Impact of Life Events on Short-Term Change in BMI in Early and Middle Childhood

Introduction

Increasing interest has been attributed to novel associations with childhood obesity and their role in a child's development. One such area is how psychosocial stressors may influence the onset of childhood obesity (1-4). Psychosocial stressors are associated with an increase in childhood obesity. Specific stressors that have been shown to be associated with an increased risk of childhood obesity include adverse childhood events (5), low socioeconomic status (6-8), poor family functioning (9, 10), problems in the household (11), or childhood abuse (12). Psychosocial stressors may affect a child's weight through different mechanisms: for example, through changes in the child's health behaviors, such as dietary intake or physical activity level, or through biological processes such as a stress response/inflammation in response to the stressor (13-15).

One specific category of psychological stressors being studied is adverse childhood events. The definition of adverse childhood events has differed among studies but mostly has been defined as any major life event that warrants a negative reaction, mentally or physically, impacting a child's well-being (16-19). Although parent responses are most often used for younger children, adverse events have also been measured using adult recall or by asking an older child/adolescent about stressful events in his or her life. Cumulative exposure to these events is associated with higher odds of obesity in childhood (5, 16). Research on adverse childhood events has placed an emphasis on studying all adverse events in childhood, as many co-occur (20). By studying a single stressor and its association with a child's weight status, the effect may be overestimated, as other stressors may be occurring simultaneously (20).

The associations between childhood adverse events and children’s weight status have been studied longitudinally (17, 18, 21-23). These longitudinal studies explore the cumulative and long-term associations of adverse childhood events and obesity but do not focus on the short-term changes that may potentially exist. Some studies have investigated the change in BMI over the course of 4 years (21, 23), whereas others have studied the accumulation of adverse events over all of childhood (17). Although changes are seen in children’s weight status after 4 years, the short-term change in weight after 1 to 2 years of experiencing high numbers of adverse childhood events is unclear. Meta-analyses show that an accumulation of high numbers of adverse events throughout childhood leads to an increased odds of overweight in adolescence and young adulthood (5). Meanwhile, other results show that when the recall length of adverse events is less than 2 years, the odds of overweight are increased but not significantly (5). Hence, whether high numbers of adverse events have a more immediate impact on children’s weight status is still unclear. In addition, the same meta-analyses found mixed results for children’s overweight status associated with high numbers of adverse events in cross-sectional studies. But because these meta-analyses observed a single weight measurement and an adverse childhood events survey within a cross-sectional study without using multiple measurements of weight and adverse childhood events throughout childhood, there is no ability to see whether high numbers of adverse events are associated with a meaningful change in weight in 1 to 2 years (5). In addition, positive life events also were shown to be associated with increased blood pressure among adolescents, a key element of an inflammatory response (24). A stress reaction may raise blood pressure and change eating and behavioral patterns, which may also contribute to an increased weight status. To our knowledge, no other studies have investigated whether this potential stress reaction to both positive and negative life events has an influence on change in children's BMI. We hypothesize that children who experience high numbers of adverse events or total events will see an increase in their BMI within 1 to 2 years.

This study uses data from a longitudinal cohort of more than 1,300 children. It was previously shown in this cohort that an accumulation of adverse events throughout childhood increased the odds of obesity in adolescence (17). Our study differs in that the 1- to 2-year change in BMI of children experiencing high numbers of adverse life events is investigated in comparison with accumulation across all of childhood. By using data with a measure of stress assessed numerous times throughout childhood, along with multiple BMI measurements, the current study has the goal of trying to identify whether a 1- to 2-year increase in BMI exists when an excess number of adverse events occurs in a child's life during early or middle childhood.

Methods

The Eunice Kennedy Shriver National Institute of Child Health and Human Development Study of Early Child Care and Youth Development (SECCYD) was used for this analysis. The SECCYD has been described in detail elsewhere (25); in brief, it was a cohort of 1,364 families who each had a healthy, full-term child born in 1991 who was assessed regularly from birth through age 15, with assessments constituting four waves of data, each spanning approximately 4 years with three to five study visits per wave of data. Mothers and their singleton newborn infants were recruited from 10 sites across the United States between January and November 1991. Families were selected on the basis of the mothers’ work intentions for the first year of the child's life, as well as on the basis of specific socioeconomic and ethnic considerations. The sampling strategy was designed to ensure that 60% of the mothers were planning on working full-time, 20% were planning on working part-time, and 20% were planning on staying home with their children; the strategy was also designed to represent the demographics of the specific site from which each mother was recruited. The goal of SECCYD was to assess the relationship between child care arrangements and children’s development, so the differing work intentions and other considerations were chosen in hopes of recruiting a diverse cohort with differing child care arrangements. Children enrolled in SECCYD were followed using regular assessments that included measurements, interviews, and surveys with the children, parents, and other caregivers, using procedures that had been standardized across sites. These assessments were performed in laboratories at participating institutions, at the families’ homes, and at the children's schools. The study was approved by institutional review boards at all locations, and mothers gave written consent.

At three time points—54 months (wave 2), grade 3 (wave 2), and grade 5 (wave 2)—mothers were given the SECCYD Life Experiences Survey. This survey was an extension of the Life Experiences Survey developed by Sarason et al. (26) but was adapted to include more potential life events. The survey was self-administered and asked mothers to read a list of 71 events, ranging from family conflict to economic hardship, and “circle whether you have experienced the event in the past year.” If mothers responded that they experienced the event, they were asked to “circle whether the event had either a positive or negative impact on your life at the time it occurred.” The Life Experiences Survey reports on both mothers’ lived adverse events and stress as well as on the positive events that occurred in their life. The survey response reported by the mothers then allows for a measure of the children’s potential lived adverse events from the previous year. All observations from completed Life Experiences Surveys were used to allow for the most accurate distribution of life events and to create categories for those children who experienced the highest number of adverse and total events. After examining the distributions of the Life Experiences Surveys across all three measurements, the upper quartile was used as the indicator of high numbers of adverse and total events. A child was considered to be in the highest quartile of negative events if 5 or more negatively rated events were reported by the mother in the previous year and was considered in the highest quartile of total events if 10 or more life events were reported, regardless of whether they were positively or negatively rated. Children were classified at each time point; therefore, it was possible, for example, to have a child included in the highest quartile at 54 months but not in the highest quartile at grade 3 or grade 5. Because we are interested in whether change in BMI can be observed over a short time period, changing from the highest quartile group to the nonhighest quartile group would allow for this short-term change to be observed or not by comparing the change in BMI when a child experienced a high number of adverse events with when the child did not experience a high number of events.

The height and weight of each child were measured regularly. Height was measured using a yardstick taped to a door. Children were asked to remove their shoes and stand straight with their backs against the wall. Height was measured at least twice, with a third and fourth measurement being done if the two initial height measurements did not fall within 0.5 inches of each other. The height of the child was recorded as the mean of the first two measurements if within 0.5 inches of each other; otherwise height was recorded as the mean of the third and fourth measurements. Weight was measured by scale with any extra clothing, such as shoes, sweat shirts, or jackets, removed. Children's weights were measured twice, and if the two measurements were within 4 oz of each other, the average was found. If the weight measurements differed by more than 4 oz, an additional two measurements were taken, and their average was used. Height and weight from the assessment including the Life Experiences Survey and the assessment prior to the survey were included in models and analyses. Therefore, height and weight measures were from 36 months (wave 1); 54 months (wave 2); and grades 1 (wave 2), 3 (wave 3), 4 (wave 3), and 5 (wave 3). All measurements were completed at laboratories associated with the study, except for grade 4 measurements, which were completed by trained medical professionals at medical offices, if not completed at the study laboratory. BMI was calculated at each time period as weight in kilograms divided by height in meters squared. The gender-specific BMI-for-age z score using the 2000 Centers for Disease Control and Prevention growth reference was then calculated at each time point and used in the analysis (27).

Additional covariates to control for potential confounding included children’s sex, race, and ethnicity. All three were reported by mothers during the first visit of the study when the children were 1 month old. Approximately 80% of the SECCYD sample reported the child's race as white and the child's ethnicity as non-Hispanic, so race and ethnicity were combined into a single binary covariate of white/non-Hispanic and other. Age was calculated as the years between the first day of the child's birth month and the date in which the Life Experiences Survey was completed. Also included in the model and analyses was the total income to needs ratio from the same assessment as the Life Experiences Survey questionnaire. The value was calculated at each assessment based on mothers’ responses to household income and number of people living within the household. The mothers’ BMI was also included, as children’s BMI is highly associated with maternal BMI, and excess stress may lead to weight change in adults (28). Mothers provided self-reported height and weight at the children’s age 15 interview for calculation of BMI.

The sample was restricted to include only children who had measures of height and weight for the survey visit prior to mothers taking the Life Experiences Survey and for the survey visit when mothers took the Life Experiences Survey, along with a completed Life Experiences Survey. Every child included in SECCYD had the potential of three total observations, one for each Life Experiences Survey assessment. In the case of a missing Life Experiences Survey, corresponding BMI measurement, or BMI measurement from the previous assessment, the observation was dropped. If all three were present, the observation was included in the modeling. In the cases in which there was no response for maternal BMI when children were 15 years old, maternal BMI was imputed (n = 221). The imputation was done using fully conditional specification (29). The imputation model included covariates in the final model in addition to other covariates that have been shown to be associated with maternal BMI, including maternal education level, self-reported race and ethnicity, age, and self-reported income. In total, 10 imputations were done, and all analyses were stratified by imputation and then aggregated using multiple imputation combination rules to account for uncertainty between imputations (29). The results from the aggregated imputations were compared with a complete case analysis as a sensitivity test, and the results did not deviate.

The impact of negative events on BMI z score and total events on BMI z score was investigated in two separate models. The outcome of each model was predicting the estimated change in BMI z score. The final models can be written in the form , where is the observed BMI z score for the i^th individual at the j^th time point, is the highest quartile indicator, is the years since the measured Life Experiences Survey, is the interaction between quartile indicator and the years since the measured Life Experiences Survey, and represents the potential confounders. Because previous BMI z scores are included as a covariate, the outcome of predicted BMI z score gives the expected change in BMI z score for some length of time while experiencing a high number of life events. The time difference between BMI z score measurements was used with the assumption that those children in the highest quartile of negative events would have been in the highest quartile, regardless of the length of time between BMI z score measurements. Each model was then centered on 1 year from the previous BMI z score measurement to the predicted BMI z score. This was done to answer the question of whether a change in BMI z score would be seen in 1 to 2 years while experiencing a high number of negative or total life events. Developmental period (54 months, grade 3, and grade 5) was also used in an interaction term with adverse events to see whether effect modification was present. The interaction was tested in the model using a likelihood ratio test, which was compared with the model without the interaction present. These repeated measure models were run with sandwich variance estimation with the clusters as unique children. The sandwich estimation provides robust and correct standard errors even if the working correlation structure is incorrectly defined (30). All statistical modeling was performed using SAS version 9.4 (SAS Institute, Cary, North Carolina).

Results

Of the 1,364 children in SECCYD, our model included observations from 936 children at 54 months, 809 children at grade 3, and 790 children at grade 5. Approximately 79% of children were of non-Hispanic or white race or ethnicity (Table 1). There was no difference between the proportion of males and females at each time point (Table 1). Of the 1,364 children in the original cohort, 592 had complete data for three observations, 277 children had complete data for two observations, and 205 children had complete data for one observation (Table 2). There were 290 children who had no observations included in the final models; this was due to missingness on height and weight measurements or missing Life Experiences Surveys. The median number of overall life events was six, and the median number of negative life events was two (Table 3). The proportion of children considered in the highest quartile of negative events was 28.1%, 28.1%, and 25.8%, respectively, at each assessment. Because all observations of the Life Experiences Survey were included in the distribution to find the quartile cutoffs, it was possible for a child to contribute to the total number of negative events without contributing an observation at that specific assessment. This occurred if the child was missing BMI measurements from either the previous assessment or at the Life Experiences Survey assessment.

Developmental period was initially used in an interaction with the highest quartile indicator at each measurement of the Life Experiences Survey, which would allow for a difference in effect to occur at differing points in childhood. None of the time points was statistically significant (likelihood ratio test: χ² = 0.01, P = 0.91), with most CIs centered around 0, suggesting there was no difference in the effect of being in the highest quartile at 54 months, grade 3, or grade 5 for either the negative events indicator or the total events indicator (Table 4). Therefore, modeling was done without an interaction with developmental period.

When the interaction with developmental period was removed, those children included in the highest quartile of negative events had a predicted BMI z score that was 0.055 units higher at 1 year and 0.053 units higher at 2 years of experiencing high numbers of adverse events, suggesting a small gain in BMI in the first year that would remain relatively steady in the second year (Table 4). In comparison, those children not in the highest quartile of negative events had a predicted BMI z score of 0.061 units higher at 1 year and 0.016 units lower at 2 years of experiencing a lower number of adverse events, suggesting that these children experienced a similar gain in BMI after 1 year but began to return to previous weight after 2 years (Table 4). The largest difference between the predicted BMI z score between the two groups occurred at 2 years of experiencing high numbers of adverse events with a predicted difference of 0.069, although this was not statistically significant (95% CI: −0.006, 0.144). As a sensitivity analysis, we used a more extreme cutoff for the highest negative events group as the top 10%, or those reporting eight or more negative events in the previous year. The resulting model gave nearly identical results.

Similarly, the predicted change in BMI z score was compared between children in the highest quartile of total events, ignoring whether the event was rated positively or negatively, and children not in the highest quartile. At 1 year, children had a predicted BMI z score that was 0.017 units higher, and at 2 years, children had a score that was 0.023 units higher, when experiencing a high number of total events (Table 5). The children not considered in the highest quartile of total events had a predicted BMI z score that was 0.066 units higher and 0.007 units lower at 1 and 2 years, respectively (Table 5). At 2 years, the predicted difference in change in BMI z score between these two groups was 0.029 (95% CI: −0.054, 0.114).

Discussion

We examined whether high levels of negative or total life events, assessed at three time points in childhood, were associated with changes in BMI z score observed over a 1- to 2-year period. Longitudinal studies show that an accumulation of negative events over time is associated with a higher BMI in late childhood and young adulthood (5). Our analysis did not find any significant differences in the change in BMI z score when experiencing a high number of negative events in 1 or 2 years. Estimates were similar with no or small changes in BMI z score within 1 to 2 years of experiencing a large number of total events, regardless of the positive or negative nature of the event.

When children experience high numbers of negative events, body weight could be influenced in two ways. First, stress may impact metabolism or sleep, leading to weight gain (31). Second, children may change their behaviors by becoming more sedentary or eating more (32). Although behaviors can change during times of adverse events, our results suggest that it takes time for children to gain the weight associated with the change in behavior. Similar to other chronic diseases, the exposure to high numbers of adverse events may not change BMI significantly during the 1 to 2 years of experiencing those negative events; instead, it may take a longer period for the association to be seen between the high numbers of negative events and a change in children’s BMI.

Although cross-sectional studies have shown mixed results when it comes to the association of adverse childhood events and overweight, longitudinal studies agree that accumulation over childhood increases the odds of obesity (5). Our analysis differs from other longitudinal studies in that we investigated whether a change in BMI could be seen in a short time period while children are experiencing high numbers of adverse events. Our study builds upon the work of Evans et al., who used height and weight measures spaced 4 years apart and a measure of risk that included similar adverse childhood events. They found that after 4 years, there was a gain in BMI percentile in children experiencing high numbers of adverse events, defined in their study as having a higher risk score (21). Our analysis found that there was no statistically significant predicted difference at 1 or 2 years of experiencing high numbers of adverse events, although our model point estimate suggests an increase in BMI z score within those 1 to 2 years. Another study, by Lumeng et al., examining the same data from SECCYD, found that accumulation of negative events over childhood significantly increased odds of overweight by 47% at age 15 (17). Our study builds on this in that we examined the change that could be seen in 1 to 2 years instead of using the accumulation of events throughout childhood.

Our study used longitudinal data and multiple measurements per child to identify whether experiencing a high number of adverse events would predict a change in BMI z score in 1 to 2 years for young and middle-aged children. Our study, however, did rely on data that were reported by mothers and therefore may not translate specifically to the children, as the Life Experiences Survey asked about the impact the life events had specifically on the mother's life when the event happened. Our assumption is that mothers acknowledging an adverse event has occurred may also have an effect on their children in two ways. First, children will also have knowledge of the event and experience similar adversity (19). Second, mothers, who have a negative perception of the event, may change the way they parent, hence shifting the adversity they experiencing to the children (19). Additionally, this measure of adverse events is subjective. One mother may not view a negative event as severely as another, but that follows for children as well; one child may have a more negative response to a given event. Additionally, the best way to measure stress in humans is debated, and the ability to independently measure stress is questionable (33, 34). Our study used self-report to measure stress from the previous year, but this may not correlate to biological markers of stress, such as cortisol (33, 34). However, studies of higher cortisol levels have shown mixed results when looking at children from families with low socioeconomic status (35-37). Therefore, a survey measuring the adverse nature of an event may be one of the best ways to measure the total stress an individual is experiencing, because of the conflicting research on an independent measure of stress.

Although our study did not find a significant difference in predicted change in BMI in 1 to 2 years between children experiencing high numbers of adverse events compared with those who did not, the model produced does suggest, through extrapolation, that those children with high numbers of negative events will have a higher change in their BMI z scores. Further directions that research in this area could explore include using a more constant measure of childhood stress versus a yearly recall survey and studying how accumulation of many events in a shorter time period than 1 year could possibly influence a change in BMI in children. It may also be interesting to explore whether the perceived negativity of the events from the Life Experiences Survey has an association with weight gain in that more negatively rated events may be more strongly associated with change in children’s BMI.