Statistical methods

For each country, the prevalence rates of obesity were calculated by year, sex, and level of education, occupational class, and age-standardized to the European Standard Population using the direct standardization method [34]. For visual between-country comparison total prevalence rates and prevalence rates by socioeconomic group indicator were plotted as over time line charts.

Inequalities were measured by means of absolute prevalence rate differences (RD) and relative prevalence rate ratios (RR) of low versus high level of socioeconomic position. A bootstrap procedure with 1000 iterations was used to calculate 95% confidence intervals. Survey weights were available in some countries or years, but not in all. Thus, unweighted results are reported in the results section. A previous study based on the same data sources and with self-assessed health as the outcome found essentially similar results if weighted or unweigthed data were used [35].

To study the trends over time in each country and in the ensemble of countries as a whole, we employed meta-regression with random effects models, using the DerSimonian and Laird method [36]. The year of data collection was used as the only independent variable in the models and the total prevalence of obesity, the prevalence of obesity by level of socioeconomic position; the absolute (RD) and relative inequalities (RR) by socioeconomic position were included as the dependent variables. The country-specific regression parameters and their 95% confidence intervals were displayed as forest plots and meta-analyses were performed to calculate overall random effect estimates for all countries. Data were analyzed for males and females separately. All analyses were performed using Stata/SE 13.1 (StataCorp, Texas, US).