Data from three nationally representative surveys were analyzed: the Household Budget Survey–Pesquisa de Orçamento Familiar (POF) conducted in 2002/2003 and 2008/2009, and the National Health Survey–Pesquisa Nacional de Saúde (PNS), undertaken in 2013. The Brazilian Institute of Geography and Statistics (IBGE) was responsible for the surveys. The raw data and questionnaires for data collection are publicly available [13,14,15].
The sampling procedures included multi-stage stratified clusters. In the first stage of the POFs, primary sampling units (PSUs) were selected by systematic sampling proportionally to the number of households. For the second stage, households were selected by simple random sampling without replacement. Anthropometric measurements were taken from all individuals present in the selected households at the time of the interview.
The PNS sample was a subsample of the Master Sample of the Integrated Household Surveys System of the Brazilian Institute of Geography and Statistics. Cluster sampling was performed in three stages: the PSUs comprised sectors of the census and were obtained by simple random sampling among those previously selected for the Master Sample. It maintained the PSU stratification used in the Master Sample. The secondary unit included 10 to 14 households selected from each PSU, and the third unit included one person aged 18 years or more from each household who responded to the individual component of the questionnaire distributed by the PNS.
The initial databases included 182,333, 190,159, and 205,546 individuals from the 2002/2003, 2008/2009, and 2013 surveys, respectively. This study analyzed data from adults between 20 and 59 years of age of both sexes. Pregnant women were excluded, resulting in a sample comprising 234,791 adults: 89,651 (2002/2003), 100,956 (2008/2009) and 44,184 (2013). Weight and height were measured similarly in all three surveys using portable digital scales and wall-mounted stadiometers. Trained interviewers performed all anthropometric measurements and individuals were asked to remove their shoes before measurements. The classification of nutritional status was based on body mass index (BMI). Overweight and obesity were defined as BMI between 25.0–29.9 kg/m2 and greater than 30 kg/m2, respectively [16].
Data regarding education were collected in the POFs as years of education and as levels of education (in seven categories) in the PNS. In order to allow comparability between the surveys, harmonization was carried out following the International Standard Classification of Education: pre-primary (0 to 7 years of study or incomplete primary school), primary (8 to 10 years of study, complete primary school or incomplete secondary school), secondary (11 to 14 years of study, complete secondary school or incomplete tertiary school) and tertiary (greater than 15 years of study, complete tertiary school or more) [17].
The prevalence (percentage) of overweight and obesity were estimated by sex, age group (20–39 and 40–59 years), and educational level (pre-primary, primary, secondary and tertiary). Logistic regression was used to evaluate trends in the prevalence of obesity and overweight. To account for the gaps between the surveys, we encoded the 2002/2003 survey as 2, the 2008/2009 survey as 8, and the 2013 survey as 13 and used the 2002/2003 survey as the reference. All analyses were stratified by sex. Interactions of age group, educational level, and survey were assessed in order to test the effect modification in the prevalence of obesity trend by the educational level. Also, a logistic regression model was used to test the possibility of a cohort effect, including a cohort indicator variable with 4 categories (1950, 1960, 1970, and 1980). The variable outcome was obesity (yes/no) and this model was adjusted by age as a continuous variable. All analyses accounted for the sample weights and for the effect of the sample design, using SAS software version 9.4 (SAS Institute Inc., Cary, NC).