Dietary Patterns and Their Associations with Obesity in the Brazilian City of Rio de Janeiro


Instituto de Medicina Social-UERJ, Rua S. Francisco Xavier, 524, 7o andar, Bloco E, CEP 20550-012, Rio de Janeiro, Brazil. E-mail:


Objective: To evaluate the dietary patterns of adults living in the city of Rio de Janeiro, Brazil and their associations with body mass index (BMI).

Research Methods and Procedures: A survey was conducted in 1996 in a probabilistic sample of 2040 households. Weight and height were measured and food intake was based on an 80-item semi-quantitative food frequency questionnaire. Dietary patterns were identified through factor analysis.

Results: More than one-third of the adult population (20 to 60 years old) was overweight (BMI = 25 to 29.9 kg/m2), and 12% were obese (BMI ≥ 30 kg/m2). Three major dietary patterns were identified: mixed pattern when all food groups and items had about the same factor loading, except for rice and beans; one pattern that relies mainly on rice and beans, which was called a traditional diet; and a third pattern, termed a Western diet, where fat (butter and margarine) and added sugar (sodas) showed the highest positive loading and rice and beans were strong negative components. Among men, the Western diet also included deep-fried snacks and milk products with high positive values. The traditional diet was associated with lower risk of overweight/obesity in logistic models adjusted for dieting, age, leisure physical activity, and occupation (13% reduction in men and 14% reduction in women comparing the traditional and Western diets).

Discussion: Factors contributing to the effects of the Brazilian traditional diet may include low-energy density, high-dietary fiber content, incorporation of low glycemic index foods such as beans, or a relatively low food variety.


As shown in many countries, the nutritional transition in Brazil from a more traditional dietary pattern to a Western dietary pattern has been associated with an increase in the overall prevalence of overweight/obesity. Prevalence of obesity (body mass index [BMI] ≥ 30 kg/m2), in Brazil, increased from 2.5 in 1975 to 4.8 in 1989 among men, and from 6.9 to 11.7 among women (1). Increasing fat intake does not seem to explain this trend. In a recent survey, fat intake represented 28% of total energy, and this percentage was ∼26% in 1975 (2). In contrast, a steady decline in the consumption of beans has been shown for the whole country (3) (4).

Most clinical and epidemiological nutrition studies have examined energy and nutrient consumption in association to outcomes such as disease, well-being, physical activity, and smoking. However, people do not eat specific nutrients, but specific foods (5) (6) (7) and nutritionists have been advocating for a long time a change in the emphasis of the dietary recommendations from nutrients to patterns of dietary intake. Findings such as the association evident between a Mediterranean diet and low rates of chronic diseases (8) and the successful treatment of hypertension through changes in the dietary pattern (9) have increased the support for these recommendations, with the World Health Organization now suggesting that dietary allowances for populations should be based on foods instead of nutrients (5). In Brazil, where the nutritional transition is an ongoing phenomenon, it has been possible to study dietary patterns and their associations with obesity. The main aim of this study was to test the hypothesis that a traditional diet, based on rice and beans, would be protective against obesity among adults living in the city of Rio de Janeiro.

Research Methods and Procedures

A survey of adults in Rio de Janeiro was carried out in 1996. Households were selected using two-stage probability sampling. In the first stage, 60 primary sampling units (PSUs) were selected from the sampling list of the Pesquisa Nacional de Amostragem Domiciliar (PNAD; Annual National Survey of Households) of 1995. The PNAD is conducted every year by the Brazilian Institute of Statistics (IBGE). An enumeration district (ED), as defined by Bureau of the Census, is a geographic area containing ∼300 contiguous housing units. In 1995, the PNAD sampled 580 EDs in the city of Rio de Janeiro and from this list the 60 PSUs used in this study were sampled. The process of listing all housing units in the sampled EDs was done by the IBGE. In the second stage, 34 households were sampled from the PSU, by systematic sampling of a fraction (PSU size/34) of the households, totaling 2040 households. The probability of selection of primary sampling units and households was proportional to the population size. All parts of the city were included in the list. The present study included 10 PSUs inside of slums. Household is defined by IBGE as one or more rooms limited by walls and covered by a roof, which allows privacy for the residents, and with an independent way to get in.

The sample size was calculated based on a prevalence of overweight of 40% (prevalence observed in a national survey in 1989) (1), with precision of 5%, after accounting for a maximum nonresponse rate of 20%. Nonresponse rate was 11.2%, and pregnant women were excluded. Of the 3196 adults aged 20 to 60 years interviewed, 147 did not complete the food questionnaire and 460 were not available for anthropometric measurements after three visits to the household.

Height and weight were collected in the households, with the participants wearing light clothes and no shoes. Stature was measured using a platform with an attached measuring bar with precision within 0.1 cm. Skinfolds were measured using a plastic caliper (Sustacal) with the precision of 2 mm. BMI was used as a measure of obesity; 37 interviewers and 16 nutritionists were trained and tested on the reliability and validity of their measurements. Repeated anthropometric measurements including skinfolds had intraclass correlation coefficients >0.80, indicating good reliability.

Race was classified based on skin color, defined by the judgment of the interviewer as black, white, mulatto, and Asian. The mulatto group is a mix of black and white ancestry. Measurement of physical activity was based on recall of usual weekly activities including time spent walking to work or school, time and kind of usual recreational activity, and occupational physical activity in the past 2 months. Time spent in each activity was multiplied by activity factors according to the Food and Agriculture Organization (FAO)/Organizacion Mundial de la Salud (OMS)/Universid de laj Naciones Unidas (UNU) recommendation (10).

A food frequency questionnaire with 80 items and usual portions was developed for the Brazilian population, based on the most commonly consumed foods (11). For the present analysis, the 80 items were collapsed into 16 groups: rice, beans, cereals, eggs, deep-fried snacks, colas/sodas, coffee, juices, vegetables, milk products, margarine and butter, fruits, meat/pork/chicken, fish/shrimp, sweets, and alcohol. The semi-quantitative food frequency questionnaire asked subjects to estimate the frequency of food consumption using seven exclusive categories raging from three or more per day to less than once per month, and never or almost never, and the amount of fixed portions of food intake in each category. The time frame of recall was the past 12 months. For the present analysis, frequency equaled the mean value of each one of the categories multiplied by the number of portions in each intake. The usual portions have been described elsewhere (11). In short, for most items the options were soup-spoon and cup; for fruits and ready-to-eat foods, portion was the most frequent unit reported in the validation study, such as one median orange, one hamburger. An alcohol intake variable combined the consumption of wine, spirits, and beer, assuming that 360 mL of beer contained 13.2 g of alcohol, 120 mL of wine contained 10.8 g, and a drink of spirits contained 15.1 g (12).

Factor analysis through principal component in SAS was used to determine the dietary patterns, with the factors rotated by orthogonal transformation (13). The weight of the factors in conjunction with Eigen values >1 determined whether a factor should be retained. The factor loading indicates the importance of one food or food group in the definition of the pattern.

The association between dietary patterns and BMI was analyzed in a multiple logistic regression with individuals classified as normal weight (BMI = 18.5 to 24.9 kg/m2), overweight (BMI = 25 to 29.9 kg/m2), and obese (BMI ≥ 30 kg/m2 (World Health Organization classification) (14). The three-level BMI classification was the dependent variable and dietary patterns were the independent variables. Models were controlled for age, being on a weight-loss diet, leisure physical activity, and occupation. Leisure activity in the previous year was categorized as never or almost never (62.8% of men and 80.2% of women), sometimes and always (15.0% of men and 7.4% of women). Occupations in the previous 2 months were classified, according to FAO/OMS/UNU, 1985 as very-low expenditure: 125 to 150 kcal/h; low: 151 to 175 kcal/h; moderate 1: 176 to 180 kcal/h: moderate 2: 181 to 220 kcal/h; moderate 3: 221 to 440 kcal/h; and high-expenditure activity: energy expenditure >440 kcal/h. Previous analysis of these data (15) indicates that leisure activity explained most of the variation in physical activity among subjects, with only 0.3% of women and 3.6% of men reporting high-energy expenditure due to occupation. All variables were collected during household interviews.


More than one-third of the adult population (20 to 60 years old) were overweight, and 12% were obese. Prevalences of underweight and overweight were approximately the same in the three ethnic groups: black, white, and mulatto, although black and mulatto women had a nonsignificant greater prevalence of obesity compared with white women (Table 1). There was a good agreement between overweight/obesity classification based on BMI with adiposity measured by the prevalence of skinfolds greater than the 85th percentile of the U.S. population (data not shown).

Table 1.  Prevalence (%) of underweight, overweight, and obesity by race, *Rio de Janeiro, Brazil, 1996
  • BMI, body mass index.

  • *

    Asians of Japanese origin were excluded.

Underweight BMI (kg/m2) <
Normal weight BMI (kg/m2) 18.5 to 24.954.054.655.4
Overweight BMI (kg/m2) 25 to 29.934.532.134.1
Obese BMI (kg/m2) ≥308.810.39.3
p value comparing race0.89  
Underweight BMI (kg/m2) <
Normal weight BMI (kg/m2) 18.5 to 24.958.055.352.6
Overweight BMI (kg/m2) 25 to 29.926.727.727.9
Obese BMI (kg/m2) ≥3011.613.516.7
p value comparing race

Forty-five percent of the participants reported a monthly per capita income below U.S. $201 and 16% reported income above U.S. $600.

By factor analysis three major dietary patterns were identified: mixed pattern when all food groups and items had about the same factor loading, except for rice and beans; one pattern that relies mainly on rice and beans, which was called a traditional diet; and a third pattern, which was called a Western diet, where fat (butter and margarine) and added sugar (sodas) showed the highest positive loading and rice and beans were strong negative components. Among men, the Western diet also included deep-fried snacks and milk products with high positive values (Table 2). Those 460 individuals not available for anthropometric measurements had dietary patterns similar to the overall sample.

Table 2.  Factor-loading matrix and Eigen vectors for the three identified dietary patterns of urban Brazilians
Deep-fried snacks0.23−−0.07−0.06
Milk products0.11−0.160.450.32−0.08−0.02
Eigen values3.151.671.186.711.841.21

For both men and women, active dieting was associated with reduction in the frequency of food consumption except fruits, vegetables, milk products, and meat. Data with both sexes combined are presented in Table 3. More than 20% of women and ∼10% of men were on a weight-loss diet on the occasion of the survey, which may explain the gender difference. When analysis was stratified by dieting, those dieting showed for the Western pattern a negative loading value for most foods (Table 4).

Table 3.  Age- and sex-adjusted daily frequency of food portion intake and alcohol consumption (g), according to dieting status during the previous year*
  • *

    Time frame of the questionnaire.

  • Including wine, beer, and spirits.

 MeanSDMeanSDp value
Rice (soup spoon)<0.001
Beans (soup spoon)<0.001
Cereals (one-slice of bread or cake, three biscuits)<0.001
Eggs (unit)0.380.450.620.83<0.001
Deep fried snacks (unit)
Fruits (one orange, 10 units of grapes, one banana)
Coffee (one cup)
Juices (one small glass)
Vegetables (two soup spoons for most, slices for beetroot)11.510.710.622.90.48
Milk products (one small glass of milk, two slices of cheese)
Butter/margarine (one spoon = 5 g)1.00.801.20.85<0.001
Meat (4 to 6 oz)1.911.
Fish/shrimp (3 to 5 oz)0.530.780.624.50.66
Sweets (one slice, 1 oz chocolate)<0.001
Alcohol (g)8.120.613.630.10.003
Table 4.  Factor-loading matrix and Eigen vectors for the three identified patterns of urban Brazilian diets, for both sexes combined and according to presence or absence of dietary restriction
Deep fried snacks0.26−0.06−0.0010.37−0.10−0.08
Milk products0.25−0.25−0.320.290.050.009
Eigen values2.881.861.375.961.811.26

The traditional diet was associated with lower risk of overweight/obesity in logistic models adjusted for dieting, age, leisure physical activity, and energy expenditure of occupation (Table 5). Excluding those men and women who reported dieting, results were about the same (Table 5).

Table 5.  Odds ratios (ORs) and 95% confidence interval (CI) for the three identified patterns of diets associated with overweight/obesity*
 OR95% CIOR95% CI
  • *

    Logistic model with body mass index categorized as: 18.5 to 24.9, 25 to 29.9, and ≥30 kg/m2.

Mixed1.030.90 to 1.180.950.61 to 1.48
Traditional0.870.77 to 0.990.860.75 to 0.97
Western0.970.85 to 1.100.990.87 to 1.12
Mixed1.060.92 to to 1.79
Traditional0.880.76 to 1.000.810.70 to 0.94
Western0.980.85 to 1.130.970.83 to 1.15

For analyses carried out by age group (20 to 40 vs. 41 to 60 years old), with models adjusted for dieting, age, leisure physical activity, and energy expenditure of occupation (data not shown) traditional diet was associated with lower risk of overweight/obesity only in the younger group (odds associated with traditional diet was 0.84; p = 0.05 among women and 0.76; p = 0.003 among men), whereas for the older group (41 to 60 years old) these values were 0.86, p = 0.13 for women, and 0.99, p = 0.99 for men. For both age groups, the principal component analysis identified the same three major dietary patterns; however, for men the Eigen values were reduced (data not shown). This analysis included 515 men and 645 women in the 20 to 40 years old group and 351 men and 461 women in the older group.


Factor analysis of dietary patterns in Rio has identified three distinct patterns. Interestingly, the patterns were similar for men and women and even for those on a weight-loss diet, a characteristic much more frequent among women than among men. Among women, the pattern was less consistent for those items that are considered strongly related to obesity, such as fats.

The traditional diet, based on rice and beans, was the only pattern associated with a reduced BMI in the logistic regression. This finding is in accordance with a greater reduction of weight on a diet of rice and beans observed in a small clinical trial of obese women (16). In this study, 40 Brazilian women with BMI ≥ 27 kg/m2 were randomized and encouraged to eat during 2 months one of two isocaloric diets of 1800 kcal/d: rice and beans twice a day, no meat and with a macronutrient composition of 14% of energy from protein, 15% from fat, and 71% from carbohydrate; or lean meat twice a day and with 18% of energy from protein, 25% from fat, and 57% from carbohydrate. After 2 months, the rice and beans group had a weight loss of 3.8 kg (SD: 1.8) and the control group a weight loss of 1.5 kg (SD: 0.9; t test p = 0.10). The loss of follow-up was 35% in the rice and beans group and 45% in the control group. Most losses (13 of 16) occurred in the first month of follow-up. The weight loss in the first month was 2.4 in the experimental group and 0.9 in the control group (t test p = 0.04).

When analysis was carried out by age group, the association between the traditional diet and overweight/obesity was retained only for the younger group. A cohort effect with middle-aged individuals having a more homogeneous diet is a possible explanation, as shown by the reduced Eigen values among men 41 to 60 years old.

Patterns of diet may provide a more comprehensive measure of dietary exposure for epidemiological research, as indicated by a previous analysis of these survey data, which did not find any association of energy and fat intake with obesity (17). In contrast, dietary patterns are highly influenced by socioeconomic factors. Wirfalt et al. (18) showed that dietary patterns were related to age, ethnicity, and income in a factor analysis of frequency-of-use data of 110 foods for 1475 men and 780 women. We controlled the analysis only for physical activity (leisure and occupation), assuming that weight gain in adults depends on the balance between the expenditure and energy intake. Physical activity is known to explain most of the variation in expenditure among different populations (19). For the population in the present study, the level of physical activity as measured by occupation, leisure activity, and time usually spent watching television or on the computer, was extremely low, with leisure showing the greatest variation with socioeconomic level (15). Inclusion of leisure-time physical activity and occupation in the models did not change the association of traditional diet and obesity.

Few epidemiological studies have tested the hypothesis of an association between dietary patterns and obesity. In a large observational prospective study of the association of dietary patterns with plasma biomarkers of obesity (20), a Western dietary pattern correlated positively with insulin, C-peptide, and leptin.

The protective role of Brazilian traditional diet may be due to low glycemic index (GI) of beans, low-energy density of rice and beans, high-fiber intake due to beans, low-fat intake, or reduced variety of the traditional diet. The GI measures the increase of blood glucose levels due to the intake of specific carbohydrate foods (21), and an exceptionally low-blood glucose response was associated with dried beans (22). A recent review by Roberts (23) concluded that although data from long-term studies are lacking, short-term investigations indicate that consumption of high GI carbohydrates may increase hunger and promote overeating relative to consumption of items with a lower GI. Two other recent reviews have emphasized a possible role for carbohydrates, especially carbohydrate source, in the etiology of obesity (24) (25). In addition, in a nonrandomized follow-up of children attending a program of obesity treatment, children assigned to a low GI diet had a significant reduction of BMI, compared with a reduced-fat, high GI diet. Main issues in this study were the lack of randomization and the short follow-up (mean of 4 months) (26).

Beans are the main source of fiber in Brazil (3). Average fiber intake among the adults, in a population-based study was 24 g/d (17 g of insoluble and 7 g of soluble fiber), and a steady decline in the consumption has been shown for the whole country. The percentage of energy from beans changed from 8.9% in 1975 to 6.4% in 1988 (4). Leguminous diets containing significant amounts of slowly absorbed carbohydrate have been shown to decrease blood glucose (27). A recent review indicates that under fixed energy intake, either soluble or insoluble fiber intake increases post-meal satiety and decreases subsequent hunger. In addition, this review suggested, at least for short-term follow-up, that high-fiber diets decrease energy intake and body weight (28). High-fiber content can also be a marker for either lower palatability or a monotonous diet. Dietary variety within food groups and palatability were shown to be important predictors of body fatness (29).

Rice and beans are also low-energy foods that contribute to the bulk of the Brazilian diet. Because individuals tend to consume a fixed amount of food (30) (31), a large intake of low-energy food will make it harder to have an excessive energy intake. Findings from several investigations have found that low-energy-dense foods reduce the total energy intake (30) (32) (33). Therefore, U.S. women consumed 20% less energy in a low-energy-dense diet compared with a high-energy-dense diet (32).

The Brazilian traditional diet is also a low-fat diet. Fat intake has been considered a risk factor for weight gain as reviewed by Bray and Popkin (34). However, in Brazil the striking increase in obesity was associated with a minimal increase in the percentage of energy from fat (2). Also, fat intake was unrelated to BMI in this study population (17). Nevertheless, underreporting of fat intake may be one reason why fat intake was not predictive of obesity. Thus, when the Brazilian food frequency questionnaire used in this study was compared with estimated daily energy requirement (11), women, but not men, with high predicted energy expenditure often underestimated energy intake, and reported low percentage of energy from fat. The possibility of a preferential underreport of fat could explain the lack of association between fat intake and overweight.

In conclusion, the population-based findings from Rio de Janeiro indicate that the traditional combination of rice and beans as major staple foods is protective against obesity. The low-energy density of rice and beans, low glycemic index of beans, high-fiber intake due to beans, low-fat intake, or reduced variety of the traditional diet are the possible explanatory mechanisms for this association. Only long-term, controlled longitudinal studies will be able to disentangle diet in the cluster of socioeconomic-related behaviors, to prove the role of this dietary pattern in weight-reducing strategy or prevention of obesity.


This work was supported by research grants from Conselho Nacional de Pesquisa (CNPq), Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES), and Ministry of Health.