The current study was a cross-sectional in Olinda in the state of Pernambuco, Brazil. Olinda had a population of 390,000 residents and 33 state public high schools in 2018.
The data were collected from February to June 2018 in 87% (n=27) of all state public high schools with daytime classes (31 schools), which agreed to take part in the study. The two schools offering only night classes were not invited. The response rate of all invited students enrolled in the 27 schools was 37.2% (2,700). Almost all the classrooms (91%) in the involved schools participated in our data collection.
The Informed Consent Form (ICF) was signed by the parents/guardians of all participants under 18-year-old while children also provided written assent. Participants over 18-years of age consented by themselves. This research was approved by the Ethics Committee of the University of Pernambuco (N. 2.361.780).
Four trained and calibrated examiners measured body weight and height with strong inter-rater reliability for height (kappa 0.99). A portable stadiometer (Sanny®) and a self-zeroing digital scale (CAMRY) with a maximum capacity of 150 kg were used to measure height and body weight, respectively. The participants were weighed barefoot wearing socks and light clothing.
The sociodemographic and health behavior modules were adapted from the U.S. Youth Risk Behavior Survey (YRBS) questionnaire. The student’s sociodemographic situation was assessed by their age, sex, and whether they received a family allowance. The family allowance is given in Brazil to poor families and extremely poor families with per-capita monthly incomes below $170 BRL (~$45 USD) or $85 BRL (~$23 USD), respectively.
The survey also asked whether the students had experienced victimization, whether they were actively trying to lose weight, as well as self-perception of happiness (In general, how do you feel in your actual life?) on a 5-point Likert scale.
Victimization was based on at least one positive answer for these three questions: (1. During the past 12 months, have you ever been bullied on school property?; 2. During the past 12 months, have you ever been electronically bullied? (Count being bullied through texting, Instagram, Facebook, WhatsApp or other social media.); 3. During the past 12 months, how many times has someone threatened or injured you with a weapon such as a gun, knife, or club on school property?). Trying to lose weight was based in the question “Which of the following are you trying to do about your weight?”, with responses categorized as: trying to gain weight, stay the same weight or I am not trying to do anything about my weight, and trying to lose weight. Weight categories were based on cut-off z scores recommended by the WHO: underweight (< -2), normal weight (≥-2 and <+1), overweight (≥+1 and <+2), obese (≥+2).
The LGB variable was derived from adolescents who reported having sexual intercourse with a partner of the same sex or with partners of both sexes.
Our contextual variable of interest was LGB visibility in school, defined by the prevalence of LGB students in each school. The variable was dichotomized into low LGB visibility (≤ 3.01%) vs. high LGB visibility (>3.01%), where the cutoff was made at the 25th percentile.
The outcome variable was unhealthy weight-control behaviors created by combining the following three behaviors: fasting, purging, and using diet pills.
Data analysis was carried out using STATA/IC version 15.1. We performed a sequence of multilevel logistic regressions (with random intercept) for the outcome of unhealthy weight-control behaviors (fasting, purging, or using diet pills). The empty model included only the dependent variable (Model 1). The second model included only the contextual variable (Model 2) and the third model included only individual variables. In the final model, we combined the individual and contextual variables (Model 4). We tested for potential mediation of the association between LGB visibility and unhealthy weight control behaviors by experiences of victimization and self-perception of happiness (models 4a and 4b). The cross-level interaction between the contextual variable (LGB visibility) and the individual variables (LGB and gender) were also tested. Some individual variables such as obesity, trying to lose weight, victimization, and self-perception of unhappiness were controlled for attenuates the association between LGB & unhealthy weight-control behaviors.
We used the proportional change in variance (PCV) to assess changes in the random intercept term. In other words, the PCV represents the proportional change in the area level variance compared to the empty model (Model 1). We also used the Median Odds Ratio (MOR) to quantify the variation between clusters in each sequence of models. When the MOR equals 1.0, it means that there is no heterogeneity between the contexts analyzed. We also tested the fitted multilevel logistic regression model using the Deviance (2 Res log-likelihood).