Effects of participation level and physical activity on eating behavior and disordered eating symptoms in the Brazilian version of the New Moves intervention: data from a cluster randomized controlled trial

ABSTRACT BACKGROUND: Childhood and adolescent obesity is a worldwide public health concern. The New Moves program aims to change eating behavior (EB) and physical activity (PA). OBJECTIVE: To evaluate the effectiveness of an intervention and predictors of better outcomes relating to EB and PA levels. DESIGN AND SETTING: Secondary data from a cluster randomized controlled trial in 10 public schools in São Paulo, Brazil. METHODS: 270 female adolescents, aged 12 to 14 years, were analyzed. Participation levels were categorized as presence in 1 to 9 sessions or 10 to 17 sessions, or control. Effectiveness was evaluated through improvement in disordered EB (DEB) and EB. Predictors of better outcomes relating to PA levels were evaluated through clustering of individual characteristics that affected changes in PA scores. RESULTS: Participation level was not significantly associated with changes in DEB or EB. Girls with higher body mass index percentile (BMI-P) percentile tended to have increases in sedentary lifestyles through the program. Girls with less body image dissatisfaction presented higher increases in daily PA. Girls with higher BMI-P percentile and higher self-esteem showed reductions in sedentary lifestyles. The program seemed to have more effect on daily PA among older girls than among younger girls. CONCLUSIONS: This program could be used as a structured action plan in schools, with the aims of improving eating behaviors and physical activity, in addition to promoting self-acceptance. The results indicate the importance of evaluating determinants of adherence, as these metrics might influence the effectiveness and future design of lifestyle programs.


INTRODUCTION
Pediatric obesity is a worldwide health concern, and the majority of overweight or obese children live in low-to-middle income countries. [1][2][3] Studies on low-income individuals 4 and schoolbased interventions in low-to-middle income countries 5,6 have demonstrated improvements in eating behavior (EB), physical activity (PA) and body weight. Traditional preventive measures against obesity that focus on weight seem to be ineffective and harmful to the participants.
These programs contribute to eating and weight concerns, body image dissatisfaction, low selfesteem and unhealthy weight control practices. Such behaviors are considered to be risk factors for weight instability and development of eating disorders (ED). 7,8 Obesity and ED result from cultural contexts that motivate an unhealthy relationship with food, EB and PA. In addition to these factors, this cultural context discourages respect for the diversity of body size. 9 Obesity and ED share psychosocial and behavioral risk factors, which suggests that integrated interventions would lead to better outcomes. 9 These integrated interventions can include overlapping of problems and involvement of similar risk factors (diet and weight).
They make use of the economic efficiency of addressing two conditions in a single intervention. 10 The New Moves program is an integrated intervention based on social cognitive theory, which is one of the most common theoretical frameworks used in interventions that have the aim of changing EB and PA. [11][12][13] This program was designed in the United States focusing on adolescent females and it incorporates issues relating to eating disorders and obesity. It has a dynamic multi-component that includes factors that can predict body satisfaction, eating behaviors and patterns, weight control practices and PA levels. Positive outcomes were found among female adolescents in the United States, with improvements in sedentary lifestyles, eating patterns, unhealthy weight control behaviors and body/self-image. 11 Considering that none of the previous intervention programs involved an after-school approach and that none of the studies with individuals from low socioeconomic backgrounds 5,6 discussed the effect of participation level on the programs, a gap in the literature currently exists.

OBJECTIVE
We aimed to evaluate the effectiveness of an intervention and the predictors of better outcomes relating to eating behaviors and PA levels.
We hypothesized that 1) girls with higher participation level would show significant improvements in those outcomes after the phase 1 (P1) intervention; and 2) their levels of body image dissatisfaction and self-esteem, their age and their nutritional status would enable prediction of PA levels and sedentary lifestyles.

Study design
This study consisted of an exploratory analysis on a previously conducted cluster randomized controlled trial of the Brazilian New Moves program (BNMP). This program was conducted among girls aged 12 to 14 years old, at ten public schools in the central and southern areas of the city of São Paulo. 14 The randomization of the original trial was performed at the school level to prevent contamination across students, between the intervention and observation arms. There was no blinding regarding intervention assignments or assessment, but blinding was present during data analysis. The analyses followed an intention-to-treat protocol that involved a sensitivity analysis in which all the subjects were included regardless of their length of follow-up or interven- PA sessions were scheduled twice a week, and nutrition and social support components were held on those same days. All of these elements were conducted by trained healthcare professionals, including psychologists, dietitians and PA professionals.

Participation level
The participation level was assessed by categorizing the number of times a participant was present in sessions promoted by the study intervention. This was considered in terms of three groups: control group, presence in 1 to 9 sessions and presence in 10 to 17 sessions. The exploratory analysis broke the randomization, so that participants in the intervention groups who did not participate in the intervention activities were analyzed as controls.

Socioeconomic status (SES)
The "Brazilian Economic Classification" was used to categorize students based on their economic status. This classification is based on possession of items (e.g. television, radio or automobile) and the head of the family's education level. From this, a score was generated and the participants were stratified according to monthly gross family income. The sum of these scores was used to determine the family's purchasing power, which was categorized ranging from A1 to E. 16

Body mass index (BMI)
BMI was calculated as the weight in kilograms divided by the height in meters squared. Weight was measured by trained study staff using a digital scale, while height was measured with a portable stadiometer. Based on BMI-P percentiles (BMI-P) standardized according to age and sex, the participants were classified as defined by the World Health Organization (2007), 17 as underweight or at risk (BMI-P ≤ 15), normal weight (15 < BMI-P < 85), overweight (85 ≤ BMI-P < 97) or obese (BMI-P ≥ 97).

Disordered eating behaviors (DEB)
Disordered eating behaviors (DEB) were assessed using the following three scales.
Body Shape Questionnaire (BSQ) -Brazilian Portuguese version: 18 This is a self-reported scale with the aim of assessing body image dissatisfaction. It presents good internal consistency (Cronbach's alpha = 0.96) and reliability (r = 0.91; P < 0.001).
The questionnaire consists of a 34-item self-reported scale that uses six Likert categories going from "never" to "always", in which the higher the score is, the greater the degree of dissatisfaction with body image is. The scores were classified as "no dissatisfaction", "slight dissatisfaction", "moderate dissatisfaction" and "serious dissatisfaction". 19

Weight Control Behaviors Scale (WCBS) -Brazilian
Portuguese version: 20 This scale is an internally reliable and valid instrument that is used to assess factors representing healthy and unhealthy weight control behaviors. Out of the items on this scale, we used only the following nine items that assessed unhealthy weight control behaviors (UWCB): fasting, skipping meals, dieting, taking diet pills, making oneself vomit, using diuretics, using laxatives, using food substitutes like powdered/special drinks and smoking cigarettes. For each item, the participants responded with a "yes" or "no" response to indicate whether they had performed the behavior with the intent of losing weight in the past month.
The internal consistency among these nine items was adequate (Cronbach's alpha = 0.66). Through this scale, we aimed to assess unhealthy weight-control behaviors. 21 This is a 10-item self-administered scale with items used as four-point Likert categories ranging from strongly agree to disagree, in which higher scores indicate higher self-esteem. Its construct validity shows a significant positive correlation with social support and its internal consistency has been reported as 0.68. With this scale, we aimed to assess self-worth and feelings about the self.

Rosenberg Self-Esteem Scale (RSES):
Although BSQ, WCBS and RSES are self-reported scales, we noticed in a pre-test study that the girls were having difficulties in filling out these questionnaires. We therefore trained the research staff to ask questions, without giving any interpretation of these questions.

Eating behaviors (EB)
To assess EB, we evaluated changes in fruit, vegetable and sugarsweetened beverage intake, and the frequency of breakfast intake.
Fruit and vegetable intake was assessed using the following questions, "Thinking back over the past week, how many servings of fruit did you usually eat on a typical day?" and "Thinking back over the past week, how many servings of vegetables did you usually eat on a typical day?" The response options for both questions went from "none, " to "five or more servings/day. " Last week's intake of regular soda and artificial juices was assessed using the following response options, which were converted to a mean intake/day: never = 0; once or twice a week = 0.2; three or four times a week = 0.5; once a day = 1; twice a day = 2; three times a day = 3; four times/day = 4; or five or more times/day = 5.
The question relating to breakfast intake was, "During the past week, on how many days did you eat breakfast?" and the response options went from "none" to "seven days".

Physical activity level
The participants were provided with accelerometer devices

Statistical analysis
Our exploratory analysis started with an evaluation of distributions, frequencies and percentages for each of the numerical and categorical variables of this study. The categorical variables were evaluated for near-zero variation, 23 or categories with only a small percentage of response that could potentially bias our models. An extensive graphical exploratory analysis was used for both univariate analysis and bivariate associations between potential outcomes and the frequency of participation in study-related activities. Missing data were explored using a combination of graphical displays involving univariate, bivariate and multivariate methods. Imputation was performed using a k-nearest neighbors algorithm (n = 5). 24 The association between participation level and outcome measurements was assessed using generalized estimating equations that adjusted for baseline variables, to account for each school level.
In accounting for each school, we automatically controlled for the confounding of zero participation among those in the control group, compared with the intervention groups. We included both groups (control and intervention) while adjusting for their differences based To avoid overfitting, we applied a cost-complexity strategy using weakest-link pruning by successively collapsing the internal node that produces the smallest per-node increase in the cost complexity criterion. When overfitting was detected, those nodes were removed. Otherwise, they were left intact. We also provided a graphical representation of each model. All analyses were performed using the R statistical language. 25

RESULTS
Out of the overall sample (n = 270), 65.2% of all the subjects did not participate in any sessions, 15.2% participated in 1 to 9 sessions and 19.6% participated in 10 to 17 sessions. Table 1 shows the baseline characteristics of the participants in each group.
Analysis of variance (ANOVA) and chi-square analyses were conducted to determine whether the groups differed in any of the variables with continuous or categorical data, respectively.
At the baseline, these groups differed in only three variables: age, SES and breakfast habits. Regarding the demographics of age and SES, these potentially confounding variables were entered as covariates in the remaining analyses.
To evaluate the association between the participation level and the outcome measurements, we used the three categories representing the participation level by means of a generalized estimating equation to account for the clustering effect within each school.
The participation level did not result in statistically significant differences concerning outcome measurements, including body image dissatisfaction, self-esteem, unhealthy weight-control behaviors, BMI-P and EB measurements (frequency of breakfast and fruit, soda and artificial juice intake per day) ( Table 2).
Based on a previously validated sensor-based index, 22 we used generalized estimating equations to identify predictors of change in daily PA and sedentary lifestyles between the baseline and 17 weeks.
All the scores were normalized to a 0-100 scale, which was then used to determine the predicted means for the change in daily PA and sedentary lifestyles. Out of the 270 participants in the trial, 75 were excluded because of lack of data recording devices, or because they missed the orientation and training day when the devices were distributed, or because some data was found to be missing at the time of data extraction from the device. The analysis was conducted with a final sample of 195 participants, given that only these girls presented MET data. We found that the differences in participation level relating to 1 to 9 and 10 to 17 sessions gave rise to statistically significant associations with increases in sedentary lifestyle (P = 0.027 and P = 0.039, respectively) ( Table 3).

DISCUSSION
Our study demonstrated that the participation level in the BNMP did not affect DEB and EB. Also, we observed a negative effect on PA levels (an increase in sedentary lifestyles). In brief, girls with less body image dissatisfaction presented higher increases in daily PA, while for those with more body image dissatisfaction, the program had almost no effect on daily PA. The program seemed to have more effect on daily PA among older girls than Control group: participants who were absent from the program interventions; 1 to 9 sessions: participants who were present in 1 to 9 sessions of the program;     One potential protective factor against body dissatisfaction is greater amounts of PA, given that this has been correlated with satisfaction with body image, self-esteem, reduced risk of ED and psychological benefits. 29 In our sample, being an older girl, with less dissatisfaction with body image, normal BMI-P and better self-esteem was predictive of higher PA levels and less sedentary behavior. Girls with more body satisfaction may be more engaged in PA for reasons of fun and socialization, while girls who are dissatisfied with their bodies aim to change their weight and shape.
For older girls, there was better understanding of healthcare, while younger girls were more motivated to play than to practice PA.
Girls with higher BMI percentiles had more sedentary habits than did eutrophic girls, and this was a possible explanation for being overweight. Girls with higher self-esteem tended to be better motivated to exercise than did those with low self-esteem. Moreover, PA interventions were associated with increased self-concept and selfworth among children and adolescents. 30 No improvement in EB was observed, given that socioeconomic issues and food availability may influence such habits. One possible explanation for this is that marketing of food and beverages may influence children's food choices, preferences and consumption, especially with regard to foods that are high in energy density and poor in micronutrients. 31 Another possible explanation is that manufactured products, with low monetary value and low dietary value, are preferred over healthy foods. 32  communities tend to buy fewer fruits, vegetables, tubers and roots in a week than do the middle and upper classes. This may be consequent to a lack of food and nutrition education in public schools and for the overall community, which would develop skills and knowledge and allow these communities to choose and consume their foods safely and appropriately. 33 Effective procedures and designs for programs targeting multicomponent behaviors among low-income individuals 4

CONCLUSIONS
This program could be used as a structured action plan in schools, with the aims of improving eating behaviors and physical activity, in addition to promoting self-acceptance. Moreover, it could be implemented in a personalized and interactive manner, with weekly messages on mobile devices. As discussed by López-Guimerà et al., 37 in interactive programs like New Moves, it is also important to monitor adherence to activities.
With regard to future work, researchers are encouraged to address the issue of adherence in after-school programs and to involve school staff in the development of further interventions. Moreover, through adding qualitative data to evaluations, insights that would enhance the likelihood of successful intervention may be gained.