Family resilience and childhood obesity among children exposed to adverse childhood experiences in a national survey

Abstract Objective Adverse childhood experiences (ACEs) contribute to poor overall health among children with obesity. This study evaluated how one potential protective factor—family resilience—affects the association between ACEs and childhood obesity. Methods This analysis was a secondary analysis of the 2016–2018 National Survey of Children's Health (NSCH), a repeated cross‐sectional survey based on parent report. Nine ACEs were queried. Family resilience was assessed with four items (potential range 0–12). The primary outcome was child weight status. Multivariable ordinal logistic regression was used, adjusting for potential confounders and the interaction between ACEs and family resilience. Results For 49,365 children ages 10–17, the median number of ACEs was 1 (IQR 0, 2), the median family resilience score was 10 (IQR 8,12), 15.3% of children had overweight, and 15.4% of children had obesity. Among the 51.3% of children who experienced one or more ACEs, higher family resilience scores attenuated the odds of being in a higher weight category. This pattern was not observed in children with zero ACEs. Conclusions In the 2016–2018 NSCH, children ages 10–17 who were exposed to ACEs had higher rates of overweight and obesity, the odds of which may be reduced when children also have higher family resilience.

trauma. 5,6 However, social trauma does not deterministically lead to poor health outcomes. Protective factors like positive maternal mental health, neighborhood and school safety, and child resilience can reduce the risk of childhood obesity even in the face of ACEs. 7 Thus, characterizing protective factors that interrupt the pathway from social trauma to childhood obesity could augment intervention strategies designed to reduce health disparities among children exposed to ACEs.
ACEs are linked with childhood obesity through a well-established pathway. ACEs are childhood social traumas caused by abuse, neglect, violence, substance abuse, parent separation, peer victimization, community violence, unemployment, and property victimization. 8,9 These social traumas incite a physiological stress response that when experienced repeatedly, becomes toxic, altering the neural circuitry controlling these neuroendocrine (i.e., stress) responses. [10][11][12] The stress response has a complex and reciprocal relationship with healthy behaviors (e.g., healthy diet patterns, physical activity, sleep). Healthy behaviors can reset a dysregulated stress response, while stressassociated social-emotional dysfunction (e.g., poor self-regulation) can reduce the capacity to engage in healthy behaviors. This reduced capacity perpetuates or even heightens the stress response and results in negative health outcomes such as childhood obesity. 10,13 Resilience is an adaptation to adversity that "transforms potentially toxic stress into tolerable stress." 14 Specifically, family resilience "is the successful coping of family members under adversity that enables them to flourish with warmth, support, and cohesion." 15 Conceptually, family resilience is one protective factor that may be especially important for reducing the impact of ACEs on child obesity. 16,17 The hypothesized mechanisms by which family resilience potentially mitigates the impact of ACEs on childhood obesity are through the provision of (1) loving and safe parent-child relationships that reduce the impact of toxic stress and (2) stable routines (e.g., family mealtimes, quality time via family activities) that facilitate engagement in health behaviors to promote health child growth patterns. In addition, family resilience is not static, but rather something that can be promoted by targeted intervention. 18 Despite a strong conceptual relationship, associations between family resilience and childhood obesity have not been consistently reported. Notably, a recent investigation using the 2016 National Survey of Children's Health (NSCH) did not find associations between family-level resilience and lower rates of childhood obesity. 16 One potential explanation is that family-level resilience may only be activated as a protective factor for childhood obesity in the context of stressors, in this case ACEs. It may also be that family resilience may have different effects on health outcomes in some population subgroups. 19,20,21 With the goal of reducing health disparities in childhood obesity by targeting socially protective factors among traditionally minoritized communities, a careful examination of how resilience functions in communities of color should be a priority.
The purpose of this study was to examine how family resilience affects the association between ACEs and childhood overweight and obesity in a nationally representative sample of children ages 10-17. A second purpose of this study was to examine the extent to which the relationships between ACEs, family resilience, and overweight and obesity may vary by race and ethnicity. We hypothesized that children with higher family resilience scores would have an attenuated association between ACEs and childhood overweight and obesity when compared to children with lower family resilience scores. We also hypothesized that the relative strength of this attenuation would vary by a child's race or ethnicity.

| METHODS
We conducted a secondary analysis of data collected via the NSCH, a repeated cross-sectional survey of households in the United States with at least one child. 22 The purpose of NSCH is to collect data on The NSCH 2016-2018 survey administration methods are publicly available. 22 The survey was conducted in both English and Spanish and administered online and by mail. The US Census Bureau initially identified households that were likely to have a child living in the home, stratified by state. Households were then screened to confirm eligibility, and one child per household was randomly selected as the subject of the survey. The weighted overall response rate in 2016 was 53.0%, in 2017 was 37.4%, and in 2018 was 40.7%.
The respondent was a parent or caregiver in the home who could answer questions about the child's health. The sampling strategy was designed to over-sample children with special healthcare needs. To obtain population-based estimates, each child for whom a survey was completed was assigned a weight that included adjustments for the base sampling weight, non-response, the selection of single child in a household, and demographic characteristics.
The primary independent variable for this analysis is a count of ACEs experienced by the child and ranges from 0 to 9. The survey included nine items to assess a child's exposure to the following ACEs: (1) difficulty covering basic needs like food or housing, (2) parent divorce or separation, (3) death of a parent, (4) parent jail time, (5) domestic violence, (6) neighborhood violence, (7) household mental illness, (8) household drug or alcohol abuse, and (9) being treated unfairly because of race or ethnic group. The NSCH calculates a total count of ACEs for children who have at least one response to the nine ACEs items, with any missing values coded as "not exposed." 23 The primary outcome for this analysis is child weight status based on parent-reported height and weight, categorized using standardized cut-points from the CDC. 24 Four categories were designated: underweight (BMI less than the 5th percentile), normal weight (5th to 84th percentile), overweight (85th to 94th percentile), or obesity (95th percentile or above). Neither raw height and weight nor the NSCH-calculated BMI are reported in the NSCH data sets, precluding the use of a continuous weight-related outcome measure.
For the current analysis, children who were underweight were combined with children who were normal weight to create an ordinal outcome variable with three possible values: underweight/normal weight, overweight, and obesity. We calculated this summary score only for participants with complete data on family resilience, excluding children who were missing any items on the family resilience scale. The weighted Cronbach's alpha for this scale was 0.89.
To account for potential confounding, we adjusted for variables that were hypothesized to be associated with both the exposure and the outcome but not part of the causal pathway. We conceptualized potential confounders using a social-ecological framework, recognizing the potential influence of individual, family, and community level factors. Each of the covariates was coded as suggested in the NSCH Codebook. 23 At the individual level, we controlled for child age, child sex (male; female), child race/ethnicity (Hispanic; Black, non-Hispanic; white, non-Hispanic; other, non-Hispanic), and current health insurance status (yes; no). Recognizing the potential contribution of chronic health conditions, we also included a variable indicating whether a child had special healthcare needs (yes; no). At the family level, we controlled for highest educational level of an adult in the household (less than high school; high school or GED; some college or technical school; college degree or higher), the family structure (two parents, currently married; two parents, not currently married; single mother; other family type), and total number of people living in the household (2; 3; 4; 5; ≥6). We did not control for family income level, as it is assessed in the first item in the ACEs scale. At the community level, we controlled for whether the neighborhood was supportive, using an NSCH-derived dichotomous measure (yes; no) derived from three items. We also controlled for the number of available neighborhood amenities like sidewalks, parks, community centers, and so on. (0; 1; 2; 3; 4).

| Statistical analysis plan
All analyses, including descriptive analyses, were conducted using survey weights provided in the NSCH data set, following NSCH methodology for both combining surveys from multiple years and analyzing subsets of the data. For the primary outcome of child weight status, we first report the prevalence of child weight status by number of ACEs experienced and then by the combination of ACEs and family resilience scores. To evaluate how family resilience affected the association between exposure to ACEs and child weight status, we used an ordinal logistic regression (OLR) model with three terms for the number of ACEs (1, 2-3, 4-9; reference = 0), a single term for score on the family resilience scale, and three interaction terms between family resilience and number of ACEs, adjusting for the covariates described above. The categorization of ACEs into four groups was based on the distribution of responses across combinations of ACEs and family resilience scores (shown in Table S1 and Figure S1), in order to ensure adequate data density for modeling.
Because of the complexity of the OLR model, we used postestimation techniques to present results in two formats. First, we calculated odds ratios across the range of ACEs and family resilience scores. The odds compared in these odds ratios, as with all OLR models, are the odds of being in a higher versus lower outcome category under any possible dichotomization of the outcome.
Our three-level outcome has two possible dichotomizations: obesity versus not obesity; or obesity/overweight versus normal/underweight. The odds ratios we calculated give the relative odds of being in a higher versus lower weight category using either of these dichotomizations, compared to children with no ACEs and a family resilience score of 0. To obtain insight into the absolute probabilities of obesity or overweight predicted by the model, we then calculated predicted probabilities of a child having overweight or obesity across the range of ACEs exposure and family resilience scores. Unlike odds ratios, predicted probabilities from an OLR model depend on a child's covariate values. We calculated predicted probabilities for children from each race/ethnicity category, using the median age and the most common values for the other adjusting covariates.
To evaluate whether the relationship between family resilience, ACEs, and child weight status differed by race/ethnicity, we constructed an expanded model that also included three-way interaction terms between a child's race/ethnicity, number of ACEs, and family resilience score, as well as the relevant lower-order interactions.
These lower-order interactions allow the relationships between ACEs and weight status and between family resilience and weight status to vary by race/ethnicity.

| Sensitivity analysis
Because missing answers to items on the ACE screener had different patterns of missingness depending on both the nature of the ACE and the child's race/ethnicity, we also conducted a sensitivity analysis in which we included only children with complete data on at least 6 ACEs and treated missing ACE items as "exposed" in calculating the total number of ACEs.

| Missing data
NSCH used hot-deck imputation for missing values in race, ethnicity, and sex; and regression-based imputation techniques for household HEERMAN ET AL.  Tables S2 and S3. Importantly, children with special healthcare needs experienced higher rates of each ACE category compared to children without special healthcare needs (Table S4 and Figure S3).
Children who experienced higher numbers of ACEs also had higher prevalence of childhood overweight and obesity. Prevalence of overweight ranged from 14% in children unexposed to ACEs to 18% in children exposed to 4-9 ACEs, while prevalence of obesity ranged from 11% to 21% in those groups, respectively ( Table 2). The prevalence of child overweight and obesity is shown for specific combinations of ACEs and family resilience in Table 3.
As noted above, we present two model-based summaries of the results from the adjusted OLR model. Figure 3 shows that among children who experienced ACEs, higher family resilience scores were associated with lower odds of being in a higher weight category. The converse was also observed: family resilience did not have the same attenuating effect on the odds of childhood overweight and obesity among children who were not exposed to ACEs. Figure 4 shows that among children with typical covariate profiles, the predicted probabilities of obesity were higher among Hispanic and Black, non-Hispanic children at all combinations of ACEs and family resilience. The full results from the main model are presented in Table S5. The results of the additional model that adds three-way interaction terms failed to detect a race/ethnicity-based difference in the manner in which family resilience decreased the odds of a higher weight category at any level of ACEs (data not shown). In our sensitivity analysis, changing how missing ACE items were coded did not result in any substantive changes to the results (results not shown).

| DISCUSSION
In the 2016-2018 NSCH, children aged 10-17 who were exposed to ACEs had higher rates of childhood overweight and obesity, the odds of which were attenuated when children also had higher family resilience. Despite the large sample size and nationally representative sampling strategy, the confidence intervals for specific combinations of ACEs and family resilience scores remained wide. This limits the ability of this analysis to firmly conclude that family resilience mitigates the effects of ACEs on childhood overweight/obesity. However, there is a consistent trend towards a "dose response," whereby (1) children with more ACEs tend to have higher odds of overweight and obesity, and (2) among children with ACEs, children with higher scores on the family resilience scale have the most attenuation of the odds ratios between ACEs and overweight/ obesity. It is also important to acknowledge that family resilience did not have the same attenuating effect on the odds of childhood overweight and obesity among children who were not exposed to ACEs. This observation supports the proposed mechanism by which family resilience may ameliorate the physical health consequences of ACEs, as we would not expect children who did not experience the social trauma of ACEs to have need of family resilience to buffer the toxic stress response. Taken together, these results indicate the need for further study into how family resilience might buffer a toxic stress response. We cannot yet firmly suggest intervention development in this area based on the lack of precision in the effect estimates (i.e., wide confidence intervals). However, the confidence intervals do include what would be a clinically meaningful reduction in the probability of overweight or obesity, suggesting that family resilience may be protective for children exposed to ACEs.
In this sample of 10-17-year-old children, more than half of children had been exposed to ≥1 ACE and nearly 8% had been exposed to ≥4 ACEs. The incredibly high rates of exposure to ACEs in this nationally representative sample indicate the pressing need for strategies to ameliorate the long-term health effects of these social traumas in childhood. This study suggests that increasing family resilience may be one such strategy. Resilience protects children from social trauma and can be conceptualized at both the individual and family levels. 15 association between child-level components of resilience with lower prevalence of childhood obesity, but did not find an association between family-level resilience and lower rates of childhood obesity. 16 The current analysis used a similar conceptual framework as this earlier study but yielded the expected associations, perhaps in part due to the use of a larger sample size by compiling 3 years of NSCH  0  22  4  18  11  13  10  16  12   1  22  10  15  24  15  15  16  19   2-3  18  5  20  22  17  21  17  21   4-9  21  35  15  18  21  17  17  23 Note: Percentages represent the survey-weighted prevalence estimates for child overweight and obesity for specific combinations of ACEs and Family Resilience Score (possible range 0-12).

F I G U R E 3
Family resilience, adverse childhood experiences, and odds ratios for overweight/obesity. Odds ratios from the adjusted ordinal logistic regression model are shown for specific combinations of adverse childhood experience (ACE) count and family resilience score. Odds ratios are with respect to the reference categories of 0 ACEs and a family resilience score of 0. For example, children with a family resilience score of 4 and 4-9 ACEs have an odds of being in a higher versus lower weight category that is 1.72 times the odds for children with zero ACEs and a family resilience score of 0. "Being in a higher versus lower weight category" applies to either of the two possible dichotomizations: obesity versus overweight/normal/underweight, or obesity/overweight versus normal/underweight. Odds ratios in the figure are shown with 95% confidence intervals. The odds ratios reported are derived from combinations of odds ratios from the overall adjusted ordinal logistic regression model; the full output of which is shown in Table S5 resilience, limiting the ability to draw inference about those specific combinations of values. This suggests an area of future study to develop additional measures of family resilience that could more precisely measure that latent construct.

| CONCLUSION
In the 2016-2018 NSCH, children ages 10-17 who were exposed to ACEs had higher rates of childhood overweight and obesity, the odds of which were attenuated when children also had higher family resilience.

CONFLICT OF INTEREST
The authors declare no conflict of interest.

AUTHOR CONTRIBUTIONS
William J. Heerman formulated the study question, interpreted results, and drafted the initial manuscript. Lauren R. Samuels conducted the statistical analysis and created data representations.

ORCID
William J. Heerman https://orcid.org/0000-0002-9706-6860 F I G U R E 4 Predicted probabilities of child obesity by race and ethnicity. The predicted probability of child obesity is shown for specific combinations of adverse childhood experience count and family resilience, based on the adjusted proportional odds regression model. The predicted probability is shown with the 95% confidence interval. To obtain these probabilities, covariate profiles were set to the specified race/ ethnicity, with other covariates set to the median or mode. The predicted probabilities are derived from the overall adjusted ordinal logistic regression model, the full output of which is shown in Table S5