Score of ultra-processed food consumption and its association with sociodemographic factors in the Brazilian National Health Survey , 2019

This is a cross-sectional population-based study that describes the score of ultra-processed food consumption, applied in the Brazilian National Health Survey performed in 2019, and its association with sociodemographic factors in Brazilian adults (18 years or older). The score of ultra-processed food consumption was calculated by adding up the positive answers about the consumption on the previous day of 10 subgroups of ultra-processed foods frequently consumed in Brazil. The distribution of the score in the population was presented as a count. Poisson regression models were used to evaluate the crude and adjusted associations of scores equal to or higher than five subgroups of ultra-processed foods with urban/rural area, geographic region, sex, age group, schooling level, and wealth index. About 15% of the Brazilian adults reached scores equal to or higher than five. After adjustment for confounders, the prevalence of consuming five or more subgroups of ultraprocessed foods decreased linearly with age, increased linearly with wealth quintiles and it was higher in urban areas, in the Southeast and South regions (compared to the others) and in men. Public policies that reduce the consumption of ultra-processed foods with emphasis on strata of the population at the greatest risk are essential and monitoring the score of ultra-processed food consumption across studies and populations will be important to assess the success of these policies. foods from a list of 10 items, similar to the one applied by the Vigitel study. In both surveys, the questionnaires were built to include the subgroups of ultra-processed foods with the greatest participation in the daily energy intake estimated by the Brazilian Dietary Survey performed in the POF 2008-2009 conducted by the IBGE. In the PNS 2019, respondents answered “yes” or “no” to the following questions: “Yesterday, did you drink or eat: (1) Soft drink?; (2) Fruit juice drink in can or box or prepared from a powdered mix?; (3) Chocolate powder drink or flavored yogurt?; (4) Packaged salty snacks or crackers?; (5) Sandwich cookies or sweet biscuits or packaged cake?; Ice cream, chocolate, gelatin, flan or other industrialized dessert?; Sausage, mortadella or hot or hamburger Margarine, mayonnaise, ketchup or other industrialized sauces?; Instant noodles, instant powdered frozen lasagna or frozen ready-to-eat turning items most drink flavored yogurt, margarine mayonnaise, ketchup or industrialized sauces. The score of ultra-processed food consumption of each participant was calculated by adding up the positive answers given to these questions regarding consumption on the day prior to the interview, which can vary from 0 to 10 points.


Score of ultra-processed food consumption
The 2019 edition of the PNS included a module for assessing the consumption of ultra-processed foods from a list of 10 items, similar to the one applied by the Vigitel study. In both surveys, the questionnaires were built to include the subgroups of ultra-processed foods with the greatest participation in the daily energy intake estimated by the Brazilian Dietary Survey performed in the POF 2008-2009 conducted by the IBGE. In the PNS 2019, respondents answered "yes" or "no" to the following questions: "Yesterday, did you drink or eat: (1) Soft drink?; (2) Fruit juice drink in can or box or prepared from a powdered mix?; (3) Chocolate powder drink or flavored yogurt?; (4) Packaged salty snacks or crackers?; (5) Sandwich cookies or sweet biscuits or packaged cake?; (6) Ice cream, chocolate, gelatin, flan or other industrialized dessert?; (7) Sausage, mortadella or ham?; (8) Loaf, hot dog or hamburger bun?; (9) Margarine, mayonnaise, ketchup or other industrialized sauces?; (10) Instant noodles, instant powdered soup, frozen lasagna or other frozen ready-to-eat meal?" 21 . The PNS replicated the list of 13 subgroups proposed for Vigitel but turning six items (three most similar pairs) into three, without excluding any item. The consumption of fruit juice drink in can or box was investigated with fruit juice drink prepared from powdered mix, chocolate powder drink with flavored yogurt, and margarine with mayonnaise, ketchup or other industrialized sauces.
The score of ultra-processed food consumption of each participant was calculated by adding up the positive answers given to these questions regarding consumption on the day prior to the interview, which can vary from 0 to 10 points.

Sociodemographic factors
The sociodemographic variables included in this study were: area of residence (urban and rural), geographic region (North, Northeast, Central-West, Southeast, and South), sex (male and female), age (18-29, 30-39, 40-49, 50-59, and 60 years and over), schooling (none, incomplete elementary school, complete elementary school, complete high school, and complete higher education), and wealth index.  The last one was built based on principal component analysis considering data about the number of  rooms and bathrooms in the household, sewage type, assets (color television, refrigerator, washing  machine, landline, mobile phone, microwave, computer, motorcycle, Internet access, and number  of cars), and existence of monthly maid/domestic employee. The wealth index was categorized into quintiles.

Data analysis
Initially, the sample was described according to the sociodemographic variables. The frequencies (%) of consumption of each selected subgroup of ultra-processed foods on the previous day were described, with their respective 95% confidence intervals (95%CI). The distribution of the score of ultra-processed food consumption was used as a count. To assess the association between the score of ultra-processed food consumption and sociodemographic variables, the score was dichotomized into the upper fifth (approximate) of the distribution, corresponding to scores greater than or equal to five. Poisson regression models were used to assess the association between the sociodemographic factors and the outcome, estimating crude and adjusted prevalence ratios (PR) and their respective 95%CI. In the multiple regression models, the sociodemographic variables were adjusted for each other.
The microdata was obtained from the IBGE website for PNS and all analyses were performed using the Stata statistical package, version 16.1 (https://www.stata.com) using the svy command, which computes standard errors by using the linearized variance estimator, and the expansion factors or sample weights.

Results
The analytical sample of this study included 88,531 Brazilian adults. Participants were more likely to dwell in urban areas and in the Southeast Region, female and with complete high school. The mean age was 44.9 years old (Table 1). Figure 1 describes the frequency of consumption for each subgroup of ultra-processed foods on the day prior to the interview. Almost half of the sample reported having consumed margarine, mayonnaise, ketchup, or other industrialized sauces (45.9%), and about one-third of the participants reported having consumed soft drink (32.9%) and loaf, hot dog or hamburger bun (31%). Between 20% and 30% of the sample reported having consumed sausage, mortadella, or ham (27.2%), fruit juice drink in can or box or prepared from a powdered mix (24.3%), sandwich cookies or sweet biscuits or packaged cake (23.8%) and packaged salty snacks or crackers (23.3%). Less than 20% of the individuals reported consuming food from each of the three remaining subgroups on the day prior to the interview (ice cream, chocolate, gelatin, flan, or other industrialized dessert; chocolate powder drink or flavored yogurt; instant noodles, instant powdered soup, frozen lasagna, or other frozen ready-to-eat meal). Figure 2 shows the distribution of the score of ultra-processed food consumption, which is equivalent to the number of subgroups consumed on the day prior to the interview. Scores ranged from 0 to 10, but one (19.9%), two (21%), three (17.6%) and four (12.5%) were the most common; 14.8% of participants reached null scores and 14.3% achieved scores equal to or higher than five. In average, the sample reported having consumed 2.49 (95%CI: 2.47-2.52) subgroups of ultra-processed foods on the day prior to the interview (data not shown). Table 2 presents the crude and adjusted relationship between sociodemographic variables and scores for the consumption of ultra-processed foods equal to or higher than five. After adjustment for confounders, individuals living in urban areas presented a prevalence of scores for the consumption of ultra-processed foods ≥ 5 66% higher than individuals living in rural areas. This prevalence was lower in the Northeast, increasing about 30% in the North and Central-West, 66% in the Southeast and 108% in the South. Men presented prevalence of the indicator 17% higher than women (PR = 1.17; 95%CI: 1.10-1.24). Prevalence of scores equal to or higher than five decreased linearly with age and increased linearly with wealth index quintiles. Schooling was no longer associated with scores of ultra-processed food consumption equal to or higher than five after adjustment for confounders (

Discussion
The inclusion in the PNS 2019 of questions about the intake of commonly consumed subgroups of ultra-processed foods has allowed the calculation of a score of ultra-processed food consumption, as the number of subgroups consumed on the previous day, varying from 0 to 10. Margarine or industrialized sauces, soft drink, packaged bread, sausages, industrialized fruit juice, and salty or sweet cookies were the subgroups most frequently consumed. About 15% of the Brazilian adults reached scores equal to or higher than five subgroups of ultra-processed foods. After adjustment for confounders, the prevalence of five or more subgroups of ultra-processed foods decreased linearly with age and increased linearly with wealth index quintiles and was lower in rural areas, in the Northeast Region, and among women.

Figure 2
Distribution of the score of ultra-processed food consumption. Brazilian adult population (18 years Table 2 Crude and adjusted association between sociodemographic characteristics and scores for the consumption of ultra-processed foods equal to or higher than five. Brazilian adult population (18 years  The three subgroups most frequently consumed in the PNS 2019 were the same indicated by the Vigitel survey in the same year (margarine, soft drink, and packaged bread). The distribution of the score of ultra-processed food consumption (as a count) was also similar in the two surveys. However, the prevalence of consuming five or more subgroups of ultra-processed foods in the day prior to the interview was higher in the Vigitel survey (18.2, 95%CI: 17.4-19.0 vs. 14.3, 95%CI: 13.8-14.8) 19 . This variation could be explained by the different number of subgroups of ultra-processed foods included in the questionnaire of each survey, 13 for the Vigitel and 10 for PNS 2019 (six of the 13 subgroups in the Vigitel questionnaire were turned into three in the PNS questionnaire). However, the main reason for this discrepancy could be that the Vigitel sample is representative only of the population living in Cad. Saúde Pública 2022; 38 Sup 1:e00119421 the capitals, where the consumption of ultra-processed foods is higher compared to non-capitals, and the PNS is representative of all national territory, including urban and rural areas. This difference in the prevalence does not make unfeasible the main objective of both Vigitel and PNS tools, which is the monitoring the consumption of ultra-processed foods in the population.
As in our study, the prevalence of consumption of five or more subgroups of ultra-processed foods (out of 13 subgroups) in the Vigitel survey linearly decreased with age and was lower among women 19 . On the other hand, while in this study no association was observed with schooling level and a positive association was observed with quintiles of wealth, in the Vigitel survey schooling level was inversely associated with ultra-processed food consumption. This result was observed possibly due to the lack of adjustment for wealth index or another socioeconomic variable. In the Vigitel survey, schooling was used as a proxy or indicator of socioeconomic level. In fact, schooling and wealth index could reflect different aspects and other studies are necessary to confirm the relationship between schooling and consumption of ultra-processed foods, when adjusting for wealth index, for example.
Nationally representative cross-sectional studies performed in several countries have unanimously observed an inverse association between the dietary contribution of ultra-processed foods (as % of total energy intake) and age while the association with sex and income/education varied across countries 6,23,24,25,26,27 . While in Canada and the United Kingdom the consumption of ultra-processed food was higher among men, in Chile consumption was higher among women and in the United States, Mexico and Colombia no differences were observed across sexes. Similar to what we observed in this study, Chile, Mexico, and Colombia also described a positive association between ultraprocessed food consumption and income, while Canada and the United States observed an inverse association. These inconsistent results were found possibly due to differences in socioeconomic variables and other variables considered in the models used in these studies or even to differences in the mechanisms that, in each country, mediate sex and socioeconomic level to the consumption of ultra-processed foods.
This study has some limitations. As most methods of evaluating food consumption, the information from this questionnaire was self-reported, and a possibility of recall bias was identified. Also, the score of ultra-processed food consumption represents a proxy of the percentage of energy contribution from ultra-processed foods and does not refer to complete data on the consumption of these foods. However, we emphasize that a validation study conducted in 2018 with a convenience sample in the city of São Paulo (Brazil), showed good agreement (kappa coefficient: 0.72) between fifths of the score for consumption of ultra-processed foods (measured by a questionnaire identical to that used in Vigitel) and fifths of the contribution of ultra-processed foods to the total daily energy intake (measured by 24-hour food record), both calculated based on the previous day food consumption 28 . This study evaluated the performance of the 13-item questionnaire and not the PNS adapted list of 10 subgroups; however, we do not believe that the agreement would not be similar. A second validation study, performed as part of the NutriNet Brasil Cohort Study (NutriNet Brasil, https://nutrinetbrasil.fsp. usp.br/), confirm the good agreement between the score of ultra-processed food consumption (with 23 subgroups, adapted to be self-filled using the Internet) and the contribution of ultra-processed foods to total energy intake 29 .
This study follows other studies that have been applying screeners to monitor the consumption of ultra-processed foods in a quick and practical manner. Besides the Vigitel study and the NutriNet Brasil, a set of questions about the consumption of ultra-processed foods was also applied in the Brazilian National Survey of School Health (PeNSE), 2019/2020 edition 30 , then estimates will also include the adolescent school population. Also, the NOVA Screener for the consumption of ultra-processed foods, the tool developed for the NutriNet Brasil, is currently under adaptation to be used in Ecuador, India, and Senegal, which has been encouraging other countries to incorporate short questionnaires within national surveillance and monitoring and evaluation systems to broaden the assessment of ultra-processed foods in the populations.
Considering the evidence that demonstrates the harmful effect of ultra-processed food consumption on diet quality and on the risk of several chronic non-communicable diseases, public policies that reduce the consumption of those ultra-processed foods and the emphasis on strata of the population at greatest risk are essential. Monitoring the score of ultra-processed food consumption across studies and populations will be important to assess the success of these policies.