Psychological attributes and eating behaviors in 5- to 12-year-old children during periods of stress

Abstract

Objective:

The objective of this study was to investigate the association between psychological attributes and obesogenic behaviors in children during the period of COVID-19 pandemic-induced stress.

Methods:

This observational study collected data from caregivers of 5- to 12-year-old children from three diverse groups assessing sociodemographic, economic, and perceived stress, along with the Strengths and Difficulties Questionnaire (SDQ) and the Family Eating and Activity Habits Questionnaire, revised (FEAHQ-R), in October 2020 and June 2021. The outcome measures were SDQ and FEAHQ-R scores. The SDQ subscales were compared with US child norms. Linear mixed models were used to examine the association between the SDQ subscales and FEAHQ-R domains, adjusting for socioeconomic covariates.

Results:

A total of 361 families (496 children) completed the SDQ and FEAHQ-R. The SDQ subscale scores were higher than those of age/gender norms (p = 0.006–<0.001). Eating style (p < 0.001) and food stimulus exposure scores (p = 0.005–0.01) were associated with the SDQ subscales, but not satiety response. Perceived stress was a significant covariate (p < 0.01). The baseline obesity status of the children was not statistically significant.

Conclusions:

Psychological stress is associated with higher hedonic behavior in children. Food stimulus exposure and leisure access can be targeted for intervention during periods of prolonged stress.

INTRODUCTION

Childhood obesity continues to be one of the most important global problems threatening human health [1]. Designing optimal intervention and prevention strategies requires an understanding of the causal mechanisms. Youth aged 2 to 19 years had 0.24-kg/m² (95% confidence interval [CI]: 0.02–0.45) greater body mass index (BMI) gain during the COVID-19 pandemic compared with that of prior years [2]. Similarly, a large study of 5- to 19-year-old children noted a faster increase in obesity prevalence (23.8%–25.5%) during the pandemic years, with a more rapid increase in BMI z score [3]. Thus, this period provides a suitable model for understanding contributors to childhood obesity.

Food intake and appetite regulation in humans are complex processes primarily governed by the hypothalamus. The orexigenic (i.e., higher food intake) signals increase hunger and appetite leading to higher body weight. Conversely, anorexigenic (i.e., decreased food intake) signals induce satiety to decrease food intake in order to maintain or reduce body weight. The hypothalamus receives signals from hormonal and neurochemical factors from the gut and adipose tissue, along with input from the emotion-sensing corticolimbic system, prefrontal cortex, and stress hormones from the hypothalamic–pituitary–adrenal axis. Although initially protective, stress-induced cortisol responses may become harmful in the long term [4]. Dopamine and serotonin neuropeptides, influenced by emotions and stress, also play important roles in appetite regulation [5].

Emotional eating may be defined as food intake in response to negative or positive emotions rather than a physiological response to hunger [6,7]. Emotional eating can be the result of confusion between the internal states of hunger and satiety and physiological symptoms related to emotions [8]. Food affects the response and expression of positive and negative emotions, and emotions can have a powerful effect on food choices [9]. Parents play an important role in influencing children’s emotional eating behavior through modeling and parenting. Low maternal support and variations in psychological control among adolescents are associated with higher emotional eating [10]. Stress and feelings of perceived helplessness are associated with high-calorie, low-nutrient food intake through emotion-driven eating [7]. Stress-related eating behaviors are associated with adverse dietary and other health behaviors in children and adolescents [7,11,12].

Higher stress levels and emotional symptoms were reported during the COVID-19 pandemic worldwide in both children and their parents, potentially modulated by socioeconomic circumstances [13–15]. COVID-19 pandemic-related psychological stress, decreased physical activity, changes in dietary patterns, and social factors have been postulated to contribute to body weight-related changes in children during the pandemic [16,17]. We hypothesized that the COVID-19 pandemic affected the behaviors, emotions, and relationships of young children, which, in turn, would be associated with obesogenic behaviors. In order to test this hypothesis, we first quantified the psychological attributes of 5- to 12-year-old children using the Strengths and Difficulties Questionnaire (SDQ) during the COVID-19 pandemic in three demographically diverse groups in New York City, New York. Next, we assessed the relationship between psychological attributes and obesogenic behaviors measured by the Family Eating and Activity Habits Questionnaire, revised (FEAHQ-R) while accounting for a myriad of sociodemographic confounders.

METHODS

Study populations

This observational study was approved by the institutional review board of Columbia University Irving Medical Center. Parents/caretakers of 5- to 12-year-old children seen at two clinics (i.e., Inwood and Upper West Side [UWS]) serving distinct populations were contacted, and the study was posted on the publicly accessible online recruitment platform (RecruitMe) hosted by Columbia University Irving Medical Center. The Inwood clinic serves a publicly insured population of predominantly Hispanic/Latinx origin living in northern Manhattan and the Bronx, whereas the UWS clinic serves a predominantly privately insured population with diverse racial and ethnic identities and higher educational levels. The primary contact of the children from the electronic health record (EHR) was invited to participate in the survey in English or Spanish via email and/or text message. No participants were excluded based on the International Classification of Diseases (ICD) codes, and the caregiver could provide responses for up to three children in the family, regardless of age. Consent and surveys were administered via the electronic Qualtrics platform or by phone by a study coordinator. The baseline survey was completed between October and December 2020 (fall 2020). Those who completed the baseline survey were sent a follow-up survey between March and June 2021 (spring 2021) in the same school year of remote learning and prior to the change in pandemic restrictions. Biweekly reminders were sent for partial or incomplete surveys.

Measures

The baseline questionnaire collected demographic data, including age, race and ethnicity, gender of the caregivers and up to three children, and household COVID-19 infections. In order to ensure the validity of the captured data, a follow-up questionnaire was administered within the same school year (~6 months apart), when the pandemic conditions had not changed.

Assessment of psychological attributes in children

The 25-item SDQ, which was completed by the parent/guardian, was used to assess young people’s behaviors, emotions, and relationships [18]. The SDQ is a behavioral screening questionnaire that captures the positive and negative attributes of youth aged 2 to 17 years on the following five subscales with five items each: emotional symptoms; conduct problems; hyperactivity/inattention; peer relationship problems; and prosocial behavior [19]. The responses are captured on a 3-point Likert scale (i.e., “not true,” “somewhat true,” and “certainly true”) and are scored from to 0 to 2, with reverse-scoring in five questions. The publicly available SDQ was used in 8250 publications from 117 countries between 1997 and 2022. Gender- and age-specific population-based norms for the SDQ subscales from the United States are available from the National Health Interview Survey (NHIS) supplement [20]. It is a widely used dimensional measure across its full range, with each 1-point increase corresponding to an increased rate of mental health disorders.

Eating behavior and food parenting

Obesogenic behaviors of children and households were assessed using the FEAHQ-R. This publicly available, 32-item tool was devised and validated in a family-based weight management program for children (test–retest reliability > 0.75) [21]; was revised over two decades [22]; and has been used in studies in the United States, the UK, and Israel [22–24]. The questionnaire ascertains the eating and activity behaviors of the child, up to two caretakers, and the home environment on a 5-point Likert scale (0 = never to 4 = always). Selected questions (five items) were rated on a 3-point Likert scale. The FEAHQ-R comprises the following four constructs: eating habits and style (12 items); response to internal hunger and satiety cues (3 items); leisure time activity (4 items), including recreational screen time and organized and unorganized leisure time; and environmental stimulus exposure (13 items), which includes the presence and visibility of calorie-dense foods and drinks in the household. Scores were calculated separately for each family member and reflected the obesogenic exposure of the participants. The total score is considered an index of the overall inappropriateness of eating patterns, with higher total scores reflecting greater obesogenic exposure [22]. As the home environment is consistent across the children in a family, the stimulus exposure questions were asked only once in the family.

Socioeconomic profile

Four domains of the family’s socioeconomic profile were examined: parental employment and education; housing quality; financial stability; and the use of supplemental benefits. Several questions for each domain were selected from an institutional database validated for the hospital catchment area (online Supporting Information Methods). Perceived stress was ascertained using a 10-point scale for children, their parents, or caregivers.

Demographic and anthropometric data

Race and ethnicity were self-reported in the questionnaire. A report of Hispanic/Latinx ethnicity was prioritized in this category, whereas the others were grouped based on self-identified race. Height and weight data were obtained from the EHR within the prior 6 months. Obesity was defined as BMI ≥ 95th percentile using the Centers for Disease Control and Prevention (CDC) 2000 growth charts [25].

Statistical analyses

Descriptive statistics were calculated based on the distribution of the variables by site because of the differences in patient demographics across the sites. The internal consistency of the questionnaire was assessed using Cronbach α. The socioeconomic profile was divided into the following five scales: caretaker score, including number of caretakers; education and employment history; smoking and alcohol use; housing score, including physical condition of the home, housing insecurity, and rent/mortgage burden; economic stress, including finances for food, difficulty in obtaining food and shelter, ability to pay bills, use of benefits and supplemental programs, availability of transport, and how quickly food ran out; and perceived stress, including the summation of individual-level perceived stress for each family member, normalized by the number of family members. The FEAHQ-R domains were calculated for each child by summing the scores of the child and two caregivers. Two items from the leisure activity domain were reverse-scored as recommended [21,22]. The stimulus exposure domain was uniform for all children within the family. The SDQ subscales were calculated based on SDQ scoring guidelines [26]. Comparisons of the subscales at baseline by recruitment site were performed using ANOVA for continuous variables with parametric and Kruskal–Wallis tests for nonparametric distribution. In order to improve power, the R language mice package was used to impute missing data (~10%) based on missing-at-random assumption for the baseline data (online Supporting Information Methods). The models were constructed using imputed data (baseline) and empirical data (longitudinal analysis). Linear mixed models were used to define the association between the SDQ subscales and FEAHQ-R domains. In the two-level baseline model, individuals were grouped into families and families within the clinic. For the three-level follow-up model, responses by time within an individual were grouped in addition to the family and clinic, as described earlier (Figure S1). The covariates were chosen based on previous studies. The base model included age, gender, race and ethnicity, and the presence of obesity as covariates, whereas the saturated model included the socioeconomic covariates of caretaker score, economic stress, and perceived stress to adjust for these potential confounders (Figure S2). Random effects were participant, family, and clinic, except for the FEAHQ-R stimulus exposure domain, which was similar for all individuals within the family, with grouping at the clinic and family levels only. The time of assessment (baseline or follow-up) was added to the longitudinal models. The model fit was examined using diagnostic plots. The housing score and presence of obesity were not used in the final models because of the lack of statistical significance. The optimal model was selected based on model estimates, model fit plots, the ANOVA test for comparing nested models, R², and the lowest Akaike information criterion and Bayes information criterion values. Sensitivity analyses were performed using only empirical data and with only one child from each family (noncorrelated participants). A p value < 0.01 was considered significant using Bonferroni multiple testing correction for the five SDQ subscales. Statistical analysis was performed in June 2024 using R version 4.3.

RESULTS

The flow of the study is shown in Figure 1. Of the 676 participants with survey reports, 496 children (361 families) with at least 80% completion of both the SDQ and the FEAHQ-R questionnaires were included in the baseline analysis. Of these, 77% (n = 383) had follow-up data available. The UWS clinic and RecruitMe cohorts had a higher proportion of self-identified White and Asian families, whereas the Inwood clinic had more Hispanic/Latinx and Black families. The socioeconomic profile of the Inwood clinic showed higher economic stress and housing burden with lower educational/professional status of the caretakers and financial means (Figure 2A–C). These demographic and socioeconomic profiles align with the New York City 2020 census data of catchment areas for the respective clinics. There were no differences in the distribution of children by sex in the study groups and birth order. A greater proportion of children at the Inwood clinic had obesity (35.5% vs. 8.7%; p < 0.001; Table 1). Despite the sociodemographic differences, perceived stress for the family was the same across the three groups (Figure 2D).

FIGURE 1.

Study details. (A) Study workflow. (B) Study instruments.

FIGURE 2.

Socioeconomic subscale profiles for families included in the cohorts. Inwood clinic (n = 127), Upper West Side (UWS) clinic (n = 334), and RecruitMe cohort (n = 35). The significant differences are highlighted. **p = 0.01–0.001; ***p < 0.001.

TABLE 1.

Baseline demographic and socioeconomic profile of the cohorts by recruitment site.

	Inwood cinic	UWS clinic	RecruitMe	p value
Participants, n	149	338	36
Age, mean (SD), y	8.6 (3.3)	9.1 (2.4)	8.4 (2.6)	0.15
Sex, n (%), female	63 (47.0)	159 (47.5)	21 (58.3)	0.44
Obesity, n (%)
BMI < 95 percentile	68 (48.2)	198 (88.9)		<0.001
BMI ≥ 95 percentile	50 (35.5)	29 (8.7)		<0.001
Child birth order, n (%)
1	91 (67.9)	240 (71.6)	24 (66.7)	0.64
2	37 (27.6)	87 (26.0)	10 (27.8)	0.92
3	6 (4.5)	8 (2.4)	2 (5.5)	<0.001
Age group, n (%)
<8y	58 (43.3)	115 (34.3)	16 (44.4)	0.13
8–11 y	39 (29.1)	134 (40.0)	15 (41.7)	0.07
>11 y	34 (25.3)	86 (25.7)	5 (13.9)	<0.001
Families, n	91	246	24
Race, n (%)
White	22 (24.1)	144 (58.3)	15 (62.5)	<0.001
Black	12 (13.2)	19 (7.7)	3 (12.5)
Asian	2 (2.1)	23 (9.3)	0
Mixed/other	42 (46.1)	55 (22.3)	6 (25.0)
Ethnicity, n (%)
Hispanic/Latino	71 (78.0)	50 (20.2)	7 (29.2)	<0.001
Non-Hispanic	14 (15.4)	158 (64.0)	14 (58.3)
Other/unknown	6 (6.4)	39 (15.7)	3 (12.5)
Socioeconomic profile, mean (SD)
Caretaker score	3.6 (2.7)	6.6 (2.6)	6.8 (2.6)	<0.001
Housing score	7.7 (4.0)	9.4 (3.3)	9.6 (3.8)	0.002
Financial score	14.2 (9.5)	25.1 (7.0)	17.9 (8.4)	<0.001
Economic stress	24.1 (9.8)	18.7 (8.1)	19.5 (9.0)	<0.001
Perceived stress	6.2 (2.2)	6.3 (1.7)	5.8 (1.9)	0.38
Housing burden, n (%)	46 (50.5)	53 (21.4)	6 (25.0)	<0.001

Note: Bold values indicate statistical significance.

Abbreviation: UWS, Upper West Side.

SDQ

The SDQ subscales had good internal consistency (Table 2). There were no differences in the SDQ subscales for emotional symptoms, conduct problems, or hyperactivity/inattention between the study groups. Peer problems were lower in the UWS clinic compared to the other groups (Table 2). The total mean as well as the subscale scores of the study population by gender was higher than the US population norms (Table 3; Figure S2). The SDQ subscales were associated with socioeconomic profiles, particularly perceived stress (Table 4; Figure S4). There were no differences in the SDQ subscale scores from baseline to follow-up (data not shown).

TABLE 2.

Results of the SDQ and FEAHQ subscales by recruitment site.

SDQ subscales, median (IQR)	Total cohort	Inwood clinic	UWS clinic	RecruitMe	p value
	Cronbach α	n = 134	n = 344	n = 36
Emotional symptoms	0.91	2.0 (1.0 to 3.0)	2.0 (0 to 3.0)	1.0 (0 to 3.5)	0.7
Conduct problems	0.90	1.0 (0 to 3.0)	1.0 (0 to 3.0)	1.0 (1.0 to 3.0)	0.8
Hyperactivity/inattention	0.90	4.0 (2.0 to 6.0)	4.0 (2.0 to 6.0)	4.0 (2.0 to 6.0)	0.9
Peer problems	0.91	2.0 (1.0 to 4.0)	1.0 (0 to 2.0)	1.0 (0 to 2.0)	<0.001
Prosocial behavior	0.91	8.0 (5.0 to 9.0)	8.0 (6.0 to 10.0)	9.0 (7.0 to 10.0)	0.09
Externalizing symptoms	0.85	4.0 (1.0 to 7.0)	5.0 (3.0 to 9.0)	5.0 (1.8 to 7.5)	0.1
Internalizing symptoms	0.86	3.0 (1.0 to 5.8)	3.0 (1.0 to 5.0)	2.0 (0 to 5.0)	0.2
Impact	0.88	0 (0 to 0)	0 (0 to 0)	0 (0 to 0)	0.3
Total	0.89	8.0 (3.0 to 13.0)	9.0 (5.0 to 13.0)	6.5 (3.3 to 12.2)	0.3
FEAHQ-R subscale, median (IQR)	Cronbach α	n = 149	n = 339	n = 36	p value
Leisure score	0.78	3.0 (−4.0 to 9.0)	−2.0 (−12.0 to 5.0)	1.0 (−15.0 to 7.8)	<0.001
Eating score	0.71	61.5 (47.0 to 72.8)	74 (64.5 to 86.0)	72.0 (59.0 to 79.5)	<0.001
Satiety score	0.73	10.0 (7.3 to 15.0)	17.0 (13.0 to 20.0)	15.0 (15.0 to 18.0)	<0.001
Environmental score	0.72	18.0 (13.0 to 22.0)	22.0 (19.0 to 26.0)	13 (9.8 to 14.2)	<0.001
Total score	0.74	93.5 (71.2 to 112.0)	110.5 (91.0 to 128.0)	101.5 (82.8 to 110.0)	<0.001

Note: Bold values indicate statistical significance.

Abbreviations: FEAHQ-R, Family Eating and Activity Habits Questionnaire Revised; SDQ, Strengths and Difficulties Questionnaire; UWS, Upper West Side.

TABLE 3.

Comparison of SDQ subscales of the cohorts by age during COVID-19 (2020) with normative data from NHIS supplement (2001).

SDQ subscale	Total cohort			Boys			Girls
	COVID-19 (n = 496)	US norm (n = 9878)	p value	COVID-19 (n = 253)	US norm (n = 5080)	p value	COVID-19 (n = 243)	US norm (n = 4798)	p value
	Score (SD)			Score (SD)			Score (SD)
Total	9.1 (6.6)	7.1 (5.7)	<0.001	9.7 (6.8)	7.5 (5.9)	<0.001	8.5 (6.2)	6.6 (5.3)	<0.001
Emotional	2.2 (2.1)	1.6 (1.8)	<0.001	2.1 (2.1)	1.4 (1.8)	<0.001	2.3 (2.1)	1.7 (1.9)	<0.001
Conduct	1.7 (1.7)	1.3 (1.6)	<0.001	1.9 (1.7)	1.4 (1.7)	<0.001	1.6 (1.6)	1.2 (1.5)	<0.001
Hyperactivity	4.3 (2.9)	2.8 (2.5)	<0.001	5.0 (2.8)	3.2 (2.6)	<0.001	3.7 (2.8)	2.4 (2.3)	<0.001
Peer	1.7 (1.8)	1.4 (1.5)	<0.001	1.9 (1.9)	1.5 (1.6)	0.001	1.5 (1.6)	1.3 (1.5)	0.06
Prosocial	7.6 (2.3)	8.6 (1.8)	<0.001	7.3 (2.5)	8.4 (1.9)	<0.001	8.0 (2)	8.8 (1.6)	<0.001
SDQ subscale	4–7y			8–10 y			11–14 y
	COVID-19 (n = 166)	US norm (n = 2779)	p value	COVID-19 (n = 246)	US norm (n = 2064)	p value	COVID-19 (n = 84)	US norm (n = 2770)	p value
	Score (SD)			Score (SD)			Score (SD)
Total	8.7 (5.2)	7.4 (5.3)	0.05	9.9 (6.5)	7.2 (5.8)	<0.001	10.3 (6.4)	7.1 (6.2)	<0.001
Emotional	1.8 (1.8)	1.5 (1.7)	0.18	2.1 (2.1)	1.5 (1.9)	0.005	2.3 (2.2)	1.7 (2.0)	0.02
Conduct	1.6 (1.4)	1.4 (1.6)	0.26	1.8 (1.8)	1.3 (1.7)	0.007	1.6 (1.6)	1.4 (1.8)	0.26
Hyperactivity	4.3 (2.5)	3.2 (2.5)	<0.001	4.2 (3.1)	2.9 (2.6)	<0.001	4.3 (1.9)	2.7 (2.6)	<0.001
Peer	1.0 (1.2)	1.3 (1.5)	0.05	1.7 (1.7)	1.5 (1.6)	0.24	2.0 (2.0)	1.4 (1.6)	0.008
Prosocial	7.6 (2.4)	8.4 (1.9)	0.009	6.9 (2.6)	8.8 (1.7)	<0.001	7.0 (2.7)	8.7 (1.8)	<0.001

Note: Bold values indicate statistical significance.

Abbreviations: NHIS, National Health Interview Survey; SDQ, Strengths and Difficulties Questionnaire.

TABLE 4.

Results of hierarchical mixed models to assess the association of socioeconomic profile and SDQ subscales.

	n	Caretaker score		Housing score		Economic stress		Perceived stress
		Estimate (95% CI)	p value	Estimate (95% CI)	p value	Estimate (95% CI)	p value	Estimate (95% CI)	p value
Emotional score	468	−0.05 (−0.11 to 0.01)	0.12	−0.11 (−0.15 to −0.06)	<0.001	0.01 (−0.01 to 0.03)	0.26	0.19 (0.10 to 0.27)	<0.001
Conduct score	469	−0.10 (−0.17 to −0.02)	0.01	−0.11 (−0.15 to −0.06)	<0.001	0.02 (0 to 0.05)	0.06	0.37 (0.26 to 0.47)	<0.001
Hyperactivity score	491	−0.20 (−0.36 to −0.05)	0.008	−0.15 (−0.26 to −0.03)	0.01	0.03 (−0.12 to 0.08)	0.21	0.50 (0.29 to 0.72)	<0.001
Peer score	469	−0.11 (−0.21 to −0.01)	0.03	−0.15 (−0.22 to −0.07)	<0.001	0.02 (−0.02 to 0.05)	0.33	0.34 (0.20 to 0.49)	<0.001
Prosocial score	477	−0.11 (−0.19 to −0.04)	0.002	−0.07 (−0.13 to −0.10)	0.01	0.02 (0 to 0.05)	0.05	0.27 (0.16 to 0.37)	<0.001
Impact score	491	−0.16 (−0.27 to −0.04)	0.006	−0.08 (−0.17 to 0.01)	0.07	0.02 (−0.01 to 0.06)	0.20	0.42 (0.26 to 0.58)	<0.001

Note: The fixed effects covariates for the model included age, gender, race and ethnicity, time, and obesity (dichotomous), and the random effects included participant, family and clinic. The statistically significant p values are in bold.

Abbreviation: SDQ, Strengths and Difficulties Questionnaire.

FEAHQ-R

There was good internal consistency for the FEAHQ-R measurements for the individual and family (Cronbach α > 0.75). Significant differences in the FEAHQ-R domains were observed between the study groups. Thus, the Cronbach α for each domain for the cohort was lower (Table 2). At baseline, there was no significant correlation between the BMI z score and FEAHQ domains (Figure S3). In the follow-up survey, leisure activity and eating habit scores decreased, satiety response scores remained the same, and stimulus exposure scores increased, albeit without statistical significance (data not shown).

Regression analyses

In the baseline models, there was a significant association between the emotional and conduct problems SDQ subscales and eating habits (p < 0.001). The estimates were attenuated after adjusting for economic and perceived stress, each of which also had a significant association (p = 0.006 and p < 0.001; Table S1). Age was positively associated with the statistical model of eating habits (β = 0.04 [95% CI: 0.03 to 0.05]; p < 0.001). The stimulus exposure score was higher in Black individuals (p = 0.001–0.004), suggesting higher exposure to obesogenic food stimuli. The leisure activity score was nominally higher in the Hispanic/Latinx group (p = 0.03). There was no association between satiety responses and SDQ subscales. In the longitudinal models with repeated measures, there was an association between eating style/habits and SDQ subscales, except for peer and prosocial (Table 5). Perceived stress was positively associated with the eating style/habit score (β = 0.97 [95% CI: 0.6 to 1.3]; p < 0.001), although economic stress was not significant. Time was positively associated with the stimulus exposure score (β = 0.21 [95% CI: 0.1 to 0.3]; p = 0.001), suggesting that obesogenic stimulus exposure increased over time and specifically in self-reported Black individuals (p = 0.001–0.006). On the other hand, the leisure activity score was negatively associated with time (β = −0.35 [95% CI: −0.6 to −0.1]; p = 0.001), suggesting that structured and unstructured leisure time activities were more available over time. As in the baseline models, there was no association between satiety responses and SDQ subscales. The caretaker score, housing score, and gender were not significant in any model. Sensitivity analyses with empirical data aligned with these results.

TABLE 5.

Association of FEAHQ-R scores with SDQ subscales in linear mixed models with repeated measures.

SDQ subscale	Eating style/habits		Satiety response		Food environment		Leisure activity
Number of participants	478	456	476	459	483	463	461	448
	Model 1	Model 2	Model 1	Model 2	Model 1	Model 2	Model 1	Model 2
	β (95% CI)	β (95% CI)	β (95% CI)	β (95% CI)	β (95% CI)	β (95% CI)	β (95% CI)	β (95% CI)
Emotional problems	0.31** (0.13 to 0.48)	0.19* (0.01 to 0.36)	−0.05 (−0.11 to 0.00)	−0.05 (−0.11 to 0.01)	0.22** (0.06 to 0.37)	0.15 (−0.02 to 0.32)	0.20* (0.02 to 0.38)	0.18 (−0.01 to 0.37)
Perceived stress		0.97*** (0.63 to 1.31)		−0.09 (−0.19 to −0.01)		0.32** (0.13 to 0.52)		−0.12 (−0.13 to 0.76)
Conduct problems	0.50*** (0.29 to 0.71)	0.40*** (0.19 to 0.61)	−0.02 (−0.09 to 0.05)	−0.01 (−0.08 to 0.07)	0.06 (−0.14 to 0.25)	0.02 (−0.19 to 0.23)	0.27* (0.05 to 0.50)	0.26* (0.03 to 0.49)
Perceived stress		0.97*** (0.64 to 1.31)		−0.11* (−0.21 to 0.0)		0.37*** (0.18 to 0.56)		0.35 (−0.09 to 0.79)
Hyperactivity scale	0.54*** (0.40 to 0.68)	0.47*** (0.33 to 0.60)	0.04 (−0.01 to 0.09)	0.05* (0.00 to 0.10)	0.21** (0.09 to 0.33)	0.18** (0.06 to 0.31)	0.13 (−0.01 to 0.27)	0.11 (−0.03 to 0.26)
Perceived stress		0.88*** (0.55 to 1.21)		−0.12* (−0.22 to −0.02)		0.32** (0.13 to 0.51)		0.34 (−0.10 to 0.78)
Peer problems	0.11 (−0.12 to 0.33)	0.04 (−0.18 to 0.27)	0.04 (−0.03 to 0.12)	0.05 (−0.02 to 0.13)	0.07 (−0.13 to 0.26)	0.04 (−0.16 to 0.24)	0.17 (−0.07 to 0.40)	0.16 (−0.08 to 0.40)
Perceived stress		1.03*** (0.69 to 1.37)		−0.11* (−0.21 to −0.01)		0.37*** (0.18 to 0.56)		0.37 (−0.06 to 0.81)
Prosocial scale	0.03 (−0.14 to 0.20)	0.06 (−0.11 to 0.23)	0.00 (−0.07 to 0.06)	−0.02 (−0.08 to 0.04)	−0.00 (−0.15 to 0.15)	0.03 (−0.12 to 0.18)	−0.17 (−0.34 to 0.00)	−0.15 (−0.33 to 0.02)
Perceived stress		1.01*** (0.67 to 1.34)		−0.10* (−0.21 to −0.00)		0.37*** (0.18 to 0.55)		0.37 (−0.07 to 0.80)

Note: Model 1 is the base model, with covariates of age, sex, and race and ethnicity. Model 2 is a saturated model that includes socioeconomic scales (caretaker score and economic and perceived stress). Bolded values indicate statistical significance, with asterisks indicating degree of significance.

Abbreviations: FEAHQ-R, Family Eating and Activity Habits Questionnaire-revised; SDQ, Strengths and Difficulties Questionnaire.

^*p = 0.05–0.01.

^**p = 0.01–0.001.

^***p < 0.001.

DISCUSSION

This study of diverse sociodemographic cohorts identified higher behavior and emotional problems, hyperactivity/inattention, peer relationship problems, and lower prosocial behavior scores in children during a period of prolonged stress than previously defined age/gender norms. During the COVID-19 pandemic-induced period of stress, these psychological problems were associated with higher obesogenic eating behaviors, but not satiety responses. Perceived stress was a significant covariate, but not the other sociodemographic variables, highlighting the dominant role of perceived stress across diverse populations. These findings provide evidence of the prominence of hedonic behaviors during periods of prolonged stress in young children. A higher obesogenic stimulus exposure was associated with Black race and leisure activity scores in the Hispanic/Latinx group. Structural preventive measures such as access to healthier foods and the ability to seek structured or unstructured leisure activities (which may protect against obesity) require special attention in historically marginalized population groups.

Our study quantified the emotional and behavioral measures in children aged 5 to 12 years during the initial 6 months of the COVID-19 pandemic, which were validated in the follow-up measurements completed within the same school year. These results highlight the prolonged impact of COVID-19 pandemic-related stress and restrictions across the demographic and socioeconomic groups. In a systematic review of 29 studies from the first year of the COVID-19 pandemic, Racine et al. noted that, globally, one in four youth experienced depression symptoms, and one in five youth had anxiety symptoms, nearly double compared with the prepandemic period [27]. Another review of a larger number of studies (n = 120) identified that nearly 31% of youth had depressive and anxiety symptoms, and 42% of youth (from 50 studies) had sleep disturbances [28]. Mental health symptoms were worse in younger children, emphasizing the relevance of regular routines and school attendance at an early age and highlighting the prolonged disruption to youths’ daily routines, academic milestones, social interactions, potential family illnesses, and loss of caregivers. Thus, the COVID-19 pandemic served as a natural model for assessing the impact of emotional and behavioral stress on eating behaviors.

Energy regulation is a complex interplay between hedonic (i.e., orexigenic) and satiation (i.e., anorexigenic) behaviors. Our results demonstrated the uncoupling of these behaviors with an increase in obesogenic eating habits, but not satiety responses. In adults, chronic stress increases disordered eating, often with maladaptive eating behaviors characterized by overconsumption (i.e., eating beyond the point of satiety or metabolic drive) and loss of control eating of predominantly highly palatable, processed, energy-dense foods [29]. Over time, these reward-seeking circuits are reinforced, albeit with a reduction in reward duration, resulting in continued energy intake without benefits [30]. Human functional magnetic resonance imaging (MRI) studies have shown heightened activity in subcortical reward processing (i.e., the amygdala, caudate, putamen, and inferior frontal gyrus) in response to high-calorie food stimuli [31,32]. Although the exact mechanisms underlying these findings are not known, animal research suggests that stressresponsive hormones (e.g., glucocorticoids, corticotrophin-releasing hormone) and metabolic factors (e.g., insulin, ghrelin, leptin) influence dopaminergic transmission in the brain. Prolonged glucocorticoid administration increases the expression of orexigenic neuropeptide agouti-related peptide (AgRP) and neuropeptide Y (NPY) [30]. Recent animal studies have demonstrated a catecholaminergic circuit between the stress-activated nucleus of the tractus solitarius and melanocortin 4 receptor (MC4R) neurons of the paraventricular hypothalamus. In a series of experiments, Laule et al. demonstrated that stress-induced activation of nucleus of the tractus solitarius projections to the paraventricular hypothalamus inhibited MC4R neurons via nor/epinephrine release, resulting in increased secretion of AgRP/NPY [33]. They also showed an important role of these neurons in the physiological stress response. When prolonged, such responses reinforce maladaptive behaviors that promote obesity in the long term [29]. Our findings provide evidence to support these proposed mechanisms in young children. Although our study did not have access to long-term BMI data of the participants to assess changes in obesity status, many other studies have documented BMI gains and an increase in the magnitude of obesity during the pandemic [2,3,34,35].

The association between perceived family stress and obesogenic eating styles and habits underscores the importance of parental mental health in children’s lives. It is unknown whether this was the result of a lack of financial and other support resources observed in the baseline survey. Economic stress was not significant in the longitudinal models; therefore, the results may be more reflective of the parental response to perceived stress, as has been reported in other studies [36,37]. Responsive feeding is an important tool in childhood obesity management, and the American Academy of Pediatrics guidelines for the management of childhood obesity emphasize the role of family participation [38]. The family systems theory emphasizes the interconnected nature of family members and the importance of family-level interventions in both clinical practice and research [39]. The use of the FEAHQ-R allowed us to capture the role of family, highlighting the importance of food parenting during periods of enhanced stress.

We noted a positive association between age and obesogenic behavior. In a study of 428 twins evaluated ages 4 and 10 years, Ashcroft et al. noted an increase in food responsiveness and a decrease in satiety in older children [40]. These findings were replicated in a more recent report of a longitudinal cohort of 167 mother–child dyads evaluated at ages 5 and 9 to 11 years [41], highlighting the overbearing of environmental and behavioral influences on biological appetite regulation as children grow.

In this study, we observed higher stimulus exposure scores for individuals of the Black race. This indicates that self-reported Black households had more obesogenic food exposure stimuli than White households. Similarly, obesogenic leisure activity scores were higher among Hispanic/Latinx individuals. In other words, these individuals had less access to structured and/or unstructured leisure activities, which decreased obesogenic tendencies and enhanced mental health [42]. A prior cross-sectional study of 1324 US youth associated the lack of leisure time access during the COVID-19 pandemic with neighborhood safety, which may have a role in the population from the Inwood clinic [43]. However, it is also possible that these disparities were linked to the disproportionately higher COVID-19 infection rates in racial and ethnic minority families during the study period [44].

The findings of this study have important implications for interventions. The higher obesogenic behaviors observed in the FEAHQ-R are more amenable to structural interventions than biological parameters such as satiety. Reducing obesogenic eating styles is potentially more amenable to intervention by making parents aware of their children’s behaviors and targeting binge and emotional eating. Relieving inciting stress under such conditions remains an important aspect of overall intervention. At the policy level, addressing structural changes to overcome barriers to healthier food environments and/or availability of leisure activities remains important.

The strengths of this study include its relatively large sample size and the inclusion of diverse demographic, cultural, and socioeconomic groups, allowing for its applicability to many population groups. Owing to COVID-19 pandemic-related restrictions, the study was limited by internet/phone-based contact methods, as in-person access was difficult; therefore, the results are liable to nonresponse bias because only individuals and/or families with a mobile device/computer could participate. The same survey design also limited the use of descriptive items to unique outlier issues faced by families beyond the scope of the questions. The FEAHQ-R incorporates the traits of the child, caregivers, and food environment, providing a more holistic view of the obesogenic traits in the family, and is highly relevant for family-based interventions. In this study, race and ethnicity were self-reported and verified by documentation in the EHR, but the ability to interpret the results for each self-reported social group was limited. Because this study only lasted 6 months in one metropolitan area, it is not known whether the findings apply to individuals facing longer durations of stress or those who are beyond the geographic areas of the study. Contrary to our expectations, there was no significant association between obesity and BMI in the present study. We did not have access to data on changes in BMI during the study period because of clinical closures, which may have provided better information. Although the SDQ reference is representative of the US population, there could have been temporal changes since it was conducted that may make the identified differences in children’s stress levels less prominent. However, the SDQ has stood the test of time in many countries, and the differences observed are stark, such that some temporal increases are unlikely to change the interpretation of the results. We postulate that emotional and conduct behaviors mediate stress toward obesogenic behaviors. However, the current study design did not allow formal mediation analyses, as it did not meet the requirement for temporal changes. Further studies are required to address these limitations. Despite these limitations, the moderately large sample size allowed for the identification of patterns and associations that have implications in both cases of stressful emergencies and in the general management of eating behaviors and food regulation in young children. The questionnaires used in this study do not allow the exploration of different parenting styles (e.g., authoritative, authoritarian, permissive), and it may be useful in future studies to dissociate parental behaviors that may be amenable to interventions for the well-being of children.

CONCLUSION

Eating behaviors in early childhood play a key role in shaping the long-term habits and health of children, especially in relation to obesity. This study identified that prolonged psychological stressors experienced by children increased behavioral, emotional, and relationship problems and were associated with changes in their obesogenic eating behaviors but did not have any association with the biologically regulated domain of satiety responses, regardless of the presence of obesity. Perceived child and family stress plays a significant role in this association, and racial disparities exist in the current obesogenic environment ripe for individual- and policy-level intervention.

Supplementary Material

Additional supporting information can be found online in the Supporting Information section at the end of this article.

Study Importance.

What is already known?

• Chronic stress is known to influence eating behaviors in adults and is an important contributor to obesity. The most prominent stress response behaviors in adults are emotional and binge eating, often of highly palatable, high-calorie, and low-nutrient foods.

• Stress-related eating disturbances have also been observed in children, but their features and mechanisms are less well defined.

What does this study add?

• This study quantified higher behavioral and conduct problems in 5- to 12-year-old children during the first year of the COVID-19 pandemic as a chronic stress model in three diverse cohorts compared with age- and gender-based US norms.

• Regardless of sociodemographic differences, there was an association between emotional and conduct problems and eating style/habits of the children, but not satiety response. Perceived stress in the family was an important covariate.

How might these results change the direction of research or focus of clinical practice?

• Obesogenic behaviors in young children in response to stress were observed in the potentially modifiable behavioral rather than the biological satiety domain. Interventions can be tailored to the specific hedonic behaviors during chronic stress.

• This study noted racial and ethnic disparities in the structural domains of food stimulus exposure and leisure activity access. These findings have policy implications to enhance equitable access.