Secondary data analysis was conducted with the publicly available and nationally representative dataset the Household Budget Survey (HBS) of Brazil, for which data collection took place between 2017 and 2018. The Brazilian HBS is coordinated by the Brazilian Institute of Geography and Statistics on a ten-year interval, and its main purpose is to survey household budget structures (e.g., products acquisitions, services requirements, and income), nutritional status, and living conditions of families in Brazil [30].

As part of 2017–2018 HBS, a subsample of 46,164 individuals 10-years or older, living in one of the 20,112 randomly selected households (out of the 57,920 households from the original sample), were invited to answer two non-consecutive days 24-h recalls (24HR), carried out by trained research agents in the household, as well as information on current supplement use, and diet modification. The individual dietary assessment phase of the HBS, also called the Brazilian National Dietary Survey (BNDS), preserves the nationwide representativeness of the original sample, respecting the two-stage cluster sampling (census sectors and households), and geographical and sociodemographic stratifications, based on the 2010 Demographic Census [30].

For the present study, all individuals that participated in the BNDS 2017–2018 were included in the analysis. Pregnant and lactating women were excluded, and all the analyses were stratified for self-declared sex (men/women), age ranges (adolescents, 10 to 19 years/adults, 20 to 59 years/elderly, ≥ 60 years), and geographical regions (North/Northeast/Southeast/South/Midwest Brazil, Supplemental Figure S 1).

Information on urban or rural area of residency, sex, age, self-declared skin color, educational status, and household per capita income were collected at the first visit interview and were used as covariates in the present study analysis.

The BDNS publicly available data does not identify participants or households other than rural/urban locality and the geographical code of the Brazilian State. Therefore, the usage of secondary data by the present study waives approval from an Ethics Committee and is in accordance to the Helsinki Declaration and the Ethical Guidelines from the Council for International Organizations of Medical Sciences.

In the BNDS assessment, dietary data were collected through two non-consecutive days 24HR, carried out by trained research agents, following the Automated Multiple Pass Method [31], guided by a tablet app developed specifically for the structured interview [30, 32]. Most of the sample answered both 24HR (n = 38,854; 99.9%) and those who did not answer the second measurement were kept in the dataset with a single 24HR information.

Participants were asked to list all the foods and beverages consumed the day before the interview without interruptions from the research agents. After that, the participants were asked about further details on culinary techniques, amount, added items (e.g.: olive oil, butter, ketchup, sugar, salt, sweeteners, honey, sauces, grated cheese, milk cream), and occasion and place when and where each food item was consumed. At the end of the interview, participants were also inquired about the usage of nutritional supplements and the existence of any special condition that could restrict their dietary intake (i.e.: actively trying to lose weight; high blood pressure, high cholesterol, diabetes, or heart disease treatment; and others) [30]. The nutrient content of foods was determined using the Brazilian Food Composition Table (TBCA-USP), version 7.0 [33].

To conduct the analysis, five diet metrics were calculated using the first 24HR data: (1) GDQS; its two sub-metrics, (2) GDQS + and (3) GDQS -; (4) the MDD-W; and (5) the percentage of dietary caloric contribution from UPF.

To score all five dietary metrics, 24HR mixed dishes were disaggregated into ingredients using standard recipes for the Brazilian cuisine, and yield and nutrient retention factors were applied from standard references [34,35,36,37,38], further explained elsewhere [39]. Food classifications were double-checked by two researchers.

The GDQS scores the dietary daily intake of 25 food groups, in grams, according to their contribution to increase or decrease the overall quality of individual diets (ranging from 0 to 49). The 25 food groups of GDQS can be separated into the so-called “healthy foods” – comprising 16 food groups which intake increase the overall diet quality score (dark-green leafy vegetables, deep-orange vegetables, deep-orange fruits, deep-orange tubers, cruciferous vegetables, other vegetables, citrus fruits, other fruits, fish and shellfish, poultry and game meat, legumes, nuts and seeds, low-fat dairy, eggs, whole grains, and liquid oils); the “unhealthy foods” – comprising seven food groups which intake decrease the overall diet quality score (white roots and tubers, processed meat, refined grains and baked goods, sugar-sweetened beverages, juice, sweets and ice creams, and purchased deep-fried foods); and two food groups classified as “unhealthy in excessive amounts”, which optimal intake increases while excessive intake decreases the overall diet quality score (red meat, and high-fat dairy). The “healthy” and “unhealthy” food groups can be scored separately into sub metrics GDQS + (ranging from 0 to 32) and GDQS- (unhealthy food groups, including those claimed unhealthy in excessive amounts, ranging from 0 to 17), respectively. Further details on GDQS scoring methods can be found elsewhere [8].

MDD-W, on its turn, scores 1 point for the intake of each one of the 10 predefined food groups: grains, white roots, and tubers and plantains; pulses; nuts and seeds; milk and milk products; meat poultry, and fish; eggs; dark-green leafy vegetables; other vitamin A rich fruits and vegetables; other vegetables; other fruits. Those food groups were defined based on their importance for diet diversity and, consequently, adequacy of micronutrient intake (vitamin A, thiamine, riboflavin, niacin, vitamin B6, folate, vitamin B12, vitamin C, calcium, iron, and zinc), especially for women. MDD-W is usually applied in a dichotomous way, in which minimum diet diversity is achieved when at least 5 out of 10 food groups are included in individuals’ diets. For the present study, MDD-W was scored from 0 to 10, adding one point every time individual diets had one or more food items consumed in more than 15 g/day from each of the 10 predefined food groups [21].

Even though MDD-W has its main usage directed to women, it can be used as a proxy of micronutrient adequacy in other groups [21]. For this reason, MDD-W was also investigated in men in this study [21].

The GDQS performance in predicting overall probability of nutrient adequate intake was compared to that of MDD-W because the former is a well-established proxy for the probability of adequate micronutrient intake [21].

Given the importance of UPF as an indicator of unhealthy dietary habits [40], higher dietary risk for obesity and its comorbidities [41], and low adherence to the Brazilian Food Guide recommendations [42], we investigated its correlation with the GDQS. UPF compose one of the four food groups determined by the NOVA classification system [40]. For this study, foods were manually classified as pertaining to UPF category according to the NOVA classification system described in detail elsewhere [40], and the calorie from UPF was divided by the total caloric intake of the diet to generate a percentage of caloric contribution from UPF for the first day of dietary data collected.

Predicted individual usual intakes of protein, total fat, saturated fatty acids (SFA), monounsaturated fatty acids (MUFA), polyunsaturated fatty acids (PUFA), fiber, calcium, iron, zinc, vitamin A, folate, and vitamin B12 intakes, from two 24HR, as well as energy, were estimated using the National Cancer Institute (NCI) method to adjust for within-person variability [43]. Predicted nutrient intakes were adjusted for energy intake using the residual method [44].

To assess nutrient adequacy from 24HR data, the individual probability of adequate intake for protein, fiber, calcium, iron, zinc, vitamin A, folate, and vitamin B12 was calculated following the full-probability method for each nutrient, described in the Institute of Medicine guidelines [45], using the energy-adjusted nutrient intakes. Individual overall probability of adequacy was then estimated by tabulating the mean probability of adequacy across those eight nutrients.

The overall nutrient inadequacy outcome was defined as an energy-adjusted mean probability of adequacy across the eight nutrients < 0.5, based on previous validations of the GDQS [8].

Continuous variables were tested for adherence to a normal distribution using the Kolmogorov–Smirnov test. Whenever a continuous variable did not present a normal distribution, non-parametric statistical analysis was applied or the variable was categorized. Categorical variables are presented as relative and absolute frequency while continuous variables are presented as mean (standard error). All descriptive statistics took the complex sample design into consideration (survey mode).

Comparisons of means between sex were conducted using the Mann–Whitney’s test, while comparisons between age ranges and geographical regions were conducted using the Kruskall-Wallis test with Tukey HSD correction for multiple tests.

Spearman’s coefficient was used to assess correlation between GDQS and nutrient intake and for the correlation between MDD-W and nutrient intake. For each nutrient intake, to compare the performance of the GDQS with the MDD-W, Wolfe’s test was applied between the estimated Spearman’s correlation coefficients for each diet metric. Spearman’s coefficient was also used to test the GDQS and MDD-W correlation with UPF intake.

To estimate the odds for nutrient inadequacy across MDD-W and GDQS quintiles, a multiple logistic regression was applied, taking the first quintile as reference, adjusted for age ranges (adolescents, aging from 10 to 19 years; adults, aging from 20 to 59 years; and elderly individuals, aged 60 years or more) urban/rural locality, income (five categories of income), supplement use (yes/no), and recent diet modification (yes/no). The same statistical adjustments were applied to estimate linear trend across quintiles, including the quintile information as a categorical variable in the model, coded “0” for the 1st quintile, “1” for the 2nd quintile, “2” for the 3rd quintile, “3” for the 4th quintiles, and “4” for the 5th quintile. Linear trends in overall nutrient inadequacy across metric quintiles were statistically compared using regression models in which quintiles of GDQS and MDD-W were included in the same model and the parameter estimates associated with 5th quintile were compared using a Wald test (p-difference).

To investigate linear increases in the probability of overall nutrient adequacy across GDQS and MDD-W quintiles, a multiple linear regression model, adjusted for age (years), urban/rural locality, income (five categories of income), supplement use (yes/no), and recent diet modification (yes/no), was used. To check for difference between MDD-W and GDQS performance, Wald’s post-test for first-to-fifth quintile delta difference in the probability of overall nutrient adequacy was applied after the multiple linear regression model.

All tests were repeated, stratifying for sex, age range, and Brazilian major geographic regions to investigate GDQS validity across categories. Except for the application of the National Cancer Institute (NCI) method to adjust for within-person variability, which was carried on using SAS® OnDemand for Academics (SAS Institute Inc., Cary, NC, USA), all statistical analysis were conducted using Stata SE®, version 17.0 (StataCorp LLC, Texas, EUA).