Proximal and Distal Factors Associated With Obesity in Childhood: An Exploratory Structural Equation Model

ABSTRACT

Background

Obesity in childhood is complex; structural equation modeling (SEM) offers an approach to ascertain complicated relationships between weight and proximal (health behaviors, child health) and distal (family/household, community) variables. The objective of the study was to use SEM to map the influence of different etiological clusters on obesity in childhood.

Methods

Secondary analysis of baseline data from patients and parents enrolled in a multi‐site study of pediatric weight management was conducted. The primary outcome was weight status (utilizing percentage above the 95th percentile BMI, %95BMI). SEM was used to evaluate the influences of variables proximal and distal to child weight within the context of the Ecological Model of obesity in childhood.

Results

Complete data on 375 child‐parent dyads were imputed. Proximal factors (child stress and mobility, parent BMI) were significantly related to %95BMI; some distal factors (family level) did not have a direct effect on %95BMI, but did indirectly through proximal factors (such as child stress). Other distal factors (neighborhood deprivation represented by the Area Deprivation Index) were significantly related to %95BMI and family‐level factors.

Conclusions

Except for the distal factor of neighborhood deprivation, proximal factors were the drivers of weight status. Distal factors did have indirect effects via proximal factors. SEM provides a means to investigate the complex contributors to childhood obesity, and could identify key leverage points for intervention.

Abbreviations

%95BMI

percent of the 95th percentile BMI

ADI

area deprivation index

EMOC

ecological model of obesity in childhood

GOF

goodness‐of‐fit

SEM

structural equation modeling

1. Introduction

Obesity in childhood is a challenging public health problem, and the causes and contributors are complex [1]. Vandenbroek et al. illustrated eight interconnected etiological clusters pertaining to childhood obesity [2]; (1) social psychology for example, sociocultural value of food; (2) individual psychology for example, experience of stress; (3) physiology for example, prenatal risk factors; (4) individual psychology for example, level of satiety; (5) physical activity environment for example, neighborhood safety; (6) food production; (7) food consumption; and (8) individual physical activity. The influence of these clusters can occur singly or in combination at the individual, family, community and society level, as aptly described in the Ecological Model of obesity in childhood (EMOC) (Figure 1) [3]. This broader view allowed for integration of social drivers of health (housing, transportation, neighborhood deprivation) [4] with more direct influences or dynamics in the child's family [5, 6, 7, 8]. The function or dynamics of a family for example, the relationships, communication, and organization within the family unit, can affect the child's health and weight [9, 10]. For instance, previous studies on children with severe obesity have shown higher levels of impaired family functioning compared with children with lower levels [4, 7].

FIGURE 1.

Ecological model of obesity in childhood (adapted from Davison and Birch [3]).

The complexity of obesity in childhood has been driven by the interaction of risk factors at different levels of influence, making it difficult to isolate the effect of a single factor. Another challenge has been determining which factors are exposure, mediator, moderator, or confounder variables in the causal model [11]. One approach to investigate complex causal relationships is structural equation modeling (SEM), a confirmatory statistical method commonly used to ascertain hypothesized relationships among multiple variables [12]. For obesity in childhood, SEM can allow for mapping factors that are proximal to the child and influence weight directly, together with more distal factors that represent the broader context of the child's environment. Proximal factors (health behaviors, child health and wellbeing, child stress) are those immediately surrounding and potentially influencing weight status on the EMOC, while distal factors (neighborhood characteristics, family factors such as family function, home environment, parent health and wellbeing, parenting) are those representing broader, environmental or indirect influences, which can then affect or impact proximal factors, thereby indirectly influencing child weight.

Using the EMOC as a framework, the objective of this observational study in young children engaged in pediatric weight management programs was to map the influence of different etiological clusters for obesity in childhood using SEM. It was hypothesized that distal risk factors that indirectly influence weight status in children, such as family functioning, parenting, parent health, and home environment, would have a similar effect on weight status as more proximal factors likely to have a direct influence (lifestyle behaviors, child health).

2. Materials and Methods

2.1. Study Design and Participants

This study was a cross‐sectional analysis of baseline data in a prospective study to develop a prediction model of attrition from pediatric weight management, the Stay in Treatment (SIT) study (clinicaltrials.gov identifier: NCT04364282) [13]. The SIT study enrolled children aged 7–18 years of age and a parent/guardian at their initial clinic visit at a pediatric multi‐disciplinary weight management programs at four sites across the United States: the Brenner FIT (Families in Training) program at Brenner Children's Hospital, located in Winston‐Salem, North Carolina; the Optimal Wellness for Life (OWL) program at Boston Children's Hospital in Boston, Massachusetts; the Promoting Health in Teens and Kids (PHIT Kids) program at Children's Mercy Kansas City, Missouri; and the Center for Healthy Weight and Nutrition at the Nationwide Children's Hospital, in Columbus, Ohio. The study design and protocol have been previously published [13]. The study was approved by the Institutional Review Board at Wake Forest University School of Medicine (IRB00062191) with reliant approval at additional sites.

Inclusion criteria were body mass index (BMI) ≥ 95th percentile for age and sex, ability to complete study measures, and the ability to speak either English or Spanish. The parent or legal guardian were the primary adult accompanying the child to treatment. The child's primary residence must be with that parent. Exclusion criteria were diagnosis of type 2 diabetes, history of a weight‐related genetic or health condition (e.g., Prader‐Willi Syndrome), and inability to complete study measures and/or activities for example, due to intellectual disability.

2.2. Measures

The primary outcome for the analysis was weight status of the child, using percent of the 95th percentile BMI (%95BMI). Given the population of children being referred for weight management and study criteria, in which all children were at or above the 95th percentile BMI, %95BMI will provide differential levels of obesity in children, capturing those at higher categories of obesity (such as severe obesity, defined as ≥ 120% of the 95th percentile). Weight and height measurements of the child and parent/guardian were obtained using established protocols [13]. Because of the COVID‐19 pandemic, home measurement protocols supervised virtually by the study staff [14], based on Center for Disease Control and Prevention guidelines, were periodically used [15].

For this secondary analysis, measures chosen for the original study were used and were classified, a priori, as being either proximal or distal to the outcome. Proximal factors (health behaviors, child health and wellbeing, child stress) are those immediately surrounding the weight status of the child; distal factors (neighborhood characteristics, family factors such as family function, home environment, parent health and wellbeing, parenting, and general parent characteristics) are those representing broader, environmental, or indirect influences.

2.2.1. Demographic Measures

Demographic characteristics included age, race and ethnicity, and sex. Multiple socioeconomic status measures were obtained including parent education level, household income, and child insurance provider (government/Medicaid vs. commercial). Food insecurity was assessed using the 2‐item Hunger Vital Sign [16].

Measures for the original study were chosen to capture or represent elements known to be: important to adherence in pediatric medical regiments (as a marker of attrition) [17, 18], representing elements of the EMOC, or associated with childhood obesity. This secondary analysis is making use of these collected data to explore complex interactions of these variables via SEM.

2.2.2. Proximal Factors

The Family Nutrition and Physical Activity (FNPA) Screening Tool was used to assess family eating, activity, and health habits [19]. The FNPA is a convenient clinical measure used in longitudinal studies of obesity in childhood [20], with good construct validity and internal consistency. Child health and wellbeing was measured by the PROMIS Pediatric Profile 25 – Short Forms, assessing child anxiety, depression, fatigue, pain, physical function/mobility, and peer relationships [21]. Child stress was measured by the PROMIS Psychological Stress Experience Short Form 4a [21].

2.2.3. Distal Factors

The Family Assessment Device (FAD), General Functioning Subscale was used to assess family functioning [22, 23]. The Confusion, Hubbub, and Order Scale (CHAOS) assessed the home environment with regard to organization [24], and has been validated against direct observation. Parent stress was measured by the Parent‐Perceived Stress Scale [25]. Parent health and wellbeing were measured by the PROMIS‐29 Profile v2.0 [21], which assesses adult anxiety, depression, fatigue, physical function, pain, sleep, and ability to participate in social activities. General parenting was assessed by the Child Report of Parent Behavior Inventory, with parent behavior measured in three dimensions: psychological control versus autonomy, acceptance versus rejection, and firm versus lax control [26]. Child perception of the family was assessed by the PROMIS Pediatric Family Relationships measure [21]. Other family characteristics were self‐reported by the parent: parent marital status, highest level of education attainment, and insurance provider (also cross‐checked against the electronic medical record if unknown or discrepancy). Parent health literacy was captured by the Newest Vital Sign, which measures a person's ability to understand words, numbers, and forms [27]. Perceived or anticipated financial burden of program participation was measured by a valid and reliable measure of financial toxicity of treatment, minimally modified for use in a pediatric weight management program [28]. Finally, neighborhood level of deprivation was captured by linking participants' 9‐digit zip code to the American Community Survey, with data extracted to determine the Area Deprivation Index (ADI), which includes factors for income, education, employment, and housing quality [29].

Parent weight status (BMI) was considered from two perspectives: as a child‐level variable (proximal factor) representing genetic risk, and as a family‐level (distal factor), representing shared environmental risks, and done so in the analysis plan.

2.3. Analysis

Descriptive and summary statistics such as frequency count and proportion as well as mean and standard deviations were utilized where appropriate. Visualization was used to summarize data and inspect for possible outliers. Using the primary outcome (%95BMI) as the target variable, bivariate analysis result‐correlation was used to examine the association between continuous variables. Chi‐squared test and ANOVA were used to examine relationships for categorical variables.

The primary model used to analyze the association of family dynamics and weight status is the full cascading SEM. Figure 2 shows the hierarchical structure of the full cascading SEM model, a factor that potentially affects the outcome (%95BMI) and belongs to one and only one specific level of the hierarchy. The full cascading model is characterized by the feature that, except for a logically implausible effect, any individual variable within a higher level is hypothesized to affect all individual variables at a lower level. A total of 33 variables divided into three levels (Figure 2) were considered. The full list of variables is reported in Table 1. Guidance from published reports for the appropriate measures was followed. For example, if an overall summary score was available from an instrument, the overall summary score was used. For some measures that are deemed multidimensional, for example, PROMIS Pediatric Profile 25 – Short Forms and PROMIS‐29 Profile v2.0 [21], current practice and only domain‐level scores were used.

FIGURE 2.

Full cascading model for weight status in the child. Variables within each individual level are not shown for readability. Except for a logically implausible effect, any individual variable within a higher level is hypothesized to affect all individual variables at a lower level.

TABLE 1.

Variables considered for inclusion into the ecological SEM model.

	Correlation coefficient (with %95BMI)	p‐value
Child characteristics
Anxiety a	0.11	0.04
Depression	0.09	0.1
Fatigue	0.11	0.04
Mobility	−0.27	< 0.0001
Peer relationship	−0.08	0.15
Pain interference	0.1	0.07
Psychological stress	0.16	0.003
Family relationships	−0.02	0.71
Child's sex	—	0.23
Child's race	—	0.38
Child's age	0.04	0.44
Family characteristics
Family functioning (family assessment device)	−0.05	0.39
Family nutrition and physical activity	−0.15	0.005
Home environment (CHAOS score)	0.03	0.62
Parent stress	0.09	0.08
Parent anxiety	0.05	0.33
Parent depression	0.07	0.18
Parent fatigue	0.11	0.03
Parent physical function	−0.12	0.04
Parent pain interference	0.08	0.15
Parent sleep disturbance	0.09	0.09
Parent social activities	−0.05	0.37
Parenting style—Psychological control subscale	−0.09	0.1
Parenting style—Acceptance subscale	−0.02	0.73
Parenting style—Firm vs. lax	−0.06	0.22
Health literacy	0.02	0.67
Food insecurity	0.14	0.01
Parent education	−0.13	0.02
Parent income	−0.08	0.15
Parent BMI	0.33	< 0.0001
Financial burden	−0.16	0.003
Insurance (collapsed into 3 categories)	—	0.75
Environmental factor
Area deprivation index (ADI)	0.17	0.004

^a Variables that are significant at a p < 0.05 level are bolded.

The cascading model was operationalized in several stages. In Step 1, to limit the number of variables included in the final SEM model, only demographic and contextual variables that were significantly associated (p < 0.05) with the weight status (%95BMI) in the bivariate analysis were allowed. For example, distal factors of Family Assessment Device and PROMIS Pediatric Family Relationships measures were not associated with %95BMI, and therefore not included in the SEM analysis. In Step 2, a “baseline” SEM with only the significant variables identified in Step 1 under the full cascading model was tested. Goodness‐of‐fit (GOF) statistics were assessed. In Step 3, the baseline model was refined to improve the GOF of the model to the data. Residual analysis identified variable pairs that had large‐than‐usual standardized residuals (> 3.0) in the fitted covariance matrix. Because the cascading model did not specify causal links for variables within each domain (e.g., family), large residuals might arise for within‐domain variables. Association relationships for the larger‐than‐usual variable pairs were added to the model and GOF was reassessed. Additionally, when two variables were contextually similar and highly correlated, only one was considered to be included. The refinement process was iterated until arriving at a final model with acceptable GOF.

To assess the agreement between the data and the primary model, as well as for the purpose of comparing across various models, GOF indexes including AIC, BIC, CAIC, SMRS, RMSEA, CFI, TLI, and were used. Lower values of AIC, BIC, and CAIC indicate better fit but no cutoffs exist because these values are data dependent. Reference acceptable criteria for SMRS, RMSEA, CFI, and TLI were respectively < 0.08, < 0.08, > 0.90, and > 0.90 [30, 31].

Besides the primary SEM full cascading model, other models were analyzed as a comparison to the full cascading model. The two other models included: (a) a restricted cascading model, that is, a model in which factors within a higher level only affect factors at the immediate next lower level and not further and (b) multivariable regression analysis.

As the measures were all treated as observed, and no latent variable was created, the SEM analysis was equivalent to path analysis. For interpreting the path coefficients, the significance level was set at p = 0.05, and all tests were two‐sided. Measures were standardized to allow better interpretation of results. The SEM was implemented in SAS PROC CALIS.

3. Results

Table 2 shows the characteristics of the sample. Data on a total of 382 participants were collected, but four participants had missing values in BMI. The %95BMI variable also contained a few outliers. After inspecting the data, it was decided that data points with a percentage of 95 BMI > 300 would be excluded. As a result, three observations were excluded and the sample size was n = 375. Child sex was nearly even between male and female, mean age 12 years, almost half white (45.5%), and 21% identified as Hispanic. Figure S1 shows the distribution of %95BMI after the outliers were removed. No transformation was conducted for this variable after the removal of the outliers.

TABLE 2.

Demographics of the sample (N = 382).

Variable (categorical)	n (%) a
Child sex
Male	184 (48.4)
Female	196 (51.2)
Child ethnicity
Hispanic	80 (21.4)
Non‐Hispanic	294 (78.6)
Child race
Asian	3 (0.8)
Black	119 (31.1)
White	174 (45.5)
Native American	5 (1.4)
Biracial	46 (12)
Other	35 (9.2)
Parent marital status
Single	93 (26.0)
Married or domestic partnership	200 (55.9)
Widowed	11 (3.1)
Divorced	36 (10.1)
Separated	18 (5.0)
Insurance
Commercial: Employer, private	150 (39.5)
Government/public: Medicaid/Medicare	223 (58.7)
Uninsured self‐pay	5 (1.3)
Other	2 (0.5)
Household income (per year)
Less than $20,000	57 (15.2)
$20,000–$39,000	107 (28.6)
$40,000–$59,000	68 (18.2)
$60,000–$99,000	48 (12.8)
More than $100,000	69 (18.5)
Prefer not to answer	25 (6.7)
Site
Site 1	137 (35.9)
Site 2	130 (34.0)
Site 3	46 (12.0)
Site 4	69 (18.1)

Variable (continuous)	Mean (SD)
Child age (N = 382)	12.0 (2.7)
Child BMI (n = 378)	35.7 (7.7)
Child % of the 95th percentile	145.8 (27)
Parent BMI (n = 375)	37.0 (9.6)

^a Some values may not equal total N due to incomplete responses.

Table 1 shows the correlations between relevant measures, and when pertinent, the relevant measure subscales with %95BMI. Table 1 also shows the variables that were retained in the SEM analysis—child characteristics, anxiety, fatigue, mobility, psychological stress, and parent BMI (as a proxy for genetic risk); and family characteristics, parent BMI, fatigue, physical function, food insecurity score, FNPA, education, and financial burden. Demographic variables including sex, age, race (child), marital status, and insurance were not included in the final model, as these variables did not meet the inclusion criterion from the bivariate analysis.

During the refinement of the SEM for the EMOC, it was determined to not include the food insecurity score because of its high correlation and conceptual overlap with financial burden. Parent fatigue was also not included as its significant causal pathways only involved other parent reported outcomes. Figure 3 shows the final full cascading SEM model. Within‐level associations, which would have been indicated by double arrows, were not shown to improve readability.

FIGURE 3.

The fully cascading structural equation model for weight status within an ecological model of obesity in childhood. The levels include outcome (%95BMI), child characteristics (PROMIS Child's anxiety; PROMIS child's mobility; child's stress; parent's BMI‐proxy for genetic factor), family characteristics (FNPA—Family Nutrition and Physical Activity; parent education; parent physical function; family financial burden), and community and societal characteristics (ADI‐measure of area deprivation). Thickness of arrow indicates different levels of statistically significant pathway, *p < 0.05, **p < 0.01, ***p < 0.001; dotted arrow indicates statistically non‐significant pathway. The model also includes several association pathways within each level of the ecological model (not indicated in the figure for readability)‐ Child's stress and Child's anxiety (0.70***), Child's stress and Child's mobility (−0.21***), Child's anxiety and Child's mobility (−0.2***) (Child Characteristics), Parent physical function and FNPA (0.17**), parent physical function and financial burden (0.25***) (family characteristics). Associations are often depicted as double arrows in a path diagram.

Table 3 shows the GOF indexes for three models: (a) the full cascading model, (b) a cascading model in which factors within a higher level only affect factors at the immediate next lower level and not further, and (c) a multivariable regression analysis. The full cascading model met the criteria of excellent fit for all of GOF indexes including SRMR (0.06), RMSEA (0.04), CFI (0.97), and TLI (0.92). As sensitivity analysis and noted earlier, the variable parent BMI was treated both as a family level and a child‐level factor. However, from the SEM GOF analysis, it was observed that when treating parent BMI as a proxy for the child's genetic and/or environmental risk, the model performed significantly better. Consequently, parent BMI was included at the more proximal level of children's characteristics, though it is recognized that it could represent both.

TABLE 3.

Goodness‐of‐fit indexes for the full cascading model, a restricted cascading model, and the regression model.

Goodness‐of‐fit criterion	Full cascading model	No cross‐level cascading model	Regression model a
AIC	96.4	101.4	110.0
CAIC	232.6	228.1	359.6
BIC	202.6	200.1	304.6
SRMR (< 0.08)	0.06	0.06	—
RMSEA (< 0.08)	0.04	0.05	—
CFI (> 0.9)	0.97	0.95	—
TLI (> 0.9)	0.92	0.90	—

^a Many standard goodness‐of‐fit criteria do not apply to the saturated regression model.

The SEM analysis (Figure 3) showed that proximal factors including child stress, child mobility and the proxy for child's genetic risk (parent BMI) were significantly related to the %95BMI, all in the expected direction. For example, a higher level of child mobility was related to lower %95BMI (standardized effect = −0.14, p < 0.05), whereas a higher level of child stress was related to higher %95BMI (0.26, p < 0.01). None of the distal, family‐level factors had a direct effect on %95BMI, but some indirect effects were delineated (e.g., FNPA→ Child's stress → %95BMI). Finally, ADI, which represents the community and neighborhood influence on a child's health, was significantly related to %95BMI (0.19, p < 0.01) as well as to all of the family‐level factors (parent characteristics, FNPA, finances).

4. Discussion

In this secondary analysis of multi‐level data of patients and families pursuing weight management, proximal and distal variables had differential relationships with weight status. Proximal factors such as child stress were directly associated with weight status in the children. However, the hypothesis about distal variables directly influencing weight in children was largely disproven; the only distal variable directly associated with weight status in children was the ADI. It was hypothesized that distal variables, including family items such as family dynamics, relationships, parenting, and parent health and well‐being, would still have strong enough influences that they would have an association with weight status in the child, but this relationship was not observed. However, there were distal variables (family health habits, parent health) that had direct associations with proximal variables, such as the FNPA measure and child stress and anxiety.

As demonstrated in the conceptual model (Figure 1), it was expected that the environment, as depicted by social drivers of health, has a significant impact on many aspects of child health [32]. In this study, the ADI, representing the distal variable of housing, income, and education, had strong associations with other factors (FNPA, parent education, and parent physical function), and %95BMI, reflecting how neighborhoods can impact individual (child) and group‐level (family) behaviors. This analysis found differential effects of parent BMI, reflecting it both as a family‐level factor (a proxy for family health behaviors) and genetic risk (obesity being highly heritable). Both effects of parent BMI are supported by the literature. For instance, the risk of obesity in the offspring was significantly higher when one parent is affected by obesity, and even higher when two parents were affected [33]. Given recent understanding of obesity as a disease and genetic contributors to obesity, as well as environmental contributors to parent BMI, treating parent BMI as a proxy for genetic risk may be the best approach in this situation.

Child‐level factors of stress and mobility had direct relation to child weight, fitting with the present evidence supporting those relationships. Several studies support physiologic and psychologic interactions between stress and weight [34], as also noted in the recent obesity guidelines of the American Academy of Pediatrics [4]. This relationship was especially noted during the recent COVID‐19 pandemic, with chronic stress resulting in increased weight at population levels [35]. Stress was identified as important in the original study, with general psychosocial health impacting attrition from weight management [36, 37, 38]. Impaired mobility was also found to be associated with increased weight status, as seen in other studies [39, 40, 41]. Higher weight in children is known to impact motor skills and gain, and reduce the ability to move comfortably. Greater body weight can alter normal gait patterns and increase stress on bones and joints, leading to pain and decreased activity.

The results of this cross‐sectional analysis support the complexity of obesity in childhood, reflecting the complex interaction of environment, families, behavior, and genetics. A more focused, prospective study aimed at identifying the mechanisms of environmental influence on family behaviors and dynamics would further help to identify key leverage points to exert change at the public health level. An example would be behavioral interventions to change family health habits; if habits are associated with home chaos or parenting, then the interventions should include elements of parent training and family therapy. For children who live in neighborhoods with higher levels of neighborhood deprivation, including home assessments that specifically identify and intervene on factors that involve family dynamics and behaviors may be warranted. From a public health perspective, large‐scale epidemiologic studies could find linkages between obesity in childhood and environmental measures, family factors (distal), family health, and health behaviors (proximal), which could focus interventions on the areas most likely to yield the greatest improvements in health.

There are several limitations to this study. This was a cross‐sectional study and a secondary analysis. The measures chosen were for a study of attrition, and not necessarily for this complex SEM analysis. These measures of family dynamics have not been widely used in obesity research focused on the pediatric population, likely not exhaustively capturing elements of individual relationships across complex family structures. As this was a convenience sample, it was not specifically designed to measure the impact of other aspects of the family, such as family structure (dual vs. single parent, one vs. numerous siblings). Measures were representative of proximal and distal factors, chosen by investigators, and may not fully represent some constructs, such as parent BMI as proxy for genetic risk. The COVID‐19 pandemic may also have affected aspects of the study, including the wellbeing of participants as it relates to health habits and stress. Although this was a large sample of treatment‐seeking children and families recruited from four different sites, the children had significant obesity and may not be representative of all children with obesity. However, given the complexities of obesity treatment focused on the family, the study was an initial step to explore this type of modeling, and demonstrated the concept of proximal and distal factors, highlighting the complexity of obesity in childhood, particularly environmental and familial influences on weight in children.

Weight in childhood is influenced by many biological and environmental factors, including genetics, health behaviors, family, and the surrounding environment. In this sample of children and families enrolled in multidisciplinary weight management, measures representing multiple levels of the EMOC to explore differential associations of proximal and distal factors on weight were utilized. Proximal factors, such as child stress, were directly associated with weight status. The only distal variable directly associated with weight status in children was the ADI. A prospective study is needed to further investigate the complexity of obesity in childhood, taking into account the multiple levels of interacting variables and how they influence child and family weight and health.

Author Contributions

Joseph A. Skelton: conceptualization, investigation, funding acquisition, writing – original draft. Erinn T. Rhodes and Sarah E. Hampl: conceptualization, investigation, writing – original draft. Ihuoma Eneli: investigation, writing – review and editing. Edward H. Ip: conceptualization, methodology, software, formal analysis, writing – original draft. All authors approve of the final version for submission and agree to be accountable for the work.

Funding

This study was funded by the National Institute of Nursing Research (Grant R01NR017639).

Conflicts of Interest

Dr. Joseph A. Skelton is Editor‐in‐Chief of the Journal Childhood Obesity (Mary Ann Liebert Inc. Publishers) and is a writer and reviewer for UpToDate. Drs. Erinn T. Rhodes, Sarah E. Hampl, Edward H. Ip, and Ihuoma Eneli have no conflicts to disclose.

Supporting information

Figure S1: Distribution of the variable percentage of the BMI 95th percentile (%95BMI).

Skelton, Joseph A. , Rhodes Erinn T., Hampl Sarah E., Eneli Ihuoma, and Ip Edward H.. 2026. “Proximal and Distal Factors Associated With Obesity in Childhood: An Exploratory Structural Equation Model,” Obesity Science & Practice: e70124. 10.1002/osp4.70124.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.