardized z-scores at age 8 or 12 years, continuous), dietary quality (HEI-2010, continuous), average daily caloric intake (continuous), mother’s pre-pregnancy BMI (continuous), and duration of any breastfeeding (continuous) to examine the robustness of our results. We did not include these variables in the main analysis models, because they are unlikely to be confounders (Figure S1). We performed the statistical analyses using RStudio version 1.4.1103.³²

Results

At the study visit, adolescents were 12.4 years old on average (range: 11.0 - 14.1 years). Fifty-three percent were females, 38% were non-Hispanic Black, and 8% had CDI-II scores indicative of possible depression (Table 1). Mothers of these adolescents were, on average, 29.1 years old at delivery (range: 18.7 - 42.4 years). Sixty-one percent of the adolescents had in utero exposure to secondhand smoke, and 11% had active tobacco smoke exposure as measured by gestational serum cotinine.

Food avoidance behavior scores generally did not vary across categories of the covariates (Table 1). In contrast, food approach behavior scores were 1 to 2 points higher among adolescents who were non-Hispanic Black, had higher pubic hair stage, lower household income, higher depressive symptom scores, or were exposed to tobacco smoke in utero compared to the respective referent groups (Table 1). Females scored about 0.2 to 0.3 points higher on Emotional Overeating, Desire to Drink, Satiety Responsiveness, and Emotional Undereating than males (Table 2). In addition, Emotional Undereating subscale was positively correlated with specific food approach subscales, particularly Emotional Overeating (Table S2).

Table 2.

Descriptive statistics of eating behaviors and cardiometabolic characteristics in the HOME Study adolescents at age 12 years: Stratified by sex (N=185).

	All Median (25th, 75th)	Males Median (25th, 75th)	Females Median (25th, 75th)
Child Eating Behavior Questionnaire Subscales
Food responsiveness	2.4 (1.8, 3.0)	2.4 (1.6, 2.8)	2.4 (1.9, 3.2)
Emotional overeating	2.0 (1.3, 2.5)	1.8 (1.1, 2.3)	2.0 (1.5, 2.8)
Enjoyment of food	4.0 (3.5, 4.5)	4.0 (3.5, 4.5)	4.0 (3.8, 4.7)
Desire to drink	2.3 (1.7, 3.0)	2.0 (1.7, 3.0)	2.3 (2.0, 3.3)
Satiety responsiveness	2.4 (2.0, 3.0)	2.4 (2.0, 2.8)	2.6 (2.2, 3.0)
Slowness in eating	2.3 (2.0, 2.8)	2.3 (2.0, 2.8)	2.3 (1.8, 2.8)
Emotional undereating	2.5 (1.8, 3.0)	2.3 (1.5, 3.0)	2.5 (2.0, 3.0)
Food fussiness	2.5 (2.0, 3.3)	2.5 (1.8, 3.2)	2.5 (2.0, 3.3)
Cardiometabolic Characteristics
HOMA-IR	2.8 (2.0, 4.3)	2.4 (1.7, 3.2)	3.5 (2.5, 5.2)
Triglycerides to HDL ratio	1.4 (1.1, 2.0)	1.4 (1.0, 1.8)	1.4 (1.1, 2.1)
Adiponectin to leptin ratio	1.4 (0.6, 4.3)	2.6 (1.0, 6.0)	1.0 (0.4, 2.3)
Systolic blood pressure (z-score)	−0.6 (−1.1, 0.1)	−0.6 (−1.1, 0.0)	−0.7 (−1.1, −0.1)
Cross-sectional area of fat inside the abdominal cavity (cm2)	44.4 (32.3, 55.4)	48.3 (41.3, 57.0)	34.8 (26.3, 53.5)

Note: The Child Eating Behavior Questionnaire subscale scores range from 1 (never) to 5 (always). HOMA-IR: Homeostatic model assessment of insulin resistance. HOMA-IR = fasting insulin (mIU/L) × fasting glucose (mg/dL) / 405. Systolic blood pressure was standardized by age, sex, and height according to the Fourth Report on the Diagnosis, Evaluation, and Treatment of High Blood Pressure in Children and Adolescents. HDL: high-density lipoprotein cholesterol.

On average, cardiometabolic risk scores were higher in adolescents who were non-Hispanic Black, at a more advanced pubertal stage, at risk of depression, less physically active, or were born to mothers who had lower income, were under 25 years or above 35 years at the time of delivery, or actively smoked during pregnancy (Table 1). In this sample, females had higher HOMA-IR, lower adiponectin to leptin ratio, and a smaller cross-sectional area of fat inside the abdominal cavity than males (Table 2).

Certain food approach behaviors were associated with higher cardiometabolic risk scores (Figure 2). A one-unit increase in Emotional Overeating and Food Responsiveness subscales was associated with 1.34 (95% CI = 0.67, 2.01) and 0.99 (95% CI = 0.41, 1.57) units increase in cardiometabolic risk scores, respectively (Table S3). For individual risk components, the strongest associations were between these food approach subscales and leptin to adiponectin ratio, cross-sectional area of fat inside the abdominal cavity, and HOMA-IR (Figure 2, Table S3).

Figure 2:

Adjusted differences in cardiometabolic risk scores and individual components with a 1-unit increase in the Child Eating Behavior Questionnaire subscales: The HOME Study (N=185). From multivariable linear regressions adjusted for adolescent age, sex, race/ethnicity, pubic hair stage, depression, physical activity, maternal age, maternal income, and gestational serum cotinine concentrations. HOMA-IR, triglycerides to HDL ratio, leptin to adiponectin ratio, and cross-sectional area of fat inside the abdominal cavity had right-skewed distributions and were log₂-transformed in the models. Results are presented as ß values and respective 95% confidence intervals. CEBQ: Child Eating Behavior Questionnaire. CM: cardiometabolic. Visceral fat: cross-sectional area of fat inside the abdominal cavity (cm²). HOMA-IR: homeostatic model assessment of insulin resistance. HDL: high-density lipoprotein cholesterol. SBP: systolic blood pressure. FR: food responsiveness. EF: enjoyment of food. EOE: emotional overeating. DD: desire to drink. SR: satiety responsiveness. SE: slowness in eating. EUE: emotional undereating. FF: food fussiness.

Among food avoidance behaviors, adolescents with lower Satiety Responsiveness and higher Emotional Undereating scores had higher cardiometabolic risk scores (Figure 2). For instance, a one-unit increase in the Satiety Responsiveness scale was associated with 0.79 units (95% CI = −1.59, 0.00) decrease in cardiometabolic risk scores (Table S3). This association was mainly driven by an association between Satiety Responsiveness and leptin to adiponectin ratio (Figure 2, Table S3).

Additional adjustment for age 12-year BMIz attenuated the associations between eating behaviors and cardiometabolic risk towards null (Figure 3, Table S4). In contrast, adjusting for age 8-year BMIz weakened the associations.

Figure 3.

Adjusted differences in cardiometabolic risk scores with a 1-unit increase in the Child Eating Behavior Questionnaire subscales: The HOME Study (N=185). Model 1 (left panel) adjusted for adolescent age, sex, race/ethnicity, pubic hair stage, depression, physical activity, maternal age, maternal income, and gestational serum cotinine concentrations. We further adjusted for adolescent body mass index (age- and sex-standardized BMI z-scores) at age 12 years (middle panel) or age 8 years (right panel) to examine potential mediation by adolescent BMI. CEBQ: Child Eating Behavior Questionnaire. CM: cardiometabolic. FR: food responsiveness. EF: enjoyment of food. EOE: emotional overeating. DD: desire to drink. SR: satiety responsiveness. SE: slowness in eating. EUE: emotional undereating. FF: food fussiness. Results are presented as ß values and 95% confidence intervals.

Generally, the associations between adolescent eating behaviors and cardiometabolic risk scores did not differ by sex (Table S5). The associations did not vary substantially from the main analysis when further adjusted for pre-pregnancy BMI, duration of any breastfeeding, adolescent diet quality, and average daily caloric intake (Table S4).

Discussion

We examined cross-sectional associations between eating behaviors and cardiometabolic risk in adolescents from the HOME Study. Our analysis showed that food responsiveness, emotional overeating, and emotional undereating were associated with higher cardiometabolic risk scores, while satiety responsiveness was associated with lower cardiometabolic risk scores. Additionally, these eating behaviors were related to individual cardiometabolic risk components, including leptin to adiponectin ratio, cross-sectional area of fat inside the abdominal cavity, and HOMA-IR. These associations were attenuated when we further adjusted for adolescent BMI, suggesting that adiposity may mediate these associations.

In this study, adolescents with higher food responsiveness and emotional overeating behavior scores tended to have greater cardiometabolic risk. This finding is consistent with Warkentin and colleagues’ study, which found that food approach behaviors at age seven were associated with greater cardiometabolic risk at age ten.⁵ Moreover, a meta-analysis by Kininmonth and colleagues found that these food approach behaviors were positively associated with adiposity measures in children.⁶ Since the present study and most prior ones were cross-sectional, prospective research is necessary to investigate if food approach behaviors early in life increase the risk of undesirable cardiometabolic outcomes. This knowledge could help guide intervention efforts to modify eating behaviors before the onset of excess adiposity or poor cardiometabolic health.

We also found that greater satiety responsiveness, the ability to regulate food intake, was associated with lower cardiometabolic risk. This observation agrees with the study by Warkentin and colleagues, who found that satiety responsiveness at age seven was associated with lower cardiometabolic risk at age ten.⁵ Prior research also linked satiety responsiveness with lower child BMI.⁶ Satiety responsiveness enables individuals to stop eating when feeling full and control the amount of food they eat and is related to specific FTO genotypes.³³ Future behavioral intervention studies could assess whether enhancing satiety responsiveness reduces adolescent cardiometabolic risk.

Interestingly, we found that adolescents with more emotional undereating behavior (EUE) tended to have greater cardiometabolic risk scores. Research on children’s emotional undereating and cardiometabolic health is inconsistent, with some studies reporting a negative association with weight and others finding no association.⁶ Several mechanisms may explain our observation of EUE and greater cardiometabolic risk. First, it is possible that adolescents with higher cardiometabolic risk experience more body shape dissatisfaction, which leads to EUE.^34,35 Second, our finding may be due to the clustering of EUE with emotional overeating behavior (EOE). In our cohort, EUE was moderately positively correlated with EOE (r = 0.48, p-value < 0.001), suggesting that adolescents who emotionally under-eat tend also to emotionally over-eat, which together may increase their cardiometabolic risk.³⁶ Emotional eating describes the tendency to change eating behaviors in stressful situations, thus explaining the positive correlation between EUE and EOE.³⁷ Moreover, both types of emotional eating can be inherited and learned jointly and share common etiologies like parenting behaviors, family environment, and children’s emotion regulation skills.^16,38 Adolescents are at particularly high risk for emotional eating.³⁹ Thus, additional research is necessary to understand the impact of EUE on adolescent cardiometabolic risk with future studies examining the clustering of emotional eating behaviors and its relation to cardiometabolic risk.

Furthermore, we found specific associations of eating behaviors with leptin to adiponectin ratio, visceral fat, and HOMA-IR. We speculate that this may be because insulin resistance in adipose tissue is associated with increased chemokine and adipokine production, which regulate inflammation and is also associated with visceral fat.⁴⁰ This observation suggests that leptin to adiponectin ratio, visceral fat, and HOMA-IR might be more sensitive cardiometabolic risk markers related to eating behaviors than SBP and TG to HDL ratio.

Adjusting for adolescent BMI at age 12 years attenuated associations between CEBQ and cardiometabolic risk scores towards the null, whereas adjusting for BMI at age 8 years did so to a lesser extent. This finding suggests that adiposity may mediate the association between eating behavior and cardiometabolic risk. However, given the cross-sectional nature of our data, we could not disentangle the temporal relations between these variables. Longitudinal studies are thus necessary to properly examine mediation by adolescent BMI in the association between adolescent eating behavior and cardiometabolic risk.⁴¹

This study has strengths and limitations. First, the cross-sectional nature of this study precluded us from investigating the directionality of these associations. Prior studies showed that eating behaviors (i.e., emotional overeating) had bi-directional relations with weight status across childhood.^42,43 Thus, future research should use longitudinal data and evaluate whether the associations between adolescent eating behaviors and cardiometabolic risk are also bi-directional. Second, this study could not investigate the associations between adolescent eating behaviors and clinical outcomes of cardiometabolic diseases. Instead, we assessed adolescents’ cardiometabolic risk with five well-known biomarkers predictive of future cardiovascular diseases, including fasting serum biomarkers. However, the risk score calculation assumes that each parameter contributes equally to one’s overall cardiometabolic risk, and specific parameters may be stronger predictors of cardiometabolic risk than others.²⁶ Thus, the pattern of results for individual cardiometabolic components may be more informative than the summary risk score. Future studies should determine if shifts in the distribution of these biomarkers related to appetitive behaviors increase the risk of clinically significant cardiometabolic disease. Third, the CEBQ relies on caregivers’ perceptions and may introduce measurement error. Nonetheless, this questionnaire has shown good internal consistency, test-retest reliability, stability, and criterion-related validity against objective eating behavior measures.^15,16 Fourth, some of our results may be exploratory as we made multiple comparisons and thus need to be interpreted with caution. However, we observed consistent associations between multiple measures of eating behaviors and cardiometabolic risk, thus, reducing the likelihood of chance findings. Moreover, we focused on the precision and magnitude of the associations and did not specify a threshold for statistical significance, as recommended by the American Statistical Association and others.^44,45 Lastly, our analyses adjusted for a comprehensive set of potential confounders, but unmeasured confounders, such as sleep duration, should be considered in future research.

Despite the limitations, our findings and prior studies suggest that certain eating behaviors may increase adolescent cardiometabolic risk. Growing evidence indicates that interventions targeting working memory training may help reduce emotional eating.⁴⁶ Mindfulness-based interventions also show the potential to change unhealthy eating behaviors, such as binge eating⁴⁷ and emotional eating.⁴⁸ With mindfulness, individuals become more aware of satiety and less reactive to emotional and external triggers for eating (e.g., food advertising and large portions).⁴⁹ Further research should also examine the determinants of adolescent eating behaviors, such as food engineering,⁵⁰ early life stressors, and social factors.

This cross-sectional study indicates that adolescent eating behaviors, specifically food responsiveness, emotional overeating, and emotional undereating, were associated with higher cardiometabolic risk. In comparison, satiety responsiveness was associated with lower cardiometabolic risk. Future studies should validate these findings in prospective cohorts and determine if promoting healthy eating behaviors may improve cardiometabolic health among youth.