Latent Profiles of Health Behaviors in Rural Children with Overweight and Obesity

Abstract

Objective

The objectives were to identify profiles of school-age children with overweight and obesity (OW/OB) from rural counties based on patterns of diet, activity, and sleep, to examine demographic predictors, and to examine whether profiles were differentially associated with psychosocial functioning.

Methods

Participants included 163 children (M_age = 9.8) and parents. Children wore accelerometers to assess physical activity and sleep duration. Consumption of fruits and vegetables (F/V) and sugar-sweetened beverages (SSB) was assessed with a food frequency questionnaire. Self-report of emotional, social, and academic health-related quality of life (HRQOL), peer victimization, social skills, and social problem behaviors was collected, as well as parent-report of HRQOL. Latent variable mixture modeling (LVMM) was conducted.

Results

Sleep did not significantly contribute to profile differentiation and was removed. Four profiles emerged: (a) Low F/V + Low SSB + Low activity, (b) Low F/V + Low SSB + Moderate activity, (c) High F/V + High SSB + Low activity, and (d) Moderate F/V + Moderate SSB + High activity. Older children were more likely to be in profile 1. After controlling for child age, parents of children in profile 1 reported significantly lower child social HRQOL than parents of children in profiles 2 and 4. Children in profile 4 reported experiencing significantly lower victimization than those in profile 3.

Conclusions

There are subgroups of rural children with OW/OB that engage in various combinations of healthy and unhealthy behaviors. LVMM has the potential to inform future interventions and identify needs of groups of children with OW/OB.

Introduction

In the United States, pediatric overweight and obesity are considered epidemics. Nationally, 16.6% of children have overweight and 18.5% have obesity (Fryar et al., 2018), with higher prevalence in rural areas (Johnson & Johnson, 2015). These rates are concerning because children with overweight and obesity are at greater risk for physical and psychosocial health detriments, including cardiovascular disease, type 2 diabetes, depression, and poorer quality of life (Sahoo et al., 2015). Further, overweight and obesity as well as associated negative health outcomes often persist into adulthood and largely contribute to increased health expenditures (Centers for Disease Control and Prevention (CDC), 2016; Trasande & Chatterjee, 2009).

Although it is important to note that the etiology is complex with multiple biological, genetic, environmental, and epigenetic contributors, there are a few key health behaviors that are linked to overweight and obesity (Centers for Disease Control and Prevention (CDC), 2016; Gurnani et al., 2015). Several dietary factors are associated with overweight and obesity, including low fruit/vegetable (F/V) and high sugar-sweetened beverage (SSB; e.g., soda, sports drinks) consumption. In fact, a direct relationship has been found between consumption of SSB and overweight and obesity in youth (Keller & Della Torre, 2015). Further, research suggests that when children eat more F/V, they also eat less high-fat/high-sugar foods that could contribute to weight gain (Epstein et al., 2001). Additionally, physical inactivity is associated with overweight and obesity. Together, this results in an energy imbalance that contributes to excess weight gain (Centers for Disease Control and Prevention (CDC), 2016; Gurnani et al., 2015). Another associated health behavior is sleep; in particular, shorter sleep duration is associated with greater risk of overweight and obesity (Wu et al., 2017). Several mechanisms have been posited to explain this relationship, including alterations in hormones that in part regulate appetite (i.e., leptin and ghrelin), and increases in waking time and fatigue, which can impact calorie consumption and physical activity (Wu et al., 2017).

Beyond links with overweight and obesity, diet, physical activity, and sleep are also associated with psychosocial functioning. For example, negative emotions and stressful events are associated with emotional eating, a higher consumption of sweet and fatty foods, and a lower intake of F/V (Michels et al., 2012). On the other hand, healthier dietary habits are associated with fewer academic and behavioral problems (Shi et al., 2013). Regular physical activity is associated with better self-esteem and academic achievement and lower risk for depression, anxiety, distress, and emotional disturbance (Ahn & Fedewa, 2011; Fedewa & Ahn, 2011). Lastly, sleep difficulties are correlated with reduced quality of life and academic performance and increased emotional distress and behavioral problems (Medic et al., 2017).

Due to their relationships with excess weight and psychosocial functioning, many pediatric weight-management interventions aim to modify these health behaviors (e.g., Gurnani et al., 2015). In particular, goals for behavior change often include increasing F/V intake, reducing SSB intake, increasing physical activity, and increasing sleep duration. It is likely that children with overweight and obesity engage in various combinations of healthy and unhealthy behaviors. However, there is no sufficient evidence to suggest which subgroups exist in this population or which combinations of behaviors may be differentially associated with psychosocial outcomes. Knowledge about how health behaviors cluster within a sample of children with overweight and obesity and the associations with psychosocial outcomes could allow for more tailored intervention strategies (Berlin et al., 2018).

One method to examine homogenous groups within a larger heterogeneous population is latent variable mixture modeling (LVMM). LVMM is a person-centered analytic approach that uses data patterns to identify groups based on similarities and differences (Berlin et al., 2014, 2018). This method does not rely on predetermination of groups, as is common with some clustering techniques; instead, individuals are probabilistically assigned to profiles based on data-driven modeling techniques, and the latent nature allows for adjustment for measurement error and membership uncertainty (Berlin et al., 2014, 2018).

Several studies have examined profiles of health behaviors, such as diet and physical activity, in youth, and systematic reviews report that three to seven profiles were identified (Leech et al., 2014; Parker et al., 2019). Further, these reviews found that profiles differed based on demographic variables, such as age, sex, and body mass index (BMI; Leech et al., 2014; Parker et al., 2019). Among studies that examined activity and diet, profiles indicative of healthier diets and higher physical activity levels were associated with greater educational aspirations, school engagement, and motivation and better self-concept and relationships. On the other hand, profiles with moderate diets and lower physical activity levels were associated with lower self-regulation, body satisfaction, and perceived health (Parker et al., 2019).

Few studies have included sleep as a factor within profiles of health behaviors associated with psychosocial functioning. Champion et al. (2018) examined profiles of six self-reported health behaviors (i.e., binge drinking, smoking, sleep duration, physical inactivity, F/V intake, and sitting time) in a sample of young adults. The authors identified three unique profiles that differed in the proportion of females and males, and they found that profiles with multiple unhealthy behaviors were associated with greater psychological impairments (i.e., distress, anxiety, and depression symptoms; Champion et al., 2018). In addition, although not using a LVMM approach, Nuutinen et al. (2017) utilized k-means analysis to examine clusters of health behaviors (i.e., screen time, sleep duration, sleep quality, F/V intake, and physical activity) and their associations with the risk for overweight status in a group of adolescents. Results revealed a healthy group (moderate/high sleep duration, physical activity, and vegetable intake, along with other health behaviors) and an unhealthy group (low sleep duration, low physical activity, low F/V intake, along with other health behaviors), as well as a few other profiles that varied in prevalence of girls and boys and differed based on health behaviors not included in the current analyses (Nuutinen et al., 2017).

The current study addresses several gaps in the literature. This is the first identified study to explore health behavior profiles using LVMM in a rural sample of school-age children with overweight and obesity. Furthermore, most studies examining health behavior profiles have not included sleep in analyses (Leech et al., 2014), which the current study included. Additionally, most studies examining profiles of health behaviors relied solely on self-report measures (Parker et al., 2019). The current study addresses this gap by utilizing objective reports of physical activity and sleep. This study also extends the literature by examining psychosocial correlates of the profiles. In particular, Parker et al. (2019) called for more exploration of interpersonal correlates, which was a focus of the current study.

The current study had several aims and hypotheses. First, the study utilized LVMM to identify profiles of school-age children with overweight and obesity based on patterns of diet, physical activity, and sleep. Second, the study aimed to explore demographic predictors of profiles. It was hypothesized that child age, sex, and weight status (BMI z-score; BMIz) would predict profile membership. Third, the study aimed to explore whether profiles were differentially associated with psychosocial variables after controlling for predictors. It was hypothesized that better psychosocial functioning (i.e., emotional, social, and academic functioning) would be associated with profiles with healthier behaviors, such as higher physical activity, healthier diet, and more sleep (Parker et al., 2019). Lastly, the study examined whether statistical differences between profiles were also associated with clinical differences.

Methods

Participants

The current sample is a part of a randomized controlled trial (RCT) examining the effectiveness of two, 12-month healthy lifestyle interventions relative to an education condition, for school-age children with overweight and obesity and their families from rural counties in North Central Florida (Janicke et al., 2019). Initially, 269 children were enrolled in the study at baseline, as well as their parents; 31.3% of the overall sample was from racial/ethnic minority groups, and 48.9% of the sample had an annual income below $40,000.

Eligibility criteria for children to participate in the RCT included being between the ages of 8–12 years of age and having a BMI at or above the 85th percentile for their age and sex. Parents/guardians needed to be under 75 years of age, and families had to live within designated rural counties. Families were excluded if they had restrictions to diet/activity, medical contraindication (e.g., heart condition), or medication contraindication (e.g., corticosteroids). See Janicke et al. (2019) for detailed methods.

Given limits to handling missing data for indicator variables with analyses, children had to have complete data. A total of 163 children (60.6% of the full sample) had complete diet and physical activity data, and 132 (49.1% of full sample) had complete diet, physical activity, and sleep data.

Procedure

Children and families were recruited by sending materials to homes through mailing lists, schools, physicians’ offices, and community organizations, and by sharing information via media press releases. Families were encouraged to call the study office to learn more about the study and participate in a screening for eligibility. Eligible families were scheduled for “pre-screening” visits in their county extension offices, where they completed informed consent/assent and provided demographic and medical history information; height and weight were also measured. Families who continued to be eligible were scheduled for a baseline visit within 3 weeks of the start of the intervention; health measurements were taken, psychosocial questionnaires were completed, and treatment assignments were distributed. Data in the current study were from the baseline visit. The governing institutional review board approved all procedures.

Measures

Fruit/Vegetable and Sugar-Sweetened Beverage Intake

The Block Kids Food Frequency Questionnaire 2004 (NutritionQuest, Berkeley, CA) is a 77-item measure that assesses child dietary intake in the past month. Parents and children worked together to indicate the frequency and quantity of each item, and standardized visual prompts were provided to facilitate estimation. Validity studies comparing this measure to 24-hr diet recalls have found variable results; however, given the barriers of administering recalls, the food frequency questionnaire is considered a reasonable and acceptable measure (Cullen et al., 2008). Variables used in the current study included the average daily servings of F/V and the average daily calories from SSB.

Physical Activity and Sleep Duration

The Sensewear Armband Accelerometer (Bodymedia, Inc., Pittsburgh, PA) was utilized to obtain objective measurement of moderate and vigorous physical activity and sleep duration. The device utilizes a bi-axial accelerometer and sensors to measure thermal flow, skin temperature and response, and air temperature. Instructions were to wear the device on the non-dominant upper arm for 24 hr for 7 days, except when swimming or bathing. To be considered a valid day, participants were required to wear the device for at least 16 hr. Participants with at least one valid weekend day and three valid weekdays were included in analyses for the current study. The Sensewear Armband Accelerometer was programmed with demographic information (i.e., sex, age, height, weight, and handedness), and data were collected at a sampling rate of 32 Hz. Data were binned into one-minute epochs and downloaded using the Sensewear Professional Software (version 7.0; Body-Media Inc.).

For physical activity, energy expenditure was quantified with device-specific proprietary algorithms (Arvidsson et al., 2009). Minutes were classified as activity categories based on metabolic equivalents (METs), with >3 METs being moderate and >6 METs being vigorous. Average moderate physical activity (in minutes) and vigorous physical activity (in minutes) per day were used for analyses. When compared to doubly labeled water, the Sensewear Armband has been shown to adequately measure energy expenditure in children (Arvidsson et al., 2009). However, studies have found that the device under- or over-estimates energy expenditure (Calabro et al., 2013; Dorminy et al., 2008; Lopez et al., 2018); similar limitations have also been found with other accelerometers, including those frequently used in research settings. Of note, the greatest validity concerns seem to be related to the detection of sedentary and light activity in children; therefore, these categories were not examined (Lopez et al., 2018; Stålesen et al., 2016).

Sleep and wake times and positionality were identified using device-specific proprietary algorithms. These data were used by research staff to hand-compute total sleep time (in minutes), which was utilized for analyses. Specifically, sleep onset was noted when at least three consecutive epochs were measured as “lying down” and “asleep” (as measured by device-detected positionality and sleep/wake status). Sleep offset was denoted when the participant had five consecutive epochs of “awake” time, with no subsequent “asleep” periods. Total sleep time (in minutes) was determined by summing the number of epochs between sleep onset and offset and averaging across days. Validation studies support the use of the armband to measure sleep at the group level for children when compared to overnight polysomnography (Soric et al., 2013).

Psychosocial Health-Related Quality of Life

The Pediatric Quality of Life Inventory (PedsQL 4.0; Varni et al., 2001) is a measure of health-related quality of life (HRQOL) that was completed by child participants and their parents. The self- and parent-report versions consist of 23 items that assess functioning. The instructions ask completers to indicate how much of a problem each item has been during the past 1 month using a 5-point Likert scale. Emotional, Social, and Academic HRQOL subscales were used in analyses. Scores range from 0 to 100, with higher scores indicating better HRQOL. This measure has demonstrated good reliability and validity (Varni et al., 2001). In the current study, the internal consistencies for the Emotional, Social, and Academic HRQOL domains for children were 0.75, 0.79, and 0.67, respectively. Internal consistencies for the Emotional, Social, and Academic HRQOL domains for parents were 0.82, 0.76, and 0.78, respectively.

Peer Victimization

The Social Experience Questionnaire-Self-Report (SEQ-S; Crick & Grotpeter, 1996) is a 15-item measure used to assess a child’s experience of peer victimization. Children indicated how frequently each item happened to them over the last month on a 5-point Likert scale. This measure consists of three subscales: Overt Victimization (i.e., intentional acts of physical aggression from a peer), Relational Victimization (i.e., acts from peers targeted at damaging social reputation or relationships), and Prosocial Acts (i.e., peers engaging in helping/caring behavior that benefits the respondent). Overt and Relational subscales were used in the current analyses. Scores range from 5 to 25, with higher scores indicating more victimization. Evidence for the SEQ-S supports adequate test–retest reliability, concurrent validity, and stable factor structure (e.g., Storch et al., 2005). The internal consistencies for Overt and Relational Victimization in the current study were 0.84 and 0.76, respectively.

Social Skills and Problem Behaviors

The Social Skills Improvement System Rating Scales (SSIS-RS; Gresham & Elliott, 2008) is a 75-item measure of children’s social skills (e.g., cooperation) and problem behaviors (e.g., bullying). Children were instructed to rate how true each statement was about them using a 4-point Likert scale. Social Skills and Problem Behaviors subscales were used in current analyses. Higher scores indicate higher Social Skills and Problem Behaviors. This measure is an updated version of the Social Skills Rating System (Gresham & Elliott, 1990). The SSIS-RS has high internal consistency and convergent validity, as well as good content, structural, and cross-cultural validity (Cordier et al., 2015; Gresham et al., 2011). In the current study, internal consistencies for the Social Skills and Problem Behaviors subscales were 0.94 and 0.88.

Weight Status

Research staff measured child height and weight to the nearest 0.1 cm and 0.1 kg using a stadiometer and portable digital scale. Three measurements were averaged, and BMIz was calculated (CDC, 2011) and used as a predictor.

Data Analytic Plan

Preliminary Analyses

For the overall sample, descriptive statistics were explored, and t-tests were performed to clarify any differences in included participants and those excluded due to missing data.

Model Specification

Mplus (Version 7.4; Muthén & Muthén, 1998–2012) was used to conduct LVMM analyses. Indicator variables inserted into the models to identify profiles included moderate physical activity, vigorous physical activity, F/V intake, SSB intake, and sleep duration, and an unrestricted model was estimated. Univariate entropy values were explored to determine how much each indicator variable contributed to classification of the profiles and were removed if necessary (Asparouhox & Muthén, 2018).

Model Estimation

One-, two-, three-, and four-profile models were fit to the data with maximum likelihood estimation and no missing values.¹ Models utilized multiple start values (4000 initial-stage, 100 second-stage), and the best log-likelihood was replicated (Berlin et al., 2014). Models were able to run successfully without imposing restrictions.

Model Selection and Interpretation

To determine the best fitting model, a number of indices were evaluated, including Akaike Information Criterion (AIC; Akaike, 1987), Bayesian Information Criterion (BIC; Schwartz, 1978), and sample size adjusted BIC (ssBIC; Sclove, 1987)²; smaller values of each indicate better fit. Entropy was also explored to quantify the classification accuracy; higher scores indicate more accurate classification (Celeux & Soromenho, 1996). Finally, the number of participants in each profile was considered to determine the benefit of adding extra profiles with few participants (Lubke & Neale, 2006). Descriptive statistics for each profile was performed.

Profile demographic predictors (i.e., child age, sex, and BMIz) were examined using Vermunt’s 3-step approach (Bakk et al., 2013) with maximum likelihood estimation with robust standard error. Results are presented as odds ratios (OR).

Psychosocial Correlates

Psychosocial correlates (i.e., emotional/social/academic HRQOL, overt and relational victimization, social skills and problem behaviors), while controlling for covariates, were examined using the manual BCH method (Muthén & Muthén, 2012) with maximum likelihood estimation with robust standard error. Results are presented as Wald tests of equal means.

Minimal Clinically Important Difference

For the profiles that differed based on statistical significance on outcomes, minimal clinically important difference (MCID) was calculated to clarify if the differences were also clinically significant (Copay et al., 2007). By examining the MCID, it clarifies whether profile differences in variables are greater than what would be expected with measurement error, thus indicating a clinically significant difference. If the difference in scores between profiles is greater than the standard error of measurement, it is considered clinically meaningful (Copay et al., 2007).

Results

Preliminary Analyses

The total sample at baseline with complete data included 163 children between the ages of 8 and 12 (M = 9.77, SD = 1.43; 55.8% girls). The reported racial makeup was 67.3% White, 13.6% Black/African American, 11.7% Biracial, and 7.4% other. Regarding ethnicity, 84.0% reported being non-Hispanic. Mean BMIz was 2.19 (SD = 0.37; ranged from 1.21 to 2.87). No differences were found in demographic characteristics or outcome variables for those included in the study and those excluded due to incomplete data on indicator variables (all p > .05).

Primary Analyses

Model Specification and Estimation

The following indicators were initially inserted in the 1-, 2-, 3-, and 4-profile models with the sample of 132 children: moderate physical activity, vigorous physical activity, F/V intake, SSB intake, and sleep duration. However, when exploring univariate entropy, it was observed that sleep duration contributed little to profile classification (0.06–0.15). Therefore, sleep duration was removed as an indicator variable (Asparouhox & Muthén, 2018), leaving moderate physical activity, vigorous physical activity, F/V intake, and SSB intake as indicators with a sample of 163 children.

Model Selection and Interpretation

Results from the 1-, 2-, 3-, and 4-profile models are presented in Table I. Results indicated that models with two, three, and four latent profiles fit better than a unitary model. Further, all criteria favored the four-profile solution, so the four-profile model was retained. The descriptive statistics of the profiles are presented in Table II.

Participants in profile 1 were characterized by consuming low servings of F/V and calories from SSB and engaging in low moderate and vigorous activity (“Low F/V + Low SSB + Low activity”; n = 44). Profile 2 included the largest number of children and was characterized by consuming low servings of F/V and calories from SSB and engaging in moderate activity (“Low F/V + Low SSB + Moderate activity”; n = 66). Of note, profiles 1 and 2 had similar levels of F/V and SSB intake, but profile 2 had about 4 more minutes of vigorous activity and 58 more minutes of moderate activity.

Profile 3 was characterized by consuming the highest servings of F/V and calories from SSB and engaging in low activity (“High F/V + High SSB + Low activity”; n = 26). Notably, participants in this profile were the closest to the recommended F/V intake, but they also reported consumption of approximately 255 extra calories per day from SSB. Average minutes of moderate and vigorous physical activity were similar to profile 1.

Profile 4 was characterized by consuming moderate servings of F/V and calories from SSB and engaging in high levels of activity (“Moderate F/V + Moderate SSB + High activity”; n = 27). Specifically, participants in this profile reported consumption of more F/V and SSB than profiles 1 and 2, but not as much as profile 3. Further, participants had considerably more moderate/vigorous activity than participants in other profiles (about 186 min of moderate and 21 min of vigorous).

Demographic Predictors

Child age was a significant predictor of profile membership (ps < 0.05). In particular, older children were significantly more likely to be in profile 1 than 2 or 3 (OR = 1.67 and 1.56, respectively). Child BMIz approached significance (p = 0.052), with children in profile 4 having lower BMIz than children in profile 1 (OR = 0.24), though not statistically significant. Child sex was not a significant predictor of profile membership (ps > 0.05).

Psychosocial Correlates

The next step was to determine the associations between the profiles and psychosocial correlates, after controlling for child age, which was the only statistically significant predictor of profile membership.³ The results indicated that parents of participants in profile 1 (Low F/V + Low SSB + Low activity) reported significantly lower social HRQOL for their children than the parents of participants in profiles 2 (Low F/V + Low SSB + Moderate activity) and 4 (Moderate F/V + Moderate SSB + High activity; χ² = 4.27, p = 0.04 and χ ² = 10.51, p < 0.01, respectively). Additionally, participants in profile 4 (Moderate F/V + Moderate SSB + High activity) self-reported significantly lower relational victimization than the participants in profile 3 (High F/V + High SSB + Low activity; χ² = 5.32, p = 0.02). There were no significant differences between profiles on other psychosocial outcomes (p > .05).

Finally, MCID was examined. MCID for parent-report of child social HRQOL is 9.59. The calculated mean differences between profile 1 and profiles 2 and 4 were 7.95 and 13.03, respectively. Therefore, there is a clinically meaningful difference in profiles 1 and 4, but not profiles 1 and 2. MCID for child-report of relational victimization is 2.02. The calculated mean differences between profiles 3 and 4 were 2.87, indicating a clinically meaningful difference.

Discussion

The current study aimed to fill gaps in the literature by using LVMM to identify and describe profiles of health behaviors in a sample of school-age children with overweight and obesity. Further, the study included objective measurements of physical activity and sleep and examined self- and parent-report of psychosocial correlates. It was hypothesized that child age, sex, and BMIz would predict profile membership and that profiles would be differentially associated with psychosocial variables.

Contrary to expectations, sleep duration did not significantly contribute to profile differentiation. However, it would be incorrect to conclude that sleep duration is not an important health behavior to consider. Instead, a more accurate interpretation is that, in this sample, sleep duration did not distinguish profiles over and beyond the differentiation by diet and physical activity. It is possible that sleep-related variables other than sleep duration may have more impact on these profiles. For example, higher sleep onset latency, more sleep disturbances, recurrent awakenings, and lower sleep efficiency are factors that have been associated with overweight and obesity in children, independent of sleep duration (Fatima, 2016). Future studies should examine profiles of these sleep variables in children with overweight and obesity to better understand how they group together and are differentially associated with outcomes.

The following four profiles were identified in this population of rural children with overweight and obesity: Low F/V + Low SSB + Low activity; Low F/V + Low SSB + Moderate activity; High F/V + High SSB + Low activity; Moderate F/V + Moderate SSB + High activity. Overall, this finding indicates that there are subgroups of children with overweight and obesity that engage in various combinations of healthy and unhealthy behaviors. Of note, children with overweight and obesity across all profiles did not report adequate F/V intake, though this is consistent with national data of children regardless of weight status (Kim et al., 2014). The recommended servings per day is five (Rogers & Motyka, 2009), but the mean across profiles ranged from about 1.5 to 3.5.

The hypothesis that child sex, age, and BMIz would predict profile membership was partially supported. In line with previous studies (Leech et al., 2014; Parker et al., 2019), child age was found to predict membership, with older children being more likely to be in the profile characterized by low caloric intake and low physical activity. This finding is not surprising given changes in health behaviors that occur as children age. For example, a previous study found that when comparing 9-11-year olds to 6- to 8-year olds, the older group was less likely to meet physical activity guidelines (Fakhouri et al., 2013). Similar patterns have been found for F/V intake (Albani et al., 2017). Contrary to hypotheses, child sex was not found to predict profile membership in the current sample. Several studies have found that a higher proportion of males were in profiles with high physical activity, whereas a higher proportion of females were in profiles with low physical activity (Leech et al., 2014). The lack of finding may have been due to the unique sample of children with overweight and obesity from rural counties, or it could have resulted from a lack of power to detect differences. In line with recommendations from Leech et al. (2014), it is recommended that future studies examine health behavior profiles in large enough samples to analyze boys and girls separately to determine similarities and differences in profiles. Child BMIz was also not found to be a significant predictor of profiles (though it approached significance), possibly because all participants had overweight and obesity and variability was small. Race/ethnicity was not examined as a predictor given the primarily White and non-Hispanic sample; however, this is recommended in future studies.

The hypothesis that the profiles would be differentially associated with psychosocial variables was partially supported. Parent-reported social HRQOL for children was found to be different across profiles. Specifically, profile 1 (Low F/V + Low SSB + Low activity) had statistically and clinically lower levels of child social HRQOL than profile 4 (Moderate F/V + Moderate SSB + High activity) and statistically lower levels than profile 2 (Low F/V + Low SSB + Moderate activity). These findings suggest that the differences in physical activity levels may have contributed to the differences in parent-reported social HRQOL for children. Similarly, profile 4 (Moderate F/V + Moderate SSB + High activity) had statistically and clinically lower levels of relational victimization than profile 3 (High F/V + High SSB + Low activity), suggesting that high activity may be a protective factor for relational victimization in children with overweight and obesity. Interestingly, no statistical differences were found in child-reported HRQOL subscales or in social skills/problem behaviors.

Findings from the current study indicate that differences in physical activity levels in children with overweight and obesity are related to differences in several social functioning variables. These findings are similar to those by Parker and colleagues (2019), who concluded that better scores on psychosocial factors were associated with higher physical activity levels, despite levels of other health behaviors. It may be that children in the high activity profile were participating in sports teams with other children, which has been shown to be related to higher parent-reported HRQOL in other studies (e.g., Vella et al., 2014). The combination of physical activity engagement and the social context (e.g., teamwork, accountability) present with sports teams may be protective for children with overweight and obesity.

Several clinical and research implications can be drawn from these findings. First, children with overweight and obesity across all profiles reported inadequate F/V intake. This finding is consistent with national dietary intake data from children of all ages regardless of weight status (Kim et al., 2014). However, F/V intake is an important aspect of overall diet quality and has been shown to be an independent contributing factor for prevention of many chronic diseases, such as hypertension and other cardiovascular diseases, and should therefore continue to be addressed along with other dietary factors within lifestyle interventions (Boeing et al., 2012). Second, given that older children were more likely to be in a profile of low activity, increasing physical activity, particularly in older children, is an important intervention target. Additionally, physical activity seems to be a protective factor in regard to social HRQOL and relational victimization for school-age children with overweight and obesity. Intervening to increase physical activity may be particularly important for this population, given higher rates of social difficulties, particularly relational victimization, for children with excess weight (Wang et al., 2010). Lastly, given that the one-profile model was not a good fit for the data, LVMM can be a useful analytic tool when examining health behaviors in this population. Data from LVMM may be valuable in intervention research using adaptive designs (e.g., SMART designs) by informing decision-making and identifying needs of specific groups of children.

Findings of the current study must be considered in the context of several limitations. First, the sample was from rural counties and was primarily Caucasian and non-Hispanic. Additionally, the data were cross-sectional. Future studies should examine health behavior patterns in children with overweight and obesity from a representative sample and over time. Regarding diet variables, there are known limitations of self-report assessments (Farshchi et al., 2017); parents and children may over- or under-report intake, leading to inaccurate estimations. Regarding physical activity, the device used in the current analyses has limitations in its ability to accurately estimate energy expenditure compared to criterion measures (Calabro et al., 2013; Dorminy et al., 2008). Although the means of the profiles may be over- or under-estimations, the patterns relative to other children in the sample are likely accurate (e.g., children in profile 4 had drastically more physical activity than children in other profiles). Additionally, given the small sample size and number of indicators, as well as limitations in the Sensewear software for estimating sedentary and light behavior, these other activity categories were not explored in the current analyses. Additionally, a large number of participants were excluded due to incomplete data. Multiple imputation was not utilized in the current study due to concern that this approach would mask later identification of profiles (Enders, 2010). Lastly, the current study did not have data on the types of moderate and vigorous physical activity the participants were engaging in. Future studies should consider including a combination of objective measurements and self-report measurements of physical activity to aid in ability to draw conclusions.

Despite these limitations, the study offered several strengths. First, children from rural counties and with diverse socioeconomic statuses were included, samples which are typically underrepresented in research. This was also the first study that examined health behaviors, including sleep, in school-age children with overweight and obesity. It also utilized objective measurements of physical activity and sleep and both child- and parent-report of psychosocial variables. Finally, the study examined several forms of social functioning in relation to the profiles, which was called for by Parker et al. (2019).

In conclusion, the current study found that this sample of children with overweight and obesity from rural counties is heterogeneous, with health behaviors (i.e., F/V intake, SSB intake, moderate, and vigorous physical activity) clustering in unique ways. This study also provided information about predictors and psychosocial correlates of the profiles. In particular, older children were more likely to be in the profile characterized by low intake and low activity. Additionally, physical activity drove statistical and clinically meaningful differences in psychosocial functioning, particularly parent-report of child social HRQOL and child-report of relational victimization. Future studies should examine these patterns in a large nationally representative sample, over time, with objective and subjective assessments of physical activity, and with additional predictors and outcomes.

Funding

This study was funded by National Institute of Diabetes and Digestive and Kidney Diseases [#R18DK082374].

Conflicts of interest: No conflicts to declare.

Footnotes

References