Modeling Body Mass Index Z-score Lability During Childhood as a Function of Child Temperament and Genetic Risk for Obesity

Abstract

Objective:

This study aimed to examine if child genetic risk for obesity and temperament (i.e., negative affectivity, effortful control) accounted for stability versus lability in children’s weight status (BMI z-score) over time.

Methods:

We followed 561 adopted children (42% female; 56% Caucasian, 13% African American, 11% Latino, and 20% other) and their birth and adoptive parents from birth to age 9 years. The multilevel location-scale model was used to examine if child genetic risk for obesity and temperament were related to differences in level and lability in child BMI z-score over time.

Results:

For the full sample, higher levels of child negative affectivity were associated with greater BMI z-score lability, while higher levels of effortful control and children’s mean-level BMI z-score were related to less lability across childhood. Additional analyses examined associations within groups of children with healthy versus overweight/obesity weight statuses. Within the healthy weight status group only, better effortful control was associated with more stable BMI z-scores, while genetic risk for higher BMIs was associated with more labile BMI z-scores.

Conclusions:

These findings provide insights into factors that can be harnessed to redirect unhealthy trajectories as well as factors that may challenge redirection or maintain a healthy trajectory.

Introduction

Childhood obesity is a primary threat to healthy child development in the USA, where 1 in 5 children experience obesity (1). Existing research has indicated that genetic and behavioral factors are associated with children’s healthy versus unhealthy weight status (2, 3). Other research has focused on predictors of patterns of body mass index (BMI) z-score over time, identifying children who start to enter on BMI z-score trajectories that ultimately lead to overweight or obesity at early ages (4). However, there is little information about lability in children’s weight status: what predicts fluctuations between healthy and unhealthy BMI z-scores versus stability in BMI z-scores over time? Identifying factors that are associated with more weight status stability may inform prevention efforts that seek to maintain healthy BMI z-scores over time, as well as identify factors that may work against efforts to reduce unhealthy BMI z-scores. Conversely, identifying factors associated with low weight status stability may provide insight into how unhealthy BMI z-score trajectories could be rerouted. The current study addresses genetic and behavioral correlates of weight status lability versus stability.

Typically, body mass index (BMI; weight/height²) is used for adolescents and adults to diagnose overweight and obesity. However, BMI z-scores assess the extent to which children deviate from sex- and age-specific BMI norms for healthy weight. Although for some children BMI z-scores remain relatively stable throughout childhood, there is also great intra-individual variability (i.e., lability) during childhood (5). In the current study, lability in children’s BMI z-scores reflects the degree to which children’s weight status fluctuates above and below age-related trends over time. The use of a z-score enabled us to estimate how much children’s BMI also deviated from age- and sex-normed changes in BMI over time. Therefore, lability captures systematic variations that can be explained by individual characteristics, and not simply time-specific measurement errors (6). With multiple assessments over varying lengths of time, lability estimation can be separated from measurement error (6). For adults and adolescents, lability in weight status can be caused by intentional bouts of dieting and food restriction (e.g., “weight cycling”) (7). Within this research literature, more labile BMIs are associated with increased risks for binge-eating and depression (8), metabolic syndrome/issues (e.g., high blood pressure, hypertriglyceridemia, and low-high density lipoprotein-cholesterol) (9, 10), type 2 diabetes (11), and all-cause mortalities(12, 13). Consequently, BMI z-score lability is associated with negative health outcomes.

In the current study, we focus on young children whose weight fluctuations are not easily explained by intentional weight loss. Within this context, lability in children’s weight status could imply fragile internal regulatory systems that maintain balanced growth and/or imply enhanced sensitivity to environmental changes that can alter their BMI z-scores. Although BMI z-score lability has not been studied extensively, lability versus stability in physical and mental functioning are often indicators of underlying dysregulation of key biological or behavioral systems that increase risks for poor health (14). For example, greater lability in infants’ heart rate (15), in adolescents’ self-esteem (16), in adolescents’ and adults’ mood (17, 18), and in older adults’ cognitive performance inconsistency (14), are predictive of poor physical or mental health outcomes.

A tendency towards greater weight lability could also forecast struggles with maintaining a healthy weight even if children currently have a healthy weight. This possibility is supported by existing studies (19, 20). For example, one study examined child BMI lability between 2 and 6 years and found that higher BMI lability was associated with overweight status during adulthood in girls (20). Similarly, another study found that greater lability in BMI over a 6-month period predicted greater amount of weight gained from 6 to 24 months in a sample of nonobese young women even after controlling for baseline BMI (19). It is important to note, however, while lability may present a challenge for maintaining healthy weights over time, low lability may also lock children who already have an unhealthy weight status into chronically overweight or obesity trajectories over time. In other words, the clinical implications of lability may be different for children who show an overweight or obesity status than for those with healthy weights.

Few studies have sought to understand sources of BMI z-score lability vs. stability amongst children who show a healthy or unhealthy weight status over time. The current study aims to examine lability versus stability in weight status across childhood, and the degree to which heritable weight predispositions and child temperament characteristics account for lability versus stability. A growing body of research indicates genetic influences on child BMI z-scores (21). However, it is not clear if genetic risk for obesity is related to BMI z-score lability over time. Child temperament characteristics may also be correlated with child BMI z-score lability. For example, in previous research children with higher negative affectivity were more likely to eat in response to negative emotions (versus hunger cues), and exhibited more unhealthy eating behaviors, including emotional overeating (“comfort eating”), emotional undereating, food avoidant eating behaviors, and periodic mixtures of emotional overeating and undereating (2, 22, 23). Therefore, heightened negative affectivity could disrupt the internal regulation of eating, and lead to more labile BMI z-scores over time (23). Children’s effortful control levels could also be related to BMI z-score lability. Effortful control refers to the capacity to regulate one’s reactivity to stimuli by allocating attention flexibly and inhibiting a dominant response to activating a subdominant response (2, 24). In contrast to negative affectivity, previous research found that children with higher effortful control were more likely to attend to satiety cues, executed more restraint over food choices, and were less likely to exhibit unhealthy eating behaviors, all of which may lead to more stable and healthy BMI z-score growth trajectories (2, 25, 26).

Last, the current study explores if associations between children’s genetic risk for obesity, temperament characteristics and BMI z-score lability depend on children’s average weight status. For example, children with an overweight/obesity weight status and with high genetic risk for obesity may have more difficulty leaving the high BMI z-score trajectory, and may exhibit stable high BMI z-scores over time compared to other children. However, it is also possible that for children with healthy weight status, higher genetic risk for obesity may be associated with more labile BMI z-scores as they may have more difficulty maintaining a healthy BMI z-score. Likewise, associations between child temperament characteristics and BMI z-score lability may also depend on children’s average weight status. For example, children with overweight or obesity may be more likely to engage in comfort eating in response to negative affect compared to children with healthy weight status (27). Therefore, negative affectivity may lead to more BMI z-score lability amongst children with overweight or obesity, but not amongst children with a healthy weight status.

The current study used a prospective adoption design to: (a) examine factors (i.e., genetic risk for obesity, child negative affectivity and effortful control) related to lability in BMI z-scores from ages 2 to 9 years for the full sample (aim 1), and (b) explore the effects of genetic risk for obesity and child temperament characteristics on BMI z-score lability separately for children with overweight or obesity weight status and children with healthy weight status (aim 2). In the adoption design, birth parents provided genes but not the rearing environment to the child, thus links between the adopted child and their birth parent(s) are best explained as genetic influences. The adoption design enables us to examine the contribution of genetic risk for obesity to child BMI z-score lability versus stability. To our knowledge, this is the first study that has examined factors (i.e., genetic risk for obesity, child negative affectivity and effortful control) related to lability in BMI z-scores using an adoption design.

Methods

Sample.

Participants were 561 adopted children, their birth parents, and adoptive parents from the Early Growth and Development Study (EGDS) (28). EGDS is a prospective parent-offspring adoption study, which is domestic adoption in the United States. Forty-two percent of the children were female and 56% were Caucasian, 19% were multiracial, 13% were African American, 11% were Latino, and 1% were other. On average, infants were placed at 6 days after birth (range = 0–91 days). Ethical approval was obtained by the institutional review boards of the University of Oregon, The Pennsylvania State University, and George Washington University, and all individuals provided written informed consent before participating. A more detailed description of demographic information, recruitment, and assessment is available in Table 1 and Supplemental Information.

Table 1.

Participants Demographics

Variable	Birth mother	Birth father	Adoptive parent 1	Adoptive parent 2
Age at childbirth, year, mean (SD)	24.35 (6.03)	26.08 (7.77)	37.43 (5.59)	38.29 (5.83)
Race and/or ethnicity, %
White	70.7%	70.1%	91.6%	89.7%
African American	13.5%	12.3%	3.7%	4.9%
Hispanic or Latino	6.1%	9.0%	2.0%	2.0%
Multiethnic	4.7%	5.2%	0.9%	1.1%
Other	5.0%	3.4%	1.8%	2.3%
Median education level	High School	High School	4-y College	4-y College
Median annual household income	$25,000 – $40,000	$25,000 – $40,000	$125,000 –$150,000	$125,000 –$150,000

Measures.

Child Body Mass Index (BMI) Z-score.

Child weight and height were obtained via adoptive parents’ reports and extracted from medical records from child ages 2 to 9 years. Adoptive parents’ reports and medical records within each age range (i.e., age 2–3, 3–4, 4–5, etc.) were highly correlated (weight: r = .90; height: r = .75). Child BMI z-score adjusted for sex and age was calculated based on the US CDC 2000 norms (29). Children with fewer than 3 assessments of height and weight were removed from further analysis (n = 109), and the final analysis included 452 children. On average, there were 10 weight and height assessments per child, and assessments spanned from child ages 2 to 9 years. For aim 2, we used children’s mean-level BMI z-score averaged across assessments to classify children into two weight status groups based on the CDC BMI-for-age 85th percentile cut-off for overweight, i.e., z-score of 1.04 or above: Overweight and obesity group (n = 91), and healthy weight group (n = 361). For the height and weight data attrition, we found no appreciable differences in the study variables and demographic variables across groups where children had 3 or more assessments of height and weight versus children had fewer than 3 assessments (adjusted p values = .0025) (A detailed description of the targeted missing data analysis is available in Supplemental Information).

Child Age.

Child age was calculated for each child at each assessment as the difference between the assessment date and the child’s birthdate. We then centered child age at age 2.

Genetic Risk for Obesity.

Birth mother’s BMI was used to index children’s genetic risk for obesity. Specifically, birth mother’s BMI was assessed based on their self-report of height and weight 6 times from post-partum 5 months to 9 years. Scores were averaged across multiple assessments to create a composite score for birth mother’s BMI.

Child Negative Affectivity and Effortful Control.

Negative affectivity and effortful control were measured using adoptive parents’ reports on the Negative Affectivity subscale (α = .72) and Effortful Control subscale (α = .72) from the Very Short Form of the Children’s Behavior Questionnaire (CBQ) (30) at 4.5 years. The scores were averaged across primary and secondary caregivers to create an index score for child negative affectivity and effortful control, respectively (negative affectivity: r = .39, p < .001; effortful control: r = .52, p < .001).

Control Variables (Covariates).

At child age of 5 months, we assessed child sex, adoptive parents’ household income, perinatal risk index, and adoption openness. Child sex was coded as 0 (male) or 1 (female). Perinatal risk index was created based on both birth mother reports on an adaptation of the McNeil-Sjӧstrӧm Scale for Obstetric Complications (31) and coded medical records (32). Adoption openness was assessed based on a composite index of birth mother’s and adoptive parents’ perceived adoption openness ranging from very closed to very open.

Of note, all the time-invariant predictors, including genetic risks for obesity, child effortful control, child negative affectivity and control variables were sample-mean centered before data analysis.

Data Analysis.

The analyses were conducted in a multilevel modeling framework that accommodated the nested nature of the data where repeated measures of BMI z-score were nested within persons. Specifically, we used the multilevel location-scale model (33) that decomposed change processes into patterns of mean (level) and intra-individual variations around the mean (lability). The model simultaneously examined normative developmental trends in child BMI z-score, lability in child BMI z-score, and how predictors and covariates were related to differences in level and lability in child BMI z-score for the full sample (aim 1), and separately for children with overweight or obesity weight status and children with healthy weight status (aim 2). Models were fitted using SAS software, Version 9.4, using PROC NLMIXED estimation. Incomplete data was treated using missing at random assumptions.

At the level of child BMI z-score, we examined predictors, including genetic risk for obesity, child negative affectivity and effortful control, of the level of child BMI z-score. These analyses controlled for child age, age², and age³ to remove any systematic age-related trends in BMI z-score over time. Since child sex, household socioeconomic status, and perinatal risk have been found to be associated with childhood obesity (34, 35), these variables were also controlled for in analyses. Last, we controlled for adoption openness to account for possible confounds due to adoptive parents’ perceived knowledge of or contact with the birth parents. At the level of lability of child BMI z-score (intra-individual variations), we examined predictors of the variance in child BMI z-score. These analyses controlled for children’s average BMI z-score across assessments, the sparsity of assessment (e.g., whether children have dense versus sparse assessments) indexed by the standard errors of the assessment ages, child sex, adoptive parents’ household income, perinatal risk index, and adoption openness. The equations for the model are illustrated in Equation 1.

Results

Full Group Analyses

Child BMI z-score:

On average, child BMI z-score followed a cubic trajectory across childhood, from ages 2 to 9 years (β_linear = .55, p < .001; β_quadratic = −.16, p < .001; β_cubic = .01, p < .001), such that child BMI z-score increased from ages 2 to 4 years, followed by continuous decreases until age 8 years, and then a continuous increase after age 8 years. After controlling for the age-related trend in child BMI z-score, birth mother’s BMI was positively associated with child BMI z-score ($\gamma$_{birthmother’s BMI} = .02, p = .005), indicating that children with a genetic tendency towards a higher BMI also showed higher levels of BMI z-score. Child negative affectivity was negatively associated with child BMI z-score ($\gamma$_{negative affectivity} = −.20, p = .004), such that children with higher negative affectivity showed lower BMI z-scores. Child effortful control was not associated with child BMI z-score ($\gamma$_{effortful control} = .07, p = .386). For the control variables, only child sex was significantly associated with child BMI z-score, such that girls had lower BMI z-scores compared to boys ($\gamma$_sex = −.21, p = .039).

BMI z-score lability (versus stability):

Tests of the main hypotheses indicate that, after controlling for the sparseness versus density of assessments, child effortful control was negatively associated with extent of lability in child BMI z-score (α_{effortful control} = −.10, p = .024). This suggests that children with high effortful control tended to have less weight status fluctuations from ages 2 to 9 years, while children with low effortful control showed more lability. Child negative affectivity was positively associated with lability in child BMI z-score (α_{negative affectivity} = .25, p < .001). Thus, children with higher negative affectivity showed more deviations around their own average BMI z-scores over time than other children. Regarding genetic risk for obesity, birth mother’s BMI was not associated with extent of lability in child BMI z-score (α_{birthmother’s BMI} = 0, p = .395). Moreover, children with higher mean-level BMI z-scores showed less BMI z-score fluctuations over time (α_{average BMI} = −.41, p < .001). Last, several demographic control variables were associated with child BMI z-score lability. Child sex, adoptive parents’ household income, and perinatal risk were all negatively associated with lability in child BMI z-score (α_{child sex} = −.24, p < .001; α_{household income} = −.06, p < .001; α_{perinatal risks} = −.02, p = .041). The results indicate that children who were female, lived in a household with higher incomes, or had higher perinatal risk scores tended to have less fluctuation in BMI z-score from ages 2 to 9 years, compared to children who were male, lived in a household with lower incomes, or had lower perinatal risk scores.

Subgroup Analyses: Children with Healthy Weight Versus Overweight/Obesity Weight Status

We re-examined the original model separately for children in the overweight and obesity group and children in the healthy weight group. The results indicate that higher child negative affectivity was associated with more lability for children in both overweight and obesity ((α_{negative affect} = .40, p < .001) and healthy weight (α_{negative affect} = .14, p = .002)) groups. In contrast, child mean-level BMI z-score over time was negatively associated with lability in child BMI z-score for both groups (overweight and obesity: α_{mean BMI} = −.40, p = .005; healthy weight: α_{mean BMI} = −.62, p < .001). Effortful control was negatively associated with lability within the healthy weight group (α_{effortful control} = −.14, p = .008), and unrelated to lability within the overweight and obesity group (α_{effortful control} = .09, p = .378). Last, birth mother’s BMI was positively associated with lability within the healthy weight group (α_{birthmother’s BMI} = .01, p = .008), but not within the overweight and obesity group (α_{birthmother’s BMI} = −.01, p = .354). Several control variables, including child sex, adoptive parents’ household income, perinatal risk, and adoption openness, were negatively associated with BMI z-score lability within the overweight and obesity group (α_{child sex} = −.61, p < .001; α_{household income} = −.13, p < .001; α_{perinatal risk} = −.06, p = .003; α_{adoption openness} = −.22, p = .006), but not within the healthy weight group. More detailed information about the model results is provided in Tables 2 and 3.

Table 2. Children in the Overweight and Obesity Weight Status Group:

Results of the Multilevel Location-Scale Model Examining Predictors and Control Variables of the Lability in Child BMI Z-score from Ages 2 to 9 Years.

	Fixed Effects: ($\mathit{\gamma }$)	Fixed Effects: (95% CI)	Within-Person Variance var(a)	Within-Person Variance (95% CI)
Intercept	−1.17***	[.92, 1.42]	−.72*	[−1.34, −.10]
Predictors
Birth mother’s BMI	0	[−.02, .01]	−.01	[−.03, .01]
Child effortful control	.06	[−.11, .24]	.09	[−.11, .28]
Child negative affectivity	.08	[−.05, .22]	.40***	[.25, .55]
Control Variables
Child sex	−.35*	[−.61, −.08]	−.61***	[−.93, −.29]
Adoptive parents’ household income	−.01	[−.07, .04]	−.13***	[−.19, −.08]
Perinatal risk index	0	[−.04, .03]	−.06**	[−.10, −.02]
Adoption openness	−.02	[−.14, .11]	−.22**	[−.38, −.06]
Child age	.72***	[.58, .86]	-
Child age2	−.29***	[−.24, −.15]	-
Child age3	.01***	[.01, .02]	-
Child average BMI z-score	-		−.40**	[−.67, −.12]
	Random Effects var($u$)
Variance Intercept $\left({\sigma }_{u0}^2\right)$	.73***
Variance child age (linear slope: $\left({\sigma }_{u\text{1}}^2\right)$)	.05***
Covariance between intercept and linear slope	−.17***

Note. CI = Confidence Interval.

^*p < .05.

^**p < .01.

^***p < .001.

Table 3. Children in the Healthy Weight Status Group:

Results of the Multilevel Location-Scale Model Examining Predictors and Control Variables of the Lability in Child BMI Z-score from Ages 2 to 9 Years.

	Fixed Effects: $(\gamma )$	Fixed Effects: (95% CI)	Within-Person Variance var(a)	Within-Person Variance (95% CI)
Intercept	−.23**	[−.36, −.09]	−1.18***	[−1.49, −.88]
Predictors
Birth mother’s BMI	.02*	[.003, .03]	.01**	[.003, .02]
Child effortful control	.02	[−.11, .15]	−.14**	[−.24, −.04]
Child negative affectivity	−.13*	[−.25, −.01]	.14**	[.05, .24]
Control Variables
Child sex	−.03	[−.20, .13]	−.08	[−.19, .04]
Adoptive parents’ household income	−.01	[−.04, .03]	−.02	[−.05, .01]
Perinatal risk index	−.01	[−.04, .01]	−.01	[−.02, .01]
Adoption openness	0	[−.09, .08]	0	[−.06, .06]
Child age	.45***	[.37, .54]	-
Child age2	−.14***	[−.17, −.11]	-
Child age3	.01***	[.01, .01]	-
Child average BMI z-score	-		−.62***	[−.70, −.54]
	Random Effects var($u$)
Variance Intercept $\left({\sigma }_{u0}^2\right)$	.71***
Variance child age (linear slope: $\left({\sigma }_{u\text{1}}^2\right)$)	.03***
Covariance between intercept and linear slope	−.10***

Note. CI = Confidence Interval.

^*p < .05.

^**p < .01.

^***p < .001.

Sensitivity Analysis

For the sensitivity analyses included, we re-examined the model using only BMI z-score assessments from ages 4.5 to 9 years to elucidate whether child temperament characteristics at age 4.5 years would predict subsequent lability in children’s BMI z-scores from ages 4.5 to 9 years. The sample included 384 children with 3 or more BMI z-scores from ages 4.5 to 9 years. The results were generally consistent with the findings reported above. More detailed information about the results is provided in Supplementary Information.

Discussion

This is the first study to examine genetic risk for obesity and child temperament characteristics as predictors of lability in BMI z-score from toddlerhood to preadolescence. In the current study, lability was defined as children’s deviations (e.g., ups and downs) from their own developmental patterns, while the use of z-score enabled us to estimate how much children’s BMI also deviated from age- and sex-normed changes in BMI over time.

Because this study used a parent-offspring adoption design, we were able to examine the degree to which genetic and perinatal risks predicted BMI z-scores and lability, independent of postnatal factors. Our findings identified genetic influences on children’s weight status, as indexed by associations between the adoptees’ birth mothers’ BMI and the adoptees’ BMI z-score. Moreover, for children with healthy weight status, having a birth mother with a higher BMI was associated with more labile BMI z-scores. This finding suggests that genetic tendencies towards higher BMIs may make it difficult for children to maintain a stable healthy weight over time than other children. Therefore, children at higher genetic risk for overweight and obesity who showed healthy BMI z-scores may still struggle and need to be followed carefully over time. Perinatal risk was related to lower lability for children with overweight or obesity weight status only, increasing their risks for staying on the overweight or obesity weight trajectories. However, for children with healthy weight status, perinatal risk was not associated with lability in BMI z-score. This finding requires further exploration as it may indicate a generalized risk for health problems or physiological dysregulation.

For all children, having a higher mean-level BMI z-score was related to more stable BMI z-scores over time. This pattern of findings suggests that as children showed higher BMI z-scores and deviated from healthy BMIs, their weight status also started to stabilize. These findings reinforce the importance of attending to deviations from healthy weight status during early childhood to prevent the stabilization of unhealthy and high weight trajectories.

Our findings also indicate that child temperament characteristics played important roles in children’s BMI z-score lability. Higher negative affectivity was associated with higher BMI z-score lability. Across childhood, higher levels of negative affectivity may pose challenges to maintaining weight status. Previous work found that children with high negative affectivity tended to eat in response to their emotional states (e.g., negative emotions) instead of satiety cues, compared to children with low negative affectivity (2, 22, 36). More specifically, children with high negative affectivity were more likely to engage in comfort eating (emotional overeating), food avoidant eating behaviors and periodic mixtures of emotional overeating and undereating in response to negative affect. This pattern of behavior could shift the regulation of eating from hunger and satiety cues to emotional cues, and lead to more labile BMI z-scores over time.

In contrast to negative affectivity, higher levels of effortful control were related to greater stability within the whole sample. However, additional analyses suggest that this association was primarily driven by children who, on average, had healthy BMI z-scores. Previous studies found that children with higher effortful control tended to eat in response to satiety cues and executed more restraint over food choices (e.g., choosing more healthy food to eat) (2, 25, 26). Therefore, better effortful control could enable children to regulate food intake to meet caloric needs, and decrease overeating. These tendencies could underlie more stable BMI z-scores over time (2, 25, 26). However, child eating behaviors were not examined in the current study and future work is needed to examine whether child unhealthy eating behaviors mediate the associations between effortful control and lability in BMI z-score. It is not clear why effortful control and BMI z-score lability were only associated within the healthy weight status group. Both groups demonstrated similar levels of effortful control, therefore these findings were not explained by mean level differences in effortful control. Given the relatively small number of children in the overweight or obesity group (n = 91), it is possible that we did not have statistical power to detect small effects. Nevertheless, our findings suggest that other factors, such as negative affectivity and mean-level BMI z-scores were stronger predictors than effortful control within this subgroup. These findings suggest that different treatments may be needed for children who have already enjoined a trajectory characterized by overweight or obesity weight status versus prevention efforts that help children maintain healthy weights. For example, amongst children who already experience overweight or obesity, many intervention programs already target caloric intake and physical activity(37–39). The current study suggests that attention to children’s negative affectivity is warranted and may be needed to maintain improvements in BMI over time. For children who demonstrate healthy weights, universal prevention programs that incorporate practices to lower negative affectivity and promote stronger effortful control skills could help stabilize healthy weight over time, especially for children at higher genetic risks for obesity.

The findings from this study should be interpreted considering the following limitations. First, the use of BMI z-score may not be the most optimal measure for assessing children with very high BMIs (above the 97th percentile) or for examining longitudinal changes in adiposity, especially for children with very high BMIs.(40, 41) Specifically, due to the skewed distribution of child BMIs in the CDC growth charts, very high BMIs are compressed into a narrow range of z-scores.(41) Consequently, children with very high BMIs are less likely to exhibit significant change over time compared to other children. Of note, we conducted a thorough examination of our study sample and found that only 6 (1.3%) children had an average BMI percentile above the 97% percentile, indicating persistent severe obesity throughout childhood. The small proportion of children potentially affected by the compression bias of BMI z-score strengthens our confidence in the robustness of our current findings. Second, this study relied on parent reports of child temperament, although we combined reports from primary and secondary caregivers to minimize any rater bias. However, including data from additional sources, such as observational ratings and teacher reports, could provide a more comprehensive assessment of child temperament. Third, child weight and height were obtained via both adoptive parents’ reports and extracted from medical records to maximize the number of BMI assessments. However, it is important to note that adoptive parents’ reports of child weight and height may be subject to rater biases, such as measurement errors or perceptual and response biases. However, the study team used growth charts and placed them in the families’ homes in the correctly measured positions, to maximize the reliability and validity of parent-reported height. It is worth mentioning that there was a strong correlation between adoptive parents’ reports and the information documented in medical records within each age range. Fourth, for genetic risk for obesity, we only included BMI information from birth mothers assessed 6 times from post-partum 5 months to 9 years, as a large proportion of missing data (> 60%) for birth father. However, since our study only captured half of the child’s genetic risk, we expect the results to be attenuated. Fifth, children’s medication use could potentially influence their weight status and negative affectivity, thereby potentially affecting their weight status lability (e.g., increasing BMI z-score lability). However, in this study, we did not have children’s medication use information and did not control for it, which is another limitation. Future research is needed to replicate the current study while controlling for children’s medication use.

Conclusion

This study expands our current understanding of factors that affect children’s weight development, not only at the level of BMI z-scores but also the stability versus lability in BMI z-scores across childhood. Findings from the current study may inform prevention programs that focus on promoting child effortful control and reducing child negative affectivity to help children stabilize healthy weight trajectories over time. Moreover, findings provide information on factors (e.g., negative affectivity) that could limit the effectiveness of intervention efforts for children with chronic overweight/obesity weight status and factors (e.g., effortful control) that maintain persistent treatment effects in reducing child overweight and obesity weight status over the first nine years of life.

Study Importance.

What is already known about this subject?

• Previous research has found that child genetic risk for obesity and child temperament (i.e., negative affectivity, effortful control) are associated with children’s weight status (BMI z-score).

• However, it is not clear if these factors account for stability versus lability in children’s weight status over time.

What are the new findings in your manuscript?

• First study examines child genetic risk for obesity and child temperament as predictors of lability in BMI z-score from toddlerhood to preadolescence.

• For children with overweight or obesity weight status, heightened negative affectivity was related to more lability, while higher mean-level BMI z-score was related to less lability. Similar associations were found for children with healthy weight status. For this group of children, however, higher genetic risk for obesity and lower levels of effortful control were associated with less stable weight status.

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

• This study expands our current understanding of factors that affect children’s weight development, not only at the level of BMI z-scores but also the stability versus lability in BMI z-scores across childhood.

• Findings from the current study may inform prevention programs that focus on promoting child effortful control and reducing child negative affectivity to help children stabilize healthy weight trajectories over time.

APPENDIX

Equation 1. The model simultaneously examined normative developmental trends in child BMI z-score, lability in child BMI z-score, and how predictors and covariates were related to differences in level and lability in child BMI z-score for the full sample (aim 1), and separately for children with overweight or obesity weight status and children with healthy weight status (aim 2):

$$BMIzscor{e}_{it}={\beta }_{0i}+{\beta }_{1i}Ag{e}_{it}+{\beta }_{2i}Ag{e}_{it}^2+{\beta }_{3i}Ag{e}_{it}^3+{\epsilon }_{it}$$ (1)

where the observed child BMI z-score at assessment $t$ for person $i$, $BMIzscor{e}_{it}$, was modeled as a function of a person-specific intercept, ${\beta }_{0i}$; a person-specific linear slope indicating rate of change with age ${\beta }_{1i}$; person-specific quadratic and cubic slopes ${\beta }_{2i}$ and ${\beta }_{3i}$, and residual error, ${\epsilon }_{it}$. that is assumed normally distributed with person-specific variance, ${\sigma }_{ei}^2$. The person-specific coefficients describing the BMI z-score trajectory were in turn modeled as:

$$\begin{gathered} {\beta }_{0i}={\gamma }_{00}+{\gamma }_{01}BMBM{I}_i+{\gamma }_{02}E{C}_i+{\gamma }_{03}N{A}_i+{\gamma }_{04}Se{x}_i+{\gamma }_{05}HouseholdIncom{e}_i+ \\ {\gamma }_{06}PerinatalRis{k}_i+{\gamma }_{07}AdoptionOpennes{s}_i+u_{0i} \end{gathered}$$ (2)

$${\beta }_{1i}={\gamma }_{10}+u_{1i}$$ (3)

$${\beta }_{2i}={\gamma }_{20}$$ (4)

$${\beta }_{3i}={\gamma }_{30}$$ (5)

$$\begin{gathered} {\sigma }_{ei}^2=a_0^{2}exp(a_{1}BMBM{I}_i+a_{2}E{C}_i+a_{3}N{A}_i+a_{4}MeanBM{I}_i+a_{5}Se{x}_i+a_{6}HouseholdIncom{e}_i+ \\ {a}_{7}PerinatalRis{k}_i+a_{8}AdoptionOpennes{s}_i) \end{gathered}$$ (6)

where ${\gamma }_{00}$, ${\gamma }_{10}$, ${\gamma }_{20}$, and ${\gamma }_{30}$ indicate the prototypical child’s developmental trajectory of BMI z-score; ${\gamma }_{01}$ to ${\gamma }_{07}$ indicate how interindividual differences in children’s level of BMI z-score are related to parent and child characteristics; $u_{0i}$ and $u_{1i}$ are unexplained interindividual differences in level and age-related linear change; $a_0$ is the expected lability (within-person variance) for the prototypical child; and $a_1$ to $a_7$ indicate how interindividual differences in lability are related to parent and child characteristics. Of particular interest for the hypotheses are $a_1,a_2$, and $a_3$, indicating the association of BMI z-score lability with birth mother’s BMI (genetic risk for obesity), child effortful control, and negative affectivity. In the model building process, $u_{2i}$ and $u_{3i}$ terms were included in Equations 4 and 5, but were removed from the final model because their variances were not discernably different than zero.