Four principal dietary patterns were identified among the 2,034 young adults of the Ribeirão Preto birth cohort, denoted healthy, traditional Brazilian, bar and energy-dense. The variance explained by these factors (20.92%) was lower than that reported in other studies which also retained four components [17, 25]. The decision to include a large number of food items in the principal component analysis may have contributed to the lower percentage of explanation of total variance [9, 12]. In the adjusted analysis, women and individuals with higher schooling (≥12 years) in adult life showed greater adherence to the healthy pattern. The highest adherence to the traditional Brazilian pattern was detected for men, mullatos, households with ≥2 members, and for those who have children, while individuals with higher schooling in adulthood (≥12 years), higher family income in adulthood (≥20 MW) and higher family income at birth (≥6.1 MW) showed lower adherence. The bar pattern was positively associated with male sex and low schooling (≤8 years). The energy-dense pattern was not associated with any of the variables investigated. Social mobility was associated with the traditional Brazilian pattern. Men and women who were not poor at birth and remained so in adulthood showed lower adherence to this pattern.
The dietary pattern denoted healthy explained the greatest percentage of variance. Its composition is similar to that of other patterns reported in the literature with the same denomination [5, 10, 26–28] or even with others such as ‘prudent’ [15, 25, 29], ‘nutrient dense’  and ‘health aware’ . This pattern is characterized by the presence of foods rich in vitamins, minerals, fibers, and unsaturated fats and with low contents of sugars, trans and saturated fats . The traditional Brazilian pattern involved traditional foods of the Brazilian diet, as it was also observed in other studies [3, 6, 14, 15]. In a similar study conducted on young adults (23 years) from Pelotas, Rio Grande do Sul, Brazil, Olinto et al.  defined a similar pattern as ‘common Brazilian’, and Gimeno et al. , in a study of adults aged 30 years or older from Ribeirão Preto, called it ‘popular’. The ‘traditional’ denomination is habitually used in studies from other countries, but the foods that compose this diet vary according to the culture of each country .
The dietary pattern denoted bar consisted of alcoholic beverages and foods commonly served in bars. No pattern with this denomination was found in the literature. However, some studies have identified factors containing alcoholic beverages in their composition [9, 16, 26, 31, 32]. The fourth pattern identified was called energy-dense because it is rich in sugars and saturated and trans fats. Patterns with similar compositions have been designated in various ways in other studies: ‘obesogenic’ , ‘energy dense’ , ‘snack’  and ‘dairy foods and desserts’ .
It should be pointed out that many studies have identified a pattern defined, in most cases, as western [25, 26, 33], which represents a combination of the energy-dense and bar patterns considered in the present study. Rezazadeh, Rashidkhani and Omidvar  defined this combination as ‘not healthy’. Positive associations between patterns with compositions similar to those of the energy-dense and bar patterns and unfavorable health outcomes such as obesity have been described in some studies [9, 14, 30].
The differences in composition between patterns with the same denomination found in the literature, and the different denominations of factors of similar composition may be explained by the subjectivity inherent to the methods of principal component analysis . In a review of this topic, Newby and Tucker  pointed out the following major decisions that should be made by an investigator which could influence the results: 1) which method to use for the investigation of food intake; 2) whether or not to group the food items; 3) how to treat the variables (grams/milliliters, % total caloric value, etc.); 4) number of retained factors; 5) starting from what value the factor loading will be considered to be important for the component; 6) how the retained patterns will be denominated. Newby and Tucker  also emphasize the importance of considering the nutritional composition of the foods that constitute each retained factor when attributing names to the dietary patterns identified. In the present study, the authors followed this recommendation.
Female sex and schooling in adulthood ≥12 years were found to be associated with the healthy dietary pattern. In contrast, men showed more adherence to the traditional Brazilian and bar patterns, as also observed in other studies that identified patterns of compositions similar to those studied here [16, 26]. Differences in diet quality between sexes have been reported and it has been well established in the literature that women adhere more to healthy dietary patterns than men [9, 11, 17, 25, 26]. According to Newby and Tucker , this can be explained in part by the preoccupation of women with a good physical shape. A positive association between high schooling and income, and healthy food patterns has also been confirmed in other studies [11, 17, 26, 27].
Mullatos showed greater adherence to the traditional Brazilian pattern, as it was also reported in other studies [3, 16]. The fact that beans, the most important food in this pattern, were the principal diet component for those with low socioeconomic status during the colonial period of Brazil might explain this result. Also in agreement with other Brazilian studies [16, 17], better socioeconomic levels at present (schooling and family income) and at birth (family income) showed an inverse association with this dietary pattern. The determinants of food choices are complex , and although price is not the only factor guiding these choices, a low income can limit them [11, 34]. Except for beef, the foods composing the pattern called traditional Brazilian in the present study are of low cost. It should also be pointed out that only four food items presented an important positive factor loading in this pattern, suggesting monotony in the diet of less privileged socioeconomic classes, as also observed in a similar study .
Low schooling in adulthood was associated with the bar pattern, in contrast to other studies that identified patterns containing alcoholic beverages in their composition, where those of high schooling tended to show greater adherence to this pattern [16, 30, 35]. Low schooling may contribute to unhealthy food choices, since education allows people to obtain information about health, especially healthy dietary patterns, with consequent improvement of dietary habits .
Although none of the socioeconomic or demographic variables investigated showed association with the energy-dense pattern in the present study, female sex and higher schooling in adulthood and income in other population groups were associated with patterns of similar composition [16, 17, 30].
Social mobility was not associated with the healthy, bar and energy dense patterns. An interaction between sex and social mobility was detected only for the traditional Brazilian pattern. The category ‘not-poor - not poor’ showed lower adherence to the traditional Brazilian for both men and women. Similarly, young adults in the Brazilian city of Pelotas, of both sexes who were in the highest income tertile at birth and in adulthood (referred to as ‘never poor’), showed lower adherence to the ‘common Brazilian’ pattern compared to those who belonged to the lowest tertile of income at birth and remained there in adulthood, termed ‘always poor’. In the Pelotas study, the category ‘poor - not poor’ also showed greater adherence to the ‘processed food’ pattern compared to those who were ‘always poor’ . Mishra et al. , studying British adults, found that those who held non-manual occupational activities and whose parents were also engaged in non-manual occupations at the time of their birth had healthier eating patterns than those who remained in the category manual as well as their parents.
In our study, the higher the number of people in the household the greater the adherence to the traditional Brazilian pattern. Also the adherence to this pattern was greater if the individual has had children. Demographic factors that take into account the number of household members have also been associated with some dietary patterns in other studies. For example, in Tehran, housing size > 20 m2/head was positively associated with a healthy dietary pattern .
The results of the present study should be compared with caution to those of other investigations since methodological differences exist between studies and food choices depend on socioeconomic and cultural factors of each population. Thus, the results of the present investigation cannot be generalized and only represent the dietary patterns of the cohort studied, but not of the entire population of young adults from Ribeirão Preto, São Paulo, Brazil. This is due to the fact that the cohort comprises a small age group born in a couple of years (1978/79). Furthermore, the population composition changed from birth to adulthood due to migration.
Our study has some limitations. It should be pointed out that there were selective losses when the group of individuals followed up in the fourth phase of the study was compared to subjects who were not followed up. The follow-up rates were slightly higher for women and for individuals whose mothers had higher schooling at the time of their birth. Although statistically significant, these differences were small . Additional limitations are those inherent to the methods of dietary surveys and the subjectivity of the PCA method for the definition of dietary patterns . Furthermore, PCA do not assume exactly an ordinal response option. In addition, the variability explained by the four retained components was relatively low (20.92%).
The FFQ used in this study was not validated in the present population, which is a limitation. However, it has been validated in second and third generation Japanese-Brazilian adults from São Paulo who have similar dietary habits to those of our population [22, 36].
However, dietary surveys are widely used in epidemiological studies and, in the present investigation, the FFQ was applied by nutritionists with the assistance of a photo album in order to facilitate the estimation of portion sizes. In addition, all subjective choices typical of the PCA method were based on scientific knowledge and on an extensive review of the literature.
Thus, the present study allowed the identification of four major dietary patterns in this population of young adults and the identification of socioeconomic and demographic differences associated with food choices. In addition, it was possible to understand how socioeconomic factors existing at birth were associated with dietary patterns in young adulthood.