Estimating the share of ultra-processed foods in Brazilian municipalities.

OBJECTIVE

To estimate the caloric share of ultra-processed foods (% UPF) in the 5,570 Brazilian municipalities.

METHODS

The estimation of % UPF in municipalities was performed using a statistical prediction model based on data from 46,164 individuals aged over >10 years who participated in the Household Budget Survey (HBS 2017-2018). Multiple linear regression was used to estimate the average % UPF (measured through two 24-hour dietary recalls) based on predictor variables (sex, age, income, education, race/color, urbanity, federative units, and geographic location). The model's adequacy was assessed through residual analysis and by comparing predicted values with those directly measured in POF 2017-2018 using Lin's concordance correlation coefficient (CCC). The linear coefficients obtained from the multiple linear regression model were applied to the sociodemographic data from the 2010 Census (measured similarly to POF) to estimate the % UPF for each municipality.

RESULTS

The statistical model proved adequate, showing normally distributed residuals and a CCC of 0.87, indicating almost perfect agreement. There was heterogeneity in the distribution of % UPF estimates, ranging from 5.75% in Aroeiras do Itaim (PI) to 30.5% in Florianópolis (SC). % UPF estimates were higher (>20%) in municipalities from the South region and the state of São Paulo. Capitals had higher estimates of caloric contribution from ultra-processed foods compared to other municipalities in their states.

CONCLUSIONS

The predictive model revealed differences in % UPF among Brazilian municipalities. The generated estimates can contribute to monitoring ultra-processed food consumption at the municipal level and support the development of public policies focused on promoting healthy eating.