We considered all live births in 1993 (N = 5,265) in the city of Pelotas, Brazil, to be eligible for a birth cohort study, and included 5,249. There were 81 multiple births (1.5%). We visited subsamples of this cohort at 1, 3 and 6 months of life and at 1, 4, 6, and 9 years of age. Detailed information about the study is available elsewhere [8,9]
. In 2004–2005 and 2008–2009, when the cohort participants were on average 11 and 15 years of age, respectively, we carried out follow-up visits aiming to find all participants of the cohort.
In both follow-ups, we obtained all information and measurements through interviews performed in households and conducted by trained interviewers, using the same methodology at each time point.
We performed adolescents' nutritional status assessment using body mass index (BMI) for age in z
-scores. We calculated BMI based on two measures of weight and height and used the mean value in the analysis. We measured weight using a digital scale with precision of 100 g (SECA, Birmingham, UK). We measured height using an aluminum stadiometer with precision of 1 mm, made for the study. Interviewers underwent standardization testing 
before beginning fieldwork and every 2 months afterward to determine repeatability and validity of measurements. We adopted the World Health Organization standards for children and adolescents aged 5–19 years 
. Adolescents with BMI index for age >2 z
-score were classified as obese. We created four situations based on the combination of the presence of obesity at 11 and 15 years of age, categorized as: non-obese at age 11 and 15 years; non-obese at age 11 years and obese at age 15 years; obese at age 11 years and non-obese at age 15 years; and obese at age 11 and 15 years.
We collected independent variables during the 11-year-old follow-up, except for the variable of sex, which we collected in the perinatal study. Other variables used were adolescent's perception of own weight, body dissatisfaction, and practice of weight loss dieting. We determined socioeconomic position (SEP) employing an index of assets constructed using principal component analysis, based on a list of 18 socioeconomic indicators. The sum of each individual variable was later transformed into a continuous variable and categorized into tertiles.
We determined body dissatisfaction by comparing desired body image with perceived body image. We assessed body image using the rating scale from Stunkard et al. 
. This scale contains nine male and female figures ordered from the smallest to the largest silhouette. First, adolescents chose the figure they believed looked most like them; afterward, they elected the silhouette they wished to have. If the answer to both questions was the same silhouette, the adolescent was considered to be satisfied with his body image. Otherwise, the youth was considered to be dissatisfied with his body image when he wished to have a smaller or larger silhouette. Perception of own weight was categorized as very fat, fat, normal, thin, or very thin. We later grouped these options into three categories: fat, normal, and thin. We obtained the variable of weight loss dieting through adolescents' report of having gone on some type of diet or dietary restriction to lose weight in the past 12 months.
We analyzed data using Stata software, version 12.0 (Stata Corp., College Station, TX). The descriptive analysis included absolute numbers and percentages or means for all study variables. We performed multivariate analysis using Poisson regression with robust variance 
and present the effects as prevalence ratios. We analyzed changes in BMI category in the 11- to 15-year period according to the independent variables adjusted for SEP. All analyses were stratified by sex. p
< .05 was assumed for two-tailed tests.
The Ethics and Research Committee of the Medicine College at the Federal University of Pelotas approved the study. Parents or guardians signed a written informed consent form authorizing youths to participate in the study.