The whole world is experiencing an obesity epidemic, affecting individuals of all ages, in all social strata and ethnic groups(Reference Stein and Colditz1, Reference Jeffery and Utter2). Although obesity is related to genetic, metabolic, behavioural and environmental influences, its rapid increase suggests that behavioural and environmental influences, rather than biological changes, are the major underlying cause(Reference Stein and Colditz1–3).
There has been a marked shift recently in the composition of the diet, with an increase in per capita energy availability, higher consumption of soft drinks and larger portion sizes(Reference Jeffery and Utter2), as well as higher intakes of fat and added sugar, accompanied by a dramatic fall in total cereal and fibre intakes(Reference Stein and Colditz1, Reference Popkin4).
Rice and beans used to be the staple foods in Brazil, but recent data show a decrease in their intake, whereas there has been a 400 % increase in the consumption of industrialized products such as soft drinks(5). Trends between 1974 and 2003 in Brazil also indicated high consumption of sugar, total fat and saturated fat, and insufficient intakes of vegetables and fruits(5). During the same period, the prevalence of obesity in Brazil rose from 2·8 % to 8·8 % among adult men, and from 7·8 % to 12·7 % among women(6).
Ecological studies in the USA have found a strong positive association between the intake of refined carbohydrates in the form of corn syrup and the prevalence of type 2 diabetes and obesity, and a negative association with dietary fibre intake(Reference Gross, Li and Ford7). Data from twenty countries found a positive association between the prevalence of overweight and obesity and the proportion of energy from fat(Reference Bray and Popkin8).
Our aim in the present study was to investigate the correlation between consumption of refined carbohydrates and fats and obesity prevalence among adults in the state capitals of Brazil.
Methodology
An ecological study was conducted, using as units of analysis the adult population living in twenty-six state capitals of Brazil in 2003. Data were obtained from the 2002–3 National Family Household Budget Survey (HBS; ‘Pesquisa de Orçamentos Familiares’), conducted by the Brazilian Institute for Geography and Statistics from July 2002 to June 2003 on a probabilistic national sample of 48 470 households, throughout the year. A two-stage cluster sampling with stratification by urban and rural areas of the states, as well as stratification by the average household schooling, was employed. The primary sampling units were selected by systematic sampling proportional to the number of households. Households were selected by simple random sampling. The household interviews were distributed over the twelve months of the year. The sample was designed to provide representative estimates at the national, regional and state and capital levels(5).
The HBS obtained information about the family food purchases for each household during seven consecutive days. All monetary and non-monetary expenses for food and beverage consumption were registered, including each product acquired, amounts and portion sizes. The short period of reference – one week of data collection – only allows for estimations of groups of families. Crude weights of purchased foods were transformed into energy and nutrients with the use of Brazilian(Reference Franco9, 10) and international(Reference Souci, Fachmann and Draut11) food composition tables.
The relative contribution of foods and food groups was expressed as the proportion (%) of total energy. Fruits and vegetables were also analysed in grams, assuming that weekly family purchase was a good estimation for these foods. Other items such as beans and rice are usually bought on a monthly basis. There were initially 820 000 foods and beverages in the data set which were grouped into 214 food items based on the frequency of purchase and the major constituents of the foods. These food items were further combined into fifteen groups including three groups of staple foods: (i) cereals and cereal products; (ii) beans, soya, lentils and garbanzo beans; (iii) roots; (iv) milk and milk products; (v) eggs; (vi) meat and meat products; (vii) fruits and natural fruit juices; (viii) vegetables; (ix) sugar and soft drinks; (x) butter and animal fats; (xi) oils and margarine; (xii) alcoholic beverages; (xiii) ready-to-eat meals including lasagne, pizza, soups, frozen rice and chicken, meat and vegetables, and other industrialized frozen dishes; (xiv) oleaginous seeds; and (xv) condiments. For analysis, we excluded the eggs, oleaginous seeds and condiments groups as they had a low frequency of family purchase. Our grouping is quite similar to the Data Food Networking (DAFNE) food classification system used in Europe(12).
Prevalence of obesity was estimated as the proportion of the population with BMI ≥ 30·0 kg/m2. Height and weight were measured in the households. Height was measured to the nearest 0·5 cm using a wall-mounted stadiometer and body weight was measured using a calibrated digital scale with maximum capacity of 150kg and precision of 100g. Analysis included all individuals aged 20–59 years, excluding pregnant and breast-feeding women.
Extreme values of BMI – below 13·0 kg/m2 and above 50·0 kg/m2 – were considered to be due to measurement error and were excluded. The gender-specific age-standardized prevalence of obesity was calculated for each state capital, using as standard the 2003 Brazilian adult population. However, Palmas, the capital of a recently created state, was excluded from the analysis owing to its unstable population structure, mainly among males.
Spearman correlation coefficients between foods and food groups and the prevalence of obesity were calculated separately for men and women. Statistical significance was set at the 5 % level (α = 0·05). All P values obtained were two-sided. All analyses were performed taking into account the complex sample survey design of HBS, using the SPSS statistical software package version 13·0 (SPSS Inc., Chicago, IL, USA).
Results
The age-standardized prevalence of obesity was 9·75 % and 10·77 % among women and men, respectively, ranging from 5·11 % to 13·61 % among women and from 5·22 % to 17·57 % among men. The highest prevalence was in Macapá (north region) for men (17·57 %) and in João Pessoa (north-east region) for women (13·61 %). Women were fatter than men in all south-east capitals and in most cities of the other regions (Table 1).
Table 1 Age-standardized prevalence of obesity among adults aged 20–59 years, by sex, in Brazil and selected Brazilian state capitals, 2002–3
Overall mean daily household energy availability for the twenty-six capitals was 7577 kJ (1811 kcal) per capita, with great variability in the relative contribution of foods and food groups to the total energy availability (Table 2). For the cereals group the mean ranged from 26·41 % to 46·07 %; for vegetal oils, from 7·86 % to 17·09 %; for meats, from 8·00 % to 21·74 %; and for sugar and soft drinks, from 7·43 % to 14·44 %.
Table 2 Mean household relative contribution (%) of foods and food groups to total energy availability and total energy availability per capita in Brazil and selected state capitals, 2002–3
Source: National Family Budget Household Survey/Brazilian Institute for Geography and Statistics.
Spearman correlation coefficients between foods and food groups and the prevalence of obesity are shown in Table 3. Among men, neither foods nor food groups were correlated with the prevalence of obesity. Among women, there was a statistically significant positive correlation between the sugar and soft drinks group and the prevalence of obesity (Fig. 1). When sugar and soft drinks were analysed separately, there were weaker correlations of borderline statistical significance, with soft drinks showing a higher correlation with the prevalence of obesity than sugar (r S = 0·401; P = 0·04 for soft drinks and r S = 0·350; P = 0·08 for sugar). Ready-to-eat meals, which usually have a high fat content, were also significantly positively correlated with obesity prevalence among women (r S = 0·393; P = 0·047).
Table 3 Spearman’s correlation (r S) between the relative contribution of some foods and food groups to total energy intake (%) and intake in grams for fruits and vegetables and the adult age-adjusted prevalence of obesity in twenty-six state capitals of Brazil, by sex, 2002–3
Fig. 1 Scatter plot for the association between the relative contribution of sugar and soft drinks to total energy intake (%) and the age-standardized prevalence of obesity among women aged 20–59 years in twenty-six state capitals of Brazil, 2002–3
There was no correlation between obesity and root foods, although for potatoes this correlation was positive (r S = 0·396; P = 0·04). No correlation was found for cereals or beans. Pasta showed a fair positive correlation, although with a borderline statistical significance (r S = 0·352; P = 0·078).
Fruits and vegetables were analysed as percentage of energy contributed to total energy intake and also by the acquired quantity in grams. Although the correlation was positive (r S = 0·337; P = 0·093), there was no significant correlation with the prevalence of obesity.
A secondary analysis was made controlling for socio-economic status, i.e. average schooling of all adults (20–59 years) and household per capita income. The partial correlations were materially unchanged after this adjustment.
Excluding from the analysis those two cities that appeared to be more influential in the association between soft drinks and obesity, the association was reduced and was still statistically significant (r S = 0·481; P = 0·001).
In order to deal with the many comparisons evaluated, a Bonferroni correction (dividing the significance level (5 %) by the number of correlations (forty-eight) was done. This conservative analysis still showed an association between sugar and soft drinks and the prevalence of obesity in women (r S = 0·597; P = 0·001).
Discussion
There is no clear explanation for the lack of association we found between food group purchase and obesity in men. The prevalence of obesity is increasing within the adult population in Brazil, especially among women; however, men are catching up(6). Women appear to be more susceptible to obesity. Therefore, at the beginning of the obesity epidemic in Brazil, differences by sex were approximately 5 % in 1975, and this difference increased to 8 % in 1989(Reference Sichieri, Coitinho, Leão, Recine and Everhart13). From the most recent data reported in the present study, the gap between the sexes is highly reduced (1 %). Studies from the USA also have shown that as the changes increase, the gender difference diminishes or even disappears(Reference Sichieri, Coitinho, Leão, Recine and Everhart13).
That adult men eat out more frequently than women may explain the decreasing gender difference. A study conducted in Rio de Janeiro in 1996 showed that 50 % of adult men’s meals and snacks were eaten away from home, without utilization of home food(Reference Sichieri14), whereas the corresponding figure for women was about 20 %. Eating out is recognized as a limitation of HBS. HBS data from the USA, in 1995, revealed that 29 % of meals and snacks were consumed away from home(Reference Lin and Frazao15). As a strategy to overcome this limitation of the data source, we used the relative contribution of food to the total energy intake as our main explanatory variable. However, this cannot account for differences in the composition of meals eaten at home and away from home, and therefore the results for women are probably more reliable than those for men.
Despite these limitations, HBS have been widely used to assess the food availability of populations, allowing between-country comparisons and providing a detailed description of the dietary choices of the population, as well as of population subgroups(Reference Naska, Vasdekis and Trichopoulou16). Also, cross-sectional studies of the association between dietary factors and obesity are highly prone to reverse causality and ecological strategies may be a way of avoiding the under-reporting of eating that occurs in all surveys of food intake, as indicated by studies using doubly labelled water(Reference Neuhouser, Tinker and Shaw17).
The main result of the present study is the positive correlation between consumption of sugar and soft drinks and prevalence of obesity among women. These correlations were weaker when analysed separately, with soft drinks, per se, having a higher correlation than sugar.
The mean relative contribution of sugar and soft drinks to total energy intake in all twenty-six Brazilian capitals was 13·4 %, whereas the WHO/FAO (2003) recommendation(18) is that the consumption of refined sugar must represent a maximum of 10 % of total energy intake. This increased consumption of soft drinks and sugar-sweetened fruit drinks is a critical element in the shift in dietary patterns worldwide(Reference Ludwig, Peterson and Gortmaker19–Reference Popkin and Nielsen21), and our result suggests that populations with a high purchase of soft drinks and sugar have a higher prevalence of obesity.
High consumption of sugar-sweetened drinks was associated with increased energy intake and obesity in children in an observational prospective study in conducted in the USA(Reference Gross, Li and Ford7). Also, soft drink consumption among US children and adolescents displaces milk and fruit juice consumption(Reference Harnack, Stang and Story20).
Shifts in the availability of sugar, based on FAO data between 1962 and 2000, show an annual increase of 310 kJ (74 kcal) per capita per day in the consumption of sugar and a positive correlation of sugar intake with per capita income and proportion of the urban population(Reference McCrory, Fuss, Saltzman and Roberts22). Sugar consumption has been linked with industrialization and the proliferation of processed food and beverages(Reference Popkin and Nielsen21).
Time trends (1974–2003) in metropolitan areas of Brazil indicate a decline in the consumption of basic traditional foods, such as rice and beans; notable increases (up to 400 %) in the consumption of processed food items, such as cookies and soft drinks; maintenance of the excessive consumption of sugar; and continuous increases in the content of total fat and saturated fat in the diet(Reference Levy-Costa, Sichieri, Pontes Ndos and Monteiro23).
Analysis of Brazilian HBS has also been conducted in relation to the price of food groups. One of these studies, conducted in the city of São Paulo from 1990 to 1996, classified food groups into industrialized and non-processed foods and revealed that expenditure on the non-processed group decreased by 35 % during the period, although these changes could not be fully explained by variations in product prices(Reference Barreto and Cyrillo24). Therefore, differences in food expenses were not based solely on income. On the other hand, income appears to have an influence on fruit and vegetable purchases. A recent analysis based on the National HBS 2002–3 showed that a reduction in price would increase the purchase of these items, which are not frequently consumed(Reference Claro, Carmo, Machado and Monteiro25). In line with this result, a population-based study in the municipality of São Paulo, based on telephone interviews, indicated that consumption of foods indicative of an unhealthy diet such as sugars and fats was inversely associated with fruit and vegetable intake among subjects of both genders(Reference Figueiredo, Jaime and Monteiro26).
It is not completely understood why and how sugar consumption is associated with obesity and a review on this subject did not show a direct association between obesity and sugar consumption(Reference Hill and Prentice27). Other studies suggest that the glycaemic index (GI), defined as the area under the glucose response curve after consumption of 50 g carbohydrate from a test food divided by the area under the curve after consumption of 50 g carbohydrate from a control food, is associated with an increase in insulin levels and C-peptide excretion(Reference Gross, Li and Ford7, Reference Ludwig28, Reference Ludwig, Majzoub, Al-Zahrani, Dallal, Blanco and Roberts29). Thus, the hyperinsulinaemia associated with high-GI diets might promote weight gain by preferentially directing nutrients away from oxidation in muscle and towards storage in fat(Reference Ludwig28). It is also suggested that consumption of sugar-sweetened drinks could lead to obesity because of imprecise and incomplete compensation for energy consumed in liquid form(Reference Ludwig, Peterson and Gortmaker19). A study among twelve obese male teenagers demonstrated that consumption of high-GI foods induces hormonal and metabolic changes possibly causing obesity(Reference Ludwig, Majzoub, Al-Zahrani, Dallal, Blanco and Roberts29).
In parallel with the increase in sugar intake, the refining process has changed the composition and quality of carbohydrates. Processing grains into white flour increases the energy density by 10 %, reducing the content of fibre by 80 % and also the protein content by almost 80 %(Reference Gross, Li and Ford7).
The mean relative contribution of vegetal oils to total energy observed in the present study was 12·80 %, varying from 7·86 % to 17·09 %. This group was not found to be correlated with obesity, although margarine showed a fair association among women. These results may suggest that other factors, rather than fat consumption, are responsible for body fatness. Another possibility is that fat consumption occurs mainly away from home. Adults who usually eat away from home have higher total energy and fat intakes, and lower fibre consumption(Reference McCrory, Fuss, McCallum, Yao, Vinken, Hays and Roberts30). A study of restaurant food and body fatness showed that frequency of consumption of restaurant food was positively associated with body fatness, independently of educational level, smoking status, alcohol intake and physical activity(Reference McCrory, Fuss, Saltzman and Roberts22).
The present study focused on the relationship between family food availability and the prevalence of obesity in twenty-six state capitals of Brazil. The major limitation of an ecological study is the difficulty in controlling for confounding factors(Reference Medronho31, Reference Morgenstern32). We chose capitals instead of states or other units of analysis as a way to minimize errors in this inference, using data from homogeneous population groups(Reference Morgenstern32). The selected capitals have quite similar percentages of jobs requiring high energy expenditure, ∼3 %(Reference Gomes, Siqueira and Sichieri33), and also the proportion reporting regular physical activity during leisure time in each capital was ∼20 %(Reference Anjos34).
In conclusion, increasing intake of refined carbohydrates, mainly from soft drinks, may have a role in the prevalence of obesity in Brazil. The present survey corroborated the results of other studies and suggests that effecting changes in family purchase patterns may be a strategy to reduce obesity.
Acknowledgements
J.C.P.L. was funded by a Fellowship-Master degree grant from the Brazilian Coordinating Center for Training University Level Personnel (CAPES). No author had any conflict of interest relating to the study. J.C.P.L. was responsible for the statistical analysis and writing of the manuscript. R.S. participated in the study’s conception, data interpretation and writing of the manuscript. A.J.L.C. participated in the study’s conception, data interpretation and writing of the manuscript.
