Factors associated with extreme weight loss behaviors among adolescents of Northeastern Brazil: a hierarchical approach

Abstract The objective of this article is to identify the prevalence and evaluate the factors associated with extreme weight loss behaviors among adolescents of Northeastern Brazil. Cross-sectional study with 2,439 adolescents from Sample 2 (2015) of the National School-based Health Survey. Extreme weight loss behaviors, such as self-induced vomiting, laxative use and use of medicines or formulas (outcome) and independent variables were evaluated using a self-administered electronic questionnaire. The statistical analysis was performed following a hierarchical conceptual model. The prevalence of extreme weight loss behaviors in the sample was 12.1%. Among the factors that showed significant association with the outcome, we emphasize the administration model of the private school (PR = 0.62; CI = 0.46-0.84), bullying related to body appearance and for other reasons (PR = 1.62; CI = 1.19-2.20), forced sexual intercourse (PR = 2.65; CI = 1.90-3.69), insomnia (PR = 1.84; CI = 1.43-2.37), and be perceived as fat or very fat (PR = 1.90; CI = 1.50-2.42). Moderate prevalence of extreme weight loss behaviors was identified among adolescents. Socioeconomic factors, exposure to violence, mental health, and body image were associated with the adoption of these behaviors.


introduction
Adolescence is the period of life in which intense physical, psychological, and behavioral changes occur 1 .Besides dealing with the changes already expected for that phase, adolescents are faced with beauty standards and extreme appreciation of appearance, resulting in unreal internalization and consequent increase in insecurity, anxiety, and body dissatisfaction 2 .
The search for a body pattern socially accepted can lead the adolescent to adopt extreme weight loss behaviors (EWLB), such as restrictive diet and fasting, use of food substitutes, smoking more cigarettes, use of laxatives, diuretics, and medications as well as induction of vomiting [3][4][5][6] .
The assessment of such behaviors varies between studies.In Brazil, Leme et al. 7 , in a cluster randomized controlled study, identified in adolescents three categories to classify weight control behaviors, including healthy ones (exercising, eating more fruits and vegetables, eating fewer fatty foods, eating fewer sweets), unhealthy ones (skipping meals, eating little and fast), extreme unhealthy behaviors (using diuretics and smoking) and, lastly, other weight control behaviors (medicines, meal substitutes).In subsequent studies with Brazilian adolescents, this classification was changed to a broader concept, already established, grouping all categories of unhealthy weight control behaviors.For example, Dunker & Claudino 8 adopted the definition of New Moves Project. 9and Ferreira et al. 10 applied an adaptation of the Hay scale 11 .
The involvement of adolescents with EWLB is a serious concern with clinical relevance, because of its contribution to the etiology of eating disorders and obesity, in addition to adverse physiological effects, such as esophagitis, gastric rupture, and functional gastrointestinal disorders 12,13 .
In recent decades, there has been a worrying prevalence of EWLB among adolescents in both developed countries (6.2% to 12%) 14,15 and those in development (9.6% to 30.7%) 16,17 .In Brazil, a study conducted with the three surveys (2009,  2012, and 2015) of the National School-Based Health Survey (PeNSE), recorded a temporal trend of increased prevalence of EWLB among adolescents living in Brazilian capitals, with estimates of 6,4% in 2009, 9.0% in 2012 and 10.1% in 2015 4 .
The increased occurrence of EWLB in adolescence is explained by various individual (biological and psychological) and socio-cultural factors identified as potential risks for the adoption of these behaviors.Evidence shows that body image dissatisfaction, followed by overweight/obesity, are determinants for adoption of EWLB 6,18 and that the higher frequency of family meals is associated with lower chances of adopting these behaviors 19 .
In this direction, in northeastern Brazil cultural values guide the way the northeastern relates to food and weight, also, is the region that has the largest coastline and high climatic temperatures most of the year, which is associated with greater body exposure with the adoption of light clothing for boys and girls 20,21 .Thus, it is assumed that northeastern adolescents are more vulnerable to social pressures to adapt their body image to a socially established standard, which can differentiate the way EWLB develops.Therefore, the dissolution of cultural boundaries and the resulting homogenization of values relativized this premise, emerging as one more element to justify the conduction of this study on this theme in this region.
Then, considering the complex and multifactorial nature of EWLB and the need to deepen knowledge on this topic, this study was guided by the hypothesis that socioeconomic and demographic characteristics, family, individual, and psychological behaviours are related to the development of EWLB in adolescents.Thus, the present study was developed to assess, based on PeNSE 22 of 2015, the prevalence and factors associated with unhealthy behaviors in adolescents of northeastern Brazil.methods This is a cross-sectional study, using data from Sample 2 of the PeNSE 2015 edition, collected from a complex sampling design 22 .This study followed the recommendations of the Strengthening the Reporting of Observational Studies in Epidemiology -STROBE.
The national sample consisted of students who attended from 6 th to 9 th grade of elementary school and from 1 st to 3 rd grade of high school in public and private schools located in urban and rural areas of the five geographic regions of Brazil.This study adopted the section of the Northeast region with 3,230 adolescents.Data from 791 questionnaires were excluded due to the absence of the variable maternal education, so 2,439 adolescents were included in the analysis.The design of the study sample selection is shown in Figure 1.
Data collection was performed with electronic devices (smartphones) in a self-administered electronic questionnaire.The researchers from the Brazilian Institute of Geography and Statistics (IBGE) distributed the devices to the students on the day of the interviews and introduced them to how to use them.In addition to the questionnaire, anthropometric measurements of weight and height were performed to assess the body mass index (BMI) of students.PeNSE's data is in the public domain and details about the sampling strategy and other research information can be found in the report of its third edition held in 2015 22 .This nationwide research was approved by the Research Ethics Committee of the Ministry of Health of Brazil, approval nº 192/2012, referring to registration n o 16,805, from CONEP/ MS, from 27/3/2012.
The outcome variable, EWLB, was constructed based on two simple questions: "Did you vomit or take laxatives to lose weight or avoid gaining weight?" (yes/no) and "have you taken any medicine, formula, or other weight loss product without medical follow-up?"(yes/no) 22 .A negative answer classified the participant with absence (0 = reference) of EWLB, and if the answer was affirmative for at least one of the questions was considered presence (1) of these behaviors in adolescents.
The exposure variables represent categories related to socioeconomic and demographic factors, family, individual and psychological behavior, and are described in Chart 1.
Initially, the descriptive analysis was performed to characterize the study population using frequencies.Then, the Pearson's chi-square test was performed to compare the prevalence of EWLB according to the exposure variables.The statistical analysis was performed following a predefined conceptual model like that suggested by Victora et al. 25 to map the relationships proposed between the EWLB and the variables of interest.This structure is organized in blocks and distributed in three levels (Figure 2).
Poisson regression models were adjusted to identify potential factors associated with EWLB.The initial model included all Level 1 variables (Figure 2).Those statistically significant (p < 0.05) in this first model contributed to the adjustment of subsequent levels of analysis.The second model included all Level 2 variables and those that were significant (p < 0.05) in the initial model.Then, a new model was adjusted with the variables of Level 2 that were significant (p < 0.05), in addition to the variables selected in the initial model and all variables of Level 3.After checking the statistical significance (p < 0.05) for the variables of Level 3, the final model was ad- justed.Also, the variables of economic indicator and race/skin color were maintained in all steps of the analysis, although they were not statistically significant (p < 0.05).Statistical analyses were performed using the software R (R Version 4.1.2-"Bird Hippie") 26 .

Results
Of the 2,439 adolescents, 51.6% were female, 54.4% were aged 15 to 19 years, and 70.0%belonged to the public school system (Table 1).The prevalence of EWLB was 12.1%, with 8.5% reporting self-induced vomiting or use of laxatives and 7.1% using medicine or other products to lose or maintain weight.Other characteristics of the sample and prevalence of EWLB are described in Table 1.
The results of the simple Poisson model indicated that among the factors of the distal level, being female (PR: 0.60; CI: 0.46-0.79)and aged 15 to 19 years (PR: 0.56; CI: 0.41-0.77)were negatively associated with EWLB.There was also a negative association between mothers having completed high school or college and adolescents studying in private schools (Table 2).
Among the factors of the intermediate level, the prevalence of EWLB is about 2.09 (CI: 1.56-2.82)times higher in adolescents who suffered bullying related to body appearance and for other reasons compared to their peers who did not suffer psychological violence.More expressive and positively significant estimates were also observed when assessing the sexual abuse suffered (PR: 3.08; CI: 2.22-4.26)by the adolescent and the outcome of interest.There was no significant association between the other variables at this level and the occurrence of EWLB (Table 2).
In the final model, after adjustment, it was observed that female (PR = 0.65; CI = 0.52-0.81),aged 15 to 19 years (PR = 0.56; CI = 0.45-0.70),private school (PR = 0.62; CI = 0.46-0.84),while representative of the distal level decreased the prevalence of adoption of EWLB.Regarding the intermediate level, it was found that experiencing violence as being a victim of bullying related to body appearance and for other reasons (PR = 1.62;CI = 1.19-2.20)and being forced to have sexual intercourse (PR = 2.65; CI = 1.90-3.69)showed positive and significant associations with the occurrence of EWLB.Regarding the factors of the proximal level, having insomnia (PR =

Discussion
This study evaluated the prevalence and factors associated with the adoption of EWLB among adolescents from northeastern Brazil.The estimated prevalence of EWLB among adolescents (12.14%) in this study was higher than that ob-served throughout Brazil in the three PeNSE surveys, 6.4% in 2009, 9.0% in 2012, and 10.1% in 2015 4 or among Spanish (3.26%) and North American adolescents (5.03%) 27 , but it was lower than the estimates for adolescents from southeastern Brazil (31.9%) 28 .The variability of the prevalence found may have been influenced by the sociocultural aspects inherent to each location and methodological differences adopted by the studies to assess the outcome.These results of the estimated prevalence of EWLB related to body and weight should raise concerns, especially in adolescence, considered an important start- Among the variables explored in the final model, the results indicated that studying in private school, being a victim of bullying related to body appearance and for other reasons, being forced to have sexual intercourse, having insomnia, and self-perception of excess weight were associated with EWLB among the adolescents assessed.
Considering the final model adjusted by the economic indicator, self-reported race/skin color, and all the variables of the model, this study showed that the school administration (private) reduced the prevalence of EWLB.Similar results 28,29 was presented, in which the presence of these behaviors was higher among adolescents enrolled in public schools.These results may be related to the socioeconomic conditions of these students, which is a hypothesis to be tested in future studies, since studying in public school is a proxy for worse economic conditions and previously it was supposed that these behaviors were more frequent in higher income populations.However, it has often been observed that people from lower social classes are more vulnerable to pressures to adapt to a socially established body standard 30 .
The results also show that adolescents who experienced appearance-related bullying were more likely to adopt EWLB and this is in line with previous findings among youngsters 31,32 .
Mostly, adolescents who have been victims of appearance-related bullying, feel more uncomfortable with teasing, which contributes to body dissatisfaction and EWLB.This result is particularly relevant, and this situation should be monitored, as other studies have also shown an association between bullying and problems related to emotional distress, such as loneliness, anxiety, insomnia, sadness, depression, post-traumatic stress disorder, and suicidal thoughts 33 , showing the extent of this problem and its serious immediate and long-term health consequences.
Furthermore, the results reveal significant associations between experiences of sexual abuse and the outcome of interest.There are numerous  This mechanism is related to body dissatisfaction, which acts as a mediator throughout the process since the victims of this violence have a sense of discomfort and dissatisfaction with their self-image seeking body changes through inadequate eating behavior [34][35][36] .Early interventions are essential so that a specific diagnostic proposal is built to better serve populations at risk and minimize the damaging effects of this violence.
Regarding the psychosocial indicators, having insomnia increased the prevalence of the outcome of interest, aligning with the result presented in other studies, in which adolescents with sleep duration of at least 8 hours per day act as a protective factor against EWLB for both sexes 37 .It is believed that this association can be explained by dissatisfaction with body image, because poor sleep quality, in addition to causing hormonal changes, increases physical inactivity and the time available for food intake mainly of high energy density foods and low nutritional value, and even prolonged exposure to television, which may lead adolescents to present distorted body image, thus favoring the adoption of EWLB 38 .
Consistent with previous studies 5,6,7,18,39 self-perception of excess weight was significantly associated with adolescent involvement with EWLB.In contemporary society, adolescents with a greater possibility of involvement in strategies for weight loss probably have a high level of body image dissatisfaction, such as stress with appearance.The hypothesis that body image perception may have a greater impact on EWLB that favors weight loss is reinforced by considering, for example, that normal-weight adolescents who overestimate their nutritional status are more likely to adhere to fasting practices, self-induced vomiting, or use of laxatives and over-the-counter drugs when compared with those with normal weight and self-assessment in accordance with their real weight 28 .
This study, for analysis of PeNSE data, is a pioneer in Brazil in evaluating the factors associated with EWLB with a hierarchical approach and a large region of the country.The results suggest potential implications for public health, yet this study has some limitations that should be considered for the interpretation of the results.First, as PeNSE is cross-sectional research, there is the possibility of causality bias because the relationship between outcome and exposure variables was estimated in a single moment.The temporal sequence of events was not considered, and the cause and effect could not be identified.Second, the sample studied was from a single Brazilian region and, therefore, is not representative of all adolescents in the country.However, the sample size was adequate (with adequate power and effect size) for the analyses and attested conclusions.Third, PeNSE 22 presents: the use of self-report for the research of extreme weight loss behaviors with the possibility of underreporting, because the adolescent can be inhibited, not reporting accurately, and may be subject to classification errors.
Despite the limitations of this study, it can be concluded that among the adolescents assessed there is a worrying prevalence of EWLB, and that studying in private school, being a victim of bullying related to body appearance and for other reasons, being forced to have sexual intercourse, having insomnia and self-perception of weight excess are important factors associated with these disorderly behaviors.The impact of these results on public and mental health involves the fact that early identification of EWLB can help prevent more serious health problems that burden the public health system.
In this perspective, these results, as well as all others from PeNSE, should be widely disseminated to subsidize actions to be implemented, expanded, and/ or reformulated.In addition, these findings can help to inform policies and programs for the prevention of EWLB and protection of young people, to monitor the magnitude and temporal trends of these factors, as well as to evaluate the actions addressed to this population group.The EWLB affects the physical and mental health of adolescents and therefore requires an integrated approach to coping with this problem that involves educators, health professionals, parents, and the community in general.These initiatives should focus on health promotion and protection, comprehensiveness, and intersectorality.

Figure 2 .
Figure 2. Hierarchical conceptual model of factors associated with EWLB among adolescents from northeastern Brazil.Brazil, 2015.
23 )t 1. Description of independent/secondary variables according to their characteristics.Brazil, 2015.from the data of ownership of household items, vehicles, number of bathrooms with shower, Internet access in the household, and the presence of a maid.The score was categorized into tertiles, according to the distribution observed in the sample studied by Levy et al.23 ) exposure to violenceVictim of bullying Sporadic behavior (never, rarely, sometimes) (reference); Yes (often, always) Forced to have sex No (reference); Yes lifestyle Physical activity Physically active (300 min/week) (reference); Insufficiently active (< 300 min/week) Sedentary behavior Non-sedentary (2 hours/day sitting) (reference); Sedentary (> 2 hours/day sitting) Source: Authors.

table 1 .
Percentage distribution of the characteristics of the sample and prevalence of EWLB among adolescents from northeastern Brazil.Brazil, 2015.

table 1 .
Percentage distribution of the characteristics of the sample and prevalence of EWLB among adolescents from northeastern Brazil.Brazil, 2015.

table 2 .
Crude prevalence ratios and final hierarchical model of factors associated with EWLB among adolescents of northeastern Brazil.Brazil, 2015.

table 2 .
2imple Poisson model.2Adjustedmultiple models.PR = prevalence ratio; 95%CI = 95% confidence interval.* Final model adjusted by the variables economic indicator and race/skin color and all statistically significant variables, Akaike -AIC (1654,69).Crude prevalence ratios and final hierarchical model of factors associated with EWLB among adolescents of northeastern Brazil.Brazil, 2015. 1